Desarrollo Economico - London University

Economics of development
M. Vesal
EC3044
2014
Undergraduate study in
Economics, Management,
Finance and the Social Sciences
This subject guide is for a 300 course offered as part of the University of London
International Programmes in Economics, Management, Finance and the Social Sciences.
This is equivalent to Level 6 within the Framework for Higher Education Qualifications in
England, Wales and Northern Ireland (FHEQ).
For more information about the University of London International Programmes
undergraduate study in Economics, Management, Finance and the Social Sciences, see:
www.londoninternational.ac.uk
This guide was prepared for the University of London International Programmes by:
M. Vesal, PhD candidate, London School of Economics and Political Science.
This is one of a series of subject guides published by the University. We regret that due to
pressure of work the author is unable to enter into any correspondence relating to, or arising
from, the guide. If you have any comments on this subject guide, favourable or unfavourable,
please use the form at the back of this guide.
Acknowledgements
This subject guide draws heavily from my experience as a class teacher for Development
Economics courses at LSE and I am indebted to the lecturers of these courses. I would like
to thank Oriana Bandiera for helpful comments on the outline and Gharad Bryan and Rocco
Macchiavello for reviewing the first draft. I would also like to thank the University of London
International Programmes editorial team, especially Iain Sharpe and Michele Greenbank.
Finally, I would like to thank Donald Verry and Sarah Douglas for coordinating the writing of
the subject guide. Needless to say that all errors remain my own responsibility!
University of London International Programmes

Publications Office
Stewart House
32 Russell Square
London WC1B 5DN
United Kingdom
www.londoninternational.ac.uk
Published by: University of London

© University of London 2014
The University of London asserts copyright over all material in this subject guide except where
otherwise indicated. All rights reserved. No part of this work may be reproduced in any form,
or by any means, without permission in writing from the publisher. We make every effort to
respect copyright. If you think we have inadvertently used your copyright material, please let
us know.
Contents
Contents
Chapter 1: Introduction........................................................................................... 1
1.1 Route map to the guide............................................................................................ 1
1.2 Introduction to development economics.................................................................... 1
1.3 Syllabus.................................................................................................................... 2
1.4 Aims of the course.................................................................................................... 3
1.5 Learning outcomes for the course............................................................................. 3
1.6 Overview of learning resources................................................................................. 3
1.7 The examination .................................................................................................... 10
Part 1: Cross-country differences and macro models of development................. 13
Chapter 2: Overview of development basic facts................................................. 15
2.1 Introduction .......................................................................................................... 15
2.2 Concept of development......................................................................................... 16
2.3 Measuring development ........................................................................................ 16
2.4 Methods ............................................................................................................... 21
2.5 Overview of the chapter.......................................................................................... 26
2.6 Reminder of learning outcomes ............................................................................. 26
2.7 Test your knowledge and understanding................................................................. 26
Chapter 3: Factor accumulation............................................................................. 27
3.1 Introduction .......................................................................................................... 27
3.2 The Solow model.................................................................................................... 28
3.3 Testing the Solow model......................................................................................... 35
3.4 Development accounting ....................................................................................... 37
3.5 Summary................................................................................................................ 39
Chapter 4: Endogenous growth and poverty traps............................................... 41
4.1 Introduction .......................................................................................................... 41
4.2 Endogenous growth theories.................................................................................. 42
4.3 Externalities, complementarities and increasing returns to scale............................... 46
4.4 Poverty traps.......................................................................................................... 49
4.5 The role of history................................................................................................... 52
4.6 Summary................................................................................................................ 52
Chapter 5: Inequality and growth......................................................................... 55
5.1 Introduction .......................................................................................................... 55
5.2 Definition............................................................................................................... 56
5.3 Measurement......................................................................................................... 56
5.4 Inequality and growth............................................................................................ 59
5.5 Summary................................................................................................................ 66
i
EC3044 Economics of development
Chapter 6: The role of institutions........................................................................ 69

6.1 Introduction .......................................................................................................... 69
6.2 What are institutions?............................................................................................ 70
6.3 Institutions and development: macro evidence........................................................ 71
6.4 Property rights and development: micro evidence ................................................... 77
6.5 Summary................................................................................................................ 80
Part 2: Markets in developing countries............................................................... 83
Chapter 7: Education............................................................................................. 85
7.1 Introduction .......................................................................................................... 85
7.2 A model of educational decision-making................................................................. 86
7.3 Promoting schooling demand ................................................................................ 87
7.4 Schooling infrastructure ......................................................................................... 89
7.5 Teachers’ incentives................................................................................................ 95
7.6 Summary................................................................................................................ 96
Chapter 8: Health and nutrition............................................................................ 99
8.1 Introduction .......................................................................................................... 99
8.2 Demand for health............................................................................................... 100
8.3 Nutrition and income............................................................................................ 103
8.4 Provider incentives and health infrastructure......................................................... 104
8.5 Within-household health inequalities.................................................................... 106
8.6 Summary.............................................................................................................. 107
8.7 Reminder of learning outcomes ........................................................................... 108
8.8 Test your knowledge and understanding............................................................... 108
Chapter 9: Land and agriculture.......................................................................... 109
9.1 Introduction ........................................................................................................ 109
9.2 Why agriculture?.................................................................................................. 109
9.3 Land.................................................................................................................... 110
9.4 Tenancy contracts................................................................................................. 111
9.5 Land reform......................................................................................................... 115
9.6 Summary.............................................................................................................. 117
9.7 Reminder of learning outcomes ........................................................................... 118
9.8 Test your knowledge and understanding............................................................... 118
Chapter 10: Credit............................................................................................... 119
10.1 Introduction....................................................................................................... 119
10.2 Role of credit...................................................................................................... 120
10.3 Limits of credit markets....................................................................................... 121
10.4 Informal credit.................................................................................................... 122
10.5 Microfinance...................................................................................................... 124
10.6 Summary............................................................................................................ 128
10.7 Reminder of learning outcomes ......................................................................... 129
10.8 Test your knowledge and understanding............................................................. 129
Chapter 11: Insurance.......................................................................................... 131
11.1 Introduction ...................................................................................................... 131
11.2 Basic model of insurance.................................................................................... 132
11.3 Empirical evidence on consumption smoothing................................................... 134
11.4 Limits to insurance............................................................................................. 135
ii
Contents
11.5 Informal insurance: credit as insurance................................................................ 138

11.6 Summary............................................................................................................ 140
Part 3: The state and the process of development.............................................. 143
Chapter 12: Infrastructure and development...................................................... 145
12.1 Introduction ...................................................................................................... 145
12.2 Infrastructure and the role of government........................................................... 146
12.3 Impact of infrastructure on development............................................................. 147
12.4 Impact of transport infrastructure........................................................................ 149
12.5 Impact of information technology ...................................................................... 150
12.6 The impact of water facilities............................................................................... 153
12.7 Summary............................................................................................................ 155
Chapter 13: Trade policy...................................................................................... 157
13.1 Introduction ...................................................................................................... 157
13.2 Theories of trade................................................................................................. 158
13.3 Trade liberalisation.............................................................................................. 161
13.4 Trade policy........................................................................................................ 163
13.5 Summary............................................................................................................ 166
Chapter 14: Environment and development....................................................... 169
14.1 Introduction ...................................................................................................... 169
14.2 The impact of the environment on outcomes....................................................... 171
14.3 Mitigation strategies........................................................................................... 175
14.4 Summary............................................................................................................ 179
Chapter 15: Taxation and development.............................................................. 181
15.1 Introduction ...................................................................................................... 181
15.2 Tax structures in developing countries................................................................. 182
15.3 Why tax structures are so different?.................................................................... 184
15.4 Fiscal capacity ................................................................................................... 188
15.5 Informal taxation................................................................................................ 189
15.6 Summary............................................................................................................ 190
Chapter 16: International aid.............................................................................. 191
16.1 Introduction ...................................................................................................... 191
16.2 Forms of development assistance........................................................................ 192
16.3 Does aid promote growth?................................................................................. 194
16.4 Unintended consequences.................................................................................. 196
16.5 Summary............................................................................................................ 197
Appendix 1: Sample examination paper............................................................. 199
Appendix 2: Examiners’ commentary.................................................................. 203
iii
Notes
iv
Chapter 1: Introduction
1.1 Route map to the guide

This subject guide is your study companion for the subject EC3044
Economics of development. The aims of the subject guide are to:
• provide a framework for the study of development economics
• introduce various topics in Economics of development listed in
the syllabus
• present the material in a structured format to facilitate learning
• guide you towards various learning activities and exercises
• encourage you to take an active learning approach by reading the
essential material, undertaking learning activities and participating in
online activities (e.g. discussion) via the VLE.
This guide is divided into 16 chapters, including this introductory chapter.
Chapters start with an introduction to the topics covered and provide a
theoretical framework to understand the relevant issues. Presentation of
the empirical evidence, mostly from journal articles, forms the basis of
policy discussions in the rest of each chapter.
Chapters contain a variety of learning activities to facilitate understanding
of methods and concepts. It is extremely important that you try all
activities in the subject guide before looking up the answers on the VLE.
You should also try the exercises at the end of referenced chapters from
the course textbook, Development economics (1998).
Although Development economics (1998) is set as the textbook for this
course, it does not cover the material in all the chapters. Instead you are
expected to read journal articles to fill in the gaps and to get a better
understanding of the empirical evidence.
In discussing empirical research we extensively use methods of
establishing causality. Various forms of regression analysis, like
instrumental variables, fixed effects and difference-in-differences, are used
throughout the course.
1.2 Introduction to development economics

Economic development is the process of gradual improvement in the
material well-being of individuals. At the macro level, development
economics studies why some countries have high standards of living, while
others do not. At the micro level, development studies the functioning
of markets in low income countries, with the ultimate goal of addressing
market failures and lifting individuals out of poverty.
Almost all topics in economics have a counterpart in development
economics. A quick look at the syllabus shows the wide range of topics
covered in this guide. This, however, is only a selection of topics from an
array of development issues. You might be taking courses in international
trade or labour economics and you can probably identify intersections of
topics covered here with those courses. So, why do developing country
issues warrant a separate course of analysis?
The context of developing countries proves to be challenging for
conventional economic models due to the prevalence of market failures.
Although textbook models of economic behaviour are useful starting
1
points for thinking about issues in developing countries, the situation

often requires a careful investigation of the plausibility of assumptions in
a developing context. For example, we will see that poor households are
unable to access formal credit from banks and instead rely on alternative,
and sometimes more expensive, sources like moneylenders. This is
because the degree of information asymmetries and strength of contract
enforcement makes formal credit unviable.
1.3 Syllabus
The course is divided into three parts. In the first part, we cover macro
models of development and focus on explaining cross-country income
differences (Chapters 2 to 6). The second part investigates different
markets in developing countries with an emphasis on understanding
the market failures and potential corrective policies (Chapters 7 to 11).
The last part discusses the distinct role of the state in the process of
development (Chapters 12 to 16). This course leaves out two important
strands of literature: ‘firms and industrial development’ and ‘conflict’. Our
treatment of some of the other topics (e.g. trade policy) does not do justice
to the vast number of journal articles in the area. The syllabus for this
course is broken into three parts:
Part 1: Cross-country differences and macro models of
development
• Concepts and measurements of economic development and the
characteristics of developing countries.
• Models of economic growth and development including endogenous
growth theories and multiple equilibria models and their potential in
explaining income disparities across countries.
• Role of history and institutions in shaping current economic outcomes
and explaining development.
Part 2: Markets in developing countries
• Understanding demand for education, role of education infrastructure,
and the incentives of education providers in developing countries.
• Demand for health and nutrition in developing countries, provision of
public health services and the issue of sex imbalances.
• Importance of agriculture and land reform in shaping the lives of the
poor.
• Labour markets in developing countries and understanding the impact
of labour relations on poverty.
• Forms of agricultural contracts.
• Forms of credit and insurance markets in developing countries.
• Microfinance and its impacts on the lives of the poor.
Part 3: State and the process of development
• Infrastructure and its impact on development, globalisation and the
role of trade policy, environment and development.
• Taxation and development, informal economy and tax evasion.
• Development aid and its effectiveness in improving outcomes.
2
1.4 Aims of the course

The aims of this course are to:
• discuss key issues in the process of economic development
• enhance students’ ability in applying economic models to study
development problems
• discuss the relevant empirical literature with an eye toward forming
policy recommendations.
1.5 Learning outcomes for the course

At the end of this course, and having completed the Essential readings and
activities, you should be able to:
• outline the main theories and concepts in development economics
• select and analyse economic theories relevant for issues in economic
development
• summarise and evaluate empirical work in development economics
• compare and contrast empirical work for the purpose of designing
policy for a specific context.
1.6 Overview of learning resources

1.6.1 The subject guide
The subject guide is your first point of reference in this course but it is not
the only one. Passive reading of the guide and skimming through model
answers for questions is not a useful strategy. The guide provides ample
opportunities for active learning by providing activities that link you to
other learning resources.
Each chapter of the guide starts with the aims and learning outcomes.
These are listed so you explicitly know what you should learn by the end
of the chapter. I recommend that you start by reading the subject guide
rather than the Essential readings. You should try to go through activities
and questions throughout the chapter. These might include reading some
references, looking at data on the web or answering a question. Feedback
on activities is provided on the virtual learning environment (VLE).
After reading through the chapter you should go through the essential
readings. The subject guide will help you to decide which parts of the
readings to focus on. If you find parts of the guide hard to understand,
you should particularly try to read the reference on that part (whether it is
Essential or Further reading).
At the end of the chapter, you should review the learning outcomes and go
through sample questions to make sure you have a solid understanding of
that chapter. These questions are very similar to the examination questions
and you will have access to suggested solutions both for activities within
chapters and questions at the end of the chapter.
1.6.2 Essential reading

Detailed reading references in this subject guide refer to the editions
of the set textbooks listed below. New editions of one or more of
these textbooks may have been published by the time you study this
course. You can use a more recent edition of any of the books; use
the detailed chapter and section headings and the index to identify
3
relevant readings. Also check the VLE regularly for updated guidance
on readings.
For most chapters in this guide the following text will be used as
Essential reading:
Ray, D. Development economics. (Princeton NJ: Princeton University Press,
1998) [ISBN 9780691017068].
For micro development topics, an indispensable book is:
Banerjee, A. and E. Duflo Poor economics: a radical rethinking of the way to fight
global poverty. (New York: PublicAffairsTM, 2011) [ISBN 9781586487980].
You should buy copies of both these books.
We rely on Ray (1988) as the theoretical foundation for the topics
available in the textbook but for micro evidence and more recent topics
we rely on Banerjee and Duflo (2011) and journal articles. This book
provides an amazingly informative summary of recent development
research that tries to understand the lives of the poor. The book webpage
at pooreconomics.com provides data and other useful resources.
Each chapter uses two or three journal articles as Essential readings to
complement the textbook. As this might be the first time you read research
articles, the subject guide will help you work through the articles by
providing questions and learning activities throughout. Reading articles is
a skill you need to develop. When reading, it is useful to think about the
following questions:
1. What question is the article trying to answer (question)?
2. Why is the question interesting (motivation)?
3. How does the article answer the question (method)?
4. What are the main results (findings)?
5. What are potential problems with the interpretation of results put
forward by the authors (limitations)?
6. What potential policy conclusions can you draw from the findings
(policy)?
One useful reading strategy is to start with a careful reading of the
introduction of the article, then skim through the tables and figures and
read the conclusion. After this you should delve into the article and try to
understand its main message. This is a useful strategy because most of the
articles assigned for this course will have a thorough introduction in which
they discuss all six questions mentioned above.
You should note that you are not expected to understand all the technical
details of the articles but you should be able to understand and intuitively
explain their results and potential limitations. Throughout the guide we
will highlight the technical details that are required. You can find the list of
Essential readings at the beginning of each chapter but below we provide a
full list of all papers used as Essential readings in the course:
Acemoglu, D, S. Johnson and J.A. Robinson, ‘The colonial origins of
comparative development: an empirical investigation’, American Economic
Review 91(5) 2001, pp.1369–401.
Ahuja, A., M. Kremer and A. P. Zwane, ‘Providing safe water: evidence from
randomized evaluations’, Annual Review of Resource Economics 2(1) 2010,
pp.237–56. Available at http://scholar.harvard.edu/files/kremer/files/
annurev.resource.012809.103919.pdf
Banerjee, A., P. Gertler, and M. Ghatak, ‘Empowerment and efficiency: tenancy
reform in West Bengal’, Journal of Political Economy 110(2) 2002,
pp.239–80.
4
Banerjee, A. and E. Duflo, ‘Addressing absence’, Journal of Economic Perspectives

20(1) 2006, pp.117–32.
Banerjee, A.V. and E. Duflo ‘The experimental approach to development
economics’, Annual Review of Economics 1(1) 2009, pp.151–78. A version of
this article is available at http://www.nber.org/papers/w14467
Banerjee, A.V. and E. Duflo, ‘Giving credit where it is due’, Journal of Economic
Perspectives 24(3) 2010, pp.61–79.
Banerjee, A.V., E. Duflo, R. Glennerster and C. Kinnan ‘The miracle of
microfinance? Evidence from a randomized evaluation’, National Bureau of
Economic Research, Working Papers No. 18950, 2013.
Bernard, A.B., J.B. Jensen, S.J. Redding and P.K. Schott ‘Firms in international
trade’, Journal of Economic Perspectives 21(3) 2007, pp.105–30.
Besley, T. ‘Property rights and investment incentives: theory and evidence from
Ghana’, Journal of Political Economy 103(5) 1995, pp.903–37.
Burgess, R., O. Deschenes, D. Donaldson and M. Greenstone ‘Weather and
death in India’, Working Paper, 2011.
Burgess, R., M. Hansen, B. Olken, P. Potapov and S. Sieber, ‘The political
economy of deforestation in the tropics’, Quarterly Journal of Economics
127(4) 2012, pp.1707–54.
Burnside, C. and D. Dollar, ‘Aid, policies, and growth’, American Economic
Review 90(4) 2000, pp.847–68.
Chaudhury, N., J. Hammer, M. Kremer, K. Muralidharan, and F. Halsey Rogers
‘Missing in action: teacher and health worker absence in developing
countries’, Journal of Economic Perspectives 20(1) 2006, pp.91–116.
Chen, S. and M. Ravallion ‘The developing world is poorer than we thought,
but no less successful in the fight against poverty’, Quarterly Journal of
Economics 125(4) 2010, pp.1577–625.
Deaton, A. ‘Understanding the mechanisms of economic development’, Journal
of Economic Perspectives 24(3) 2010, pp.3–16.
De Mel, S., D.McKenzie, and C. Woodruff ‘The demand for, and consequences
of, formalization among informal firms in Sri Lanka’, American Economic
Journal: Applied Economics 5(2) 2013, pp.122–50.
Donaldson, D. ‘Railroads of Raj: estimating the impact of transportation
infrastructure’, American Economic Review (forthcoming). A version of this
article is available at http://economics.mit.edu/files/6038
Duflo, E. ‘Schooling and labor market consequences of school construction in
Indonesia: evidence from an unusual policy experiment’, American Economic
Review 91(4) 2001, pp.795–813.
Easterly, W. ‘Inequality does cause under development: insights from a new
instrument’, Journal of Development Economics 84(2) 2007, pp.755–76.
Easterly, W. ‘Can the West Save Africa?’, Journal of Economic Literature 47(2)
2009, pp.373–447.
Field, E., ‘Entitled to work: urban property rights and labor supply in Peru’,
Quarterly Journal of Economics 122(4) 2007, pp.1561–602.
Galor, O. and J. Zeira, ‘Income distribution and macroeconomics’, Review of
Economic Studies 60(1) 1993, pp.35–52.
Glaeser, E., R. LaPorta, F. Lopez-de-Silanes and A. Shleifer ‘Do institutions cause
growth?’, Journal of Economic Growth 9(3) 2004, pp.271–303.
Gordon, R. and W. Li ‘Tax structures in developing countries: many puzzles
and a possible explanation’, Journal of Public Economics 93(7–8) 2009,
pp.855–66.
Hsieh, C-T. and P.J. Klenow, ‘Development accounting’, American Economic
Journal: Macroeconomics 2(1) 2010, pp.207–23.
Jensen, R., ‘The digital provide: information (technology), market performance
and welfare in the South Indian fisheries sector’, Quarterly Journal of
Economics 122(3) 2007, pp.879–924.
Mankiw, N.G., D. Romer and D. Weil ‘A contribution to the empirics of
economic growth’, Quarterly Journal of Economics 107(2) 1992, pp.407–37.
5
Melitz, M.J. and D. Trefler, ‘Gains from trade when firms matter’, Journal of
Economic Perspectives 26(2) 2012, pp.91–118.
Nunn, N. ‘The importance of history for economic development’, Annual Review
of Economics 1(1) 2009, pp.65–92. Available at http://goo.gl/dgt3YG
Piketty, T. and N. Qian, ‘Income inequality and progressive income taxation in
China and India 1986–2010’, American Economic Journal: Applied Economics
1(2) 2009, pp.53–63.
Ravallion, M. ‘The mystery of the vanishing benefits: an introduction to impact
evaluation’, World Bank Economic Review 15(1) 2001, pp.115–40.
Strömberg, D., ‘Natural disasters, economic development, and humanitarian
aid’, Journal of Economic Perspectives 21(3) 2007, pp.199–222.
Townsend, R., ‘Consumption insurance: an evaluation of risk-bearing systems
in low-income economies’, Journal of Economic Perspectives 9(3) 1995,
pp.83–102.
Udry, C. ‘Credit markets in northern Nigeria: credit as insurance in a rural
economy’, World Bank Economic Review 4(3) 1990, pp.251–69.
1.6.3 Further reading

Please note that as long as you read the Essential reading you are then free
to read around the subject area in any text, article or online resource. You
will need to support your learning by reading as widely as possible and by
thinking about how these principles apply in the real world. To help you
read extensively, you have free access to the VLE and University of London
Online Library (see below).
The subject guide also provides a list of Further readings that could guide
you in exploring the subject further. Sometimes the results from Further
readings are referred to in the subject guide. This material is required and
if you have difficulty understanding the interpretations, you should consult
the original source. The following is the list of all Further readings in this
guide:
Acemoglu, D., and S. Johnson, ‘Unbundling institutions’, Journal of Political
Economy 113(5) 2005, pp.949–95.
Aker, J.C. and I.M. Mbiti, ‘Mobile phones and economic development in Africa’,
Journal of Economic Perspectives 24(3) 2010, pp.207–32.
Aleem, I. ‘Imperfect information, screening, and the costs of informal lending:
a study of a rural credit market in Pakistan’, World Bank Economic Review
4(3) 1990, pp.329–49.
Angrist, J.D. and J.S. Pischke, ‘The credibility revolution in empirical
economics: how better research design is taking the con out of
econometrics’, Journal of Economic Perspectives 24(2) 2010, pp.3–30.
Ashraf, N., O. Bandiera and K. Jack, ‘No margin, no mission? A field experiment
on incentives for pro-social tasks’, CEPR Discussion Paper no. 8834, 2013.
Banerjee, A.V. and E. Duflo, ‘The economic lives of the poor’, Journal of
Banerjee, A.V. and E. Duflo, ‘Inequality and growth: what can the data say?’,
Journal of Economic Growth 8(3) 2003, pp.267–99.
Banerjee, A.V. and A.F. Newman, ‘Occupational choice and the process of
development’, Journal of Political Economy 101(2) 1993, pp.274–98.
Banerjee, A.V., E. Duflo, R. Glennerster and D. Kothari, ‘Improving
immunization coverage in rural India: a clustered randomized controlled
evaluation of immunization campaigns with and without incentives’,
Working Paper 2010.
Bardhan, P. and D. Mookherjee ‘Determinants of redistributive politics:
an empirical analysis of land reforms in West Bengal, India’, American
Economic Review 100(4) 2010, pp.1572–600.
Besley, T. and R. Burgess, ‘Halving global poverty’, Journal of Economic
6
Besley, T. and R. Burgess, ‘Land reform, poverty reduction, and growth:

evidence from India’, Quarterly Journal of Economics 115(2) 2000,
pp.389–430.
Besley, T. and M. Ghatak ‘Property rights and economic development’ in
Rodrik, D. and M. Rosenzweig Handbook of development economics (Oxford;
Amsterdam: Elsevier, 2010) Volume 5, Chapter 68, pp.4525–95.
Besley, T. and T. Persson, ‘Taxation and development’ in Auerbach A.J.,
R. Chetty, M. Feldstein and E. Saez (eds) Handbook of public economics.
(Oxford; Amsterdam: Elsevier, 2013) Volume 5, Chapter 2, pp.51–110.
Bjorkman, M. and J. Svensson, ‘Power to the people: evidence of a randomized
field experiment on community-based monitoring in Uganda’, Quarterly
Journal of Economics 24(2) 2009, pp.735–69.
Burnside, C. and D. Dollar ‘Aid, policies, and growth: reply’, American Economic
Review 94(3) 2004, pp.781–84.
Caselli, F. ‘Accounting for cross-country income differences’ in Aghion P. and
N.D. Steven Handbook of economic growth. (Oxford; Amsterdam: Elsevier,
2005) Volume 1, Part A, pp.679–741.
Caselli, F., G. Esquivel, and F. Lefort ‘Reopening the convergence debate: a new
look at cross-country growth empirics’, Journal of Economic Growth 1(3)
1996, pp.363–89.
Chaudhury, N., J. Hammer, M. Kremer, K. Muralidharan and F. Halsey Rogers
Cohen, J., and P. Dupas ‘Free distribution or cost-sharing? Evidence from a
randomized Malaria prevention experiment’, Quarterly Journal of Economics
125(1) 2010, pp.1–45.
Comin, D., W. Easterly and E. Gong ‘Was the wealth of nations determined
in 1000 BC?’, American Economic Journal: Macroeconomics 2(3) 2010,
pp.65–97.
Cull, R., A. Demirgüç-Kunt and J. Morduch ‘Microfinance meets the market’,
Das, J., J. Hammer, and K. Leonard ‘The quality of medical advice in low-
income countries’, Journal of Economic Perspectives 22(2) 2008, pp.93–114.
Davis, D.R. and D.E. Weinstein ‘Bones, bombs, and break points: the geography
of economic activity’, American Economic Review 92(5) 2002, pp.1269–89.
Deaton, A., ‘Income, health, and well-being around the world: evidence
from the Gallup World Poll’, Journal of Economic Perspectives 22(2) 2008,
pp.53–72.
Deaton, A. ‘Instruments of development: randomization in the tropics, and the
search for the elusive keys to economic development’, Proceedings of the
British Academy, 2008 Lectures 162 (Oxford: Oxford University Press, 2009)
pp.123–60.
Dell, M., B. Jones and B. Olken ‘Temperature shocks and economic growth:
evidence from the last half century’, American Economic Journal:
Macroeconomics 4(3) 2012, pp.66–95.
Devoto, F., E. Duflo, P. Dupas, W. Pariente and V. Pons ‘Happiness on tap: piped
water adoption in urban Morocco’, American Economic Journal: Economic
Policy 4(4) 2012, pp.68–99.
Dinkelman, T. ‘The effects of rural electrification on employment: new evidence
from South Africa’, American Economic Review 101(7) 2011, pp.3078–108.
Duflo, E., R. Glennerster and M. Kremer ‘Using randomization in development
economics research: a toolkit’, Handbook of development economics 4 2007,
pp.3895–962.
Duflo, E. and R. Pande ‘Dams’, Quarterly Journal of Economics 122(2) 2007,
pp.601–46.
Duflo, E., R. Hanna and S. Ryan ‘Incentives work: getting teachers to come to
school’, American Economic Review 102(4) 2012, pp.1241–78.
7
Duflo, E., M. Greenstone, R. Pande and N. Ryan ‘Truth-telling by third-party

auditors and the response of polluting firms: experimental evidence from
India’, Quarterly Journal of Economics 128(4) 2013, pp.1499–545.
Easterly, W., R. Levine and D. Roodman ‘Aid, policies, and growth: comment’,
American Economic Review 94(3) 2004, pp.774–80.
Ebenstein, A. ‘The consequences of industrialization: evidence from water
pollution and digestive cancers in China’, Review of Economics and Statistics
94(1) 2012, pp.186–201.
Frazer, G., ‘Used clothing donations and apparel production in Africa’, Economic
Journal 118(532) 2008, pp.1764–84.
Galiani, S., P. Gertler and E. Schargrodsky ‘Water for life: the impact of the
privatization of water services on child mortality’, Journal of Political
Economy 113(1) 2005, pp.83–120.
Giné, X., R. Townsend and J. Vickery ‘Patterns of rainfall insurance
participation in rural India’, World Bank Economic Review 22(3) 2008,
pp.539–66.
Goldberg, P.K. and N. Pavcnik ‘Distributional effects of globalization in
developing countries’, Journal of Economic Literature 45(1) 2007 pp.39–82.
Goyal, A. ‘Information, direct access to farmers, and rural market performance
in Central India’, American Economic Journal: Applied Economics 2(3) 2010,
pp.22–45.
Graff Zivin, J. and M. Neidell ‘Environment, health, and human capital’,
Journal of Economic Literature 51(3) 2013, pp.689–730.
Greenstone, M. ‘Estimating regulation-induced substitution: the effect of the
Clean Air Act on water and ground pollution’, American Economic Review
93(2) 2003, pp.442–48.
Hsieh, C-T and P. Klenow ‘Misallocation and manufacturing TFP in China and
Jayachandran, S. ‘Liquidity constraints and deforestation: the limitations of
payments for ecosystem services’, American Economic Review 103(3) 2013,
pp.309–13.
Jensen, R. ‘The (perceived) returns to education and the demand for schooling’,
Kleven, H.J., M.B. Knudsen, C.T. Kreiner, S. Pedersen and E. Saez ‘Unwilling
or unable to cheat? Evidence from a tax audit experiment in Denmark’,
Econometrica 79(3) 2011, pp.651–92.
Kremer, M., J. Leino, E. Miguel and A. P. Zwane ‘Spring cleaning: rural water
impacts, valuation and property rights institutions’, Quarterly Journal of
Economics 126(1) 2011, pp.145–205.
Kuziemko, I. and E. Werker ‘How much is a seat on the Security Council worth?
Foreign aid and bribery at the United Nations’, Journal of Political Economy
114(5) 2006, pp.905–30.
La Porta, R., F. Lopez-de-Silanes and A. Shleifer ‘The economic consequences of
legal origins’, Journal of Economic Literature 46(2) 2008, pp.285–332.
Lin, J. Y. ‘Rural reforms and agricultural growth in China’, American Economic
Review 82(1) 1992, pp.34–51.
Maccini, S. and D. Yang ‘Under the weather: health, schooling, and economic
consequences of early-life rainfall’, American Economic Review 99(3) 2009,
pp.1006–26.
Meyer, B.D. ‘Natural and quasi-experiments in economics’, Journal of Business
and Economic Statistics 13(2) 1995, pp.151–61.
Miguel, E. and G. Roland ‘The long-run impact of bombing Vietnam’, Journal of
Development Economics 96(1) 2011, pp.1–15.
Morduch, J. ‘The microfinance promise’, Journal of Economic Literature 37(4)
1999, pp.1569–614.
8
Morduch, J. ‘Income smoothing and consumption smoothing’, Journal of

Murphy, K., A. Shleifer and R. Vishny ‘Industrialisation and the big push’,
Journal of Political Economy 97(5) 1989, pp.1003–26.
Newell, R.G., W.A. Pizer and D. Raimi ‘Carbon markets 15 years after Kyoto:
lessons learned, new challenges’, Journal of Economic Perspectives 27(1)
2013, pp.123–46.
North, D.C. ‘Institutions’, Journal of Economic Perspectives 5(1) 1991,
pp.97–112.
Nunn, N. and L. Wantchekon ‘The slave trade and the origins of mistrust in
Africa’, American Economic Review 101(7) 2011, pp.3221–52.
Nunn, N. and N. Qian ‘U.S. Food Aid and Civil Conflict’, American Economic
Review (forthcoming).
Olken, B.A. and M. Singhal ‘Informal taxation’, American Economic Journal:
Applied Economics 3(4) 2011, pp.1–28.
Pavcnik, N. ‘Trade liberalization, exit, and productivity improvements: evidence
from Chilean plants’, Review of Economic Studies 69(1) 2002, pp.245–76.
Qian, N. ‘Missing women and the price of tea in China: the effect of sex-specific
earnings on sex imbalance’, Quarterly Journal of Economics 123(3) 2008,
pp.1251–285.
Ray, D. ‘Uneven growth: a framework for research in development economics’,
Redding, S.J., D.M. Sturm and N. Wolf ‘History and industry location: evidence
from German airports’, Review of Economics and Statistics 93(3) 2011,
pp.814–31.
Rosenzweig, M.R. and K.I. Wolpin ‘Natural “natural experiments”’, Journal of
Economic Literature 38(4) 2000, pp.827–74.
Schultz, T.P. ‘School subsidies for the poor: evaluating the Mexican Progresa
poverty program’, Journal of Development Economics 74(1) 2004,
pp.199–250.
Stern, N. ‘The economics of climate change’, American Economic Review 98(2)
2008, pp.1–37.
Subramanian, S. and A. Deaton ‘The demand for food and calories’, Journal of
Political Economy 104(1) 1996, pp.133–62.
Udry, C. ‘Risk and insurance in a rural credit market: An empirical investigation
in northern Nigeria’, Review of Economic Studies 61(3) 1994, pp.495–526.
Verhoogen, E.A. ‘Trade, quality upgrading, and wage inequality in the Mexican
manufacturing sector’, Quarterly Journal of Economics 123(2) 2008,
pp.489–530.
1.6.4 Online library

In addition to the subject guide and the Essential reading, it is crucial that
you take advantage of the study resources that are available online for this
course, including the VLE and the Online Library.
You can access the VLE, the Online Library and your University of London
email account via the Student Portal at:
http://my.londoninternational.ac.uk
You should have received your login details for the Student Portal with
your official offer, which was emailed to the address that you gave on
your application form. You have probably already logged in to the Student
Portal in order to register. As soon as you registered, you will automatically
have been granted access to the VLE, Online Library and your fully
functional University of London email account.
If you have forgotten these login details, please click on the ‘Forgotten
your password’ link on the login page.
9
The VLE
The VLE, which complements this subject guide, has been designed to
enhance your learning experience, providing additional support and a
sense of community. It forms an important part of your study experience
with the University of London and you should access it regularly.
The VLE provides a range of resources for EMFSS courses:
• Self-testing activities: doing these allows you to test your own
understanding of subject material.
• Electronic study materials: the printed materials that you receive from
the University of London are available to download, including updated
reading lists and references.
• Past examination papers and Examiners’ commentaries: these provide
advice on how each examination question might best be answered.
• A student discussion forum: this is an open space for you to discuss
interests and experiences, seek support from your peers, work
collaboratively to solve problems and discuss subject material.
• Videos: there are recorded academic introductions to the subject,
interviews and debates and, for some courses, audio-visual tutorials
and conclusions.
• Recorded lectures: for some courses, where appropriate, the sessions
from previous years’ Study Weekends have been recorded and made
available.
• Study skills: expert advice on preparing for examinations and
developing your digital literacy skills.
• Feedback forms.
Some of these resources are available for certain courses only, but we
are expanding our provision all the time and you should check the VLE
regularly for updates.
Making use of the Online Library

The Online Library contains a huge array of journal articles and other
resources to help you read widely and extensively.
To access the majority of resources via the Online Library you will either
need to use your University of London Student Portal login details, or you
will be required to register and use an Athens login:
http://tinyurl.com/ollathens
The easiest way to locate relevant content and journal articles in the
Online Library is to use the Summon search engine.
If you are having trouble finding an article listed in a reading list, try
removing any punctuation from the title, such as single quotation marks,
question marks and colons.
For further advice, please see the online help pages:
www.external.shl.lon.ac.uk/summon/about.php
1.7 The examination

Important: the information and advice given here are based on the
examination structure used at the time this guide was written. Please
note that subject guides may be used for several years. Because of this
we strongly advise you to always check both the current Regulations
for relevant information about the examination, and the VLE where you
10
should be advised of any forthcoming changes. You should also carefully

check the rubric/instructions on the paper you actually sit and follow
those instructions.
Structure
The time allowed in the exam is three hours. The exam has two
sections. In Section A you will have to answer eight true/false questions
and provide a justification for your answer. Simply writing true/false will
get you no credit. In Section B you will have to answer two long questions
from a choice of four. A Sample examination paper and an Examiners’
commentary are provided at the end of this subject guide.
Advice
You should follow all the excellent advice to candidates published annually
as an introduction to the Examiners’ commentaries.
In addition, the following advice is valuable:
1. Prepare thoroughly for the examination by attempting the activities/
questions in the textbooks and in this subject guide.
2. If you are unsure of what a question is asking, for example because
of doubt about its interpretation, state the assumptions you made to
answer the question.
3. Even though a question may not specifically ask for defining key
terms, it is a good idea to state the definition of concepts used in the
question.
4. The introduction to the Examiners’ commentaries mentions the
importance of key words. Be prepared for words that tell you what to
do – for example: analyse, describe, discuss, outline, examine, assess,
suggest, propose – and make sure you follow the instruction. Do not
write down all you know about a topic in the hope that some of it is
relevant.
5. In many cases, good answers will require diagrammatic or
simple algebraic analysis to complement verbal reasoning. The
complementary dimension is important: good diagrams can often save
much verbal explanation but free-standing diagrams, however well-
drawn and labelled, do not communicate sufficient information to the
Examiners. Diagrams and algebraic presentations need to be explained
in the text of the answer. Symbols and abbreviations, other than those
that are widely known and used (for example, ‘Y’ for output; ‘S’ for
saving; ‘FDI’ for private foreign investment; ‘IMF’ for International
Monetary Fund), should be defined/spelled out when you first use
them in the examination.
The examination paper for EC3044 Economics of development
involves economic analysis. The Examiners expect analytical explanations
and/or suggestions for policy based on economic theory and empirical
evidence. Thus, even if it is not explicitly mentioned, you need to
demonstrate knowledge of theories and empirical evidence in answering
questions.
It is also a paper involving economic principles and their application, and
therefore the following general guidance is offered. You are not expected
to learn and reproduce vast quantities of statistical data. Authors include
statistics to introduce, illustrate or highlight particular issues, often prior to
presenting possible analytical explanations of the data. To the extent that
you wish to refer to statistics in your examination answers, broad orders of
11
magnitude will suffice. You should, however, be familiar with the results and
conclusions of Essential readings.
Remember, it is important to check the VLE for:
• up-to-date information on examination and assessment arrangements
for this course
• where available, past examination papers and Examiners’
commentaries for the course which give advice on how each question
might best be answered.
12
Part 1: Cross-country differences and macro models of development
Part 1: Cross-country differences

and macro models of
development
13
Notes
14
Chapter 2: Overview of development basic facts
Chapter 2: Overview of development

basic facts
2.1 Introduction
2.1.1 Aims of the chapter
The aims of this chapter are to:
• introduce the concept of development
• explain various ways of measuring development
• introduce main methods used in studying development issues.
2.1.2 Learning outcomes

By the end of this chapter, and having completed the Essential reading and
• explain and evaluate various measures of economic development
• recall and state key facts about development such as large income
disparities and the staggering number of people who live in poverty
• explain the role of theory and empirics in development and recognise
the difference between correlation and causation.

Ray (1998) Chapter 2 and Appendix 2.
Banerjee and Duflo (2011) Chapters 1 and 2.
Banerjee, A.V. and E. Duflo ‘The experimental approach to development
economics’, Annual Review of Economics 1(1) 2009, pp.151–78. A version of
this article is available at www.nber.org/papers/w14467
Chen, S. and M. Ravallion, ‘The developing world is poorer than we thought,
but no less successful in the fight against poverty’, Quarterly Journal of
Economics 125(4) 2010, pp.1577–625.
Deaton, A., ‘Understanding the mechanisms of economic development’, Journal
of Economic Perspectives 24(3) 2010, pp.3–16.
Ravallion, M., ‘The mystery of the vanishing benefits: an introduction to impact
evaluation’, World Bank Economic Review 15(1) 2001, pp.115–40.

Angrist, J.D. and J.S. Pischke ‘The credibility revolution in empirical economics:
how better research design is taking the con out of econometrics’, Journal of
Banerjee, A.V. and E. Duflo ‘The economic lives of the poor’, Journal of
Besley, T. and R. Burgess ‘Halving global poverty’, Journal of Economic
Deaton, A. ‘Income, health, and well-being around the world: evidence from
the Gallup World Poll’, Journal of Economic Perspectives 22(2) 2008,
pp.53–72.
Deaton, A., ‘Instruments of development: randomization in the tropics, and
the search for the elusive keys to economic development’, Proceedings of
the British Academy, 2008 Lectures, 162. (Oxford: Oxford University Press,
2009) pp.123–60.
Duflo, E., R. Glennerster and M. Kremer ‘Using randomization in development
economics research: a toolkit’, Handbook of Development Economics 4 2007,
pp.3895–962.
15
Rosenzweig, M.R. and K.I. Wolpin, ‘Natural “natural experiments”’, Journal of

Economic Literature 38(4) 2000, pp.827–74.
2.1.5 Works cited

Acemoglu, D. ‘Theory, general equilibrium, and political economy in
development economics’, Journal of Economic Perspectives 24(3) 2010,
pp.17–32.
Dunning, T. Natural experiments in the social sciences: a design-based approach.
(Cambridge: Cambridge University Press, 2012).
Helliwell, J.F. ‘How’s life? Combining individual and national variables to
explain subjective well-being’, Economic Modelling 20(2) 2003, pp.331–60.
Sen, A. ‘The concept of development’ in Hollis C. and T.N. Srinivasan Handbook
of development economics. (Oxford; Amsterdam Elsevier., 1988), Volume 1,
Chapter 1, pp.9–26.
Stock, J.H. and M.W. Watson Introduction to econometrics. (Harlow: Pearson
Education, Limited, 2012).
Wooldridge, J. M. Introductory econometrics: a modern approach. (Cengage
Learning, 2012).
2.1.6 Overview
In the first section of this chapter we discuss the concept of development.
In section 2.3 we look at various ways of measuring development. GDP
(gross domestic product) per capita is introduced as a core measure for
comparing living standards across countries. We next present facts on
large income gaps between developing and developed countries and
the heterogeneity of growth performances. We briefly discuss poverty,
inequality and other ways of measuring well-being. In section 2.4 we
discuss the importance of theory and the role of empirical analysis in
development research.
2.2 Concept of development

Development could refer to conditions that identify the quality of material
life. It could also refer to the processes of change and improvement that
raise living standards. Development is definitely not a pure economic
phenomenon but in this course we only focus on the economic aspects
of development. This includes a range of issues in health, education,
inequality and poverty. You should not confuse the concept of development
and the various measures that look at aspects of development (covered in
the next section). For a deeper discussion of the concept of development
and its roots in economic thinking, you are encouraged to read Sen
(1988).
2.3 Measuring development

In order to have an understanding of the relative performance of
countries, regions or individuals, we need to measure development.
Development is inevitably a multidimensional phenomenon. You can
think of many important aspects of material well-being that indicate levels
of development. Economists have long focused on income and income
growth as measures of economic development. This is, however, no longer
the case. Many studies and international organisations widely report other
indices of development.
In this section we look at income, poverty, inequality and other measures
of well-being used in development.
16
2.3.1 Income and growth

GDP is the core measure of aggregate income used to measure total
production from a country or region over a given period of time (usually
a year, sometimes a quarter). You should have a quick catch up with what
you have learned from EC2065 Macroeconomics about GDP and be
able to distinguish between it and gross national product (GNP) .
GDP per capita (divided by population) is, therefore, the primary measure
of average individual income in a given country. As income is spent
on purchasing goods and services, GDP per capita could be a measure
of living standards. Development is concerned about comparisons of
living standards over time and across countries. In order to make GDP
comparable over time, we need to account for inflation. As prices go up
(inflation), the purchasing power of income falls. By choosing a base year
and using appropriate price indices, you can convert nominal GDP to real
or constant GDP. This will then capture the real change in living standards
relative to the base year for a given country.
To compare living standards across countries we need to convert GDP
figures into a common currency (e.g. US $). But what is the exchange
rate for this conversion? One option is to simply use the average exchange
rates from currency markets in a given year. There are certain problems
with this method. Most importantly, exchange rates are affected by
tradeable goods and services, while GDP includes both tradeable and
non-tradeable goods and services. If non-tradable goods and services are
‘cheaper’ in developing countries, using an exchange rate conversion GDP
would overstate the income gap (i.e. developing countries would appear to
be poorer than they actually are).
One way to overcome this problem is to use physical quantities produced
in each country and aggregate them using the set of relative prices in a
given country (e.g. the USA). In other words, this method calculates how
much more material is produced in a country if goods have a relative value
similar to the USA. Data on quantities is hard to come by. Therefore this
idea is usually implemented by adjusting existing GDP figures using an
appropriate price index.
The World Bank started the International Comparison Program (ICP) to
collect price data on individual items across a wide range of countries.
This price data allows the construction of purchasing power parity (PPP)
conversion rates. These conversion rates show the cost of purchasing a
given basket of goods in different countries relative to the USA. You can
also refer to the World Bank page on ICP for more detailed information:
http://siteresources.worldbank.org/ICPEXT/Resources/ICP_2011.html
Activity 2.1
Here we try to develop a simple example to understand how PPP conversions help make
GDP figures more comparable across countries. Consider two countries that only produce
haircuts and potatoes. Potatoes are freely traded (without any cost) but haircuts aren’t.
Quantities produced and corresponding prices in the two currencies are reported in Table 1.
Country A Country B
Quantity Price (R: Rupee) Quantity Price (D: Dollar)
Potatoes 10 ton 100 R/ton 60 ton 10 D/ton
Haircuts 100 10 R/each 100 4 D/each
Table 1: Quantities produced and prices in 2013.
17
a. Calculate the GDP for each country in its own currency. Then use the exchange rate
(1 dollar buys 10 rupees) to convert country A’s GDP into dollars. What is the ratio of
country B’s GDP to that of country A?
b. Now use the PPP principle to calculate country A’s GDP in dollars. What is the ratio of
two countries’ GDP using this method?
c. Is it the haircuts or the potatoes that make the two methods different?
2.3.2 How large are the income gaps?

Once you have made GDP figures comparable across time and countries,
you can start looking at income gaps between developing and developed
countries. You should get a clear idea about the income gap between
the rich and the poor countries to appreciate the role of development
economics. Ray (1998) Chapter 2 provides some historical figures but
for an updated view of the situation, you should try one of the online
data sources. Activity 2.2 asks you to look at the staggering income gap
between the USA and the poorest country in the world.
Activity 2.2
Use World Development Indicators (WDI: http://data.worldbank.org/data-catalog/world-
development-indicators) to find the poorest and richest countries in the sample in 1980.
How large is the rich country’s income relative to the poor country? Find the poorest
country in 1990, 2000, and 2012 and calculate the ratio of US income to the poorest
country’s income in each year. Is the gap between the USA and the poorest country
shrinking or growing? You should use the variable from WDI that measures constant GDP
per capita in units of PPP dollars to look at income.
Figure 2.1 shows the evolution of real GDP per capita measured at PPP $ for
India, Kenya and the USA. The figure shows the staggering income gaps.
While US income was around $42,000 per capita in 2010, Kenyans had an
income of just $1,500. In other words, the average income of US citizens
was 28 times higher. The graph also shows the growth performance of these
countries. While the USA has experienced a continuous increase in income,
GDP per capita in the Kenya did not change much over the sample. Average
income in India, however, has increased more than threefold over the
plotted time.
GDP per capita (PPP, constant 2005 international $)
45000
3500
India
Kenya
3000
USA
40000
2500
35000
2000
30000
1500
1000
25000
1980 1990 2000 2010

Year
Figure 2.1: Evolution of GDP per capita in the USA (right axis), India and Kenya
(left axis).
Source: World Development Indicators, The World Bank.
18
Table 2.1 shows the average growth rate of a selected number of countries
over 33 years. The USA had a steady growth of around 1.7 per cent
and as the last column shows, it takes about 42 years for US income to
double if it grows at the average rate (Activity 2.3 asks you to calculate
this for China). On the other hand, China and India have experienced
rapid growth during the past few decades which led to a reasonable
average growth rate of around 8.9 and 4.3 per cent respectively. It seems
that at current rates the gap between these countries and the USA is
shrinking. The last four countries chosen from Africa show mixed growth
performances. Uganda and Ethiopia have grown at reasonable average
rates of 2.4 and 1.8 per cent respectively, but Kenya and the Democratic
Republic of Congo (formerly Zaire) had poor growth. Average income
in Zaire has declined in the past decades while it was stagnant in Kenya,
leading to a widening gap between these countries and the USA.
Country Average growth during Years it takes for income to double
1980 – 2012 (per cent)
USA 1.7 42
China 8.9 8
India 4.3 16
Kenya 0.3 204
Uganda 2.4 29
Ethiopia 1.8 39
Zaire –2.3 –
Table 2.1: Average growth rates for a sample of countries.
Activity 2.3
China has grown extraordinarily over the past few decades. Assuming China sustains its
current growth rate of 9 per cent over a long time, calculate the number of years it takes
for China to double its income. WDI reports China had a GDP per capita of $6,819 in
2010 when the equivalent figure in the USA was $42,000. If the US economy continues
to grow at 2 per cent and China sustains a growth rate of 9 per cent, calculate the
number of years it would take China to catch up with US income.
How reliable are PPP comparisons?

Even with PPP conversions, GDP differences across developed and
developing countries could overstate the real gap in living standards.
There are two reasons for this. First, home production may account for a
larger part of production (and consumption) in developing countries. For
example, childcare is a service which the majority of families purchase
from a provider in developed countries but in developing countries
children stay at home and either siblings or parents look after them.
Although in both cases a utility generating service is delivered, in one case
it is transacted in the market and hence included in GDP, while in the other
it is home production and not included in national accounts.
Activity 2.4
Can you think of other examples of home production that are more common in
developing countries? How does the process of development change the nature of home
production?
19
The second reason why PPP-converted GDP ratios might overstate the
real gap is the presence of the informal economy or black market. The
share of businesses that are not registered with the government to avoid
paying taxes and other duties is much larger in developing countries. This
means a much larger share of economic transactions are not registered
and usually carried in cash. In the presence of an informal economy, the
GDP (which relies on formal sources) underestimates actual production.
It is however unlikely that these concerns could bridge the large gaps in
incomes described above.
► Stop and read

Ray (1998) pp.7–21 on measuring income and growth (sections 2.1 and 2.2).
2.3.3 Poverty and inequality

Poverty is a keyword you come across frequently in development
literature. In one sense development economics is the study of methods for
improving the livelihoods of the poor. But how do we define poverty and
how are countries compared on this dimension?
A common definition of poverty is the headcount ratio (HCR). This
measure calculates the ratio of the number of individuals living below a
level of income relative to the total population. In other words, it shows
the percentage of individuals living below a poverty line.
Each country has its own ‘national poverty line’. Higher income countries
have higher poverty lines. To make international comparisons, the World
Bank adopted a global absolute poverty line of existing on $1 a day. This
is close to India’s poverty line and it is hard to imagine how individuals
could live on so little. Table 3 in Chen and Ravallion (2010) shows the
staggering number of individuals living below the $1 a day poverty
line. Over time the situation has improved in many countries, with the
exception of Sub-Saharan Africa where 40 per cent of people still live on
less than $1 a day.
► Stop and read

Banerjee and Duflo (2011) Chapters 1 and 2 provide an excellent summary of the
choices poor individuals face in their everyday life. Chen and Ravallion (2010) estimate
the number of the poor below an internationally comparable poverty line.
There is a close relationship between poverty and growth. If the fruits of

income growth are distributed evenly among the population, we would
expect growth to eradicate poverty. Indeed, from the discussion above you
can see that countries/regions that experienced good growth in the past
decades also reduced the percentage of their population living in absolute
poverty, while countries in Sub-Saharan Africa with a poor growth
performance also maintained 1980 poverty levels.
Growth, however, does not necessarily translate into poverty reduction.
Growth could be compatible with maintaining or even deteroriating
poverty levels when inequality worsens. Inequality is also of intrinsic
interest as a direct measure of development. We discuss various measure
of inequality and the interplay of inequality and development in Chapter 5
of this guide.
► Stop and read

Ray (1998) pp.21–25 on inequality and growth (section 2.3).
20
2.3.4 Measuring well-being

The ultimate goal of development is to increase material well-being
and life satisfaction but so far we have focused on income. The central
question is whether income is a good measure of human well-being and
living standards. In other words, does an increase in income lead to an
improvement in various aspects of well-being?
It is useful if we distinguish between subjective and objective well-being.
When you ask individuals whether they are satisfied with their lives, you
are measuring subjective well-being. On the other hand, when you measure
things like life expectancy, infant mortality or years of education, you are
choosing a set of objective variables and hence looking at objective well-being.
In an effort to quantify societal objective well-being, in 1990 the United
Nations Development Programme (UNDP) started to report a composite
index known as HDI (Human Development Index). This Index combines the
three elements of life expectancy, education and income and comes up with
an objective measure of societal well-being across countries of the world.
Activity 2.5
UNDP has a useful website (http://hdr.undp.org/en/data) that allows visualisation of HDI
data. Use the HDI data to create a graph that shows HDI versus GNI per capita (or log of
GNI per capita) for countries of the world in 2012.
a. What does this graph tell you about the appropriateness of GDP as a measure of
well-being?
b. Thinking about how HDI is constructed, do you think HDI is a good measure of well-
being?
Measuring subjective well-being is much more difficult, but the general

pattern is that people in developed countries are generally more satisfied
with their lives. This however varies greatly with age. Read Deaton (2008)
if you are interested in knowing more about subjective well-being and how
it changes with the levels of income and age of people.
HDI and several other measures of well-being are strongly correlated with
incomes across the countries of the world. Therefore it seems reasonable to
focus on income in cross-country comparisons of development. But, as you
will see during this course, the story of development is not that simple. In
order to understand the causes of development we need to delve into each
dimension of development separately and discuss why outcomes are hugely
different across countries and how we can improve them.
► Stop and read

Ray (1998) pp 25–32 on Human Development Index (section 2.4).
2.4 Methods
In this section we provide an overview of the methods used in studying
development economics. We offer a discussion of why theory is important
and explain how empirical studies inform our thinking about development
issues.
2.4.1 Role of theory

In studying development issues it is important to start with a theory. Theory
is not necessarily a mathematical model but is a careful articulation of
mechanisms that link factors to outcomes. There are four roles that theory
could play in development:
21
1. It maps the relevant factors and their interactions with each other and
with outcomes.
2. It clarifies the assumptions under which government intervention is
needed. In other words, it helps identify the market failures that justify
policy making.
3. Theory provides quantitative implications that could be tested using
data.
4. Theory provides a framework for understanding empirical work. If you
are interested in the role of theory you should read Acemoglu (2010).
2.4.2 Empirical methods

Once you have a theoretical understanding of issues you need to test
whether the predictions are held in the data. Furthermore, theory could
outline potential policies for improving outcomes. In order to assess the
effectiveness of such policies you need to engage in empirical evaluation
using data. Econometrics is a powerful tool in testing predictions
and evaluating alternative policies. It is, however, very important to
understand the pitfalls and limitations of econometric analysis.
This course will draw heavily on empirical journal articles that use various
econometric techniques. Our emphasis is on intuitive understanding of
what could go wrong with econometric methods and the assumptions
required for causal interpretation of the results. However, you need to
have good understanding of basic regression analysis, which we refer to as
ordinary least squares (OLS).
► Stop and read

For basics of statistical analysis and simple regression you should read Ray (1998)
Appendix 2. If you find the concepts here difficult, you should consult an econometrics
textbook. We recommend Stock and Watson (2012), Wooldridge (2012), or Dunning
(2012). You should read the first few chapters on the basics of multivariate regression
analysis.
Often in development economics, we are interested in knowing the

relationship between two variables. For example, we want to know whether
building schools increases educational attainment or whether getting a loan
helps with entrepreneurial activities. Regression analysis is the natural tool
for dealing with these questions. Let us consider a simple regression of y on
x. The following equation is called a regression specification and shows the
dependent variable y is regressed on a constant and another variable x. The
final term shows the error term in the regression.
yi=α + βxi + ϵi
The coefficient β shows the increase in y if x is increased by 1 unit. In reality
we do not know the true relationship between the variables and we need
to estimate β. The estimated coefficient is usually indicated by a hat above
the parameter (e.g. β̂ ). The subscript i identifies the unit of analysis (i.e. the
level for which we have collected data on y and x). We might have country
level data, district level data or individual level data. Furthermore, we
might have several observations for one unit of analysis. For example, we
might observe countries over time. This is usually represented by multiple
subscripts.
The key question we have to deal with is how the estimated coefficient is
related to the true (unobserved) parameter of interest. Econometric
methods like regression rely on observed correlations in the data. In this
simple regression, an estimated coefficient is the ratio of the covariance of
22
y and x to the variance of x.

cov ( y, x)
=
var (x)
If, for any reason, the two variables show positive covariance in the
data, the estimated β will be positive. But whether this implies that the
true effect of interest is positive or not is a step that requires further
assumptions. In principle we would like to have an unbiased estimator
of the true parameter. This means drawing several samples from the
population under investigation and calculating β̂ should, on average, give
the true parameter β, i.e. E[β̂] = β. To give some insight into when and
why this desirable property might be satisfied, we start by making the
problem more specific.
2.4.3 Programme evaluation problem

Let us focus on a policy intervention that tries to improve children’s
education by giving cash to poor households. Households above the
national poverty line are ineligible for the programme. Our goal is to
estimate the effect of cash transfer on children’s test scores. Let us denote
YiT to be child i’s test score if the household has received the treatment
(cash). Similarly denote YiC to be child i’s test score when the child’s
household did not receive the treatment (no cash). We would like to
estimate the treatment effect (i.e. YiT – YiC ), but we never observe the same
household, at the same time, both treated and not treated. Therefore,
one of the values in the treatment effect is unobserved. This shows the
impossibility of estimating individual treatment effects. But is there a way
in which we can estimate the average treatment effect? In other words,
can we estimate E[YiT – YiC] using a sample of households in treatment and
control groups?
A natural estimate for the average treatment effect is to compare average
test scores for households receiving the cash to those not receiving it. This
is shown below:
D = E[YiT | cash] – E[YiC | no cash] = E[YiT | T ] – E[YiC | C ] (1)
The term E[YiT | T ] shows the average test score for households in the
treatment group who received the treatment. Similarly E[YiC | C ] shows
the average test scores for households in the control group that did not
receive the treatment. Both of these objects are observable in the data
we collected, therefore we can calculate D. In order to see the relation
between D and the average treatment effect, let us add and subtract E[YiC |
T ] in (1) to get
D = E[YiT | T ] – E[YiC | T ] + E[YiC | T ] – E[YiC | C ] (2)

= E[YiT – YiC | T ] + E[YiC | T ] – E[YiC | C ]
The first term on the right-hand side of (2) shows the average treatment
effect for the population of households who are treated. The rest of the
expression shows selection bias, that is, the difference between average
test scores for households in the treatment group that are not given cash
transfers (E[YiC | T ]) and average test scores for control households that
are not given cash (E[YiC| C ]). Selection bias captures the difference
between average outcomes in the absence of any treatment (and hence
due to other factors). D captures the average treatment effect only when
the selection bias is zero.
The estimator is equivalent to running a regression as follows:
Yi = α + DTi + ϵi
23
where Yi is test scores for household i, and Ti is a dummy variable that is

equal to 1 if the household has received cash and zero otherwise. Ignoring
selection bias would result in a biased estimate of the treatment effect. In
the current example, being poor was used to allocate cash grants. In the
absence of treatment we expect poor children to have lower test scores,
therefore we expect the selection effect to be negative and we therefore
underestimate the effect of cash grants on test scores.
Activity 2.6
A rural bank offers low interest loans to poor households in a village. After one year
the bank runs a local survey and finds poor households which borrowed money have
on average $100 more income compared to poor households that did not borrow from
the bank. It concludes that loans improve the income of the poor and therefore the
government should subsidise credit to the poor.
1. T he data clearly indicate that higher income is positively correlated with borrowing
from the bank, but does it imply that the causal effect of borrowing is higher income?
2. W
hy is it problematic to compare the income of borrowers to non-borrowers? Which
households are more likely to borrow from the bank? How does this affect the
reliability of the estimates obtained above?
► Stop and read

Ravallion (2001) for a discussion of issues in programme evaluation.
2.4.4 Randomised controlled trials (RCTs)

‘Observational data’ usually involves confounding factors that are hard
to control for. Issues like selection could render the results of empirical
studies useless for making policy recommendations. In response to
concerns about causality, the development literature has moved from
cross-country empirical studies and survey-based micro work to within-
country randomised controlled trials (RCTs) in recent years.
In RCTs researchers and policy makers work together to see which policies
work and why. Since RCTs randomly assign individuals to treatment and
control groups, there is no a priori reason to believe the treatment and
control should have different outcomes except because of the treatment
(policy). Therefore by designing careful experiments, researchers could
identify the causal effect of policies. In other words, the selection effect
in equation (2) will be zero because, in the absence of the treatment, we
expect the average outcomes for control and treatment group to be the
same: E[YiC | T ] = (E[YiC| C ]. Therefore, D above will give a valid estimate
of average treatment effect.
There are certain difficulties with running RCTs:
1. RCTs are very expensive and, therefore, are usually implemented at a
small scale (a few hundred villages).
2. Randomisation may not be feasible in some contexts due to ethical
considerations.
3. Implementation of RCTs is not straightforward and failed RCTs might
be even worse than observational data for getting useful insights.
Failure could arise from several issues. For example, there might be
mix-up between treatment and control group due to migration.
4. Behaviour of subjects might be different simply because they know
that they are being observed (Hawthorne and John Henry effects).
24
5. Since experiments are often small, it might be difficult to generalise

the results to populations outside the study area.
6. Experiments might not be able to capture general equilibrium effects
because of their size and local focus.
7. Short-run responses to a treatment might differ from long-run effects.
Experiments often capture the former while from a policy perspective
the long-run effect might be more important.
Early RCTs focused on programme evaluation and were, therefore,
subject to most of the criticisms. However, recent experiments start with
a theoretical framework and try to disentangle the theoretical effects
of interest. If successful in identifying mechanisms for the working of
programme effects, RCTs could hold useful lessons for contexts beyond the
one studied.
► Stop and read

Banerjee and Duflo (2009) and Duflo et al. (2007) for a discussion about randomised
experiments, their merits and potential issues. To learn more about other criticisms of
RCTs, see Deaton (2008, 2010).
2.4.5 Methods using observational data

Before RCTs became popular in development literature, most of the studies
relied on observational data. In order to overcome the problem of selection
bias, observational studies need to justify their identification strategies. There
are several ways that selection bias could be reduced. The first method is to
rely on simple regression (OLS) and include sufficient controls so selection is
fully taken care of. With controls we would not need to assume (E[YiC | T ] =
E[YiC | C]). Instead we assume average outcomes would have been the same
in the absence of the treatment conditional on the controls in the regression
E[YiC | T, Wi ] = E[YiC | C, Wi]. In our example, we include parental education,
initial household income and other relevant variables. This modified
assumption might be easier to satisfy.
A variant of including additional controls is to use fixed effects. Fixed
effects are dummy variables that are included as controls in the simple
regression. For example, in the cash transfer programme outlined above, we
might worry that certain provinces have better test scores because of other
characteristics (e.g. more schools per student). To correct for this we can
include province fixed effects (i.e. one dummy for each province that is equal
to one for the given province and zero otherwise). Province fixed effects
control for observable and unobservable province characteristics that affect
all individuals similarly in a given province and do not change over time.
A second method is to rely on natural experiments. A natural experiment
is a situation where a variable of interest is changing in a seemingly
exogenous manner due to a policy reform, a threshold rule or similar
sharp changes. Natural experiments create plausibly exogenous variations
that might be useful in identifying causal effects. We provide more
discussion of specific empirical methods for identifying causal effects
using observational data in the following chapters: instrumental variables
approach in Chapter 6 (section 6.3.1) and difference-in-differences
estimation in Chapter 7 (section 7.4.1)
► Stop and read

Angrist and Pishke (2010) and Rosenzweig and Wolpin (2000) for a discussion of how
natural experiments could be used in identifying causal effects.
25
2.5 Overview of the chapter

In this chapter we introduced you to the concept of development and
discussed various ways of measuring development. GDP per capita is the
key measure of living standards used in cross-country comparisons. We
discussed the staggering income disparities between countries of the world
and the potential to improve the lives of millions of people.
You should now watch the following TED talks:
1. Hans Rosling shows the evolution of countries of the world in terms
of global health indicators and income. This provides an excellent
overview of development over the past decades. See: www.ted.com/
talks/hans_rosling_shows_the_best_stats_you_ve_ever_seen.html
2. Esther Duflo, a leading scholar in development RCTs, talks about
poverty and the role of RCTs in informing policy at: www.ted.com/
talks/esther_duflo_social_experiments_to_fight_poverty.html
2.6 Reminder of learning outcomes

Having completed this chapter, and the Essential reading and activities,
you should be able to:
• explain and evaluate various measures of economic development
• recall and state key facts about development such as large income
disparities and the staggering number of people who live in poverty
• explain the role of theory and empirics in development and recognise
the difference between correlation and causation.
2.7 Test your knowledge and understanding

1. A rural bank offers low-interest loans to poor households in village
A. One year later it starts the same programme in village B. To assess
the effectiveness of lending as a policy for fighting poverty, the bank
runs a survey of households in both villages at the start of the second
year of operation. The sample includes individuals from village A who
borrowed money in the past year and those who did not. It also includes
the individuals from village B who have applied for a loan but not
yet received it. It also includes households from village B that haven’t
applied for a loan. The following table shows the average income of
borrowers and non-borrowers in villages A and B at the time of the
survey.
Average income ($) Village A Village B

Borrowers 1,000 1,200
Non-borrowers 900 1,000
Discuss whether it is possible to estimate the causal effect of borrowing

on income from this survey. From the variety of estimates you can get
from the table, which one is more reliable? Which group do you think
would form a more valid control group for the borrowers in village A?
Be clear about the assumptions you need to put a causal interpretation
on the estimates.
Reminder: Feedback to activities in this chapter are available

on the VLE.
26
Chapter 3: Factor accumulation
3.1 Introduction
• present the Solow model of economic growth in order to understand
the role of factor accumulation in development
• discuss empirical evaluation of the Solow model and the relevance of
factors of production in explaining income gaps
• introduce development accounting and explain its difference from
growth regressions.

• analyse and solve for the steady state of the Solow model
• recognise and describe key predictions of the Solow model
• critically evaluate empirical tests of the Solow model
• define development accounting and state key conclusions on the role
of factors in explaining income gaps.

Ray (1998) Chapter 3.
Hsieh, C-T and P.J. Klenow, ‘Development accounting’, American Economic
Journal: Macroeconomics 2(1) 2010, pp.207–23.
Mankiw, N.G., D. Romer and D.Weil ‘A contribution to the empirics of economic
growth’, Quarterly Journal of Economics 107(2) 1992, pp.407–37.

Caselli, F. ‘Accounting for cross-country income differences’ in Aghion, P. and
N.D. Steven, Handbook of economic growth. (Oxford; Amsterdam: Elsevier,
2005) Volume 1, Part A, Chapter 9, pp.679–741.
Caselli, F., G. Esquivel and F. Lefort ‘Reopening the convergence debate: a new
look at cross-country growth empirics’, Journal of Economic Growth 1(3)
1996, pp.363–89.
Hsieh, C-T and P. Klenow, ‘Misallocation and manufacturing TFP in China and
3.1.5 Works cited

Banerjee, A.V. and E. Duflo ‘Growth theory through the lens of development
economics’ in Aghion, P. and N.D. Steven Handbook of economic growth.
(Oxford; Amsterdam: Elsevier, 2005) Volume 1, Part A, Chapter 7,
pp.473–552.
Barro, R.J. and J. Lee ‘A new data set of educational attainment in the world,
1950–2010’, Journal of Development Economics 104 (September) 2013,
pp.184–98.
Heston, A., R. Summers and B. Aten ‘Penn World Table Version 7.1’, Center
for International Comparisons of Production, Income and Prices at the
University of Pennsylvania, November 2012.
27
3.1.6 Overview
In Chapter 2 we saw that the income gap between rich and poor countries
is staggering and does not seem to be closing. In this chapter we begin our
quest for understanding why some countries are able to produce so much
output while others struggle to feed themselves.
Firms combine inputs to produce goods and services using a production
function. Adding up all production taking place in the economy yields
aggregate output (e.g. GDP). It might be that poor countries are poor
because of lower levels of production inputs. This story suggests the
way to develop is to encourage accumulation of production factors. Poor
countries should save more and invest more in their capital stock in order
to develop.
In this chapter we consider the Solow model, a theoretical foundation
for assessing the relevance of the factor accumulation view of
underdevelopment (section 3.2). The theoretical analysis lays the
foundation for an empirical assessment of the importance of factor
endowments in explaining observed income gaps between the rich and the
poor countries (section 3.3). Finally, we introduce development accounting
as an alternative approach for looking at the explanatory power of factor
endowments (section 3.4).
3.2 The Solow model

In its original form, the Solow model is a model of capital accumulation.
The model helps us to understand conditions for continuous growth of
economies through the accumulation process. It also shows the limits of
contributions of capital accumulation to the growth process. Although
most of our discussion is concerned with accumulation of capital, the
model is easily extended to incorporate any accumulative factor of
production. This, however, does not change the key insights of the model.
Let us start by thinking more carefully about the production process.
Production takes place by combining inputs in a special way. To produce
vegetables, farmers use land, water, seeds, fertiliser, pesticides, labour,
tractors, etc. Each farmer might use a different technology and combine
inputs in a different way, leading to different levels of production. In
economics the production function models this process. We usually focus
on a few factors of production and make them explicit in the production
function and keep other factors implicit in the form of a productivity
parameter or residual. For example, a production function for vegetable
farming could be as follows:
yi = fi (ki, li)
where yi is the quantity of produce by farmer i, ki is the capital
(machinery) used, and li is labour employed in the production process.
All other factors mentioned above are left implicit in the production
function (fi). If farmer j uses fertilisers and farmer i does not, then they
will have different production functions (fi ≠ fj). This means that, using
the same level of capital and labour, farmer j will have a different level of
production from farmer i.
Economic growth is concerned about aggregate (total) production in
a given country. Therefore we need to sum up the monetary value of
all production to arrive at aggregate output. Similar to the farming
production function, we can use an aggregate production function to
show the link between aggregate output and aggregate levels of input use
(factors). For example:
28
Y = F(K, L)
shows the relationship between aggregate output Y and total capital, K,
and total labour, L as two factors of production. We will naturally assume
that expansion of inputs leads to expansion of output. There are many
other factors of production that are not explicitly listed but influence
aggregate output by changing the shape of the production function F. One
country might have a more efficient judiciary and hence higher security of
investment leading to more production given similar capital and labour in
another country.
Technical note
The assumption required for existence of the aggregate production function is existence
of perfect capital and labour markets. These assumptions might not hold, especially in
developing country context. If you are interested you should read Banerjee and Duflo
(2005).
Activity 3.1
The Cobb–Douglas function is a commonly used production function. A general Cobb–
Douglas production function takes the form of Y = AKαLβ. Let us assume β = 1 – α and
think about a situation where all countries have access to this aggregate production
function. In this activity we want to see how large the gap in capital endowments should
be to explain an income per capita gap of 10 folds (assume the labour force is equal to
the size of the population).
a. Derive the per capita output as a function of per capita capital stock.
b. Assume α = 0.5. How large should the gap in capital per capita stock be to explain a
tenfold gap in output per capita?
Economic growth is the expansion of output over time. This could happen
through the expansion of factors of production (factor accumulation) or
through enhancement of the production function (e.g. technical progress).
The Solow model focuses on the process of capital accumulation and
its contribution to economic growth. In each period a constant fraction
of aggregate income is saved and the rest is consumed. Assuming the
economy is closed (no trade or external capital), saving is equal to
investment. On the other hand, capital depreciates over time (wear and
tear of machinery). Therefore the change in the capital stock can be
modelled as follows:
.
K = sY – δK (1)
.
K shows the time derivative of the capital stock (i.e. change of capital
stock), which is equal to savings minus depreciation of existing capital.
Here s is the saving rate and δ is the depreciation rate.
Since per capita output (income) is the key determinant of individuals’
material wellbeing, we need to rewrite the model in terms of per capita
variables. For the moment let us assume everyone in the economy is
working (i.e. the whole population is the labour force) and that the labour
force is growing at a constant rate of n. Furthermore, assume that the
production function remains constant over time (i.e. we do not have any
technical progress). In what follows the lower case variables show per
worker versions of the original variables. Dividing (1) by L and using the
. . . .
fact that K/L = k + (L/L)k = k +nk , we can derive the per worker version of
(1):
.
k = sy – (δ + n)k (2)
29
This says that the change in capital per worker is equal to savings per
worker (sy) minus depreciation of capital per worker ((δ + n)k). In addition
to the physical depreciation of capital, the addition of new labourers
reduces capital per worker and hence works in a similar way to physical
depreciation. When savings per worker is larger than depreciation of
.
capital per worker, we have capital per worker accumulation ( k > 0 )
.
leading to growth of output per worker ( y > 0 ). If the two terms on the
right-hand side of (2) are equal, capital and output per worker remain
. .
constant ( k = y = 0 ). This situation defines the steady state of the economy.
To move forward with the analysis, we need to impose two further
assumptions on the production function. First, we assume the production
function exhibits constant returns to scale (CRTS). This assumption states
that scaling-up factors of production would increase output by the same
scaling factor. Technically it allows us to write output per worker only as a
function of capital per worker as illustrated below.
Y F(K, L) F K
y= = = ,1 = F(k, 1) = f(k)
L L L
Using this in equation (2) gives equation (3) that determines changes in
capital per worker as a function of capital per worker. This is known as the
Solow equation.
.
(3)
k = sf (k) – (δ + n)k
Activity 3.2
Why is it that no change in capital per worker results in no change in output per worker?
Does the same conclusion go through if the production function improves over time?
(Hint: use the production function.)
Activity 3.3
Does the constant returns to scale assumption make sense at the firm level? To answer
this, think about a factory adding more machines and labour (scaling up). Does it make
sense to assume that output increases proportionately? Why might the CRTS assumption
be more plausible at the aggregate level?
Second, we assume the production function exhibits diminishing returns to

capital and labour. This assumption states that fixing labour, the marginal
product of capital is decreasing in capital. In other words, the additional
output produced from an increase in capital diminishes as the capital stock
increases, while other factors remain constant. This assumption implies the
production function, F(K,L), is concave with respect to capital and labour.
Furthermore, with a little bit of algebra you can see that f (k) is going to be
concave in capital per worker too.
Activity 3.4
Show that the Cobb-Douglas production function, Y = AKαL1-α, exhibits diminishing
returns to labour and capital. Does the per worker production function derived from this
also show diminishing returns to capital per worker? (Assume α ∈ (0, 1).)
3.2.1 Steady state

The steady state of Solow model is when capital per worker does not
change over time. A graphical representation of Solow equation – (3) – is
particularly useful here. In Figure 3.1 the straight line is the depreciation
line and corresponds to the second term in equation (3). The figure also
30
shows the per worker production function, y, and the saving line, sy. Both
of these curves are concave, reflecting diminishing returns assumption.
The change in capital per worker is equal to the difference of savings and
depreciation lines. Point A is the steady state where the two curves
intersect and capital per worker remains unchanged at k*. Starting from an
initial capital per worker equal to k1, investment (saving) is less than
depreciation and capital per worker declines until we return to k*. Starting
from an initial capital per worker equal to k0, investment is larger than
depreciation leading to an increase in capital per worker until we reach k*.
y
y*
(n + δ)k
A } net reduction in k
sy
net addition to k
}
k0 k'0 k* k1 k
Figure 3.1: Solow equation and the steady state level of capital per worker.
At the steady state, capital per worker is constant and therefore from the
production function, output per worker remains constant at y* = f (k*). This
result says that in the absence of technical progress (i.e. the time-invariant
production function) the long-run growth rate of output per worker is zero.
This result critically relies on the diminishing marginal return assumption.
Adding to the capital stock increases output, but at a decreasing rate.
Eventually the addition to capital per worker is just enough to compensate
for depreciation and therefore growth falls to zero. Zero long-run growth
is, however, not consistent with the observed growth rates for several
developed countries. For example, the USA has been growing at an average
rate of 2 per cent in the last century. Later we present an extension of
Solow model that features positive long-run growth.
Activity 3.5
Let us assume the aggregate production function takes the Cobb-Douglas form
Y = AKαL1-α
where A is total factor productivity (TFP) and captures the aggregate efficiency of
combining capital and labour in the economy.
Find the steady state level of capital per worker and output per worker in the Solow
model with this production function.
3.2.2. Comparative statics

Our initial goal was to understand whether factor endowments could
explain the observed income gaps. We learned so far that in the steady
state of the Solow model, output per worker does not grow. To explain
income gaps we consider two possibilities. First, it might be that countries
are away from their steady states and are in transition to reach their
corresponding equilibria. We deal with this possibility later under the
convergence hypothesis. The second possibility is that countries are at
their steady states and observed income gaps are due to steady state
31
differences. In this section we explore this possibility and try to answer

the following question: what parameters could potentially explain large
income gaps?
The Solow diagram (Figure 3.1) could be used to study the impact of the
model parameters on the steady state. Figure 3.2 shows how we can analyse
a rise in the savings rate. An increase in the savings rate from s to s1 shifts up
the saving curve to s1 y. Starting from the old steady state at point A, savings
now exceed depreciation. Therefore capital per worker increases until we
reach point B at the intersection of the new saving curve and depreciation
line. At the new steady state, output per worker is unambiguously higher
because it is increasing in capital per worker and k*1 > k*.
Activity 3.6
Do individuals care about output per se? If we think individuals care about consumption,
could you say whether consumption per worker has increased when the savings
rate increased based on Figure 3.2? What is the savings rate that gives maximum
consumption per worker?
y
y*1
y* (n + ∂)k
B s1y
sy
A
k* k*1 k
Figure 3.2: Impact of a rise in saving rate on steady state output per worker
As another example, consider an increase in the population growth rate.
This makes the depreciation line steeper because now capital per worker
is reduced at a faster rate because it is spread over a faster growing
labour force. At the original steady state, investment is now lower than
depreciation of capital per worker. Therefore the capital per worker
shrinks until we reach the new steady state with smaller capital and
output per worker.
Activity 3.7
Draw the Solow diagram for the increase in population growth rate.
Changes in both savings rate and population growth have level effects.
They change the steady state level of capital and output per worker. They
do not, however, impact on the long-run growth rate of these variables.
Regardless of the parameter values, the long-run growth rate is zero in
the Solow model. Total output, however, grows at the same rate as the
population to keep output per worker constant. Question 2 in Test your
knowledge and understanding at the end of the chapter asks you to
explain changes in response to alterations in the the depreciation rate. You
can try this question before moving on.
32
The comparative static analyses suggest that steady state income gaps
across countries could be due to the parameters of the Solow model.
Poor countries are poor because they have lower saving rates, higher
depreciation, higher population growth, and worse production functions.
We will see below if this is a realistic description of the empirical evidence
but it does not seem that this is a useful conclusion. If low saving rates,
high depreciation and population growth are responsible for low incomes,
then how should developing countries change these? These are the
parameters in the model and are taken to be exogenous. Therefore we
cannot make recommendations about them based on the model.
Activity 3.8
In Activity 3.1 we ignored the fact that economies might be at a steady state determined
by Solow parameters. This activity asks you to revisit conclusions derived there under
steady state conditions. Again assume a Cobb-Douglas production function. In Activity 3.5
you derived the steady state level of capital and output per worker under this assumption.
Now assume that two countries have the same TFP, capital share, depreciation and
population growth rates (A, α, δ, n are equal). What ratio of saving rates could explain
a 10–fold gap in steady state output per worker across the two countries? What is the
implied gap in capital per worker across the two countries? Is this reasonable? (Assume
α = 0.5 if needed.)
► Stop and read

Ray (1998) Chapter 3, pp.47–71 (sections 3.1, 3.2, and 3.3).
3.2.3. Technical progress

The result that, in the absence of technical progress, long-run growth
rate of output per worker is zero, is in contrast to the experience of many
developed countries. For example, US output per capita on average has
grown by 2 per cent each year over the past century. The zero long-
run growth rate could be changed in two ways. First, we can replace
the diminishing marginal returns assumption with a constant returns
assumption. This implies adding more capital always results in the same
proportional increase in output and there is no reason for growth to die
out. In other words, the savings curve is a straight line in this case and
if savings initially exceed depreciation, it will always be higher than
depreciation and growth continues forever. This is what you see in Harrod-
Domar model. Many of the endogenous growth theories also feature some
sort of constant marginal returns to sustain non-zero, long-run growth
rates (we will see these models in Chapter 4).
The second method is to consider a variant of the Solow model where
the production function exogenously expands over time. In other words,
the technology used to produce output from capital and labour improves.
The steps taken to solve the model in the presence of technical progress
are similar to what we did for the original model. At the steady state,
output per worker will grow at the exogenously given rate of technical
progress. This solves the empirically implausible conclusion of zero steady
state growth, but the revised model is silent about how countries could
influence the rate of technical progress, thus limiting its usefulness in
articulating policy.
► Stop and read

Ray (1998) Chapter 3, pp.71–73 (section 3.4 on technical progress).
33
3.2.4. Unconditional vs. conditional convergence

What are the predictions of the Solow model if countries are not at their
steady states? Let us start by looking at the Solow diagram in Figure 3.1
and consider two countries with exactly the same parameter values and
therefore the same steady states. Now consider a case where country 1
starts from an initial capital per worker equal to k0 and country 2 has a
higher initial capital per worker at k'0. Although country 1 is poorer in this
example (smaller output per worker), it has a higher growth rate. This is
because at k0 the gap between the savings curve and the depreciation line
is larger than the gap at k’0 and therefore the change in capital per worker
is larger. Therefore, country 1 will have a faster growth rate compared
to country 2 and eventually it catches up with country 2. Both countries
will converge to the common steady state at point A. This is unconditional
convergence. It states that countries converge to the same steady state
regardless of their initial level of development. Furthermore, poorer
countries grow faster than rich ones, until they catch up.
In the absence of any estimates for the parameter values it is logical to
start from a baseline of equal parameters, but some empirical regularities
go against the unconditional convergence hypothesis. For example, we
have seen in Chapter 2 that some very poor countries remain poor and
have lower growth than the rich countries in the sample.
We now explore the case where different countries might have different
parameter values. Figure 3.3 shows a situation where two countries are
the same except for their saving rates. Country 1 starts from a lower level
of capital per worker at k0 while country 2 starts from k'0 which is higher
than country 1’s initial capital per worker. In contrast to unconditional
convergence, here country 2 has a higher growth rate because the gap
between the savings curve and the depreciation line is larger for country 2.
What matters for the transitional growth rate is the distance of countries
to their own steady states. A poor country with a low steady state might
have slower growth than a rich country which is still far from its steady
state. As the two countries converge to their own steady states, output per
capita growth rate falls to the rate of technical progress. This implies no
convergence in levels of output per worker but convergence in growth rates.
y
y*2
y*1 (n + ∂)k
s 2y
} A
sy
}
k0 k'0 k*1 k*2 k
Figure 3.3: Conditional convergence.

In summary, unconditional (absolute) convergence states that, regardless
of the initial level of output per worker, countries converge to the
same steady state level of output per worker. Conditional convergence,
34
in contrast, states that countries converge to their own steady states

depending on their rates of savings, depreciation, population growth and
technological progress. Countries further from their own steady state
grow faster and in the steady state, countries could have different levels
of output per worker. Assuming a similar rate of technical progress would
result in convergence in long-run growth rates.
Activity 3.9
Using World Development Indicators, plot a scatter of average GDP per capita growth rate
during 1960–2010 versus initial GDP per capita in 1960 for all countries in the sample. Is
there any evidence of unconditional convergence? Now restrict your sample to a) OECD
countries, and b) low income countries. Is there any evidence of convergence among each
set of countries? Explain your results. Note: GDP in PPP $ is available only from 1980
from this source, therefore you should use GDP in constant $ instead.
► Stop and read

Ray (1998) Chapter 2, pp.74–84 (section 3.5 on convergence up to subsection 3.5.6).
3.3 Testing the Solow model

How much of the observed income gaps across countries could be
explained by the accumulation story offered above? Mankiw et al. (1992)
provided the first empirical test of Solow’s predictions. In this section we
review their results.
.
k = sk α – (δ + n + g)k (4)
‹
‹
Before facing the model with the data we need to derive an empirical
specification from the model. Let us start with equation (3) and
assume a Cobb-Douglas production function with labour augmenting
technological progress Y = Kα(AL)1–α. This is called labour augmenting
because technology seems to expand the effective units of labour (or its
productivity). The Solow equation for this model could be written as
follows:
where k is capital per efficiency units of labour (K / AL) and technology
‹
grows at a constant rate of g. Assuming countries are at their steady states

we can derive an expression for output per worker in terms of the model
parameters. First we find .the steady state level of capital per efficiency
unit of labour by setting k = 0 in equation (4):
‹
1
s
k* = ( n + δ + g )1 – α
‹
(5)
Taking logs of the original production function and using (5) in that we
can see the steady state level of output per worker is:
Y α
log = log A +
L 1– α (log s – log(n + δ + g)) (6)
Equation (6) can be written as an empirical specification for the

determinants of output per worker across countries as follows:
Y = a + α log s – α
log log (n + δ + g) + ϵi (7)
L i 0 1– α i
1– α
where i shows we are looking at country i, a0 is the regression constant
derived from the average level of technology, and ϵi is an error term that
reflects random technological differences (not captured by a0). Under the
assumptions of the Solow model, and if countries are at their steady states,
equation (7) suggests output per worker is positively related to the savings
35
rate and negatively related to rates of population growth and depreciation.

Furthermore, the coefficient estimates of the second term and the third
term in equation (7) should be of similar magnitude and opposite sign.
Using early versions of Penn World Tables (PWT), Mankiw et al. (1992)
regress log of GDP per worker in 1985 on log of average investment to
GDP ratio as a proxy for saving rate (s) and log of average population
growth (n) plus 0.05 (they assumed δ + g is constant across countries and
equal to 0.05). Interestingly, the final coefficient estimates of the saving
and depreciation terms are almost equal in magnitude and opposite in
sign (as predicted theoretically). But the implied capital share, α, from the
estimated coefficients is not close to the reasonable estimate of 0.33 except
for the OECD sample.
Activity 3.10
To get causal coefficients on covariates in equation (7), the saving rate and depreciation
term must be exogenous with respect to the error term. Do you think the assumption
of an exogenous saving rate is plausible? What factors could influence the saving rate?
Would these factors also affect GDP per worker? Given these, could you still argue
coefficient estimates in Mankiw et al. (1992) reflect the causal effect of variables on GDP
per worker?
Activity 3.11
An important factor absent in equation (7) is human capital. Does ignoring human capital
in the empirical estimation of the Solow model imply an under- or overestimate of the
influence of savings rate on growth rates?
Mankiw et al. (1992) show that 59 per cent of the cross-country income
differences is explained by the two variables included in the regression.
Although this shows the Solow model has good explanatory powers, it is
not clear what we can learn from this exercise. For example, this analysis
says poor countries are poor partly because they do not have enough
investment. But the critical question is how they could encourage more
investment.
Many factors that are omitted from the analysis above could affect the
saving rate and output per worker. For example, financial development
affects the ease of getting business and consumption credit, which in turn
improves the investment environment and could result in higher saving
and investment. At the same time, better access to credit could enhance
consumption smoothing and hence better occupational decisions by
households that could improve national income. The positive correlation
between the omitted financial development factor, investment rate
and output biases the estimated coefficient of investment rate upward.
Therefore the results of Mankiw et al. (1992) should be seen as simple
correlations without any causality implications, leaving the question of
growth mechanisms unanswered (see Caselli et al. (1996) for a discussion
of omitted variable bias and endogeneity in growth regressions).
In Activity 3.9 you worked with data and looked at convergence. In order to
study convergence in a more systematic way we need to model the dynamics
of output per worker rather than just the steady state relationships as in
equation (6). Furthermore, we need to control for each country’s steady
state. We have seen in equation (6) that the steady state is a function of rates
of saving, depreciation, population growth and technical progress. Once we
control for the steady state, countries further from the steady state should
have higher growth rates. This is the intuition for specification (8) used by
Mankiw et al. (1992) to test conditional convergence.
36
log yTi – log y0i = a0 + a1 log si + a2 log (ni + δ + g) + a3 log y0i + ϵi (8)
The dependent variable in equation (8) is average growth in GDP per
worker over a given period. The first three terms on the right-hand side
of (8) control for the country-specific steady states and a3 is related to
the convergence hypothesis. If countries further from their steady state
have higher growth rates, then we would expect a3 to be negative (zero
or positive coefficient would go against convergence). From the previous
analysis it is clear that a1 is expected to be positive while a2 should have a
negative sign but these are different from coefficients derived in (7).
Mankiw et al. (1992) find that the sign of the coefficients are as predicted
and the implied convergence rate is about 2 per cent. This implies that
it takes about 35 years for a country to bridge half of the distance to its
steady state level of GPD per worker.
Activity 3.12
Do you think we can put a causal interpretation on coefficient estimates from ordinary
least square estimation of equation (8)? Outline a potential endogeneity story based on
omitted variables that results in biased estimates of a3.
► Stop and read

Mankiw et al. (1992) and Ray (1998) Chapter 3, pp.84–90 (subsection 3.5.6 till end of
chapter).
3.4 Development accounting

The analysis we have seen in Mankiw et al. (1992) is an example of
growth regressions. It uses a regression framework to estimate the
contribution of factors of production and other institutional and policy
variables on economic growth. The difficulty of overcoming endogeneity
concerns (e.g. omitted variables, measurement error and reverse
causality) in growth regressions implies these regressions are at best
showing correlations in the data. Growth accounting is an alternative
approach that, instead of estimating parameters, assumes some reasonable
values for parameters and tries to see how much of the growth could be
explained by measured expansions in factors of production.
Development accounting is similar to growth accounting in terms of
techniques used but focuses on a cross-section of countries, whereas
growth accounting considers a single country over time. Development
accounting assesses how much of the variation in output per worker across
countries could be explained by differences in factors of production. It
is an accounting exercise because it is silent about the mechanisms that
could shape factor contributions and builds on calibrations of factor shares
(e.g. capital share, α) as opposed to estimation of such variables from cross
country data.
To get an idea of what development accounting does, let us start by
assuming a Cobb-Douglas production function as follows:
Y = AKα(LH)1–α
where K is physical capital, L is labour, H is human capital and A is total
factor productivity (TFP) and captures the efficiency of inputs’ use.
Rewriting the production function in per worker terms and taking logs
results in
log y = log A + α log k + (1 – α)log h (9)
as usual lower case variables show per worker versions of original ones.
37
Once capital share, α, is calibrated (usually it is assumed to be 0.33) we

can measure everything except TFP. In other words, TFP is backed out as
a residual similar to Solow residual in growth accounting. Caselli (2005)
provides a discussion of how you can measure k and h using existing data
from Penn World Tables (Heston et al., 2012) and educational attainment
datasets (e.g. Barro and Lee, 2013).
The evidence from this exercise suggests that at least 50 per cent of
differences in per capita income cannot be explained by differences
in physical and human capital (Caselli, 2005). In other words, TFP
differences explain more than half of the income gaps. The critical
question is, of course, why TFP varies so dramatically across countries.
Development accounting is silent on this question simply because TFP is
the unexplained residual in this framework.
Activity 3.13
Explain the difference between development accounting and growth regressions. How
could you reconcile the fact that Mankiw et al. (1992) were able to explain 59 per
cent of income differences using physical capital only but Caselli (2005) states physical
and human capital cannot explain more than 50 per cent of the income gaps in a
development accounting exercise?
If we rely on a narrow technological interpretation of TFP it is hard

to see how a single measure of aggregate technology could vary so
much across countries. Poor countries may not have access to the best
available technologies, but many machines are imported and huge gaps in
technology seem unreasonable. TFP, however, is not all about technology.
It includes all other inputs of production, apart from those measured (e.g.
human and physical capital for most studies). Quality of management
practices, protection of property rights, efficiency of labour markets,
regulation of entry, transport infrastructure and openness to trade are only
a few variables from the long list of factors exerting an influence on TFP.
In the upcoming chapters of the guide we will investigate several of these
factors. Here we discuss a recent literature that relies on within-country
heterogeneity of firm-level TFP as an explanation for large gaps in cross-
country aggregate TFPs.
If capital is allocated efficiently across sectors the marginal product of capital
should be equalised. Therefore sectors with higher productivities would
absorb more capital and expand, while sectors with lower productivity
attract little capital. Market failures could prevent reallocation of capital
from low to high productivity sectors. This implies that even if countries
have identical best technologies in each sector, they might still have different
aggregate productivities due to inefficient allocation of capital. To better
understand this concept, let us work through the following activity.
Activity 3.14
Consider country 1 that has two sectors, A and B. There is only one firm in each sector
that uses capital to produce a homogenous output. The production function in sector i is
yi = Ai kαi, where Ai is the sectoral productivity and AA = 2 AB = 2. Furthermore, share of
capital, α, is 1/3.
a. Assume in country 1 there is one unit of capital and 1/9 of it is invested in sector
A and the rest is used in sector B. Find the aggregate productivity parameter in
this economy, assuming the aggregate production function is y = Akα, where
y = yA + yB and k = kA + kB. [hint: find total output by adding each sector’s output
using the sectoral production functions, then use the aggregate production function
to back out A.]
38
b. Consider country 2 with an exactly similar structure to country 1 except for the
sectoral distribution of capital. Here 8/9 of the capital is used in sector A and the rest
is employed in sector B. Find the productivity parameter for this country.
c. What is the ratio of productivities in the two countries? By how much is country 2
richer?
d. Repeat the analysis in c) assuming AA = 4AB = 4.
Another source of inefficiency is within-sector misallocation. Looking at

narrow industries, one would expect more productive firms to expand at
the expense of low productivity firms until the marginal product of capital
(and other factors) is equalised across firms. Presence of market failures
again prevents equalisation of marginal products. For example, Hsieh and
Klenow (2009) show that reducing within-industry productivity dispersion
in China and India to the level observed in the USA would increase
aggregate TFP in the respective countries by 30 and 59 per cent.
► Stop and read

Hsieh and Klenow (2010) and Caselli (2005) pp.679–98 (sections 1, 2 and 3).
3.5 Summary
In this chapter we reviewed the Solow model to assess the role of factor
accumulation in explaining large income gaps between developing and
developed countries. With diminishing marginal returns to capital, steady
state growth would converge to zero unless we have TFP improvements. The
Solow model is, however, silent about factors impacting on TFP.
Growth regressions confirm most of the theoretical predictions of the Solow
model, but endogeneity and omitted variable issues prevent us from putting
a causal interpretation on these results. Development accounting literature
suggests a big part of income differences cannot be explained by differences
in factors of production (physical and human capital). Furthermore, TFP
itself matters for factor accumulation decisions. A higher TFP increases the
returns to physical and human capital accumulation. Therefore, TFP not only
has a direct effect on income gaps but also indirectly affects the income gap
by changing incentives for factor accumulation. Notice that the development
accounting exercise assigns the indirect role of TFP to factor endowments.

• analyse and solve for the steady state of the Solow model
• recognise and describe key predictions of the Solow model
• critically evaluate empirical tests of the Solow model
• define development accounting and state key conclusions on the role
of factors in explaining income gaps.

1. What are the implications of the Solow model for international
aid? Suppose the economy is in steady state, what is the impact of
providing cash aid? What if the economy is away from the steady?
39
2. What is the effect of an increase in the depreciation rate on steady

state capital per worker and output per worker in the Solow model?
How does it impact on total output? Does this change affect the long-
run output per worker growth rate? What about growth rate of total
output?

on the VLE.
40
Chapter 4: Endogenous growth and poverty traps
Chapter 4: Endogenous growth and

poverty traps
4.1 Introduction
• present an introduction to endogenous growth theories
• discuss complementarities and increasing returns as two fundamental
issues in endogenous growth theory and poverty trap models
• introduce models of poverty traps and discuss empirical evidence on
existence of such phenomenon in developing countries
• discuss the relevance of historical events in shaping current outcomes.

• outline core differences between endogenous growth theories and the
Solow model
• define complementarities and relate them to real economic examples
• analyse models of poverty traps and explain the role of different
assumptions in shaping conclusions
• outline the role of history in development.

Ray (1998), Chapter 4 and 5.
Nunn, N. ‘The importance of history for economic development’, Annual Review
of Economics 1(1) 2009, pp.65–92. Available at http://scholar.harvard.edu/
files/nunn/files/nunn_are_2009.pdf

Comin, D., W. Easterly and E. Gong ‘Was the wealth of nations determined
in 1000 BC?’, American Economic Journal: Macroeconomics 2(3) 2010,
pp.65–97.
Davis, D.R. and D.E. Weinstein ‘Bones, bombs, and break points: the geography
of economic activity’, American Economic Review 92(5) 2002, pp.1269–89.
Miguel, E. and G. Roland ‘The long-run impact of bombing Vietnam’, Journal of
Murphy, K., A. Shleifer and R. Vishny ‘Industrialisation and the big push’,
Nunn, N. and L. Wantchekon ‘The slave trade and the origins of mistrust in
Africa’, American Economic Review 101(7) 2011, pp.3221–52.
Redding, S.J., D.M. Sturm and N. Wolf ‘History and industry location: evidence
from German airports’, Review of Economics and Statistics 93(3) 2011,
pp.814–31.
4.1.5 Overview
The unfortunate prediction of the Solow model was that growth fades
away as the economy approaches a steady state. Incorporating technical
progress in the Solow model maintains long-run growth but since the rate
41
EC3044 Economies of development
of technical progress is exogenous, the model does not provide insights

into factors influencing long-run growth.
In the first section of this chapter, we introduce endogenous growth
theories. These models replace the diminishing marginal returns
assumption in Solow with constant marginal returns in one way or another
in order to deliver non-zero long-run growth. With diminishing returns to
capital, additions to the capital stock result in a decreasing contribution
to output. As the capital stock grows, eventually the increase in output
yields just enough investment to offset depreciation, resulting in no
further increase in output. In contrast, in endogenous growth theories the
assumption of constant returns to accumulable inputs (factors like physical
and human capital) ensures a proportional contribution of such factors
to output. Simultaneous accumulation of all such factors would maintain
long-run growth, and output growth is always enough to sustain higher
than depreciation investments in factors, leading to a balanced growth
path in the long run (see section 4.2).
In section 4.3 we introduce the idea of complementarities and argue that
their presence could result in endogenous growth and multiple equilibria.
Section 4.4 introduces the idea of poverty traps. We present an example
of coordination failure and discuss the big push theory, outlining the role
of complementarities and market failures in these models. We also present
empirical evidence on the existence of poverty traps. Section 4.5 discusses
the role of history in shaping economic outcomes.
4.2 Endogenous growth theories

Endogenous growth is in contrast to the assumption of exogenous
technical improvement in the extended Solow model. Here long-run
growth is a result of endogenous factors in the model as opposed to
assumed exogenous technical progress. The rate of long-run growth
therefore is determined inside the model.
Let us start with the simplest possible endogenous growth model, the AK
model. In this model, diminishing returns to capital is replaced by constant
returns and capital accumulation results in sustained growth. In the AK
model, capital is the only factor of production and increases in capital lead
to a proportional increase in output (i.e. capital output ratio is fixed). The
production function takes the following form
y = Ak (1)
where y is output per worker, k is capital per worker, and A is the inverse of
capital output ratio. A constant fraction of output is saved each period and
invested in the capital stock. Natural wear results in depreciation of capital
at a constant rate of δ and the labour force grows at an exogenous rate of
n. Therefore we have the standard accumulation equation as follows:
.
k = sy − (n + δ)k (2)
This model is very similar to Solow and in fact you can see the only
difference is that we have replaced α = 1 in the Solow model with a
Cobb-Douglas production function. The steady state of the AK model is,
however, markedly different from the Solow model. To see this, replace
(1) in (2) to get
.
k = sAk − (n + δ)k (3)
k
k = sA − (n + δ)
(4)
The right hand side of (4) is composed of constant parameters leading to a
fixed growth rate for the capital stock. If sA – (n + δ)>0 capital per worker
42
grows at this rate. There is no transitional dynamics here and the economy
is at its steady state growth rate starting from any initial level of capital
stock.
Activity 4.1
By taking logs from both sides of the production function, show that the growth rate of
output per worker in the AK model is equal to that of capital per worker.
The assumption of constant returns to physical capital is hard to justify in

the AK model. There are good reasons to believe that adding capital when
there is little existing capital should have high returns but increasing the
stock when it is already high should result in a small change in output.
Therefore we need to go beyond this simple model to create endogenous
growth under more reasonable assumptions.
4.2.1 Human capital

So far we have considered labour as a homogenous input that
exogenously expands due to population growth. Individuals, however,
spend considerable resources on education and health. While there is
intrinsic value to health and education, these investments contribute
to the productive abilities of individuals. Ignoring the contribution of
accumulated skills to production could be the reason for the lack of
sustained growth in the Solow model.
Skilled workers have higher productivities because they have better
knowledge and ability to carry out production tasks (e.g. working more
efficiently in teams and with sophisticated machinery). Skilled workers
also increase the marginal product of capital. Therefore additional
investments in capital yield higher returns. Simultaneous accumulation
of human and physical capital could overcome diminishing returns and
maintain non-zero steady state growth.
Many models incorporate human capital accumulation to generate
endogenous growth. We briefly discuss a simple extension of the model
presented in Ray (1998). We consider a model where physical and human
capital enter a Cobb-Douglas production function with constant returns to
scale
Y = AKαH 1−α (5)
where Y is output, K is physical capital, H is human capital, and A is time-
invariant total factor productivity. We still assume that the labour force
grows at an exogenous rate of n. In per worker terms, the production
function is as follows (divide by L)
Y = Akαh1−α (6)
where small case variables show per worker versions of the corresponding
variables. Physical and human capital accumulation are similar to capital
accumulation in Solow as follows:
.
k = sk y − (n + δ)k (7)
.
h = sh y − (n + δ)h (8)
where sk and sh are saving rates for physical and human capital, and δ is
a common depreciation rate. Equation (8) says that the change in human
capital is equal to investment minus depreciation. A constant share of
output is allocated to human capital investment (sh y) while the existing
stock of human capital depreciates due to population growth and loss of
skills.
43
Activity 4.2
Derive equation (8) from the aggregate human capital accumulation equation
.
H = sh Y − δH . This is very similar to what we did in the previous chapter to derive (7).
In steady state, output per worker grows at a constant rate. Looking at the
production function in equation (6), you should observe that this is only
possible if physical and human capital also grow at exactly the same rate.
Therefore physical and human capital output ratios must remain constant
in steady state. To solve for steady state growth rate, divide equation (7)
and (8) by k and h respectively and use the production function, equation
(6), to get:
.
y
k
k
= sk k
− (n + δ) = sk A ( kh )1−α − (n + δ) (9)
.
y
h
h
= sh h
− (n + δ) = sh A ( hk )α − (n + δ) (10)
Given that steady state growth rates for physical and human capital are
identical, we can derive the physical to human capital ratio as follows by
equating (9) and (10):
sk
h
k
= sh (11)
Replacing (11) into (9) and (10) gives the steady state growth rate of the
economy
. . .
y
y = k
k
= h
h
= Asαk s1−α
h − (n + δ) (12)
Assuming the right hand side is positive, we get endogenous growth
that continues forever. Although physical and human capital exhibit
diminishing returns on their own, there are constant returns to physical
and human capital together in this model. Therefore, a balanced
expansion of both forms of capital allows proportional expansion of
output. This is why we get endogenous growth here.
Activity 4.3
In the AK model we argued it is not reasonable to assume constant returns to physical
capital. Is it plausible to assume constant returns to human and physical capital above?
This model shows an interesting feature of endogenous growth models.

Changing parameters like saving rates, depreciation and capital share
have growth effects. For example, from (12) you can see that increasing
the saving rate for physical capital increases steady state growth rate.
Furthermore, the convergence result in the Solow model is overthrown
and initial income gaps are maintained in steady state. Countries with
higher saving rates, lower population growth and lower depreciation
continue to grow faster even in the steady state. This should not be
surprising because, as we discussed in Chapter 2, the convergence result
relies on diminishing return assumption.
The AK model also featured a similar property and, in fact, the two models
are more similar than it appears at first sight. The following activity helps
you discover this point.
Activity 4.4
Rewrite the production function in (6) as follows
1−α
y = A ( kh ) k
Now assume we are in the steady state. Why is it that in steady state the production
function in (6) is equivalent to the AK production function? How could you redefine the
productivity parameter here to have exactly an AK production function?
44
► Stop and read

Ray (1998) Chapter 3, pp.99–107 (sections 4.1, 4.2, and 4.3).
4.2.2 Technical progress

A large number of articles in the endogenous growth literature
deliver sustained long-run growth through purposeful investments in
technological advancement. In these models firms invest in enhancing the
productivity parameter in the production function. If sufficient resources
are spent on technical progress it could compensate for the falling
marginal product of capital and result in non-zero long-run growth. In
other words, these models again target the diminishing return assumption
and replace it with constant returns to technology and capital.
Firms can either invest in developing new blueprints (innovation) or in
adopting existing ones (imitation). Depending on the position of a country
with respect to the world technology frontier, modelling one or the other
might be more relevant. Innovation or imitation could happen along two
dimensions. They could increase the number of available goods or improve
the quality of existing products. The key in all models of technical progress
is that firms choose to spend resources on research and development
(R&D) in the hope of discovering a new variety or an improved product.
While innovation is a costly process, copying from existing ideas is
relatively cheap. With perfect competition, no firm invests in R&D because
once the innovation is available, competitors can easily copy and beat the
innovator on price, leading to negative economic profits for the innovator.
This necessitates a departure from perfect competition. Innovators need
to get positive economic profits from selling their innovations to cover the
upfront cost of R&D. Investors weigh net present value of future profits
against costs of innovation and choose to invest in developing new ideas
until the marginal cost of doing so is equal to the marginal benefit.
Activity 4.5
Explain why under perfect competition and free entry no firm has an incentive to engage
in R&D activities.
Microfounded endogenous growth models usually feature many sectors

and profit maximisation for heterogenous firms. Therefore we are not
going to cover any of them in detail. Once solved, however, they resemble
the simpler models as discussed above. The key features in both models
is constant returns with respect to accumulable inputs. An exercise at
the end of this chapter asks you to solve a model of a product-variety
expansion with the following production function:
Y = N 1−α k α .
Here N is the number of available varieties in the economy and K is
capital. Firms divert part of their output to invest in R&D and develop
new varieties. This will increase the number of available varieties used
in the production. Therefore, capital is spread over more varieties and,
although each variety (sector) could exhibit diminishing returns to capital,
in aggregate there are constant returns to the combination of the number
of varieties and capital.
► Stop and read

Ray (1998) Chapter 4, pp.107–11 (section 4.4 up to 4.4.4).
45
4.3 Externalities, complementarities and increasing

returns to scale
One way to get constant returns is to incorporate externalities in the
production function. When someone’s choice affects the payoff of another
person, we say that the decision has created an externality. If the payoff
is increased, we refer to it as a positive externality and if it is decreased,
it is a negative externality. For example, when an entrepreneur introduces
a new production technique to an industry, other firms active in the
sector will become aware of the technique and receive an information
benefit from an entrepreneur’s action. Having more information, they
can better plan for adoption of new technologies and hence receive a
positive externality. It is called an externality because the entrepreneur
himself would not care about potential benefits to others and therefore the
benefits are external to him.
Activity 4.6
Think about the process of human capital accumulation. What are potential externalities
attached to the decision of investing in one’s education? Are these positive or negative
externalities?
4.3.1 Learning by doing

To understand how the presence of externalities could influence the
growth process let us focus on a learning by doing externality. There are
a number of firms in the economy and each uses the same production
technology with diminishing returns
Yi = AKαi L1−α
i
(13)
where Yi is output produced, and Ki and Li are the amount of capital and
labour employed by firm i. A is a common productivity parameter that,
in this model, we assume to be dependent on aggregate capital in the
economy
A = A0Kη (14)
This assumption creates a positive externality for firms’ capital
accumulation decisions on everyone’s productivity. One firm investing
in its own capital stock delivers a benefit to all existing firms through
an increase of productivity parameter. For example, if more farmers use
tractors and fertiliser, the remaining farmers get a better knowledge of
how to use these inputs (learning from others).
Under perfect competition, it can be shown (see Activity 4.8) that the
aggregate production function takes the following form
Y = AKαL1−α (15)
where Y is total output, K is total capital and L is total labour employed by
firms. Using (14) in (15) results in
Y = A0Kα+ηL1−α (16)
Although we assumed diminishing returns at the micro level, the
aggregate production function exhibits increasing returns to scale.
Increasing returns creates the possibility of sustained and even
accelerating growth which suggests a divergence between developed and
developing countries. Countries with higher capital stock have higher
productivity. This compensates for some of the fall in the marginal product
of capital due to higher capital stock.
46
Activity 4.7
Why does the production function in (16) represent increasing returns to scale?
Activity 4.8
In this activity, you will show that the aggregate production function can be derived from
the firm level production function.
a. Write down the profit maximisation problem for firm i and derive the first order
conditions. You can assume a common real wage of w and a common real capital
rental rate of r.
b. How does the profit-maximising capital-labour ratio differ across firms?
c. (optional) Now write aggregate output as the sum of all output produced by
individual firms, and use the results for capital labour ratios from above to derive a
production function in terms of total capital and labour.
(Hint: Y = ∑iYi and you are asked to find Y as a function of K = ∑i Ki and L = ∑i Li)
► Stop and read

Ray (1998) Chapter 4, pp.112–19 (subsection 4.4.4 up to end of section 4.4).
4.3.2 Complementarities
The positive externality from capital accumulation in the model above is
in fact a complementarity. The capital accumulation decision of each firm
affects the accumulation decision of others. If one firm decides to invest,
everyone will enjoy a higher marginal product of capital and increase
their investments. In other words, capital accumulation by one firm
increases the gains from accumulating more capital for others. This is a
complementarity because not only is the level of utility (profit) affected
but the relative value of alternative options (investment) has also changed.
When a firm invests more it increases the value of the investment for other
firms and therefore they are inclined to choose this option (relative to the
no investment option).
Activity 4.9
Define a complementarity and explain how it differs from an externality.
Complementarities have a ubiquitous presence. Consider the example of

the adoption of high yield variety (HYV) seeds. Farmers need to use the
right amount of fertiliser and other inputs to get the most benefit of HYVs.
Therefore, adopting HYVs involves costly experimentation with inputs.
Once farmers learn the right amount of inputs, HYV are much better than
traditional seeds. If most of the farmers in an area have already adopted
HYVs, neighbouring farmers can see the results of their experimentation
and learn from their mistakes. On the other hand, if no one has chosen
HYVs, farmers need to carry out several rounds of experimentation to
learn the right amount of inputs.
Activity 4.10
Why is the adoption of HYV seeds an example of a complementarity? Try to be specific
about the complementary actions.
Now think about the cost of adopting the traditional crops. Since everyone
is familiar with these crops, farmers know which seeds are suitable for
which plots and they have learned the right level of inputs too. Figure 4.1
shows the cost of adoption of HYVs and traditional seeds as a function of
47
the number of farmers using them. We have drawn the curves so that the
cost of adopting traditional crops is always higher than HYVs crops for a
given number of users. Historically, only traditional crops were available so
many have already chosen them. Since most of the farmers are using the
old varieties, the cost of adopting a traditional crop is lower (point A) than
the cost of learning about HYV (point B). A new farmer, deciding on what
to cultivate next year, will choose to go with the old varieties. This results
in non take-up of HYV (or slow take-up if we change the model to allow
for some heterogeneity among farmers).
Cost of adopting crops
A
traditional crops
HYV crops
HYV farmers traditional farmers
Number of users
Figure 4.1: Cost of adoption in the presence of complementarities.
In effect, the model described here has two equilibria: one in which
everyone is doing traditional agriculture with low levels of production
and the other where everyone has adopted the HYV with a high level of
production. If we start with a situation where there are no traditional
farmers, HYVs are the better option because they involve lower adoption
costs. But starting from an existing pool of traditional farmers (e.g. at
point A), the HYVs are not taken up.
In this framework, history plays a key role in the process of development.
We have path dependency and it matters what initial endowment and
practices countries have. This is in contrast to Solow where we had
convergence, and so the initial level of capital did not matter. The
presence of complementarities also provides a justification for government
intervention. If the government gives a temporary subsidy for adoption
of HYV, then we might move from the old variety equilibrium to the HYV
equilibrium as you will see in Activity 4.11.
Activity 4.11
How could a temporary subsidy for adoption of HYV change the equilibrium of the
economy? First think about the impact of the subsidy on Figure 4.1 curves, then try to
argue why a temporary subsidy is enough and we do not need permanent support.
► Stop and read

Ray (1998) Chapter 5, pp.131–47 (sections 5.1 and 5.2).
4.3.3 Increasing returns

Increasing returns to scale (IRS) is a situation where scaling up all
production inputs results in a more than proportionate increase in output.
The presence of IRS results in decreasing marginal and average costs
of production. Therefore, firms with higher levels of production have a
cost advantage over new entrants and smaller firms. Further expansion
48
of output and stealing the market from competitors is profitable because

it leads to greater utilisation of IRS and reduction of average costs. This
characterises the case of a natural monopoly. Other entrants are deterred
because they cannot undercut the monopoly’s price and still survive.
Activity 4.12
Explain why the presence of IRS in an industry is not compatible with competitive
markets.
► Stop and read

4.4 Poverty traps

Poverty traps are an undesirable equilibrium situation. They are traps
because they are difficult to escape from. Individuals without any assets
or skills could be in a poverty trap because they earn low incomes and
this income is insufficient to be invested in the acquisition of productivity
enhancing skills. Countries with poor institutions might be in a poverty
trap because they cannot produce enough output to fix their institutions.
Institutions are, however, a critical determinant of aggregate productivity.
Maintaining poor institutions keeps the country’s income down.
4.4.1 Coordination failure

Coordination failure could result in poverty traps. Consider an example
with two firms using an old technology to produce two goods. Firms can
choose a new technology by paying an upfront cost of F (e.g. training
workers to operate the new machines). Furthermore, investment in new
technology creates a positive externality on the other firm (as in the
learning model you saw earlier). Table 4.1 shows the payoff structure
for these firms under the two scenarios of investing or not investing in
the new technology. When both continue with the old technology and
do not invest, they both get zero profits (this is a normalisation). When
firm 1 invests, it pays the upfront cost of F. If firm 2 does not adopt
the technology, firm 1’s payoff is σ – F but if firm 2 also invests in new
technology, the externality results in an additional gain and increases firm
1’s payoff to π – F. On the other hand, if firm 1 decides not to invest in
new technology, but firm 2 does, firm 1 receives a payoff of 1. We assume
σ – F < 0 and π – F > 1, so investment is an inferior decision if the business
partner does not invest and it is superior if the partner decides to invest.
Firm 2
Invest Do not invest
Invest (π – F, π – F) (σ – F, 1)
Firm 1
Do not invest (1, σ – F) (0,0)
Table 4.1: Payoffs under the two scenarios of investing or not investing.
Activity 4.13
Explain why the externality in Table 4.1 is a complementarity too.
This simple game has two pure strategy Nash equilibria. There is a good
equilibrium where both firms invest in the new technology. Conditional
on firm 2 investing in the new technology, firm 1 prefers to invest as well,
because π – F > 1. Symmetrically, conditional on firm 1 investing in the
new technology, firm 2 prefers to invest too because π – F > 1. The other
49
equilibrium is where no one invests in the new technology. When firm 2

does not invest then firm 1 prefers to withdraw from investment because σ
– F < 0 and vice versa. Therefore either both firms invest or both withdraw
from investment. This shows the classic big push argument, although
industrialisation by one firm is not viable (σ – F < 0); when both firms
industrialise they can get a higher profit and generate more income (see
next subsection).
Expectations play a key role in this classic example of coordination failure.
If each firm expects the other to invest in adopting the new technology it
will invest itself. But if expectations are for some reason set at non take-up
of new technology the economy remains at the low equilibrium.
4.4.2 Big push

The example of the shoe factory from Ray (1998, p.137) is quite telling
in motivating the big push idea. An economy that has only one big shoe
factory serving the domestic market cannot be viable, because individuals
working in the factory purchase everything, not just shoes, with their
incomes. There will not be enough demand for all the shoes produced. In
contrast, simultaneous expansion of several sectors could be viable, even
when they all cater for domestic demand. Some individuals work for the
shoe factory, others work to produce garments, etc. and with their incomes
they would also purchase the output of these factories.
Activity 4.14
Explain how the parable of the shoe factory fits the coordination failure game represented
in Table 4.1.
Murphy, et al. (1989) offer a formal model for the parable of the shoe
factory. Consider an economy where firms can use an old cottage
production method or an industrialised new technology with increasing
returns to scale. Sectors using the cottage production method are
competitive. Industrialisation involves fixed costs, paid only if subsequent
profits are sufficiently high to cover upfront costs. Once a sector is
industrialised, all production is done by the single monopolist using
the advanced technology (due to increasing returns discussed earlier).
Workers in the industrialised sectors receive a higher wage than the
workers in cottage production. Therefore, there is positive externality
of industrialisation. Once a sector is industrialised, workers receive
higher wage and therefore would increase demand for all products thus
increasing the profitability of other sectors. In fact, this is a case of a
complementarity, because industrialisation in one sector increases returns
to industrialisation in other sectors.
This model features mutliple equilibria. An equilibrium where none of the
firms decides to industrialise and everyone uses the cottage method leads
to low incomes, and a second equilibrium where all sectors industrialise
and incomes increase. As Murphy, et al. (1989, p.1004) put it,
simultaneous industrialization of many sectors can be self-
sustaining even if no sector could break even industrialising
alone.
There are two features of this model worth emphasising. First, the size
of the market matters greatly for the industrialisation decision. In the
absence of trade, producers cater for the domestic market, the size of
which depends on population and income.
50
Activity 4.15
Explain how the presence of a large export market could remove the cottage equilibrium
in the big push model.
The second feature of the model, which leads to a multiplicity of

equilibria, is the presence of market failures in the labour and capital
markets. The model assumes wages are higher in the industrialised sector.
This suggests the presence of market failure which prevents equalisation
of wages potentially due to efficiency wages or skills. In order to have the
two equilibria we also need imperfect capital markets. If capital markets
are perfect then one firm could borrow enough money to industrialise all
sectors at the same time to solve the coordination problem.
Activity 4.16
What are policy implications of the big push theory? How could the government intervene
to promote industrialisation?
4.4.3 Empirical evidence on poverty traps

We have studied several ideas that could lead to multiple equilibria and
the presence of poverty traps. But, is there any empirical evidence that
supports this view of development? Testing for poverty traps is hard,
because it is not clear how we can empirically see if a country has moved
from a bad equilibrium to a good one. Empirically it is difficult to separate
transitional dynamics from a shift in steady states.
Miguel and Roland (2011) compare the heavily bombed districts of
Vietnam during the Vietnam War to those that were not destroyed as
much. The presence of regional poverty traps could leave the heavily
bombed areas in a poverty trap because most of the capital stock was
destroyed in these areas. They test for local poverty traps by looking at
various outcomes but find no significant difference between local poverty
rates, consumption levels, infrastructure, literacy and population density.
Therefore, this study points to the absence of local poverty traps and
confirms convergence in the long run. The empirical setting, however, does
not allow identification of countrywide poverty traps (e.g. if Vietnam as a
whole has fallen into a poverty trap due to US bombing).
Activity 4.17
Explain why inter-regional trade in Vietnam could eliminate local poverty traps.
In another study, Davis and Weinstein (2002) find that Allied bombing of
Japanese cities during the Second World War did not have a significant
impact on long-run population trends. Therefore, they find evidence in
support of unique steady states for city size rather than multiple steady
states.
Redding et al. (2007) provide evidence in support of multiple equilibria
and show the temporary division of Germany resulted in a permanent shift
of German’s air traffic from Berlin to Frankfurt. Before division, Berlin
airport was the primary hub, but after division, Frankfurt adopted this
role. The reunification of the Germany did not reverse this change. This
suggests strong path dependency where, for historical reasons, the airport
hub was moved to another location and remained there.
51
Activity 4.18
Why might the case of Frankfurt airport not be very useful in providing evidence for the
existence of multiple equilibria in a cross country setting?
4.5 The role of history

History could influence current economic outcomes through various
channels. In the presence of multiple equilibria, historical events could
shift the path of development from one equilibrium to the other, resulting
in a long-lasting impact on economic outcomes. Alternatively, historical
events could change fundamental factors influencing economic growth.
For example, historical events could shape local institutions. Institutions,
however, are persistent and therefore link historical events to current
outcomes. In Chapter 6 we will see examples of studies in this area. Here
we touch the surface of a growing empirical literature that looks at the
impact of history.
History could shape social norms and expectations. For example, Nunn
and Wantchekon (2011) use the Afrobarometer survey of 2005 and link
today’s trust levels to historical slave trade in Africa. They show that
individuals whose ethnic group was exposed to more slave trade between
1400 and 1900 have less trust in their neighbours, relatives and local
government. The level of trust affects the extent of cooperation and
teamwork in society and hence could influence development.
Historical levels of knowledge and technology could affect current
economic outcomes. For example, if higher levels of technology reduce
the cost of adopting new technology, then countries with historically
higher levels of technology would be able to grow at a faster rate. Comin
et al. (2010) find that technology levels in 1000BC, 0AD, 1500AD and
GDP per capita levels are positively correlated.
► Stop and read

Nunn (2009) and Ray (1998) Chapter 5, pp.155–61 (section 5.5 to the end).
4.6 Summary
Endogenous growth theories resolve the inability of the Solow model
to generate sustained economic growth. The crucial difference between
the two paradigms is the assumption of diminishing returns to factors
of production. In Solow, consecutive increases in capital stock result in
decreasing additions to output. In the steady state, this additional output
is just enough to offset depreciation of capital. Therefore, the net addition
to capital is zero and the growth process halts. But, in endogenous growth
models, there are constant returns to accumulable factors of production.
Simultaneous investments in these factors result in additional output that
is sufficiently high to continue net increases in factors of production. This
virtuous circle yields endogenous growth.
Another paradigm that could explain wide income gaps across countries
is based on the existence of multiple equilibria. In presence of market
failures, complementarities could result in multiple equilibria. Countries
might be trapped in a bad equilibrium due to their inability to coordinate
their way out of the situation. Although models of poverty traps are
intuitive, empirical evidence is inconclusive on their existence.
52

• outline core differences between endogenous growth theories and the
Solow model
• define complementarities and relate them to real economic examples
• analyse models of poverty traps and explain the role of different
assumptions in shaping conclusions
• outline the role of history in development.

1. A toy model of product-variety expansion. Consider the following
production function
Y = N 1−αKα
where output depends on the number of available varieties and the
capital stock. sn fraction of output is invested in R&D and results
in a flow of new varieties proportional to the amount invested.
Furthermore, δ fraction of existing varieties become obsolete and
is not produced. Therefore the change in the number of varieties
.
over time is N = µsnY − δN , where μ is the yield of R&D investment.
Capital accumulation takes the Solow form with a saving rate of sk
and depreciation rate of δ. The population is assumed to be constant.
Find the steady state growth rate of output, capital and the number
of varieties. How is this model different from the human capital
investment studied in the chapter?
2. Plot a diagram for the capital accumulation equation in the AK model
and show why growth continues forever if sA – (n + δ) > 0.
3. Recognising the difficulties of HYV adoption, the government of India
decides to show how HYVs outperform traditional varieties by planting
them on a set of randomly chosen plots next to traditional varieties.
Furthermore, it hires skilled farmers to find out the right amount of
fertiliser and other inputs for HYVs. Do you expect this policy to be
effective in encouraging farmers to take up HYVs? What are the merits
of using model farms as opposed to providing a monetary subsidy for
HYV farmers?
4. Explain the idea of the big push. Why might industrialisation by one
sector not be viable but simultaneous industrialisation of several
sectors prove to be viable?
Reminder: Feedback to activities in this chapter are available on

the VLE.
53
Notes
54
Chapter 5: Inequality and growth
5.1 Introduction
• discuss the measurement of inequality
• understand the relationship between inequality and growth
• explain empirical issues in estimating the role of inequality in
development.

• explain the axiomatic approach to measuring inequality
• recognise and explain various aspects of the two-way relationship
between inequality and growth
• summarise and evaluate empirical designs for identifying causal effect
of inequality on income.

Ray (1998) Chapters 6 and 7.
Easterly, W. ‘Inequality does cause under development: insights from a new
instrument’, Journal of Development Economics 84(2) 2007, pp.755–76.
Galor, O. and J. Zeira ‘Income distribution and macroeconomics’, Review of
Economic Studies 60(1) 1993, pp.35–52.

Banerjee, A.V. and E. Duflo ‘Inequality and growth: what can the data say?’,
Journal of Economic Growth 8(3) 2003, pp.267–99.
Banerjee, A.V. and A.F. Newman ‘Occupational choice and the process of
development’, Journal of political economy 101(2) 1993, pp.274–98.
Ray, D. ‘Uneven growth: a framework for research in development economics’,
5.1.5 Works cited

Deininger, K. and L. Squire ‘A new data set measuring income inequality’, World
Bank Economic Review 10(3) 1996, pp.565–91.
Deininger, K. and L. Squire ‘New ways of looking at old issues: inequality and
growth’, Journal of Development Economics 57(2) 1998, pp.259–87.
5.1.6 Overview
In the previous chapters we had a focus on aggregate incomes. In reality,
different sectors and individuals could have heterogeneous experiences of
the growth process. Some might benefit less, others might benefit more. In
this chapter we discuss the issue of inequality and its relation to growth.
The issue of inequality warrants attention on two grounds. First, we care
about inequality from a moral perspective. Thus, targeting inequality as a
goal of intrinsic value could be a justified action. The second reason why
we study inequality is in relation to growth. There are many theoretical
reasons for back and forth links between inequality and growth (functional
role of inequality). For example, inequality might provoke destructive
55
social unrest and damage growth. Growth, on the other hand, could create
a source of taxation for redistributive policies to reduce inequality.
In this chapter, we first consider the measurement of inequality and
discuss various definitions. We then review theoretical links between
growth and inequality and critically evaluate one empirical article in this
area to better understand the challenges of identifying the causal effect.
5.2 Definition
While, at the broadest level, we might be interested in inequality in life
satisfaction, including disparities in freedoms, capabilities, and resource
enjoyment, we restrict our analysis to economic inequality in this
chapter. Despite its narrow focus, economic inequality is related to other
inequalities, but surely does not capture all differences.
There are various ways of defining economic inequality. We can look at
income inequality, wealth inequality, lifetime income inequality, etc. Each
of these measures has a different scope. For example, income inequality
shows differences in the flow of resources in the short run. It is subject
to volatilities and temporary losses and gains. On the other hand, wealth
inequality captures disparities in the stock of resources. It reflects the long-
run relative position of individuals. Although more fundamental, wealth
inequality is harder to measure. Therefore, most of the cross-country
literature focuses on income inequality.
The level of investigation also matters greatly. We can define inequality at
the level of a country, a region, a village or even a household. At each level
we try to measure the equality of resource distribution between individual
members and understand the causes and implications of such disparities.
In this chapter our focus is on country-level inequality (i.e. we look at
disparities between households in a given country).
Activity 5.1
Explain the differences between income and wealth inequality. Which one is likely to be
more important?
► Stop and read

5.3 Measurement
Let us say we want to measure whether distribution of income is more
unequal in one country relative to the other, or whether inequality has
reduced over time in a given country. Visual comparison of income
distributions is arbitrary. We need to focus on a more quantifiable measure
for cross- and within-country comparisons. We can rely on moments of
distribution that reflect the degree of income disparities but here we have
many possibilities. We can use the income gap between the worst-off
and most well-off individuals, dispersion of distribution, share of income
accrued to quintiles of distribution, etc. We rely on an axiomatic approach
to restrict ourselves to appropriate measures of inequality.
5.3.1 Axiomatic approach

The inequality literature postulates four axioms that any measure of
inequality should satisfy. These are deemed as reasonable restrictions.
The first axiom is the anonymity principle. It states that the identity of
individuals should not impact on inequality comparisons. For example, in a
56
country with three individuals and an income distribution of (y1, y2, y3) our
measure of inequality should assign the same degree of inequality for (10,
1, 100) and (100, 1, 10) income distributions.
The second axiom is the population principle, where a proportionate
increase in population size should not affect income inequality.
Considering the example above (10,1,100) is as unequal as a society with 6
individuals and an income distribution of (10,10,1,1,100,100).
The third axiom is the relative income principle. It states that only
relative incomes matter for inequality comparisons. Therefore scaling
all individuals’ income by a constant (e.g. balanced growth) should not
change the level of inequality. Therefore, (10, 1, 100) and (20, 2, 200) have
the same level of inequality because relative income is unchanged, so
10 20 , 100 200 .
= =
1 2 1 2
The fourth axiom is the Dalton principle. A distribution derived from
another only by taking away income from the poor and giving it to the rich
is more unequal. For example, the measure of inequality satisfying this
axiom should imply a higher inequality for (9,1,101) compared to (10,1,100)
because the latter yields the former distribution through a redistribution of
income from the poor with 10 unit of income to the rich individual.
Activity 5.2
Show measures of inequality that satisfy the four principles (assign lower inequality to
(5,5,3,3,2,2) income distribution in relation to (1,3,6) distribution).
► Stop and read

Ray (1998) Chapter 6, pp.173–78 (section 6.3 up to subsection 6.3.3).
5.3.2 Measures of inequality

The first three axioms allow us to look at the share of income that goes
to different shares of the population. The Lorenz curve provides a visual
representation of inequality that incorporates this feature. Sorting
individuals from poorest to richest, we calculate the share of income
accrued to various shares of the population. For example, Figure 5.1
shows the Lorenz curves for China, India and Brazil, using data from
World Development Indicators (Deininger and Squire, 1996, were the
first to provide consistent inequality measures). In Brazil, the 20 per cent
poorest individuals only get 3 per cent of national income, the 40 per cent
poorest get 10 per cent, and so on (notice that population and income
share are plotted in a cumulative way in the Lorenz curve). The figure
also shows the equality line. If everyone had the same income in a society
(egalitarian), then income shares would be identical to the population
shares resulting in the 45 degree line (e.g. 20 per cent of population would
get 20 per cent of the total income).
The Lorenz criterion states that one distribution is more unequal than
the other if its Lorenz curve is completely below the other distribution.
Although this criterion satisfies all the axioms listed above, it fails to
compare two distributions with crossing Lorenz curves. Furthermore,
we might be interested in summarising the level of inequality in a
single number rather than a graph, probably to be used as a yardstick
for measuring improvements and in regression analysis. Therefore we
consider several complete measures of inequality (i.e. a single number) in
what follows.
57
100
China 2009 India 2010
× Brazil 2009 equality
Cumulative income share 80
60
40 × 41
20 × 22
× 10
0 × × 3
0 20 40 60 80 100
Cumulative population share
Figure 5.1: Lorenz curves for China, India and Brazil.

Source: World Development Indicators, The World Bank
Activity 5.3
Which country has the highest level of inequality in Figure 5.1? Why?
The GINI coefficient is an inequality measure that could be calculated

from the Lorenz curve. It is the ratio of the area between the equality line
and the Lorenz curve to the area of the triangle under the equality line.
Under complete equality this ratio would go to zero and under extreme
inequality it approaches 1. Therefore we consider higher GINI coefficients
as indication of higher levels of inequality.
Activity 5.4
Explain how you would calculate the GINI coefficient for the countries in Figure 5.1. Does
the GINI coefficient deliver the same ranking of inequality in these countries as the Lorenz
criterion?
A set of commonly used inequality measures are Kuznets ratios. These are
defined as the ratio of the share of income of x per cent richest individuals
to the share of income for y per cent poorest individuals. We can calculate
these from the Lorenz curve too. For example, the ratio of share of income
going to the 20 per cent richest to that of the 20 per cent poorest in Brazil
is about 20 (1-0.41 divided by 0.03), meaning the 20 per cent richest earn
20 times more than 20 per cent poorest on average.
Activity 5.5
What is the ratio of share of income for the 40 per cent richest to the 40 per cent poorest
in Brazil based on Figure 5.1?
Summarising income distribution in one number is not completely

costless. Most importantly, different measures of inequality might yield
contradicting results. For example, GINI coefficient might show an
improvement while a Kuznets ratio shows a deterioration of inequality
over time or across countries. Therefore researchers often look at several
measures of inequality to make sure their analysis is robust to the
measures used.
58
► Stop and read

Ray (1998) Chapter 6, pp.178–93 (from subsection 6.3.3 till the end of chapter).
5.4 Inequality and growth

There is a complex relationship between inequality and growth. An array
of theoretical models investigate various potential links from inequality
to growth and vice versa. Our empirical knowledge of the relationship,
however, lags behind specifically because of the confounding effects
arising from the multitude of potential links. In this section we consider
some of the theoretical links and explain the difficulty of empirical work
by discussing examples of empirical studies. We start by looking at a
historically important hypothesis: the inverted U relationship between
inequality and income. Then we discuss a broader set of mechanisms
before reviewing an empirical article. In the final part we consider an
equilibrium model to see the evolution of incomes and inequality from an
initial wealth distribution.
5.4.1 Inverted U hypothesis

Kuznets first suggested there might be an inverse U relationship between
inequality and income. As its name suggests, the inverted U hypothesis
postulates a non-linear relationship between income and inequality. It
argues that the process of development is initially accompanied by rising
inequality but as the economy grows compensatory factors start to reduce
inequality. Therefore, plotting a measure of inequality on the y-axis and
GDP per capita on the x-axis, we should see that at low levels of income,
inequality rises with increases in income but after a certain level of income
the relation is reversed and higher income reduces inequality.
One potential story that could provide specific mechanisms leading to an
inverse U relationship is as follows. Think about agriculture and industry
as the two main activities in the economy. At the beginning most of the
population is working in agriculture earning similar incomes, resulting
in low inequality. The introduction of industry increases the earnings
of industrial workers but those living off agriculture still have the same
incomes. This implies a rise in inequality as average incomes go up.
However, as the industrial sector booms, compensatory factors kick in. For
example, higher income results in higher demand for everything, including
agriculture, making agriculture more profitable. The machinery produced
in industry could improve agricultural productivity thus improving
farmers’ living standards and reducing inequality.
The inverted U hypothesis has a clear prediction for the patterns in the
data and historically received considerable attention. In principle we can
take two approaches for testing the inverted U hypothesis. We can focus
on a given country and follow the evolution of inequality as the country
develops (time series). The problem here is lack of historical (inequality)
data on the countries that started poor and ended up rich. The second
approach is to look at a sample of countries at a point in time (cross
section). If the underlying relationship between income and inequality
is similar across countries we would expect rich countries to reflect what
would happen to the poor countries when they develop. Therefore, a cross
section would contain countries at different stages of development and
would allow for a test of the inverted U hypothesis.
In reality the factors influencing inequality might be different across
countries and our ability to control for all these factors is limited.
Measurement error and validity of cross-country comparisons of inequality
59
measurement add to the puzzle. The cross-country literature was, however,

very active during the 1990s and 2000s. The introduction of better quality
GINI coefficients and share of quintiles of income distribution by Deininger
and Squire (1996) gave a further boost to the cross-country literature.
Using this more accurate data, Deininger and Squire (1998) conclude that
the inverted U disappears once you control for other factors.
Activity 5.6
Using World Development Indicators dataset plot GINI coefficient versus GDP per capita
in 2000 for countries of the world. Is there an inverse U relation between the GINI and
GDP per capita?
► Stop and read

Ray (1998) Chapter 7, pp.197–211 (sections 7.1–7.2.4).
5.4.2 Does inequality cause underdevelopment?

We consider two arguments, among many, on why high inequality might
promote growth. First, low income individuals usually consume most
of their incomes (low saving rates), while high income individuals have
fulfilled their basic needs and save a larger fraction of their incomes. An
unequal society with a few wealthy individuals might have higher savings,
and hence higher growth compared to a more egalitarian economy where
everyone is poor. Second, starting a business usually involves paying for
upfront costs (capital investments, renting premises, initial advertisement,
etc.). Concentration of resources allows the well-endowed entrepreneurs
to invest in starting new businesses that promote economic growth.
To do justice to this topic, let us review two out of many theoretical
arguments that argue inequality harms growth. First, high inequality
raises demand for redistribution. This requires a progressive tax system
to fund redistributive policies. Taxes, however, diminish investment
incentives because they reduce marginal returns, lowering growth as a
result. Second, lack of resources could prevent the poor from investing in
human capital. While the rich might have started businesses, unavailability
of skilled workers keeps labour productivity at a low level (see Galor
and Zeira, 1993, model below). Therefore, theoretically it is not clear if
inequality is a cause or hindrance for development.
► Stop and read

Ray (1998) Chapter 7, pp.211–20 (subsections 7.2.4–7.2.6).
Identifying the causal effect of inequality on growth is a challenging

empirical problem. Household surveys are often the source of inequality
measurement. Even if we can settle on a measure of inequality out of
the ones discussed above, countries have different methodologies for
running the surveys, leading to the issue of cross country comparability.
Furthermore, there are several omitted variables that could be related
to inequality and growth. For example, low state capacity could result in
the absence of free public education and hence higher inequality as the
wealthy educate their children but the poor cannot. At the same time,
low state capacity diminishes rule of law and protection of property rights
with negative consequences for economic growth. In this scenario, we see
a significant correlation between inequality and growth, not because one
causes the other, but because the omitted state capacity variable affects both
inequality and growth. Finally, not only does inequality impact on growth
60
but growth also affects inequality (reverse causality). Higher growth could
increase the available resources for redistribution and lower inequality.
Measurement error, omitted variables and reverse causality all result in
endogeneity of the inequality measure in the regression. Therefore the
OLS-based coefficient estimate of the regression will be biased. Easterly
(2007) adopts an instrumental variable approach to deal with the
endogeneity of inequality (see Chapter 6 for a discussion of instrumental
variables). He uses suitability of land for sugar relative to wheat as an
instrument for inequality. The idea is based on the agronomy and historical
features of wheat and sugar plantations. Historically, sugar plantations
took up a lot of land and usually used slave labour. Landowners here had
an incentive to keep labourers poor (see evidence in Easterly, 2007) and
therefore large sugar plantations resulted in high levels of inequality. On
the other hand, wheat production was historically done by small and
medium landowners leading to a more equal society.
Activity 5.7
How does Easterly show that the relevance condition is satisfied? In other words,
what are the theoretical and empirical grounds for a positive correlation between
land suitability for sugar and inequality? (You may answer this after you have studied
instrumental variables and the Easterly article.)
While the historical evidence supports the relevance condition, we also

need to check the exclusion restriction for the validity of instrument. The
exclusion restriction states that sugar versus wheat suitability should not
impact on growth directly or indirectly through any channel other than
inequality. For example, sugar wheat suitability is highly correlated with
the share of land in the tropics, which could have an indirect effect on
income other than through inequality (e.g. through settler mortality as
you will see in Chapter 6). Easterly (2007) tries to include a set of controls
(like a dummy for being in the tropics) and argue for the plausibility of the
exclusion restriction.
Table 4 in Easterly (2007) shows the estimation results. Interestingly,
using the instrumental variables (IV) approach strengthens the significant
negative OLS coefficient for the effect of inequality on income. Assuming
the exclusion restriction is satisfied, we can interpret the IV estimates as
the causal effect of inequality on growth. Therefore, it seems the evidence
here supports a negative impact of inequality on growth.
Activity 5.8
Instrumental variable estimation corrects for the endogeneity issues in the OLS
estimation. Comparing columns (1) and (2) in Table 4 of Easterly (2007), could you
discuss whether the endogeneity has resulted in an overall upward or downward bias in
the OLS estimation?
► Stop and read

Ray (1998) Chapter 7, pp.220–3 (subsection 7.2.6) and Easterly (2007). Further readings:
Banerjee and Duflo (2003) and Ray (2010).
5.4.3 Skill acquisition, inequality, and growth

In this section we discuss a more formal model that links inequality to
development. The model helps us to clarify assumptions for a positive or
negative impact of initial inequality on development in a stylised setting.
The model is from Galor and Zeira (1993) in your Essential reading list.
61
Consider an economy producing a single homogenous good. Individuals

live for two periods: receiving bequests from parents and leaving bequests
for their children. Individuals are identical in all aspects except in terms
of the bequests they receive from parents. In the first period, individuals
can invest in education which requires a lump sum payment of h. In
return skilled individuals receive a higher wage, ws, in the second period.
Alternatively, individuals can choose to work as unskilled labourers in both
periods receiving wn. Individuals consume their earnings at the end of the
second period and leave bequests for their children.
Individuals can either use parental bequests, x, or borrow money from
capital markets to invest in education. Imperfections in capital markets
result in a borrowing interest rate, i, which is higher than the lending
interest rate, r.
Activity 5.9
What are the potential factors that generate a wedge between lending and borrowing
interest rates?
For simplicity, let us assume individuals leave a constant fraction, α, of their

wealth as bequests and consume the rest (this could be the outcome of
utility maximisation). There are three possibilities. First, individuals decide
to work as unskilled labourers. In this case, they save (lend out) parental
bequests and at the end of the second period bequeath bn for their children
bn (x) = α ((1 + r)(x + wn) + wn) (1)
This shows α fraction of end of lifetime wealth is left as bequests. The first
term in parenthesis shows the sum of the bequests received and unskilled
wage in the first period which is multiplied by (1 + r) because individuals
lend this money on the market in the first period. The second term shows
the unskilled wage in the second period.
The second and third cases are individuals who decide to invest in human
capital and become skilled workers. The second group funds h (cost of
education) from parental bequests because they have enough money (x > h)
and leaves bequests as follows
bs (x) = α ((1 + r) (x – h) + ws) (2)
The first term shows part of parental bequests is invested in education

and the rest is saved in the capital markets. The second term is the skilled
wage which is received in the second period.
The third group does not have enough funds and borrows money to pay
for educational investment (x < h). For this group the bequest left for
children is as follows
bs (x) = α ((1 + i) (x – h) + ws) (3)
There are two differences between (3) and (2). The lending interest rate
is smaller than the borrowing interest rate (i > r) and x–h is negative in (3)
while it is positive in (2). In other words, the third group borrows money
and has to repay the loan in the second period which subtracts from total
wealth.
Individuals choose to invest in education to maximise total wealth (and
hence consumption and bequests) at the end of their lifetime. Those with
enough bequests (x ≥ h) have an incentive to invest in education if their
wealth is higher than an alternative case of remaining unskilled. This
suggests the following inequality
62
(1 + r) (x – h) + ws ≥ (1 + r) (x + wn) + wn
Simplifying this leads to the following condition

ws – h(1 + r) ≥ (2 + r) wn (4)
This condition says the skilled wage is sufficiently high to cover the cost of
investing in education and the opportunity cost of not working in the first
period. We assume condition (4) is satisfied, otherwise no one invests in
education.
Activity 5.10
Consider an individual with insufficient bequests (x < h). Show that if (4) is not satisfied
this individual will not invest in education. What is the interpretation of this?
Now consider individuals with insufficient funds (x < h). Remaining

unskilled will yield end of life bequests based on (1) and investing in
education by borrowing in the first period yields (3). Therefore low
bequest individuals invest in education if the following inequality is held:
(1 + i)(x – h) + ws ≥ (1 + r)(x + wn) + wn
Simplifying this implies that individuals invest in education only if their

parental bequests are higher than f defined below.
1
x ≥ f = [(2 + r) wn + (1 + i) h – ws] (5)
i–r
In summary, two groups invest in education. Children of the rich parents

invest in education entirely from parental bequests while some of the
middle class borrow and invest in education. The poor do not invest in
education and work as unskilled labourers (i.e. those with (x < f)). But
what does this imply about bequests left for children? How does the next
generation fare relative to their parents? Do they become poorer or richer?
To analyse these questions, consider individuals receiving xt from parents,
who bequest xt+1 for their children in return. Given the above discussion
the level of bequests for next generation is determined as follows:
{
α ((1 + r)(xt + wn) + wn) if xt < f
xt+1 = α ((1 + i)(xt – h) + ws) if f ≤ xt < h (6)
α ((1 + r)(xt – h) + ws) if xt ≥ h
Here f is the crucial constant that depends on model parameters.

The relation of parental bequests with f and cost of education, h, will
determine whether the next generation invests in education. Figure 5.2
shows the dynamics of bequests in this model based on (6). The solid line
shows the dependence of next period bequests on current bequests. Steady
state level of bequests is when the last period bequest is equal to the next
period bequest (xt+1 = xt). In the diagram this will be the crossing of the 45
degree line with the bequest function. Bequest function and the 45 degree
line cross three times, therefore, there are three steady states but only xn
and xs are stable in the sense that small deviations from the equilibrium
point will return us to the equilibrium.
Activity 5.11
Why does the middle segment of the bequest function (solid line) in figure 5.2 have a
steeper slope compared to the two other segments? Do you think the first and the last
segments have the same slope? Why?
63
Let us consider the time path of individuals starting with parental

bequests equal to x0. They will invest in education (since x0 > f as shown
in Figure 5.2) and leave bequests equal to x′0 = α(1 + i)(x0 – h) + wx) for
their children. The new level of bequest is found by moving up to the
bequest function (solid line) from x0 (point A on the line). Their children
subsequently leave bequests for next generation which could be found by
following the arrows on the figure. As generations pass the individuals
starting from x0 will end up in a steady state level of bequests equal to xn.
Although some of the fathers chose to invest in education their bequests
is falling and after some time their children stop investing in education
because it is not profitable.
xt+1
xt+1(xt)
x’0 A
45º
xn f x0 g h xs xt
Figure 5.2: Next period bequests as a function of parental bequests
Activity 5.12
Draw Figure 5.2 and try to follow the bequest path for individuals with parental bequest
greater than g. Show they reach the skilled labour equilibrium at xs level of bequests.
Activity 5.13
The way Figure 5.2 is drawn requires two further assumptions: α(1 + r) < 1 < α(1 + i).
What happens if these assumptions are violated? How does the figure look like? What
would be steady state level of bequests?
So, what does this model tell us about the relation between inequality and
growth? The main implication of this model is that initial distribution of
wealth matters for steady state wealth distribution and level of income.
Assume L is total population and Lg shows the initial number of individuals
with initial bequests less than g. The steady state bequest distribution
consists of two points: one at xn and another at xs. Everyone with initial
bequest higher than g will leave enough bequests for their descendents
so they will converge to xs. On the other hand, individuals with initial
bequests less than g witness a decreasing level of wealth and converge
to the unskilled bequest level at xn. The steady state average bequest is a
weighted average of xn and xs as follows (notice wealth is equal to bequest
divided by α)
Lg Lg (7)
x= xn + (1 – ) xs
L L
Countries that start as poor end up poor in this model, because most
of the individuals do not have enough resources to invest in education
64
and remain unskilled (converge to xn equilibrium). Rich countries, with

very unequal wealth distribution, also end up poor in this model. This
is because the few rich remain at the skilled equilibrium but the poor
majority do not have enough funds to sustain education investments. Here
redistribution might pull some of the poor out of poverty while the rich
still converge to xs. This conclusion depends on the initial distribution of
wealth. In some circumstances redistribution might actually result in a
reduction of average wealth in steady state. Activity 5.14 asks you to look
into some examples. The model presented here abstracts from growth
and capital accumulation but it is not difficult to introduce those features.
Intuitively, if skilled and unskilled wages grow at different rates the steady
state average growth rate is a weighted average of these rates.
Activity 5.14
Consider an economy with 100 individuals and assume g = 1,000 $US. Half of individuals
have an initial wealth of $US500 while the other half have an initial wealth of $US2,000.
The steady state wealth for skilled labourers is $US3,000 while that of unskilled labourers
is $US500. The setting is exactly identical to the model described above and you can use
the formula derived here.
a. Calculate average steady state wealth in this society.
b. Now think about a redistribution plan that takes $US600 from the 50 per cent
of wealthy individuals and redistribute it evenly among the 50 per cent of poor
individuals. Is this associated with an increase in average steady state wealth?
c. Consider another initial wealth distribution, with 70 per cent of the poor having
$US500 and 30 per cent of the rich having $US1500. What is the average steady
state wealth? Can you think of a redistribution that results in higher average steady
state wealth?
This model is of course an abstraction and many interesting real world

features are absent. But it serves the purpose of showing the complex
equilibrium relations between credit markets, occupational choice and
inequality. When the credit markets are imperfect, reflected in i > r
and skill acquisition requires indivisible investments, given by upfront
education cost of h, the poor remain poor simply because they do not have
the means to invest in acquiring the highly valued skills.
Note the model presented here abstracts from price adjustments and the
fact that skilled and unskilled wages could adjust when labour supply and
demand for each type change. For a more advanced model you should
read Banerjee and Newman (1993).
What are the potential policy implications of this model? Improving credit
markets could be a reasonable action. After all, if the cost of borrowing
is lower or if individuals can borrow against future wages (instead of
current endowments) a larger fraction will be able to pursue education
investments. The model might also support a policy of taxing the well-off
and investing in education (e.g. free public education) but you should be
careful about this due to the complex relationship between inequality and
wealth (e.g. see Activity 5.14).
Activity 5.15
Improving credit markets is equivalent to lowering i in the model. Based on Figure 5.2,
could you predict the impact of lowering i on the steady state wealth distribution? How
do different segments of the bequest function change?
65
► Stop and read

Ray (1998) Chapter 7, pp.223–40 (subsection 7.2.7 till the end of chapter) and Galor and
Zeira (1993).
5.5 Summary
In this chapter we discussed the difficulties of comparing inequality across
countries. The axiomatic approach provides four intuitive criteria that
could be used to select appropriate inequality measures. The Lorenz curve
and GINI coefficient are the two most commonly used inequality measures
in the context of cross country comparisons.
Theoretically, inequality could affect development in many different ways.
Therefore, in order to gain insights into the role of inequality, we need to
rely on empirical studies. Yet the presence of multiple mechanisms, poses
identification challenges in estimating the causal effect of inequality on
development. We investigated an empirical article, employing instrumental
variables to identify the causal effect of inequality on income. The
conclusion is far from clear. Interpreting the negative correlation between
inequality and income in the data as causal, is in the end, a controversial
issue.
We finished the chapter by presenting a theoretical equilibrium model
of skills acquisition, inequality and development. Although very stylised,
the model suggests that credit market imperfections could play a role
in shaping the influence of inequality on economic development. In the
absence of credit market failures, inequality would not matter for growth.
Conversely, the presence of credit market frictions justifies a role for the
government to intervene and correct negative impact of inequality.

• explain the axiomatic approach to measuring inequality
• recognise and explain various aspects of the two-way relationship
between inequality and growth
• summarise and evaluate empirical designs for identifying causal effect
of inequality on income.

1. Explain the inverse U hypothesis for the relation of inequality and
income. Outline at least one set of mechanisms that could generate
such a pattern in the data.
2. A researcher is interested in estimating the impact of income
inequality on GDP per capita. She collects data on GINI coefficients
and GDP per capita from the World Development Indicators dataset for
all countries of the world.
a. She decides to estimate the impact of inequality by regressing
GDP per capita on GINI coefficients using ordinary least squares.
Outline the key challenges to this estimation strategy.
b. A historian tells the researcher that economic inequality is very
persistent and past events like slavery could have had significant
66
impacts on contemporary inequality. How might she use this

information in an instrumental variable strategy to estimate
the causal effect of inequality on GDP per capita? What are the
identification assumptions? Do you think they are satisfied?
Reminder: Feedback to activities in this chapter are available on

the VLE.
67
Notes
68
Chapter 6: The role of institutions
6.1 Introduction
• explain meaning of institutions and their role in the process of
economic development
• present empirical evidence on the role of property rights and
contractual institutions on development
• discuss limitations of empirical macro institution literature
• present examples of empirical micro institution literature.

• define institutions and identify their impact on individuals’ incentives
• discuss empirical evidence on the role of property rights and
contractual institutions in shaping economic outcomes
• critically evaluate empirical research on the impact of institutions.

Acemoglu, D, S. Johnson and J.A. Robinson ‘The colonial origins of
comparative development: an empirical investigation’, American Economic
Review 91(5) 2001, pp.1369–401.
Banerjee and Duflo (2011) Chapter 10.
Besley, T. ‘Property rights and investment incentives: theory and evidence from
Ghana’, Journal of Political Economy 103(5) 1995, pp.903–37.
Field, E. ‘Entitled to work: urban property rights and labor supply in Peru’,
Glaeser, E., R. LaPorta, F. Lopez-de-Silanes and A. Shleifer ‘Do institutions cause
growth?’, Journal of Economic Growth 9(3) 2004, pp.271–303.

Acemoglu, D. and S. Johnson ‘Unbundling institutions’, Journal of Political
Economy 113(5) 2005, pp.949–95.
Besley, T. and M. Ghatak ‘Property rights and economic development’ in Rodrik,
D. and M. Rosenzweig Handbook of development economics. (Oxford;
Amsterdam: Elsevier, 2010) Volume 5, Chapter 68, pp.4525–95.
La Porta, R., F. Lopez-de-Silanes and A. Shleifer ‘The economic consequences of
legal origins’, Journal of Economic Literature 46(2) 2008, pp.285–332.
North, D.C. ‘Institutions’, Journal of Economic Perspectives 5(1) 1991,
pp.97–112.
6.1.5 Works cited

Acemoglu, D., S. Johnson and J.A. Robinson ‘Institutions as a fundamental
cause of long-run growth’ in Aghion, P. and S.N. Durlauf Handbook of
economic growth. (Oxford; Amsterdam: Elsevier, 2005) Volume 1, Part A,
Chapter 6 pp.385–472.
Angrist, J.D. and A.B. Krueger ‘Instrumental variables and the search for
identification: from supply and demand to natural experiments’, Journal of
69
Englebert, P. ‘Pre-colonial institutions, post-colonial states, and economic

development in tropical Africa’, Political Research Quarterly 53(1) 2000,
pp.7–36.
Grier, R.M. ‘Colonial legacies and economic growth’, Public Choice 98(3–4)
1999, pp.317–35.
La Porta, R., F. Lopez-de-Silanes, A. Shleifer and R. Vishny ‘Legal determinants
of external finance’, Journal of Finance 52(3) 1997, pp.1131–50.
La Porta, R., F. Lopez-de-Silanes, A. Shleifer and R. Vishny ‘Law and finance’,
Botero, J.C., S. Djankov, R. la Porta, F. Lopez-de-Silanes and A. Shleifer, ‘The
regulation of labor’, Quarterly Journal of Economics 119(4) 2004,
pp.1339–82.
6.1.6 Overview
Complementing the previous chapters, this chapter looks at the role of
institutions in explaining underdevelopment. We draw on both cross-
country and within-country evidence on the role of institutions in
development.
In Chapters 3 and 4 we discussed several potential explanations for large
income disparities across countries. The Solow model and the growth
accounting literature showed factor endowments have some role to play
in explaining gaps but a large part of income gaps are left unexplained
as total factor productivity (TFP) differences (remember Solow residual).
Endogenous growth theories used the idea of purposeful investment
in enhancement of productivity (e.g. R&D) to provide a model of
technological development. Here countries that invest more in R&D could
grow faster. The key question that remains unanswered in what we have
covered so far is why some countries accumulate more capital and are set
on a TFP enhancing path while others fail to do so. In this chapter we try
to delve deeper and discuss the role of institutions as fundamental causes
of income differences (Acemoglu et al., 2005).
In the next section we define institutions. In section 6.3 we look at the
empirical macro literature on the role of institutions. In section 6.4 we
turn to a brief discussion at the micro level and discuss the role of property
rights in improving incomes using a within-country study.
6.2 What are institutions?

North (1991) defines institutions as ‘the humanly devised constraints
that structure political, economic and social interaction’. This definition
includes formal and informal constraints like traditions, customs,
constitutions and property rights. Institutions shape the incentive
structure and therefore the decisions of individuals in the economy. It is
important to note that de jure institutions might differ greatly from de
facto institutions. In other words, on paper many institutions might be
similar (e.g. Latin American institutions compared to US institutions) but
in practice they differ greatly. What matters for economic incentives is
effective (de facto) institutions which may not coincide with statutory (de
jure) institutions.
But why do we need institutions? Why do we need to constrain individual
actions? Institutions set the stage for human interaction. In the market,
they establish rules to allow for exchange and cooperation. Institutions
allude to profitable ways of cooperation or lack thereof for utility-
maximising individuals. For example, in the absence of formal legal
institutions, individuals lend and borrow money in their circle of family
70
and friends, because they have full information about their attitudes and
actions and can impose costly sanctions if members defect (e.g. exclusion
from social institutions). Here the social institutions (e.g. family and
friendship bonds) determine who you can trade with. Without formal
institutions it is impossible to engage in lending and borrowing beyond
family and friends.
By shaping the relative benefits of different actions, institutions delineate
skills and forms of knowledge that are conducive to individuals’ material
wellbeing. If the rules of the game honour profits of entrepreneurial
activity, individuals understand that effort is rewarded and engage in
business investment. In contrast, prevalence of theft and expropriation by
the government discourages entrepreneurship and promotes rent-seeking
skills as fertile investments for increasing individual welfare. At some level
the question of development is a question of institutional development.
Why is it that some settings provide an incentive for continuous
improvements in the rules of the game, leading to economic prosperity,
while others resist such changes?
► Stop and read

North (1991).
6.3 Institutions and development: macro evidence

There is a large empirical literature looking at the impact of countrywide
institutions on economic outcomes. Here we touch on some of the seminal
works on property rights and legal institutions. We leave out studies that
look at political institutions and their impact on economic outcomes.
The main challenge in estimating the causal impact of institutions on
economic outcomes is the endogeneity of institutions. We are interested
in estimating the causal effect of institutions on economic outcomes,
but statistical methods like regression rely on observed correlations in
the data to estimate coefficients. Omitted variables, reverse causality
and measurement error could result in spurious correlations in the data.
Simple regression would wrongly associate such correlations as the causal
impact of institutions.
Figure 6.1 shows the situation in a schematic way. While we are interested
in the impact of institutions on certain outcomes (arrow showing channel
of interest), omitted variables and reverse causality are at play. Many good
institutions, policies and outcomes are clustered. Countries with good
protection of property rights also have effective judiciaries. Conversely,
countries with a high risk of expropriation also feature ineffective
judiciaries. This makes it hard to disentangle the effect of one institution
from another. It could be that omitted variables are responsible for good
institutions and economic outcomes. For example, while the judiciary
is the fundamental cause of high security of property rights and high
incomes, regressing income on security of property and ignoring the
judiciary results in a significant coefficient for property rights.
71
Channel of interest
Institutions Outcomes
Reverse causality
Omitted factors
Figure 6.1: Interlinked variables and the difficulty of estimating causal effect of
institutions.
Activity 6.1
Explain why ignoring certain features of geography (e.g. percentage of land in
mountainous areas) in a regression of income on security of property rights could result in
the problem of omitted variable bias.
Outcomes could cause changes in institutions too. This is the reverse

causality arrow in Figure 6.1. For example, higher income generates more
tax revenue that could finance a more effective enforcement of the de jure
institutions. Even if institutions have no effect on outcomes, the reverse
causation generates a correlation between outcomes and institutions.
Simple regression of outcomes on institutions implies a non-zero effect of
institutions because the observed correlation is not zero.
Finally, measurement of institutions is not a trivial task. It is not clear
how you would measure the effectiveness of the judicial system and
security of property rights. Researchers usually use proxy variables to
estimate aspects of these broad concepts. A proxy variable is correlated
with the true measure of institutions but has some measurement error.
The most innocuous form of measurement error is the classical or random
measurement error where the error itself is not correlated with the true
variable and other covariates included in the regression. Even in this case
the estimated coefficients are biased (usually toward zero). For example, if
we are measuring the security of property rights using incidences of theft
of property in a country, then the measurement error could be unreported
incidences. Classical measurement error here is when the number of
unreported cases does not vary with the security of property rights. On the
other hand, if countries with higher security of property rights had fewer
unreported cases, measurement error is correlated with the true level of
security. This is a form of non-classical measurement error and could result
in false conclusions.
The early literature on the role of institutions was mostly qualitative
and at best identified correlations in the data (e.g. Englebert, 2000,
and Grier, 1999) but the recent literature – starting with the seminal
works of Acemoglu et al. (2001) and La Porta et al. (1997, 1998) –
uses a variety of econometric techniques to get causal estimates of the
impact of institutions. The most commonly used identification strategy is
instrumental variable approach. Before studying these articles, we briefly
discuss the rationale of the instrumental variable approach and analyse the
assumptions needed for this approach to work.
72
6.3.1 The Instrumental variable approach

Endogeneity of institutions implies the observed correlation between
economic outcomes and institutions may not necessarily be due to the
impact of institutions but could be because of reverse causality, omitted
variables, or measurement error. The instrumental variable approach
removes the endogenous correlation and focuses on the direct impact
of institutions on outcomes. The idea is to find an instrument, Z, that
is related to institutions but has no other impact on outcomes. In other
words, the effect of Z on outcomes is only through institutions, and apart
from that, it is exogenous with respect to outcomes.
Figure 6.2 augments Figure 6.1 to show the idea of the instrumental variable
approach. We need at least two assumptions for the instrumental variable
approach to work. First, the instrument must be (strongly) related to the
endogenous variable. This is the relevance condition and is shown by the
solid arrow from the instrument to institutions in Figure 6.2. Second, the
instrument must not be related to the outcome variable other than through
the endogenous variable. Dashed arrows, running from the instrument to
omitted variables or outcomes, show violations of exclusion restriction. The
exclusion restriction requires that such arrows do not exit.
Channel of interest
Institutions Outcomes
Relevance Reverse causality
Instrument Omitted factors
Exclusion
Figure 6.2: Schematic representation of the idea of the instrumental variable

approach.
The instrumental variable approach could be implemented in a two stage
process known as two stage least squares (2SLS). Under this approach the
endogenous variable is regressed on the instrument in the first stage. The
predicted values of the endogenous variable are then used in the second
stage to estimate the causal effect of interest. The first stage reflects the
strength of the relevance condition. Specifically the t-statistics of the
estimated coefficient for the instrument in the regression should show a
significant relationship (or, in case of multiple instruments, the F-statistics
for all estimated coefficients should be significant). The exclusion
restriction cannot be checked in the data and if violated could severely
bias the estimates. Researchers usually need to argue why exclusion
restriction is satisfied and often it is very hard to do so. In the context of
institutions, the presence of omitted variables and clustering of country
characteristics pose a real challenge for finding a valid instrument. We will
assess the plausibility of this assumption in the studies that follow.
Sometime the relationship between the instrument and outcome is
interesting in itself. A direct regression of outcome on the instrument
73
is refered to as the reduced form estimate. If the instrument is plausibly

exogenous but affect outcomes through many channels, the exclusion
restriction is violated. However, the reduced form is still a valid regression.
Therefore, while the instrument cannot be used to identify the causal effect
of the endogenous variable, the results of the first stage give an idea of the
importance of instrument for the outcome.
► Stop and read

Angrist and Krueger (2001) to get a better understanding of instrumental variables.
6.3.2 Property rights

We have seen that investment in capital and enhancing productivity is
crucial for economic growth. But investment payoffs are realised in future
and are subject to various risks. Apart from natural disasters and risks
involved in the market, we can think of two potential investment risks:
• theft or destruction of investment by other individuals
• expropriation by the government.
A country might have a strong government that does not allow individual
theft but expropriates anything at will. Therefore security of property rights
could be defined at two levels: security against private actors and against
the government. Checks and balances on the government would restrict
the possibility of arbitrary action. If investors believe that there is a high
probability that the government could seize their assets they would cut
back on their investments.
The colonisation of the countries of the world by the European powers in
the 18th and 19th centuries provides an opportunity to estimate the effect
of institutions. Acemoglu et al. (2001) employ an instrumental variable
approach to estimate the effect of security from government expropriation
on economic development. They argue the disease environment of the
colonies had a major impact on the strategies adopted by colonisers.
In colonies with harsh disease environments, and hence higher settler
mortality, colonisers transplanted extractive rent-seeking institutions.
Colonies with more favourable disease environments allowed for the
settlement of Europeans and transplantation of institutions securing
property rights during colonial rule. Persistence of early institutions results
in a strong relation between historical and contemporary institutions.
Activity 6.2
How do the three claims presented in Acemoglu et al. (2001) p.1370 provide evidence
for the relevance condition for using settler mortality as an instrument for contemporary
institutions?
Activity 6.3
How do Acemoglu et al. (2001) measure present-time government expropriation risk?
How do they measure economic development? Do you think the two measures are fit for
the purpose? Why?
The main specification estimated in Acemoglu et al. (2001) is the following

log yi = µ + αRi + X 'i γ + i (1)
where the dependent variable is a logarithm of GDP per capita and the
regressors include a constant, µ, protection against expropriation (R), and
a set of controls represented by the vector X and α is the coefficient of
interest.
74
Activity 6.4
Why does estimating (1) using ordinary least square (OLS) yield a biased estimate of
α? Try to tell a coherent story for a specific issue that causes the bias and intuitively
determine the direction of the bias.
Since R is endogenous in equation (1), Acemoglu et al. (2001) use settler

mortality as an instrument for protection against expropriation. The
following equation shows the first stage
Ri = ζ + β log Mi + X'i δ + vi (2)
where Mi is the settler mortality rate. Once this equation is estimated, the
predicted values for protection against expropriation are used in equation
(1) resulting in the following second stage estimation equation.
log y = µ + αRˆ + X' γ +
i i i i
(3)
where Rˆ i is calculated using the estimated parameters from (2) as follows

ˆ = ζˆ + βˆ log M + X' δˆ .
R (4)
i i i
When exclusion restriction is satisfied, the coefficient estimate for α is

consistent. Intuitively this says that on average the estimated α is going
to capture the true causal parameter when the sample is large enough.
Panel B of Table 4 in Acemoglu et al.’s article shows the results of second
stage estimation, panel A captures the first stage, and panel C demonstrate
OLS estimation of equation (1). The coefficient of interest is the one in
the first rows of panel A and C corresponding to α in equation (3) and
(1) respectively. The 2SLS estimate in column (1) which shows increasing
protection against expropriation by 1 unit (on the 0–10 scale), increases
GDP per capita by 94 log points. This is an increase of around 2.6-fold
in levels of GDP per capita (2.6 ≈ exp(0.94)). Note that the estimation
strategy (use of settler mortality) does not imply institutions are
predetermined. They are only using one source of variation in institutions
that is potentially exogenous to estimate the casual link.
Activity 6.5
The numbers reported in parenthesis under each coefficient estimate in Table 4 of
Acemoglu et al.’s article show standard errors of the estimated coefficients (you can see
this in the notes under the table). In order to see if an estimated coefficient is significantly
different from zero with 95 per cent probability, you need to divide the estimated
coefficient by its standard error to get the t-statistics. As a rule of thumb if the t-statistic is
greater than 2 (precisely 1.96) in absolute value, that coefficient is significantly different
from zero.
a. Are the estimated α from 2SLS and OLS significant?
b. Is the relevance condition satisfied? (Hint: you need to check the significance of an
estimated coefficient.)
The critical assumption in Acemoglu et al. (2001) is the exclusion

restriction. They present several pieces of evidence to support this
assumption. First, they argue that the indigenous people had immunity
against malaria and yellow fever, the two major causes for settler
mortality. Therefore the disease environment did not have as much of an
impact on the indigenous people. This argument alleviates concerns about
the direct impact of the disease environment on economic performance.
Second, they get similar results using only yellow fever as a cause of
settlers’ mortality. Since yellow fever is eradicated now it cannot have a
direct effect on contemporary GDP.
75
Activity 6.6
Glazer et al. (2004) find that there is a significant correlation between the settlement
of colonisers and human capital endowments in 1900. They argue that settlers brought
with them everything they had, including human capital. Does this pose a threat to the
identification strategy employed in Acemoglu et al. (2001)?
What are the policy implications of the results in Acemoglu et al. (2001)?
They show that institutions are very important for economic development
and gains from improving them are potentially very large. But they do not
provide any guidance on the factors that shape institutions. Furthermore,
the risk of expropriation is a measure of policy that could change in the
short run rather than a measure of a long lasting institution (Glazer et al.,
2004).
► Stop and read

Acemoglu et al. (2001).
6.3.3 Legal and contractual institutions

Another branch of macro instituions literature started with the two articles
by La Porta et al. (1997, 1998) who argue that legal protection of outside
investors against corporate insiders limits expropriation by insiders and
promotes financial development. The authors construct a measure of
investor protection and show it is highly correlated with countries’ legal
origins. Specifically, countries with common (English) law origin have the
highest protection of investors compared to countries with civil (French)
law origin. Legal traditions were introduced in many countries through the
colonisation process. Colonisers transplanted their own legal systems, with
long-lasting impacts on the legal framework of the colonies.
La Porta et al. (1998) verify significant correlations between financial
development and legal origins. Financial development is an important
determinant of economic growth. Therefore the results here point to legal
institutions and the resulting contractual environment as an important
contributor to economic development.
In effect the argument here states that the identity of the colonisers
(English, French, etc.) matters greatly for the effect of colonisation on
future development. This view is in contrast to the previous section where
the characteristics of the colonised region (disease environment) was
thought to be of crucial importance. Acemoglu and Johnson (2005) try to
assess the relative importance of these two factors in the set of colonised
countries.
Acemoglu and Johnson (2005) run a regression of GDP per capita on
measures of contractual and property rights institutions using settler
mortality and legal origins as instruments. 2SLS estimates show that
contractual institutions do not have a signfiicant effect on income while
property right institutions continue to show a significant effect (see Table
4 in Acemoglu and Johnson’s article for 2SLS results). They interpret this
conclusion in the following way. Individuals would find ways to protect
their rights even if contractual institutions are not efficient but they find
it harder to protect against government risk of expropriation, which is
reflected in the measures of property rights used.
76
Activity 6.7
Subsequent studies show legal origins are correlated with other country characteristics
like labour market regulation (Botero et al., 2004). What does this suggest about
the validity of legal origins as an instrument for financial development or contractual
institutions?
► Stop and read

Glaeser et al. (2004). Also see La Porta et al. (2008) for a review of the legal origins
literature.
6.3.4 Issues with macro evidence

The identification problem is at the core of criticisms about cross-country
institutions literature. Although many studies use instrumental variables to
identify causal effects of institutions, the exclusion restriction is unlikely to
be satisfied. The clustering of institutions makes it hard to find measures
correlated with one institution but unrelated to the remaining institutions.
The second issue is measurement of institutions at the country level.
Some of the measures employed show policy choices rather than deep
institutional variables. For example, risk of expropriation is influenced
by government policies. We have examples of dictatorships with
unconstrained powers choosing to protect private property. Despite a
huge effort by both political scientists and economists to quantify political
institutions (e.g. see the Polity IV project at: www.systemicpeace.org/
polity/polity4.htm), some aspects of institutions are hard to quantify.
Finally, it is not clear how much we can learn by focusing on macro level
institutions. There is usually great institutional variability within countries.
This is, for example, reflected in the large income differences across states
of India. Average per capita net state domestic product in Kerala was four
times larger than that of Bihar in 2004–05 (Directorate of Economics and
Statistics of the respective state governments). A focus on macro figures
masks heterogeneity within countries. Furthermore, measurement of
micro-level institutions is more transparent, making the interpretation of
results easier.
These shortcomings resulted in a shift of development literature from
macro-oriented studies to within-country micro studies. We touch on some
of the prominent micro studies of property rights in the next section.
6.4 Property rights and development: micro evidence

Even at the micro level, property rights have different aspects. Usage
rights consider the freedom of individuals in using their property
for consumption or production. Transfer rights look at the ability of
individuals to rent or sell their assets. These rights could be held against
various groups. For example, communal rights restrict outsider access to
properties of a group but within the group the elder or leader allocates the
assets to individuals (Besley and Ghatak, 2010).
Property rights shape individuals’ investment decisions for several reasons.
First, a higher security for property rights reduces risks of expropriation.
Therefore individuals are more certain they will reap the benefits of their
investments. Second, if property rights are well defined and enforced, it
is easier to use assets as collateral for borrowing and lending. This could
alleviate liquidity constraints and enhance investment. Third, transfer
rights and well enforced usage rights would expand trading opportunities.
77
Having the opportunity to sell or rent out investments enables individuals

to insure against idiosyncratic shocks. For example, when a farmer invests
in an irrigation system but receives a health shock, he will not be able to
cultivate the land and reap the fruits of his investment. However, the value
of the investment will be reflected in the rental price and renting it out to
someone else retains some of the gains of the investment (Besley, 1995).
The impact of property rights is not limited to investment. Higher security
of rights reduces the need to guard property and frees resources (money
and labour time) that could be used in productive activities. Furthermore,
individuals gain the freedom to choose the place of business activity when
the need to guard their property (house) is removed.
Activity 6.8
Explain theoretical reasons why property rights might influence levels of production.
In what follows we look at two articles estimating the effect of property

rights on production decisions: Besley (1995) and Field (2007). Besley
studied two regions of Wassa and Anloga in Ghana. Wassa is a cocoa
growing region, where tree plantation is the key investment. Anloga is
less dependent on agriculture, and grows mostly shallots on small fields.
Historically, communal rights were in place in Ghana but a transition
to individual rights was taking place in both regions at the time of
the sample. The data used by Besley cover household characteristics,
information on transfer rights, and investments made on land. The data
distinguishes between transfer rights requiring approval from village
lineage and those that do not. Besley counts the rights with and without
approval separately and includes the two counts in the regressions.
Activity 6.9
Why might investments on land impact on land rights? Suppose we use OLS to estimate
the causal effect of land rights on investments. We regress a dummy variable that
shows whether investment is made in a given field (e.g. a tree plantation in Wassa) on
the number of rights with and without approval enjoyed by the owner. What does the
dependence of land rights on investments imply about the causal interpretation of the
estimated coefficients for land rights here?
Recognising the endogeneity of land rights with respect to investment,

Besley (1995) adopts an instrumental variable approach. In Wassa, he
measures for land rights using mode of acquisition, whether the owner
has a title deed, whether there was litigation over the field, the number of
years the land has been owned, and whether there were trees at the time
of acquisition. We expect these variables to be correlated with current land
rights. But to satisfy the exclusion restriction these variables should not
impact on investment (tree plantation) other than through their impact on
current rights.
The results for Wassa, reported in Besley’s article, suggest land rights with
and without approval have a similar effect on investment. Having one
more right with approval increases the probability of investment in trees
by 12 percentage points. The coefficient is marginally significant at 5 per
cent (column 2). The table also confirms that the relevance condition is
strongly satisfied by looking at the joint significance of instruments in the
first stage (columns 3 and 4).
78
Activity 6.10
Do you think the exclusion restriction for the instruments are satisfied in Wassa? Why
might inclusion of household dummies (fixed effect) alleviate concerns about violations of
exclusion restriction?
Table 4 in Besley’s article includes household dummies to control for

household heterogeneity. The results are similar to the earlier figures, but
now rights with approval do not have a significant effect on investment.
Now an additional right without approval (individualistic) increases
probability of investment by 28 percentage points. With household
dummies the coefficients are identified, based on within household
variation in investment decisions and land specific rights. In other words,
it captures the disproportionate investments made on lands with rights,
compared to those without rights, owned by an average household. Fixed
effects control for observable and unobservable household characteristics
like knowledge and ability in growing cocoa.
The collateral-based view (better rights allowing use of property as
collateral) suggests that overall household rights matter for investment
rather than field specific rights. Once you control for household fixed
effects the collateral effect is eliminated and the coefficients on rights
variables should fall to zero. The fact that they are not zero suggests field
specific rights matter apart from overall household rights, pointing to the
importance of security argument and gains from trade.
Activity 6.11
Why does Besley (1995) not show the regression results with household dummies for the
Anloga region? Could you include household fixed effects if each household owns only
one plot? How does Besley test for the collateral based view in Anloga? What does he
conclude about the importance of collateral view?
Activity 6.12
The results of Besley (1995) confirm that communal rights lead to a significantly lower
level of investment on land and are inefficient. Therefore policy makers should promote
individual rights. Discuss.
► Stop and read

Besley (1995).
In a another study, Field (2007) uses a large titling programme in Peru to

estimate the effect of property rights on the labour supply of individuals.
The programme was rolled out between 1995 and 2003 and provided
property titles to squatters in urban areas free of charge. The theory
suggests an increase in security of property could increase labour supply
because households do not need to stay at home to protect their property.
The identification strategy employed by Field uses the gradual roll-out and
compares the difference in labour supply of squatters and non-squatters in
programme areas to the difference in labour supply of squatters and non-
squatters in non-programme areas. This is a difference-in-difference (DID)
estimation strategy and corrects for fixed differences between squatters and
non-squatters and between programme areas and non-programme areas. To
interpret the estimated coefficient as causal we need to assume that in the
absence of the programme the labour supply difference between squatters
and non-squatters would have been the same in programme and non-
programme areas. (We elaborate on the DID idea in the next chapter.)
79
Field finds a significant effect on squatters in programme areas. Squatter

households on average increased their labour supply by 13.5 hours when
the titling programme reached their neighbourhood.
Activity 6.13
What would be the effect of titling if the sense of security does not change after receiving
the title? Discuss this in relation to Table 3 in Field’s (2007) article.
Activity 6.14
Field’s (2007) results clearly show that security of property rights is of crucial importance
in the labour supply decision of households. Discuss.
Stop and read

Field (2007).
6.5 Summary
By shaping incentives, institutions have a significant impact on economic
development. Secure property rights and efficient legal systems encourage
entrepreneurial activities, while weak rule of law, expropriation
and unbinding contracts diminish investment incentives. Empirical
identification of the impact of institutions is fraught with difficulties.
Measurement error, omitted variable bias, and reverse causality are all
important challenges to identifying the causal effect of institutions on
development outcomes.
While Acemoglu et al. (2001) and La Porta et al. (1997, 1998) have
made progress on the empirical front by using the European colonisation
of many countries of the world as an instrument for contemporary
institutions, it is unlikely that instruments used in this literature satisfy
exclusion restriction.
The recent micro literature on the role of local institutions in shaping
individual outcomes in developing countries is more transparent on
the aspects being measured and more informative on policies that help
improve outcomes. Evidence from a move from communal to individual
property rights in Ghana suggests security of individual rights could play
a role in productivity-enhancing investment decisions. A large titling
programme in Peru also confirms significant impacts on the labour market
decisions of individuals.
► Stop and read

Banerjee and Duflo (2011) Chapter 10 to get an overview of the role institutions
(economic and political) play on economic development.

• define institutions and identify their impact on individuals’ incentives
• discuss empirical evidence on the role of property rights and
contractual institutions in shaping economic outcomes
• critically evaluate empirical research on the impact of institutions.
80

1. Outline key empirical challenges for estimating the effect of
institutions on economic development.
2. Define property rights and explain the difference between usage and
transfer rights.

on the VLE.
81
Notes
82
Part 2: Markets in developing countries
Part 2: Markets in developing

countries
83
Notes
84
Chapter 7: Education
7.1 Introduction
• discuss factors influencing education decisions
• understand role of policy in improving educational outcomes
• review some of the evidence and critically evaluate one study.

• explain a model of educational choice and discuss the role of relevant
factors
• recognise and outline various issues in improving educational
outcomes in developing countries
• summarise a difference-in-differences estimation and explain its
limitations
• summarise and critically evaluate empirical evidence on policies
promoting education.

Banerjee, A. and E. Duflo ‘Addressing absence’, Journal of Economic Perspectives
20(1) 2006, pp.117–32.
Duflo, E. ‘Schooling and labor market consequences of school construction in
Indonesia: evidence from an unusual policy experiment’, American Economic
Review 91(4) 2001, pp.795–813.

Duflo, E., R. Hanna and S. Ryan ‘Incentives work: getting teachers to come to
school’, American Economic Review 102(4) 2012, pp.1241–78.
Meyer, B.D. ‘Natural and quasi-experiments in economics’, Journal of Business
and Economic Statistics 13(2) 1995, pp.151–61.
Schultz, T.P. ‘School subsidies for the poor: evaluating the Mexican Progresa
poverty program’, Journal of Development Economics 74(1) 2004, pp.199–
250.
Jensen, R. ‘The (perceived) returns to education and the demand for schooling’,
7.1.5 Overview
Low human capital partly explains why income per capita is low in
developing countries. In this chapter we study the market for education
and analyse three factors that characterise these markets in developing
countries: low demand for education, poor education infrastructure
(schools) and lack of incentives for education providers (teachers).
We assess the empirical relevance of each of these factors using recent
evidence from randomised controlled trials (RCTs).
85
Until recently much of the focus was on access and enrolment. Therefore
school construction policies and promoting enrolment were top priorities.
Recently the issue of education quality has been receiving more attention,
partly because of shocking evidence on dismal test performances by
students. A recent survey shows around 35 per cent of children aged 7–14
in India could not read a first-grade paragraph, 60 per cent could not read
a simple story and 70 per cent could not do division (Banerjee and Duflo,
2011). Emphasising enrolment and ignoring the quality of education could
therefore lead to considerable waste of resources.
In this chapter we start with a theoretical model of how children (or their
parents) make educational decisions (section 7.2). We use the model
to categorise reasons for low educational attainment in three groups of
demand (section 7.3), infrastructure (section 7.4) and incentives (section
7.5). We then move on to review the empirical evidence on the importance
of each factor, drawing mostly from randomised controlled trials (RCT) in
developing countries.
► Stop and read

7.2 A model of educational decision-making

In this section we develop a very simple theoretical framework to
understand the factors that influence the number of years children attend
school and the role of government policy. Children often have little
involvement in the schooling decision. Parents decide on whether to send
their children to school and when to stop. This is particularly true about
primary education as it happens at an early stage.
In principle there might be pecuniary and non-pecuniary benefits from
education. Educated individuals can earn higher wages and enjoy better
work conditions and prestige. Parents, however, need to pay school fees
and other costs in order to educate their children. In principle parents
would care about education in and of itself (e.g. due to non-pecuniary
reasons) and about its long-term benefits. To simplify matters we abstract
from these complexities and assume parents only care about the share of
future income they receive from children in their old age. Parents choose
to educate their children to maximise their utility as follows
U(y, S) = m log y – h(S) (1)
Parental utility depends on a child’s income in adulthood y, years of
education S, and share of the child’s income accruing to the parents m.
Furthermore, income increases with years of education and we assume a
log linear relationship as follows
log y = α + βS (2)
Here β is the return to education: remaining in school for an additional

year increases income by β per cent.
Activity 7.1
Equation (2) assumes a constant return to education. In other words, an additional year
of primary education has the same marginal effect on earnings as an additional year of
secondary school. Do you think this is a reasonable assumption? Why? What might be a
more plausible assumption?
The second term in (1) is cost of education. We assume the cost of

education is a convex and increasing function of years of schooling.
86
Specifically we assume a linear first derivate for the cost function as follows
h'(S) = γ + φS (3)
The convexity assumption implies φ > 0. In other words, the marginal
cost of remaining one extra year in school increases with educational
attainment. The cost of education contains both direct costs and indirect
costs. Direct costs include school fees, uniforms, stationery, books, travel,
etc. Indirect costs include the opportunity cost of attending school (e.g.
loss of earnings).
Plugging (2) in the utility function in (1) and maximising utility then using
(3) gives the optimal years of schooling as follows
mβ – γ
S* = (4)
φ
Equation (4) captures the impact of the model parameters on optimal
years of schooling. Higher parental valuation of a child’s income (m),
higher returns to education (β) and lower cost of education (γ and φ)
result in more years of schooling.
Activity 7.2
Based on the model outlined above, what are likely explanations for fewer years of
schooling in rural areas relative to urban areas? What are potential reasons for lower
educational attainment for girls relative to boys?
7.3 Promoting schooling demand

There are various factors that could hold back schooling demand. In
this section we discuss two randomised experiments dealing with credit
constraints and low expectations about returns to education.
7.3.1 Cash transfers

We have left the financing of education costs outside the model. Clearly
families need to spend money on school fees, uniforms and travel. In the
absence of perfect credit markets, borrowing will be limited. Families
with few assets will have difficulty getting credit for children’s education
and might choose lower levels of education because they do not have the
money to cover education costs. Subsidies for education (e.g. free books,
meals, and uniforms) could alleviate credit constraints and induce more
years of schooling.
Activity 7.3
Giving monetary reward to parents for children’s attendance at school is a popular
education subsidy (e.g. Progresa in Mexico). Would it be necessary to condition the
transfer on attendance if the only reasons for children’s absence were credit constraints?
The government of Mexico implemented a means-tested education subsidy

– Progresa – to alleviate poverty and promote school attendance. A cash
subsidy was payable to eligible mothers whose children between grades 3
and 12 had 85 per cent attendance rate. The subsidy was larger for higher
grades and girls in secondary school. The government identified 495
localities as poor communities and about two-thirds of households were
determined to be eligible for Progresa. From this 314 communities were
randomly chosen to receive Progresa in 1998; the remainder received it
in 2000. Schultz (2004) uses the randomised roll out of the programme
to estimate the causal effect of conditional cash transfers (CCT) on
education.
87
Table 7.1 shows four potential cases arising from the programme
placement. Sit shows enrolment rates for group i at time t. Localities are
divided into treatment (receiving Progresa in 1998) and control (those
receiving it in 2000 and hence not receiving it in 1998). Poor households
are recipients of the transfers while the non-poor are not eligible.
Therefore, i takes four values: poor in Progresa localities (i = 1), poor in
non-Progresa localities (i = 2), non-poor in Progresa and non-Progresa
localities (i = 3, and 4 respectively). t takes five values: periods 1 and 2
correspond to the surveys done before Progresa and periods 3 to 5 are for
surveys done after 1998 but before 2000.
Poor (eligible) Non poor (ineligible)
Progresa localities (314) S1t S3t
Non-Progresa localities (181) S2t S4t
Table 7.1: Treatment and control groups.
Comparing poor households in Progresa and non-Progresa localities
after the programme was rolled out will give a causal estimate of the
programme's effect if there weren’t any pre-programme enrolment
differences between the poor in control and treatment localities.
D1 = S1t – S2t for t = 3, 4, 5
The fact that localities are chosen at random to receive Progresa implies
programme placement is orthogonal (exogenous) to area characteristics
but does not rule out pre-programme differences in enrolment among
the eligible groups. Schultz tests this by looking at enrolment difference
between poor in Progresa and non-Progresa regions in periods 1 and 2
(prior to roll out).
D p1 = S1t – S2t for t = 1, 2
The first three columns in Table 3 in Schultz's (2004) article show that
D p1 is not significantly different from zero while the post-programme
difference (D1) is positive and significant at 5 per cent from grades 1 to
6. Schultz also reports the difference-in-differences (DID) estimate of the
programme effect by subtracting pre-programme differences in enrolment
from post-programme differences across eligible (poor) households in
Progresa and non-Progresa localities.
DD1 = D1 – D1p
This estimator will remove any time-invariant differences between poor

households in Progresa and non-Progresa localities (pre-programme
differences). DID estimates suggest a positive effect of the programme but
the effect is significant only for children in grades 4 and 6. Based on DID,
enrolment increased by 3.4 percentage points across grades 1 to 8. This
corresponds to a cumulative gain of 0.66 years of schooling for targeted
recipients (around 10 per cent increase from baseline average years of
schooling).
Activity 7.4
Why do you think Progresa might have an effect on non-poor households in the localities
that received the programme? Does Progresa impact on the enrolment of non-poor
children negatively or positively?
Let us think about implications of Schultz’s results. Progresa provided an

income transfer conditional on good attendance. The transfer increases
the poor households’ income and could increase the number of years of
88
children's education through two channels. First, higher income implies

higher demand for normal goods. Children’s education increases if it is
a normal good. Second, by relaxing credit constraints the programme
allows poor households to invest in the optimal amount of education
for their children. In addition to income effects, the conditionality has
a substitution effect too. The relative price of education is reduced and
therefore families substitute toward more education.
► Stop and read

Schultz (2004).
7.3.2 Perceived returns to education

Since benefits of education are materialised in the future, parents choose
schooling based on expected benefits. This means perceived β might be
different from the real world expected returns. For example, parents might
underestimate the benefits of education because they lack the information,
potentially because they are not well educated. We can go one step
further and argue expectations about one’s social place might influence
actual outcomes. For example, there is some evidence that children from
low castes in India are expected to be under achievers (Banerjee and
Duflo, 2011). Jensen (2010), in an interesting study, randomly provides
information on returns to education and finds large significant effects
on years of schooling. The treatment group children increase years of
schooling by 0.2 year and probability of completing secondary school
increases by 5 percentage points over an average of 13 per cent.
7.4 Schooling infrastructure

Lack of proper schooling infrastructure could lower educational attainment
through many channels. Poor road conditions and unavailability of schools
increase travel time and cost (Φ in the model above, which includes
both out of pocket and opportunity costs). Low teacher-pupil ratios or
inappropriate teacher training reduces the quality of education and is
likely to reduce returns to schooling (β in the model).
Here we discuss the impacts of a large school construction programme on
the educational attainment of children in Indonesia. Funded from booming
oil revenues, the INPRES programme built 61,807 new schools between
1973 and 1979 in Indonesia, doubling the stock of schools. To maintain
the quality of education the government recruited new teachers and, as a
result, the number of teachers increased by 43 per cent. The programme
targeted areas with low enrolment levels in 1972 to increase equity across
provinces.
We are interested in estimating the impact of this massive school
construction programme on years of education. Programme targeting
creates regional variation in the level of exposure to new schools. Areas
with low enrolment rates received more schools and therefore were
expected to have larger education gains. Furthermore, the programme
benefited children aged between 2 and 6 years old in 1974 because these
children started going to primary school in 1975, around the time that the
first INPRES schools were built. By this time older cohorts had finished
primary school and therefore did not receive a benefit.
In principle we would like to know the effect of INPRES on average
years of schooling for the children affected by the programme. Table 7.2
classifies regions into high and low intensity school construction regions
89
H
and shows average years of education for each group. For example, YT
shows average years for treated cohorts (subscript T) in high intensity
regions (superscript H). Ideally we would like to know counterfactual
average years of schooling for the treated cohorts in high intensity regions
(i.e. average years of education for treated cohorts had there not been
any school construction). In other words, the causal estimate of INPRES
is the difference between the observed average years of schooling for
treated children and the unobserved average for the same children in the
counterfactual case of not receiving the treatment.
As the counterfactual is not observed we need to find a control group that
resembles the treated children and use the average years of schooling for
them as an estimate. The INPRES programme was not randomised. The
government specifically targeted low achieving areas to increase equity.
One idea for estimating the impact of the programme is to compare
average years of education for young cohorts in high and low intensity
regions (cross-sectional difference). This is DT given in the third column of
Table 7.2. The identification assumption for DT to give the causal estimate
of the programme effect is that, in the absence of INPRES, children aged
between 2–6 would have achieved the same average years of schooling in
high and low intensity regions (i.e. in the absence of INPRES DT = 0). It is
impossible to test this identification assumption because the counterfactual
situation – where INPRES was not enacted – did not happen. We can,
however, argue for plausibility (or implausibility) of this assumption.
Activity 7.5
Do you think this identification assumption is likely to be satisfied?
Another potential way to estimate the causal effect of INPRES is to compare

the average schooling for children aged 2–6 and children aged 12–17 in
1974 (cohort or time difference). This is DH in Table 7.2. The identification
assumption here is that, in the absence of INPRES, children aged 2–6 would
have achieved the same level of schooling as children age 12–17 in 1974.
This is a contentious assumption because we expect younger cohorts to
have achieved higher levels of education in the absence of any intervention.
This could be because over time more families start to appreciate the
importance of education and invest more in children’s education. Therefore
we do not expect DH to be zero in the absence of INPRES. In fact, any factor
that affects the education decisions of older cohorts relative to younger ones
will be confounded with the INPRES effect.
Intensity of school
building in region
High Low Difference
Aged 2 to 6 in 1974 YTH YTL DT = YTH − YTL
(Treatment)
Aged 12 to 17 in 1974
YCH YCL DC = YCH − YCL
(Control)
Difference DH = YTH − YCH DL = YTL − YCL DD = (Y TH − Y CH ) − ( Y TL − Y CL )
Table 7.2: Average years of schooling for treatment and control cohorts in high
and low intensity regions.
90
7.4.1 Difference-in-differences
A more sophisticated identification strategy uses the two dimensions that
determine exposure to INPRES. Neither the old nor the young cohorts
in low intensity regions were affected by the programme. Therefore, the
difference in average years of schooling for young and old in low intensity
regions (DL in Table 7.2) captures only the effect of factors that tend to
raise the education of younger cohorts relative to older ones and not that
of INPRES. On the other hand, the difference between young and old
cohorts in the high intensity regions is both due to the programme and
other confounding factors. Difference-in-differences estimator, DD, uses
DL as an estimate of the counterfactual difference between young and old
cohorts in high intensity regions in the absence of the programme.
The first differencing (DH and DL) removes fixed regional characteristics
that influence years of schooling similarly for both cohorts. The second
differencing, DH – DL, eliminates the influence of fixed cohort factors.
Therefore the identification assumption for DD is: had the programme not
been in place, the difference between average years of schooling between
the young and the old cohorts would have been the same across high
and low intensity regions. This is usually known as the parallel trends
assumption.
Activity 7.6
Another way of wording DD estimate is to start by first noting the difference between
young cohorts across regions and then subtract the difference between old cohorts across
regions. This gives the same estimator as the DD discussed above. State the identification
assumption in this case. Is it different?
Figure 7.1 shows a graphical representation of DD estimator. The four

solid dots depict average years of schooling for various groups taken from
Table 7.2. The top solid line is for the low intensity regions while the
bottom one is for high intensity regions. High intensity regions initially had
low enrolment and therefore expected to have lower average schooling.
Also, young cohorts do better and therefore the lines are upward sloping.
Young cohorts in high intensity regions attain higher education better than
expected due to the programme effect. The identification assumption says
that, in the absence of INPRES, the gap between old and young would be
the same across regions. Therefore, the hollow dot is the expected level of
average education for young people in high intensity regions.
Average years of schooling
YTL
YCL YTH
DD
YCH
old young cohort
Figure 7.1: Schematic view of difference-in-differences estimation and parallel

trends.
91
The difference-in-differences strategy can be implemented in a regression

framework as follows
yijk= α + β1 Highj + β2Treati + γHighj × Treati +∈ijk (5)
where yijk is years of schooling for individual i born in year k and region
j. High is a dummy variable that is equal to one for individuals born in
regions with a high school building intensity and zero otherwise. Treat is a
dummy that captures whether the individual was young enough to benefit
from the programme. It is equal to one for children aged 2–6 in 1974 and
zero otherwise. The coefficient γ on the interaction term captures the DD
estimate of the effect of the programme. The interaction term is equal
to one only for young cohorts born in high intensity regions and hence
measures the impact of the programme on them.
The regression allows for inclusion of other controls. One natural
extension is to include cohort and region fixed effects instead of having
one dummy that distinguishes between young and old or high and low
intensity regions. Furthermore, a regression framework could use the
whole range of variation in the school building intensity. In other words,
we can include a continuous measure of intensity that takes a different
value for each region to reflect the actual number of schools built.
Activity 7.7
This activity helps you show that the coefficient on the interaction term, γ, in (5) is in fact
capturing the DD estimator.
a. Using (5) find the average years of schooling for the four groups identified in Table
7.2. For example, for old cohorts in high intensity regions first observe High = 1 and
Treat = 0 then find an expression for years of schooling for this group based on (5).
Then work out the average for all old cohorts in high intensity region (hint: average
∈ is zero) to get YCH .
b. Find DH, DL, and DD using expressions you found in (a) and show DD = γ.
One way to assess the validity of the parallel trends assumption is to

collect data for periods before the programme implementation. In
the INPRES example, the older cohorts are not exposed to treatment.
Therefore, looking at the difference between years of schooling for very
old versus old cohorts should give a comparable estimate for high and
low intensity regions if the parallel trends assumption is valid. Figure
7.2 shows this idea. The difference-in-differences estimator for the old
versus very old cohorts is DDplacebo = ( YC – YC2 )– ( YC – YC2 ) . If the parallel
H H L L
trend assumption is satisfied we expect DDplacebo to be zero, because this

captures the difference between average years of schooling for old and
very old cohorts across high and low intensity regions. The parallel trend
requires this difference to be zero in the absence of the treatment. Well,
the comparison of very old and old is in a time where no treatment was
in place. As shown in Figure 7.2, the two lines connecting very old and
old are parallel, and hence in this hypothetical case the graph provides
evidence in support of parallel trends.
92
YTL
YCL
YTH
Average years of schooling

YC2L
DD
YCH
YC2H
very old old young cohort
Figure 7.2: Assessing parallel trends assumption with pre-reform data.
► Stop and read

Meyer (1995) for discussion of difference-in-differences estimation.
7.4.2 Impact of INPRES programme

Table 3 in Duflo (2001) shows the results of the simple difference-in-
differences strategy described above. Panel A shows that children aged
2–6 in 1974 on average received 0.12 years more education because of the
programme but this effect is not significant at 5 per cent (note t-statistics is
0.12
0.089 = 1.34
). Note the table reveals younger cohorts achieve more education
in both regions (0.47 years more in high intensity and 0.36 years more
in low intensity) and high intensity regions have lower average years of
schooling compared to low intensity regions (–1.27 years for younger
cohorts and –1.39 years for older cohorts).
The difference-in-differences estimator shows the causal effect under
the parallel trends assumption. But how plausible is the parallel trends
assumption here? There should not be any factor other than the school
building programme that affects younger cohorts differently in high
intensity region. To provide evidence in support of the identification
assumption Duflo runs a placebo experiment where she compares children
aged 12–17 to children aged 18–24 in 1974. Both of these groups were
too old to benefit from INPRES. Finding that DD estimator for this group
is zero provides evidence that the evolution of years of schooling is not
different across high and low regions. Panel B in Table 3 confirms this and
shows DD is 0.034 years with a standard error of 0.098 for this placebo
experiment.
Despite this placebo experiment we might still run into trouble if, for
example, the government started an iron supplementation programme
exactly at the time of INPRES in high intensity areas because healthier
children are more likely to stay in school. Note that even if this programme
was available to both high and low intensity regions, the parallel trends
assumption might still be violated. For example, iron supplementation
might have a larger effect in deprived areas, which could incidentally
be areas with low initial enrolment and high INPRES intensity. In this
example the observed differential increase in education for young cohorts
in high intensity areas is both due to iron supplementation and INPRES.
93
Activity 7.8
Think about a situation where the Indonesian economy was hit by a country-wide
recession exactly at the time of school building programme. Do you think such an event
would violate the parallel trends assumption? Now suppose the low intensity regions
were industrial regions and were hit harder by the recession. Could this result in a failure
of parallel trends? Do you think the DD in this scenario would over- or underestimate the
true impact of INPRES?
What do we learn from the INPRES programme? What are policy

implications of this study for other developing countries? The INPRES
programme certainly facilitated access to schools, reducing travel costs
and resulting in more years of schooling. For each additional school built
per 1,000 children, the average years of schooling went up by 0.12 years
(Table 4 in Duflo’s article). This suggests access was an inhibiting factor in
Indonesia but could this be generalised? Probably countries with schools
far apart should build more schools. But staffing and maintaining new
schools is as important. This study, however, does not inform us of the
most efficient way of improving access. For example, is it more efficient to
improve transport links or build local schools?
7.4.3 Estimating returns to education

The variation in years of schooling generated by INPRES could potentially
be used to estimate returns to education. This estimate is interesting
because it is a more general parameter than the programme impact. It is,
for example, useful to know how much an extra year of schooling is worth
when designing education policy. The following activity shows why a
simple OLS regression of wages on years of schooling is unlikely to give an
unbiased estimate of returns to education.
Activity 7.9
Suppose we have observational data for a cross section of adults living in a given
country. The data provides information on years of schooling, wages and other individual
characteristics. Why do you think running a regression like (6) does not provide an
unbiased estimate for returns to education (β)? What issues could result in endogeneity
of years of schooling in this regression?
log wagei = α + β years_of_schoolingi + γcontrolsi + ∈i (6)
Duflo looks at the impact of the INPRES programme on wages using the
same difference-in-differences strategy. Young cohorts received 2.6 per
cent higher wages due to the INPRES programme. Note, this estimate is
only valid if the parallel trends assumption is satisfied. Panel B in Tables
3 and 4 provide evidence in support of parallel trends by showing the old
cohorts do not show a differential increase in wages across high and low
intensity regions.
In order to estimate returns to education Duflo runs a regression similar to
(6) but uses exposure to INPRES as an instrument for years of schooling.
For this to be a valid estimate the exclusion restriction must be satisfied.
In other words, INPRES should only affect wages through education and
not through direct or indirect channels. Table 7 in Duflo’s article reports
the instrumental variable estimates. An additional year of schooling
increases wages by around 6.8 to 10 per cent. This estimate is in the range
of estimates found for developed countries and could help policy makers
assess the benefits of education policies.
94
Activity 7.10
Do you think the exclusion restriction is satisfied here? Could INPRES directly affect
adulthood wages? What are potential channels (other than years of schooling) that
INPRES could affect adulthood wages?
► Stop and read

Duflo (2001).
7.5 Teachers’ incentives

Learning does not naturally occur when teachers are recruited and
students are enrolled. Teachers need to show up in classes and actually
teach what they have prepared for students. Students need to be in class
and pay attention. Tuition support at school and parental help at home
are crucial for overcoming learning difficulties. The evidence shows that
teachers and students in developing countries are frequently absent
(Banerjee and Duflo, 2006, and Chaudhury et al., 2006). Teachers’
absence or lack of effort in class lowers the quality of education (i.e. 10- or
11-year olds not being able to read a simple story) and reduces returns to
education, leading to lower levels of schooling.
Teacher incentives might be an important factor in explaining absence.
Governments recruit teachers to teach in classes but lack the ability
to perfectly monitor what they do. Therefore we have a classic case of
principle-agent problem. The principle (the government) tries to raise
education standards by hiring agents (teachers) but the act of teaching
requires costly effort. In the absence of appropriate incentives, teachers
slack off. Fixed salaries and permanent positions neither reward good
teaching nor punish slackers.
Designing incentive structures is not easy. Consider a performance pay
scheme with a fixed base salary and a bonus that depends on teacher
performance. Several questions need to be addressed. Are end-of-year test
scores good measures of teacher performance? Should the government
base the reward on quantifiable measures like presence in class or more
qualitative measures like type of activities carried out in class? Is leaving
the monitoring to headteachers a good idea? Should the reward be given
to schools based on average performance or should individual teachers
be given incentives? What happens to teachers who are passionate
about their jobs (intrinsically motivated) after we introduce pay for
performance? If the reward is based on test scores, what happens to
harder-to-measure aspects of learning? These are important questions
that the recent empirical literature on incentive structures in the provision
of public services has started to investigate. We briefly review a study by
Duflo et al. (2012) out of many others.
Activity 7.11
In this activity we think about multi-tasking issues in the design of incentive structures.
Teachers carry out different tasks in class with different learning outcomes. For example,
group work could enhance the ability of students to work in teams (a useful life skill
and also a productive asset) but this task may not translate into immediate gains in
test scores. Assume we designed a pay-for-performance system where teachers are
rewarded for higher average test scores. How might this design undermine the purpose of
improving the quality of education?
95
Duflo et al. (2012), in collaboration with Seva Mandir, an NGO running

single-teacher schools in rural Rajasthan, India, designed an experiment
to estimate the impact of financial incentives on teacher performance.
Despite constant reminders from the NGO and a potential threat of
dismissal, 35 per cent of teachers were absent prior to the intervention.
They randomly assigned 57 schools to treatment and 56 schools to control
group. Teachers in the control group received the normal fixed salary
of 1,000 Rs. The treatment schools were given a tamper-proof camera
and asked to take a photo of the teacher with students at the beginning
and end of classes each day. The time and dates on the camera role were
used to calculate the number of days teachers were present. Treatment
teachers’ received 500 Rs if they attended school fewer than 10 days in a
given month and 50 Rs for each additional day (therefore, the maximum
salary in treatment schools is 1,500 Rs). The programme reduced teachers’
absence rate by 21 percentage points relative to the control group during
the 30 months after the start of the intervention. Students’ test scores also
increased significantly by 0.17 standard deviation in the treatment group.
Activity 11
Could we argue that the decrease in teacher’s absence rate is solely due to financial
incentives? If not, what are other potential mechanisms for the observed effect?
► Stop and read

Duflo et al. (2012) and Banerjee and Duflo (2006).
7.6 Summary
We started this chapter with a very simple yet useful theoretical model
that provided us with a framework for understanding the role of various
factors in education decision-making. We categorised potential education-
enhancing policies into three broad groups.
First, we discussed factors that could alter demand for schooling by
households. Two randomised experiments show the importance of
credit constraints and perceptions about returns to education. Progresa’s
conditional cash transfer in Mexico resulted in improved students’
attendance, potentially through relaxing credit constraints and the
changing price of education.
Second, we investigated the role of schooling infrastructure in education
decisions. There are many elements of infrastructure that could influence
household decision-making. We looked at the INPRES school construction
project in Indonesia. The programme increased years of schooling
significantly, potentially due to a reduction in travel costs.
Third, we considered the recent literature on the high teacher absence
rates in developing countries. The empirical studies here show that
changing the incentive structure for teachers could increase their effort
and result in improved learning outcomes. The choice of monitoring
technologies and design of the incentive structure are two challenging
issues for a successful intervention.
► Stop and listen to LSE–IGE public lecture on reforming educational systems at:
http://goo.gl/n2sMjT
96

• explain a model of educational choice and discuss the role of relevant
factors
• recognise and outline various issues in improving educational
outcomes in developing countries
• summarise a difference-in-differences estimation and explain its
limitations
• summarise and critically evaluate empirical evidence on policies
promoting education.

1. Consider the model of schooling choice presented in the chapter. How
would you change the utility function to account for the fact that
parents might attach an intrinsic value to more schooling? How does
this change the optimal number of years of schooling?
2. Duflo (2001) suggests that the government of Indonesia initiated
water and sanitation programmes as a result of the oil boom. Does
the presence of these programmes pose a threat to the difference-in-
differences identification strategy? How does Duflo control for these
programmes?
3. Suppose the government of Rajasthan decides to roll out performance
pay for teachers across the province based on the positive effects
found in the Duflo et al. (2012) study. The government, however,
decides to monitor teachers through the school headteachers instead
of using a camera. It, however, provides similar performance pay. Do
you think the government programme will be as effective as the NGO’s
intervention? What are the reasons to be sceptical?

on the VLE.
97
Notes
98
Chapter 8: Health and nutrition
8.1 Introduction
• review the patterns of health care provision in developing countries
• understand the causes of poor health and low take-up of preventive
measures
• explore the links between poverty and poor health.

• identify and analyse potential causes for poor health and low take-up
of preventive care
• evaluate the link between poverty and nutrition
• describe the patterns of health care and provision in developing
countries.

Ray (1998) Chapter 13, pp.489–504.

Ashraf, N., O. Bandiera and K. Jack ‘No margin, no mission? A field experiment
on incentives for pro-social tasks’, CEPR Discussion Paper no. 8834, 2013.
Banerjee, A.V., E. Duflo, R. Glennerster and D. Kothari ‘Improving immunization
coverage in rural India: a clustered randomized controlled evaluation of
immunization campaigns with and without incentives’, Working Paper
2010.
Bjorkman, M. and J. Svensson ‘Power to the people: evidence of a randomized
field experiment on community-based monitoring in Uganda’, Quarterly
Cohen, J., and P. Dupas ‘Free distribution or cost-sharing? Evidence from
a Randomized Malaria Prevention Experiment’, Quarterly Journal of
Economics 125(1) 2010, pp.1–45.
Das, J., J. Hammer and K. Leonard ‘The quality of medical advice in low-
income countries’, Journal of Economic Perspectives 22(2) 2008, pp.93–114.
Qian, N. ‘Missing women and the price of tea in China: the effect of sex-specific
earnings on sex imbalance’, Quarterly Journal of Economics 123(3) 2008,
pp.1251–85.
Subramanian, S. and A. Deaton ‘The demand for food and calories’, Journal of
Political Economy 104(1) 1996, pp.133–62.
8.1.5 Works cited

Behrman, J.R. and A.B. Deolalikar ‘Will developing country nutrition improve
with income? A case study for rural South India’, Journal of Political
Economy 95(3) 1987, pp.492–507.
99
Deaton, A. and J. Drèze ‘Food and nutrition in India: facts and interpretations’,
Economic and political weekly 44(7) 2009, pp.42–65.
8.1.6 Overview
Following our investigation of the determinants of human capital, this
chapter looks at health care provision in developing countries. We try to
think about choices the poor make to invest in their health and assess
whether they are trapped in health-driven poverty.
Health is a multidimensional concept and difficult to measure. Self-
reported indicators are subjective and they may not be comparable across
individuals. Therefore, most studies rely on objective health measures
like weight, height and Body Mass Index (BMI) that are easy to collect
and reflect nutritional intake. Developments in clinical technology have
resulted in it being cheap to measure other objective health conditions like
anaemia and blood pressure.
The situation in developing countries is poor across the board of health
indicators. Diseases like malaria, gastro-intestinal worms, respiratory
infections and other preventable illnesses are widespread. Children are hit
the hardest and each year several millions die of curable conditions like
diarrhoea. This is despite the fact that cheap cures like oral rehydration
solution (ORS) exist but are not used as frequently.
Investment in preventive care is far lower than curative care in developing
countries. Cheap technologies for purifying water, keeping malaria
mosquitoes away and deworming children exist. Research evidence
suggests that even the poorest families could afford to buy a bottle of
chlorine to purify water and prevent diarrhoea and other waterborne
diseases but they choose not to. Despite being free, immunisation rates are
low in developing countries.
In the next section we look at demand for health care and review the
evidence on the use of preventive care. Section 8.3 looks at the link
between income and demand for nutrition. Section 8.4 considers poor
provision and weak incentives for health providers as a source of poor
health outcomes. Here, we review two articles with an eye toward policy
making. In Section 8.5 we consider the distribution of resources within the
household and present empirical evidence on discrimination against girls
in developing countries.
► Stop and read

8.2 Demand for health

Similar to the model we have seen in the previous chapter, we can write
a simple optimisation problem to find demand for investments in health.
We can broadly divide health investments into preventive and curative
measures. Good diet (calorie and micronutrients), exercise, immunisation,
clean water, protection from malaria mosquitoes, clean environment,
sewerage and toilets are just a few examples of preventive investments
that improve individuals’ health. Medical advice, appropriate diagnosis
and adequate prescription are examples of curative investments.
Rational individuals weigh expected costs and benefits of investments in
health and choose how much of various measures to consume. Private
benefits from improved health include net present value of increased
earnings due to higher productivity, longer lifetime, avoided pain
100
and generally a higher quality of material life. On the cost side, health
investments require out-of-pocket spending, abstaining from enjoyable yet
damaging habits, and time cost of visiting a doctor or nurse.
Many complications affect demand for health. First, there are potentially
large positive externalities from improved health. Protecting an individual
from contagious diseases reduces the risk of catching these illnesses for
everyone in the neighbourhood. Presence of health externalities is the
classic case of a market failure that justifies government intervention (e.g.
free immunisation).
The second feature of health markets is imperfect information. We usually
think the benefits of immunisation or clean water are obvious but this many
not be the case. Individuals might not be aware of critical factors affecting
their health and underestimate the value of preventive investments.
Another important aspect of imperfect information is the asymmetry in
knowledge about the appropriate course of treatment. Patients do not know
what is good for them and rely on a doctor’s diagnosis. The evidence from
developing countries suggests some doctors check patients superficially and
even write prescriptions without looking at the patient (Das et al., 2008).
The direct result of this is the ineffectiveness of the health care system.
Indirectly, poor practices could build mistrust and result in low demand.
The third issue affecting health demand in developing countries is credit
constraints. Low income households may lack the funds required for
buying health services. Research evidence does not, however, favour
this mechanism. In the next section, we will see some evidence that
poor households choose to spend less than what is required on calories.
Furthermore, investment in preventive health care is not costly (e.g. free
immunisation), yet full immunisation is not very common (Banerjee et. al.,
2010).
The fourth set of factors potentially with high explanatory power is
behavioural biases. Individual decision-making is usually based on
heuristics, beliefs and recommendations from friends and family. It might be
that individuals do not believe immunisation has any benefits. Correcting
these beliefs may prove a challenging task.
Activity 8.1
One important feature of investments in health is that costs are paid at the present time
while benefits are accrued in future. Could this feature explain low demand for preventive
measures? How exactly?
In the remaining of this section we look at two randomised experiments to

understand ways of overcoming low take up of preventive health services.
8.2.1 Malaria bed nets

Many health investments have positive externalities that justify government
subsidies. Sometimes the positive externalities are so large that free
distribution is the best option (e.g. for immunisation). Bed nets for the
protection against Malaria are, however, different. After acquisition
individuals need to use bed nets in order for the benefits to materialise.
Classical economic theory suggests that usage should only be decided on
marginal costs and benefits. The price paid for an item is sunk and should
not play any role at the usage stage. Individual behaviour, however, shows
signs of sunk cost fallacy. People have higher usage of the things they have
paid more for. Furthermore, a zero price or free distribution might signal low
quality and ineffectiveness of the product that reduces subsequent usage.
101
Activity 8.2
A positive price for bed nets has selection effects too. How would charging a positive
price impact on the pool of customers buying bed nets (i.e. take-up)?
Cohen and Dupas (2010) designed an experiment to empirically evaluate

these claims. Insecticide-treated bed nets (ITNs) are proved to be
effective in reducing malaria infection and maternal anaemia but they
are expensive. Therefore it is important to understand the optimal level
of subsidy that maximises the benefits. Cohen and Dupas randomised the
offer price of ITNs in 20 prenatal clinics in Kenya. Four clinics served as
control, five clinics distributed ITNs free of charge, five charged a price
of 10 Kenyan Shillings (Ksh), three charged a price of 20 Ksh and three
charged a price of 40 Ksh. The highest price corresponds to 90 per cent
subsidy and is still 10 Ksh below the market price.
The results of the Cohen and Dupas study show that take-up of ITNs falls
by 60 percentage points when the price is increased from zero to 40 Ksh.
The usage, however, did not show a significant change for groups paying
higher prices. Therefore it seems demand is very responsive to price but
once ITNs are acquired, the usage does not depend on the price. Given
large private and social gains from ITNs they suggest free distribution of
ITNs make sense.
Activity 8.3
Cohen and Dupas (2010) use a two-stage randomised design to separate the selection
effect of a given price from the sunk cost effect. For a random set of women who decided
to buy ITNs at the clinics, they offered a lower price than the one posted. They argue that
comparing usage levels of these women with those who did not receive a discount would
tell us about the sunk cost effect on the usage of ITN. Explain carefully why a single stage
design cannot achieve the same thing and how the two stage design separates the two
effects.
8.2.2 Immunisation camps

The rate of full immunisation in India is particularly low. Banerjee et
al. (2010) designed a randomised experiment to shed light on factors
preventing households from having their children immunised. They
randomly assigned 134 villages in rural Rajasthan to three groups. Thirty
villages were assigned to treatment A where an NGO set up reliable
immunisation camps (a nurse visited the village on the same day each
month and offered immunisation); 30 villages were assigned treatment
B and received both the reliable camp and 1kg of lentils for each
immunisation visit and a set of plates for completing the immunisation
process. The remaining 74 villages received no intervention.
The results show 39 per cent of children aged 1–3 received full
immunisation in treatment B villages but only 18 per cent did so in
treatment A. The control group average was 6 per cent. The results show
that increasing the reliability of immunisation centres is important in
encouraging take-up but still the majority of children remain vulnerable.
Offering a small incentive increases take up drastically, although full
coverage is not achieved.
Activity 8.4
Why do think offering such a small incentive (1 kg lentils and set of plates) results in a 21
percentage point increase in full immunisation rates? A closer look at the results reveals
that about 80 per cent of children in treatment B villages received at least one injection
102
(out of five for full immunisation) which is almost similar to the treatment A villages. But
the fraction of children receiving more injections falls much more rapidly in treatment
A compared to treatment B. Why do you think parents do not bring in children for more
injections when they have done it for the first couple of times? Why is the situation in
treatment B villages different?
Activity 8.5
What are potential policy implications of this study? Why do a large percentage of
children remain not fully immunised in the treatment group? Could you think of an
experimental design to investigate other factors holding back immunisation rates in
Rajasthan?
8.3 Nutrition and income

One critical investment in health is through nutritional intake. Anaemia
(iron deficiency), insufficient calorie and micronutrients have damaging
effects on health and reduce workers’ productivity. This in turn could
lower wages and increase medical expenses leaving less to be spent on
nutritious food. This logic shows there might be health-driven poverty
traps. Poor individuals cannot invest in their nutritional intake which
lowers their income and further restricts their ability to afford nutrients.
Activity 8.6
Could borrowing money for a short period remove health driven poverty traps? Do you
think the poor can borrow enough money to overcome the poverty trap?
The critical question we investigate in this section is whether nutritional

intake increases as people become richer. There is certainly a positive
correlation across countries. Richer countries have higher average
caloric intake, and less prevalence of illnesses related to deficiency of
micronutrients (e.g. anaemia). But do the poor in developing countries get
enough nutrients once they become better off? In other words, how does
calorie intake change as individuals become richer in a given country?
It seems reasonable to assume that with rising incomes people will
purchase more calories. However, Behrman and Deolalikar (1987)
show that in rural South India nutritional intake does not improve upon
increases in income. This is known as the revisionist view. A weak (or
zero) relationship between income and nutrition has significant policy
implications. Policies supporting income generation would not reduce
malnutrition and policy makers need to target nutrition directly (e.g.
through school meals, price subsidies and nutritional education).
Activity 8.7
How does the revisionist position impact on our understanding of income as a good proxy
for wellbeing?
In the end the relationship between income and nutrients is an empirical

question. Subramanian and Deaton (1996) analysed a sample of 5,630
rural households from the state of Maharashtra in India using the
National Sample Survey (NSS). It asks respondents to recall consumption
(expenditure and quantity) of more than 300 items in the past 30 days.
They use calorie tables to convert physical quantities of various items into
calorie availability (for example, 1kg of potatoes has 770 calories).
Poor households in developing countries spend a large fraction of their
income on food. In the NSS sample used here, the average share of food
103
of total expenditure is 67 per cent. Subramanian and Deaton (1996)

estimate food expenditure increases by 0.75 per cent when income (total
expenditure) increases by 1 per cent. But available calories do not increase
as much. They estimate a 1 per cent increase in income raises available
calories by 0.36 per cent.
Activity 8.8
Why did Subramanian and Deaton (1996) look at calorie availability rather than calorie
intake? Does it matter which measure we use?
Do these estimates match up? A 1 per cent increase in income increases

food expenditure by 0.75 but calorie intake goes up only by 0.36 per cent
(i.e. 0.75–0.36 per cent seems to be missing). Substitution away from
poor quality food to better tasting food items is responsible for the wedge
between elasticity of food expenditure and available calories with respect
to income. Subramanian and Deaton (1996) show that as households earn
more, they tend to move from cereals with high caloric content to other
food categories like meat, fruits, vegetables and dairy. The better tasting
food items are not as rich in calorie as cereals are, and they are more
expensive as well. Therefore, the per calorie price is much higher. While
households can buy 1,000 calories with 0.64 Rs spent on cereals, they
need 11.7 Rs to buy a similar number of calories from meat.
The overall conclusion from Subramanian and Deaton (1996) is far from
the revisionist position. Calorie intake increases with income but not as
much as we would expect, possibly due to a substitution towards better
tasting food items.
Activity 8.9
You might expect the relation between calorie and income to be nonlinear. Poor
households might spend any additional income to buy more calories but as income
increases and caloric needs are satisfied, the relationship declines. Therefore, we expect
an S-shaped relationship between calorie and income. Do Subramanian and Deaton
(1996) find any evidence of this S-shape? Why?
Activity 8.10
Deaton and Drèze (2009) show a striking result. Although real GDP per capita in India
has grown at an average rate of 3.95 per cent between 1980 and 2005, per capita
calorie intake is declining. Similar patterns are observed for other nutrients and only per
capita consumption of fats is rising. How can you reconcile this with the evidence from
Subramanian and Deaton (1996)? Is it possible that the caloric needs of Indians is falling
as well? What factors could explain this declining pattern?
► Stop and read

Ray (1998) Chapter 8, pp.261–266 (section 8.3.4 on nutrition), pp.272–79 (section
8.4.2) and Chapter 13, pp.489–504 (section 13.4).
8.4 Provider incentives and health infrastructure

The government is the main player in the provision of health services
in developing countries. It seems that although physical infrastructures
have improved greatly over the past decades (Das et al., 2008), facilities
are not running efficiently. Chaudhury et al. (2006) found that 40 per
cent of primary health care providers were absent when they made an
unannounced visit in Indonesia and India. The absence rate is 35 per cent
in Bangladesh and 37 per cent in Uganda. These numbers are staggering.
104
How would a poor family feel if four times out of 10 no one could be
found at the health centre they brought their sick child to?
This evidence is consistent with what you have seen in the education topic.
Health care workers with secure positions and fixed incomes are unlikely
to put much effort into their work. The question is: what sort of incentive
structures and monitoring strategies could induce better performance? In
what follows we look at two randomised experiments on these questions.
► Stop and read

Chaudhury et al. (2006).
8.4.1 Health workers’ incentives

Health workers are involved in pro-social activities, that is to say, actions
that not only benefit the parties in the transaction but also the wider
society. Individuals with higher intrinsic motivation might prefer to
join the health sector precisely because they care more about society.
Therefore, individuals in the health sector might be different from others
in the private sector. A recent study by Ashraf, Bandiera and Jack tries to
tackle the question of optimal incentive structure for individuals involved
in pro-social activities.
Apart from financial incentives two other methods can be used to increase
effort by intrinsically motivated individuals. First, they can be given a
stake in the success of the organisation and made to feel they are part of
the system and are making a difference to the lives of the others. Second,
attempts can be made to promote their intrinsic motivation. Ashraf et
al. (2013) designed a randomised experiment in collaboration with the
Society for Family Health (SFH) in Zambia. The agents are hairdressers
and barbers who are hired to distribute female condoms to their customers
as a sideline. HIV prevalence in Zambia is 14 per cent of the adult
population and 42 per cent of hairdressers talked about HIV with their
customers before the experiment began. Agents are well aware of the
pro-social component of selling condoms because of a large information
campaign by the Ministry of Health.
Agents were randomly assigned to four groups. Control group agents were
hired on a voluntary basis. The first treatment provided a small financial
reward (10 per cent of the retail price), while the second gave a large
financial reward (90 per cent of the retail price) for sales of condoms. In
the third treatment, agents receive a star for each condom pack sold and
were to be invited to a ceremony if their sales exceeded 216 packs. The
non-financial reward group received a wall chart showing the number
of condoms sold (i.e. stars). Monitoring the sales was straightforward as
restocking requests were made to SFH.
The results of Ashraf et al.’s (2013) experiment reveals the non-financial
incentive treatment (stars) sold twice as many packs as the other groups.
Interestingly, the financial incentives did not have a significant effect on
sales. In terms of policy implication the results suggest that the design of
incentive schemes in activities that involve a pro-social component (e.g.
health and education) should not solely be based on monetary rewards.
In fact, non-pecuniary rewards might be much more effective. Before
generalising these recommendations to other contexts you should note
that the agents here had a private job and the pro-social task was only a
side business. Furthermore, even the largest financial reward was small
(one seventh of the price of a haircut).
105
Activity 8.11
The fact that the star treatment in Ashraf et al. (2013) outperformed the other groups
could be due both to leveraging the intrinsic motivation of pro-social agents and hyped
demand because of the display of the star chart in the salon. Think about an experiment
that could separate these two effects.
8.4.2 Community-based monitoring

Lack of accountability and monitoring could be responsible for health
workers’ poor performance. Involving the local community in the
process could improve outcomes because these individuals are the main
beneficiaries of the health facilities and they have ample opportunity
to monitor facilities (because they are local). However, a community’s
expectations might have been adjusted to the existing poor performance.
They do not expect much because they do not have the expectation
that facilities could run better. Even if the community thinks facilities
are running poorly, there might be a coordination failure in organising
members to act effectively and correct misbehaviour.
Bjorkman and Svensson (2009) in collaboration with a local NGO in
Uganda designed a randomised experiment to test whether community-
based monitoring could increase accountability and health outcomes. They
surveyed 50 facilities and randomly assigned 25 facilities to a treatment
group. In the treatment group the NGO held meetings with the community
and informed them about what they should expect from the facility and
the type of services they could expect to receive. In the meeting the NGO
also helped the community to organise their monitoring efforts.
The treated facilities showed signs of improvement. While no control
facility had a suggestions box, 36 per cent of the treatment facilities had
one. Use of numbered waiting cards was 16 per cent higher in treated
facilities and other facility level characteristics showed signs of improved
accountability and engagement. This resulted in great improvement in the
running of the facilities, utilisation of services and health outcomes. For
example, absence rates fell by 13 percentage points in treated facilities,
utilisation of outpatient services increased by 20 per cent, and under fives
mortality fell by 33 per cent.
Activity 12
The results of Bjorkman and Svensson’s (2009) experiment show great potential for
community involvement in improving health facilities. Can we argue that improvements
in treated health facilities are due to better community monitoring? Why or why not?
Does the distinction matter for making policy recommendations? Think carefully about the
treatment and what it does.
8.5 Within-household health inequalities

There is a large body of literature documenting gender imbalances in
developing countries. In most developed countries the male-to-female ratio
is close to one. While 50.1 per cent of the current population in Western
European countries is female, this ratio is 48.4 per cent in China and India.
There are simply many missing women (Qian, 2008). These imbalances
might reflect sex-selected abortion practices, neglect or underinvestment
in girls’ health, and other factors.
Normally we think of a household as a unit of analysis and ignore
intra-household allocation of resources. Members of a family are
either assumed to have the same preferences or that they pool all their
106
incomes. The household head or everyone together decide on the optimal

consumption bundle. In this classic setting it does not matter who earns
the income and the household problem is very similar to an individual’s
utility maximisation. This is the unitary model of household decision
making. An alternative perspective views distribution of resources as a
bargaining process whereby the share of each member depends on their
bargaining power. This is the non-unitary model of the household. Here it
matters who earns the income potentially because the earner has higher
bargaining power.
Qian (2008) presents evidence consistent with the non-unitary model
of decision making. She studied a period of agricultural reform in China
when there were increased wages for female workers in tea producing
areas and for male workers in fruit producing areas. Girls’ survival rates in
tea producing areas increased significantly after the reform was enacted
compared with areas not affected. Similarly, the increase in males’ income
in fruit producing areas worsened girls’ survival rates after the reform.
Furthermore, the findings suggest when mothers bring more income to
the household, the educational attainment of both genders improves, and
when fathers earn more income boys’ education is not affected while girls
receive less education. Qian’s results suggest that empowering women
is a potential remedy for gender imbalances in developing countries, for
example government cash transfers directed to women.
8.6 Summary
We started this chapter by thinking about potential factors that influence
demand for investments in health. It seems that the poor do not invest
as much as they should in preventive care. This is despite availability
of cheap technologies for purifying water or deworming children and
substantial subsidies on products like bed nets. We explored several
articles to better understand why take up of health technologies is low. It
seems even small positive prices discourage demand drastically (bed nets)
and small rewards encourage take-up significantly (immunisation).
We next started to explore the idea of a health induced poverty trap. The
works of Angus Deaton and co-authors show puzzling patterns in the
Indian nutritional intake. It seems that not even the poorest households
are trapped in a health driven poverty trap, as it is very cheap to acquire
the calories needed for working. Households, however, tend to buy
tastier food as they earn more income instead of investing in more
nutritious items. These studies reject the revisionist position that there
is no relation between income and calorie intake but still leave plenty of
room for food information campaigns, school meal programmes and food
supplementation facilities.
Our study of health markets continued by discussion of the critical issue
of health provision by doctors and nurses. We reviewed the evidence
on high absence rates among health workers and saw that improving
accountability through community empowerment programmes could
substantially improve the running of health centres. We also considered
the special circumstances of health workers. The study of Ashraf et al.
(2013) suggests that there might be a role for non-monetary incentive
schemes in improving health services.
We concluded the chapter by a brief look at the issue of gender
imbalances. Qian’s (2008) study suggests allocation of resources within
the household might be similar to a bargaining process where the
individual with more earning capacity is likely to have a greater say.
107
► Stop and read

Ray (1998) Chapter 8, pp.279–90 (subsection 8.4.3 till the end of the chapter).

• identify and analyse potential causes for poor health and low take up
of preventive care
• evaluate the link between poverty and nutrition
• describe the patterns of health care and provision in developing
countries.

1. Why might free distribution of deworming pills to parents be a worse
policy option than administering them directly to schoolchildren? In
answering the question, think about parents’ actions that could affect
the benefits of each policy. You may also want to use the fact that
worms are usually transmitted when children swim in contaminated
water or step in human excrement.
2. Subramanian and Deaton (1996) measured calorie intake from
physical quantities of various items reported in NSS data. Total
expenditure was also found by adding up reported expenditure.
Assume households do not have perfect recall and make reporting
errors. How does this measurement error biases the coefficient
estimate of expenditure in a regression of calorie availability on
expenditure? Is it an upward or a downward bias?

on the VLE.
108
Chapter 9: Land and agriculture
9.1 Introduction
• discuss the role of agriculture in developing countries
• explain land contracts and the implied incentive structure
• understand the principal-agent model
• assess the evidence on one land reform programme.

• list key facts about the importance of agriculture in developing
countries
• identify and explain different land contracts and the implied incentive
effects
• analyse and explain a principal-agent model
• critically evaluate empirical evidence on the impact of land reform.

Ray (1998) Chapters 11 and 12.
Banerjee, A., P. Gertler and M. Ghatak ‘Empowerment and efficiency: tenancy
reform in West Bengal’, Journal of Political Economy 110(2) 2002,
pp.239–80.

Bardhan, P. and D. Mookherjee ‘Determinants of redistributive politics:
an empirical analysis of land reforms in West Bengal, India’, American
Besley, T. and R. Burgess ‘Land reform, poverty reduction, and growth: evidence
from India’, Quarterly Journal of Economics 115(2) 2000, pp.389–430.
Lin, J.Y. ‘Rural reforms and agricultural growth in China’, American Economic
Review 82(1) 1992, pp.34–51.
9.1.5 Overview
A large share of the population in developing countries relies on
agriculture for their living. This chapter describes the main features of
agricultural markets in developing countries, emphasising the role of land
holdings and the potential impact of land reforms.
9.2 Why agriculture?

The livelihoods of many individuals in developing countries depend on
agriculture. Figure 9.1 shows the share of agricultural workers in total
employment is negatively correlated with income. Even in 2010, around
50 per cent of the workforce in India was employed in agriculture. The
share of agriculture in GDP and employment decreases as countries
develop. This concurs with the idea of industrialisation we discuss in
section 4.4.2. Although Figure 9.2 confirms the declining pattern in China
109
and India since 1980, still the lives of many depend on agriculture in both
countries.
60
Agriculture labour % of employment in 2010
0 20 40
Fitted line
6 7 8 9 10 11
GDP per capita in 2010 (PPP. constant 2005 international $)
Figure 9.1: Share of agriculture in employment and GDP per capita in 2010.
Activity 9.1
Based on Figure 9.1, could we argue that a high share of agricultural workers causes
lower GDP per capita? Why?
Share of agriculture in employment and GDP

80
60
40
20
0
1980 1990 year 2000 2010
China: labour share India: labour share

China: GDP share India: GDP share
Figure 9.2: Share of agriculture in employment and GDP in China and India.
► Stop and read

9.3 Land
Land is an important input in agriculture. Countries vary greatly in
landholding inequality. Latin American countries historically had large
plantations and currently feature high land inequality. Asian countries
have had a more equal land distribution and more small-scale farming. But
does the inequality of landholding matter?
Land inequality matters from two perspectives. First, land is an asset and
holdings reflect wealth inequality. To the extent that reducing inequality is
110
a government’s objective, it cares about land inequality. The second issue is

that land inequality has many implications for efficiency. There are several
reasons to expect higher productivity from larger farms relative to the
smaller ones. Farming requires machinery and facilities with fixed costs
(tractors, water pumps, bullocks, etc.). Larger farms have better access to
credit, making it easier to invest in such technologies. Furthermore, in the
presence of an efficient rental market you would expect better farmers to
acquire more land because they can generate more value. This results in
higher productivity of large holdings because of their higher agricultural
abilities.
The empirical evidence from developing countries, however, shows the
opposite: small farms have higher productivities. Furthermore, owner-
cultivated farms show higher productivities compared to sharecropped
farms. To understand this puzzle we need to think about the way in
which large and small farms operate. Small farms are usually cultivated
by owners while large farms use hired labour. The incentive structure for
these groups is very different. This opens up the debate that perhaps land
redistribution is an effective means of achieving equity and promoting
efficiency. Before looking at the issue of land reform, we discuss various
forms of agricultural contracts and their implied incentive structures in the
next section.
9.4 Tenancy contracts

Landowners could cultivate their own land or transfer their cultivation
rights to others through a variety of contractual forms. We consider three
of the most common agriculture contracts: wage contract, fixed rent and
sharecropping. Cultivators need to expend timely effort in combining a
variety of inputs to get a good harvest. Failure to use the right amount
of inputs at the right time results in lower production. Ploughing, tilling,
sowing, manuring and applying fertilisers and pesticides are a few
examples of tasks that need to be carried out at the right time. Effort and
diligence is, however, non-contractible because they are not perfectly
observable. Therefore, contracts are either unconditional or based on
observable output.
Activity 9.2
Assume the only input was effort and that output was a deterministic function of input. Is
effort directly observable? Could you still argue effort is non-contractible?
Activity 9.3
Are weather shocks really unobservable (e.g. rainfall during the harvest season)? Is the
farmer’s response to weather shocks observable? What other unobservable conditions
could you think of that affect agricultural output?
In a fixed rent contract the owner receives a fixed income regardless of

output fluctuations. If the crops fail or flourish due to weather conditions
the owner still receives his predetermined rent. On the other hand,
a farmer’s income depends on the weather and moves accordingly. A
sharecropping agreement specifies the share of output that goes to the
tenant and the landlord (e.g. 50–50). Therefore, landlords’ and tenants’
incomes depend on the crop yields and they both bear some of the risks.
A wage contract does not share any benefit (or loss) of the yield with
workers as they get a fixed salary, while the landlords receive all the gains
(or losses) from the harvest.
111
Sharecropping is the most commonly used form of tenancy in India,

but does economic efficiency also justifies its prevalence? What are the
incentives for the tenants to exert any effort under each of the three
contracts we discussed? In order to understand better the incentive
structure under various contractual forms, we start by outlining a simple
model.
9.4.1 A principal–agent problem

We can think of tenants as agents recruited by the landlord to work on the
farm. In this view, the landlord is the principal who offers a remuneration
scheme to the agent. The agent’s effort influences output but so do other
stochastic factors. The landlord tries to get more output by providing
incentives to the tenant to invest the optimal level of effort through
the contract structure. Effort is, however, costly for the agents and they
maximise their benefits given the contract structure. This is an example of
a classic principal–agent problem in economics. We have already discussed
the issue of teacher and health worker incentives in developing countries
and how better monitoring and remuneration schemes could change
behaviour. Here we take a closer look at this problem through a simple
mathematical model.
A tenant farms a piece of land, exerting effort e which is unobservable
to the landlord. The effort costs the tenant c(e) (you can think of this as
utility or monetary costs). Output takes only two values and could be low
or high. We normalise the low level of output to 0 and denote the high
level by H. The randomness of output implicitly reflects the presence of
shocks like sporadic rainfall not controlled by the tenant. Higher effort
increases the probability of having a better output. Specifically high output
happens with probability e (equal to the exerted effort) and low output
with probability 1 – e (note this means effort is changing between 0 and
1).
The tenant is compensated with payments from the landlord. Effort is not
observable and therefore the contract could only condition on output.
In this simple setting output takes only two values, therefore the most
general contract looks like (l,h) where l and h are payments to the tenant
when output is low (0) and high (H) respectively. Under a wage contract
the tenant receives a fixed wage l = h = w regardless of output level. With
a sharecropping contract the tenant gets a share of output and therefore
l = α · 0 = 0 and h = α · H (α is the fraction of output going to the tenant).
A fixed rent contract takes the form of l = – r and h = H – r, where r is the
fixed rent and shows up as negative because it is paid by the tenant to the
landlord.
Let us think about the socially optimal level of effort, that is to say, the
level that maximises total surplus (output minus the cost of effort). This
is known as the first best level of effort. When the land is cultivated by
the owner, output goes to the owner and he incurs the cost of effort too,
therefore owner-cultivated land achieves the optimal level of effort in
this model. The owner cultivator chooses the level of effort to solve the
following maximisation problem:
max e · H + (1 − e) · 0 − c(e)
e[0,1]
To make matters straightforward, let us make a further assumption and

model the cost of effort as c(e) = 12 ce2. Differentiation gives the following
first order condition for optimisation
H
e* = c (1)
112
This is the level of effort that results in the highest expected social return.
Higher potential gains (H) encourage the cultivator to exert more effort,
while a higher cost of effort (c) works in the opposite direction.
Activity 9.4
Could e be greater than one in this model? What is the assumption we need so that the
optimal effort is actually the one we found in (1)?
Let us now think about the optimal contract from the landlord’s
perspective. We start by looking at the tenant’s behaviour given the
contract offered. We then look at the landlord’s problem in picking the
right payments to maximise profits.
In the landlord-tenant model, the tenant cares about output to the extent
that it affects his remuneration. More precisely, the tenant chooses the
level of effort to maximise his own payoff given as follows
1
max e · h + (1 − e) · l − 2 ce2
e[0,1]
Differentiation gives the tenant’s chosen level of effort

h−l
e˜ = c (2)
Condition (2) is referred to as the incentive compatibility constraint (IC).
IC states the level of effort that will be chosen by the tenant given the
contractual payments. The condition matters because the tenant cannot
be forced to exert a certain level of unobservable effort. The only way the
landlord can influence the effort exerted is through the reward structure.
Equation (2) shows that the larger the difference between the payments
in the high and low state, the higher the level of effort. The following
exercise asks you to show that the fixed rent contract is the only one that
achieves the first best level of effort here.
Activity 9.5
Use (2) and verify that sharecropping and a fixed wage contract deliver a level of effort
strictly less than the first best (1) but a fixed rent contract achieves the first best.
The landlord maximises net expected payoff, acknowledging the fact that
he cannot influence the level of effort other than through the terms of the
contract (IC condition). He also needs to provide an interesting enough
offer. Otherwise the tenant would walk away and choose a different
activity. This is known as participation constraint (PC).
Let us assume the landlord is only interested in a fixed rent contract and
find the rent he is likely to offer. We assume the tenant’s best outside
option delivers u, therefore his utility must be at least as large as u (PC).
Using the fact that the fixed rent contract delivers the first best level
of effort, the IC condition is equivalent to setting e = e*. Therefore the
landlord maximises his net expected payoff as follows
max r
( ) ()
r 2
H H 1 H
s.t. c (H − r) + 1 − c (−r) − c c ≥ u
2
The objective function is increasing the rent while the left-hand side
of the constraint is decreasing in the rent. In other words, the landlord
benefits from a higher rent but the tenant is hurt. Therefore, the landlord
sets a rent that leaves the tenant indifferent between accepting the rental
contract and his outside option. In other words, the solution to the above
problem is found from the PC under equality. A little bit of algebra shows
113
H2
r* = −u
2c
This is a very intuitive result. The optimal rent that could be charged upon
the tenant is equal to the total expected value from cultivation ( Hc ) minus
2
H2
the cost of effort ( 2c ) and the outside option (u). In other words, the
tenant is only compensated for the cost of effort and his outside option.
Despite the fact that the fixed rent contract gives the efficient outcome,
sharecropping is the one most common in India. We discuss two potential
extensions of this model that could change the optimality of fixed
rent contract. You should, however, make yourself familiar with other
extensions in Ray (1998) Chapter 12.
9.4.2 Risk sharing

The model we just discussed assumes that both the landlord and the
tenant are risk neutral. In other words, they only care about the expected
payoffs. What would happen if either or both landlord and tenant are risk
averse? Risk averse individuals prefer a certain stream of income to a risky
one with the same expected value. The three contracts have different risk
implications.
The fixed rent contract involves the highest degree of income risk for the
tenant as he keeps all output. The landlord, however, receives a fixed rent
and does not bear any risk. A risk averse tenant may prefer contracts that
remove part of this risk. For example, in a sharecropping contract both
parties share the output. When the harvest is bad the tenant does not
need to pay out-of-pocket rent. This reduces a tenant’s income volatility
compared to the fixed rent case. There is, however, some chance of having
a low output and hence a low payoff. The wage contract shifts output risk
to the landlord and insures the tenant (worker) against any adverse shock.
Regardless of the quality of the harvest, the workers are compensated with
their wages.
Activity 9.6
Do you think poor tenants are more risk averse? Why? Given your answer, what type of
individuals might accept a wage contract rather than a sharecropping one?
The difference in risk implications of the contracts offers a potential

explanation for the popularity of sharecropping. If tenants are generally
poor and risk averse, they are willing to accept a lower expected income
that involves less risk (e.g. sharecropping over fixed rent). On the other
side, if landlords are rich and risk neutral, they are willing to provide
insurance for the tenant upon receiving a higher expected return.
9.4.3 Limited liability

So far we have assumed that the tenant is always able to pay the agreed
rent. But in reality if tenants do not have enough assets and lack other
sources of income they will not be able to pay a fixed rent. In the case of
crop failure they are left with no money and default on payments. The
inability of a tenant to respect the agreed rent due to lack of assets is
known as limited liability.
Let us revisit our simple model and think about the implication of limited
liability for the optimal contract. Assume the tenant’s wealth is zero. In
case of crop failure the tenant will not be able to pay anything. The fixed
rent contract sets l = –r < 0 which violates the limited liability condition
and therefore it cannot be a solution to the landlord’s optimisation.
114
Sharecropping is still possible because the payments are made out of the
output and the limited liability condition does not matter. This is another
reason why sharecropping is more popular than fixed rent contracts.
Activity 9.7
What would happen if there is a risk of landlord’s default? Specifically, is it still possible
to have a fixed wage contract? How about a sharecropping contract?
► Stop and read

9.5 Land reform

We have seen that large farms potentially could have lower productivity
due to incentive problems. Land redistribution could therefore both
increase equity and efficiency. There are of course theoretical reasons that
result in a trade-off between equity and efficiency in land redistribution
(e.g. fixed technology costs mentioned earlier). Some crops (e.g. sugar)
might be produced more efficiently in large plantations, while others could
be produced as efficiently in small farms. In the end, empirical work is
needed to establish whether there is a trade-off. In this section we start by
looking at the theoretical foundations of land reform and the difficulties
of implementing it. Then we consider a case of tenancy reform in West
Bengal.
9.5.1 Does land redistribution make sense?

If a government is interested in reducing inequality, does it make sense
to use land redistribution to achieve this goal? Is it not better to use
progressive tax and transfers to achieve a more equal society? Land is
different from income because it is immovable and visible. Therefore,
behavioural responses to land redistribution might be weaker. Income tax
distorts labour supply (we will see more of this in Chapter 15 on tax) and
restricts the extent of redistribution that could be achieved. Furthermore
an effective transfer system also requires identification of welfare
recipients. Landholdings could work as a tagging device too. Those with
large holdings are likely to be well-off and redistribution from these to the
individuals with little or no land reduces inequality.
The micro evidence in this chapter on low productivity of large
landholdings suggests there might be positive efficiency gains from
reducing landholding inequality. In other words, if we break up the large
holdings and give small plots to the individuals who previously worked on
the plot we basically solve the incentive problem.
Implementing land reform is, however, politically difficult and costly. Large
landlords hold strong lobbies that could topple governments or intimidate
farmers. In fact, in practice, land reforms are usually followed by social
unrest. Therefore, there might be questions about the feasibility of land
redistribution. On the other hand, seizing someone’s property and giving it
to others might reduce government credibility in protecting property rights
and discourage investment. This point should not be overemphasised
because income tax is also a form of expropriation. The difference is
income is flow of resources while land is part of wealth accumulated over
time.
In summary, the theory suggests there might be an efficiency gain from
land reform but empirical work should assess the validity of this claim.
115
9.5.2 West Bengal tenancy reform

Land reform could come in many forms. A government might put a
cap on landholding and redistribute anything beyond the cap among
landless households. Alternatively, they can regulate the rental market
and the terms of the tenancy contracts. The West Bengal land reform
was an example of the latter. Operation Barga was launched by the Left
Front government in West Bengal in 1977 to implement and enforce
dormant tenancy regulations. Under the law, tenants registered with the
Department of Land Revenue could not be evicted from sharecropped
land as long as they paid the minimum of 25 per cent of output to the
landlord. Operation Barga was a village-to-village campaign that informed
sharecroppers of the opportunity to register and benefit from this law.
Banerjee et al. (2002) use a difference-in-differences estimation strategy to
estimate the causal effect of the programme on land productivity.
Operation Barga affected land productivity through two channels. First,
registration effectively becomes tenants’ outside option, increasing their
bargaining power. This forces landlords to offer a higher share of the crop
thereby increasing tenants’ incentives to exert effort. This is likely to result
in productivity gains.
Activity 9.8
Use the sharecropping model discussed earlier to show that an increase in share of
tenant, α increases tenant’s effort and expected output.
Second, registered tenants gain protection against eviction as far as they

pay 25 per cent of output. This increased security of tenure increases
investment incentives but also reduces the landlord’s ability to punish
the tenant (by eviction) when output fails. Theoretically it is not clear
which of these dominate. Therefore the theoretical impact of West Bengal
tenancy reform on productivity is ambiguous.
Banerjee et al. (2002) show Operation Barga actually changed the
contractual terms. The share of tenants receiving more than 50 per cent
of output increased from a pre-reform level of 17 per cent to 39 per cent
after the reform.
Activity 9.9
After the reform a significant number of tenants were still unregistered. Do you think the
contractual terms for these tenants should change (e.g. their output share)? Why? Why
do you think these tenants remain unregistered?
To estimate the impact of land reform on productivity, Banerjee et al.

(2002) proposed to use Bangladesh as a control for West Bengal. They
argue that the two regions are similar except in religion and political
boundaries. In fact, the two were part of a single state before India’s
independence. The fact that the Left Front government took office in
West Bengal (but not in Bangladesh) and initiated Operation Barga was
therefore exogenous to agricultural productivity. They collected district-
level data on rice yields in the two regions from 1969 to 1993. Using a
difference-in-differences strategy, they compared rice yield differences
between West Bengal and Bangladesh before Operation Barga and after
that period. The similarity of the two regions suggests that in the absence
of Operation Barga the rice yield difference between West Bengal and
Bangladesh would have remained the same.
116
Activity 9.10
Figure 4 in Banerjee et al. (2002) shows that rice yields were very close in West Bengal
and Bangladesh up until 1979. What do you infer from this about the parallel trends
assumption? The West Bengal and Bangladesh governments differed in the speed of
adopting high yield varieties (HYV) and implementation of public irrigation projects. Could
you still argue that the parallel trends assumption is satisfied and hence the difference-in-
differences estimates capture the causal effect?
The simple difference-in-differences results presented in Table 2 of

Banjeree et al.’s article show rice yields in West Bengal were higher by 5
per cent during the 1984–93 period. The results are larger after controlling
for time-varying factors like public irrigation and adoption of high yield
varieties (Table 3). Therefore it seems that in the West Bengal land reform,
productivity has increased significantly.
Activity 9.11
Why might you think that implementation of Operation Barga was easier than other land
redistribution programmes?
► Stop and read

Banerjee et al. (2002) pp.239–242, 258–267.
9.5.3 Other evidence

Lin (1992) used province level data in China to study several agricultural
reforms initiated in 1970s. The first element of the reform was a rise in
quota and above quota prices for several products. The second element of
the reforms was a move from team-based production to household-based
farming. Before the reform, around 20 neighbouring households worked
in production teams. The reform rented out parts of the land to individual
households for periods of up to 15 years. The third part of the reform
policies was partial removal of planning restrictions for certain products.
Lin combined data on 28 provinces between 1970 and 1987. The results
show that the move from collective production to household production
increased agricultural productivity. The other two policies, however,
impact on crop choice and use of inputs but they do not seem to affect
productivity. The findings here again suggest the importance of incentives
in shaping productivity.
Besley and Burgess (2000) compiled four variables on measures of land
and tenancy reform legislation in 16 Indian states between 1961 and
1992. Combining this with National Sample Survey they show that rural
poverty declines after reforms that alter the production relations (e.g.
tenancy reforms). Land redistribution, however, does not have a significant
effect on poverty.
9.6 Summary
In this chapter we have seen that agriculture is still the source of income
for many individuals in developing countries. Land is both an important
asset and an input in agriculture. We saw the evidence on the negative
correlation between productivity and land size, despite some theoretical
arguments for a positive correlation. We worked through a principal-agent
model and showed the type of agricultural contracts that could have a
major impact on the farmers’ effort and on shaping productivity. Large
117
landholdings use wage labourers or rent out parts of their land in the form
of sharecropping arrangements. The workers or sharecroppers have lower
incentives to exert effort, resulting in low productivity. Smallholdings are
usually owner-operated and more efficient.
We discussed the arguments for land reform as a tool for achieving
efficiency and equity. Although theoretically it is ambiguous whether land
redistribution results in higher productivity, the evidence from Operation
Barga in West Bengal suggests there might be productivity gains from
tenancy reforms.

• list key facts about importance of agriculture in developing countries
• identify and explain different land contracts and the implied incentive
effects
• analyse and explain a principal-agent model
• critically evaluate empirical evidence on the impact of land reform.

1. Discuss two reasons why larger farms should be more productive
compared to smaller ones. What does the empirical evidence suggest?
2. This activity asks you to derive the optimal sharing rule in a
sharecropping contract. Consider the landlord-tenant relationship
described above, but instead of a fixed rent contract, assume the
landlord only offers a sharecropping one. What would be the share of
the tenant offered by the landlord (in the model)? Does this provide
any justification for why the 50–50 sharing rule is so common?
3. Banerjee et al. (2002) present evidence that after Operation Barga
land prices fell. Why do you think this happened? Could we argue that
tenancy reform in West Bengal resulted in land redistribution?

on the VLE.
118
Chapter 10: Credit
Chapter 10: Credit
10.1 Introduction
• present stylised facts about credit markets in developing countries
• understand the nature of market failures in credit markets
• discuss effectiveness of microfinance.

• explain stylised facts about credit markets
• analyse and discuss models of credit markets and discuss their
empirical relevance
• outline the key features of microfinance
• critically evaluate microfinance impact studies.

Banerjee, A.V. and E. Duflo ‘Giving credit where it is due’, Journal of Economic
Banerjee, A.V., E. Duflo, R. Glennerster and C. Kinnan ‘The miracle of
microfinance? Evidence from a randomized evaluation’, National Bureau of
Economic Research, Working Papers No.18950, 2013.

Aleem, I. ‘Imperfect information, screening, and the costs of informal lending:
a study of a rural credit market in Pakistan’, World Bank Economic Review
4(3) 1990, pp.329–49.
Cull, R., A. Demirgüç-Kunt and J. Morduch ‘Microfinance meets the market’,
Morduch, J. ‘The microfinance promise’, Journal of Economic Literature 37(4)
1999, pp.1569–1614.
10.1.5 Works cited

Angelucci, M., D. Karlan and J. Zinman ‘Win some lose some? Evidence from a
randomized microcredit program placement experiment by Compartamos
Banco’, National Bureau of Economic Research, Working Paper No.
w19119., 2013.
Attanasio, O., B. Augsburg, R. De Haas, E. Fitzsimons and H. Harmgart
‘Group lending or individual lending? Evidence from a randomised field
experiment in Mongolia’, Working Paper 2011.
Augsburg, B., R. de Haas, H. Harmgart and C. Meghir ‘Microfinance, poverty
and education’, National Bureau of Economic Research, Working Paper No.
18538, 2012.
Burgess, R. and R. Pande, ‘Do rural banks matter? Evidence from the Indian
Social Banking Experiment’, American Economic Review 95(3) 2005,
pp.780–95.
119
Crépon, B., F. Devoto, E. Duflo and W. Parienté ‘Impact of microcredit in rural

areas of Morocco: evidence from a randomized evaluation’, MIT, Working
Paper, 2011.
10.1.6 Overview
This chapter introduces credit markets in developing countries. We
start by reviewing the role of credit in business finance, consumption
smoothing and efficient allocation of capital in section 10.2. Section 10.3
discusses the limits to credit markets due to the presence of information
asymmetries and difficulties with enforcing repayments. While these
limitations are common to all contexts, the legal institutions in developed
countries are more effective in enforcing contracts. Furthermore, the use
of collateral requires well-defined property rights, which might not be
available in the context of rural credit markets.
Section 10.4 reviews stylised facts about informal credit markets in
developing countries. We discuss how nearly perfect information flows
among the network of family, friends and local village residents overcome
the limits of formal credit markets. Section 10.5 considers the role of
microfinance in helping the poor. We review a randomised experiment on
the impact of microfinance.
10.2 Role of credit

Credit serves three purposes. First, it helps businesses to finance fixed
start-up costs or adopt new technologies. A few wealthy businesses might
be able to fund investments out of their own pockets, but usually business
activities rely on credit for buying new machinery, renting premises or
paying for the initial advertisements. Second, credit is extensively used
to get working capital. For example, shopkeepers need to pay for stocks
in advance but selling items takes time. Unless shopkeepers have enough
savings they need to use credit to carry on selling goods (cashflow
management). Third, credit is useful in smoothing income shocks. A
household that was unable to generate any income because its head was
sick could borrow money to fund day-to-day consumption needs.
Credit matters greatly to the poor partly because they have few savings
and face greater shocks to their incomes. Many poor people borrow money
from informal sources such as landlords, moneylenders and relatives to
finance both business and consumption needs. Many poor families have
their own small businesses. They use credit extensively to make new
investments or fund their working capital. For example, fruit vendors
borrow money to purchase carts for carrying fruits or to purchase the fruits
for sale. On the consumption side, the poor use credit to smooth shocks
to their income and expenditure like agricultural shocks or idiosyncratic
health shocks.
In a world with perfect credit markets we expect the (risk adjusted)
marginal product of capital to be equalised across all activities.
Furthermore, endowments are irrelevant and whoever turns out to be the
most productive gets the funds. But this is far from what we observe in the
real world. In reality, interest rates vary greatly between individuals. While
formal banks offer very low rates, moneylenders charge extremely high
rates. Endowments (e.g. savings and assets) matter greatly: the poor have
a more limited chance of starting a business compared to the rich.
Before starting to think about credit market imperfections, you should do
Activities 10.1 and 10.2 to better understand the potential efficiency costs
of imperfect capital markets.
120
Chapter 10: Credit
Activity 10.1
Consider an economy with 100 individuals. All individuals have access to a production
function that only uses capital to produce output yi = ki½. Furthermore, the economy is
endowed with 100 units of capital.
a. Assume that one individual has 100 units of capital and everyone else is left with
no capital (extreme inequality). Calculate total output in the economy when credit
markets are broken.
b. Now consider a case where everyone has one unit of capital (full equality). Calculate
total output assuming no credit transactions takes place.
c. Calculate the marginal product of capital. How much is an individual with no capital
willing to pay to borrow money?
d. Now consider the wealth distribution in part (a) but allow for existence of perfect
capital markets. Calculate total output in the economy.
Activity 10.2
Let us think about a situation where credit markets are not working efficiently. Can you
explain why lack of credit could explain persistence of inequality and poverty? What does
this imply about policies that target distribution of endowments?
10.3 Limits of credit markets

Information or lack thereof is a critical factor in credit markets. Lenders
like to know what the chances are that borrowers will repay their loans,
but they do not fully observe all the characteristics and actions that matter.
Several issues arise from the asymmetric information between lenders
and borrowers. First, in the presence of limited liability, borrowers can
only repay loans if their projects succeed (or they have the means to pay).
This creates an incentive for borrowers to take risky actions because they
get a high return if the project succeeds but do not lose anything if it fails
(see Activity 10.3). Although lenders spend resources on monitoring,
they cannot observe borrowers’ actions perfectly. This is the moral hazard
problem: borrowers engage in risky actions that are detrimental to the
lender yet the lender is unable to verify that such actions have taken place.
The second issue is that lenders try to identify safe projects to avoid too
many unpaid loans. But at any given interest rate the risky investors get a
higher expected benefit because they do not repay when the project fails.
It is impossible for the lenders to weed out all untrustworthy borrowers
due to asymmetric information on the type of the borrowers. Therefore,
a larger fraction of risky investors take out loans. This is the problem of
adverse selection: the inability of the lender to perfectly observe project
types results in higher number of risky loans.
The third issue is strategic default, which is a form of moral hazard. Even
when projects succeed, borrowers have an incentive to hide the outcome
and pretend to have no money. If lenders lack the ability to perfectly
monitor borrowers, strategic default becomes easier.
Activity 10.3
Consider a situation where you can invest in two projects. One is a safe project that
generates $US1,200 from $US1,000 investment. The other is a risky project requiring the
same investment. But half the time it fails (0 output) and half the time it gives $US2,200
of output. You only care about expected returns (i.e. you are risk neutral).
a. Suppose you are using your own savings to finance the investment. Which investment
option do you prefer?
121
b. Now suppose you use a loan with an interest rate of 10 per cent to finance the
investment under limited liability. In other words, if the project fails you pay nothing
to the lender simply because you do not have any assets. Which investment option do
you prefer now?
c. Now assume you could pledge collateral which will be seized in case of default. Find
the minimum amount of collateral which leaves you indifferent between choosing the
risky and the safe project when the lender charges 10 per cent interest.
Collateral could work as a reassurance for lenders. The chances of losing

an asset in case of default reduces the likelihood of strategic default even
if monitoring is not perfect. The implication of this, however, is that only
individuals with some tangible assets can borrow money. The poor will not
be able to get collateral based credit. This is potentially why most of the
poor are involved in informal credit transactions.
► Stop and read

Ray (1998) Chapter 14, pp.529–32 (section 14.1).
10.4 Informal credit

While formal credit is almost absent from the lives of the poor, informal
lending and borrowing is widespread. Banerjee and Duflo (2010) report
that less than 6 per cent of borrowing was from formal sources in 12 out
of 13 developing countries they surveyed. We first review five stylised facts
about informal credit markets and then think about potential explanations
for these stylised facts, using a simple theoretical setting.
10.4.1 Stylised facts

Banerjee and Duflo (2010) identify the following stylised facts about
informal credit markets in developing countries:
Fact 1: Lending rates as high as 50 per cent per annum are common.
This is despite the fact that deposit rates are not very high. In Pakistan the
average interest rate was 79 per cent while the marginal cost of funds (for
lenders) was estimated to be 48 per cent (Aleem, 1990).
Fact 2: Interest rates vary greatly within the same credit market (e.g.
Aleem (1990) shows that in Pakistan the standard deviation of interest
rates was 38 per cent).
Fact 3: The rich are charged lower interest and receive larger loans.
Fact 4: Default rates are low and cannot explain the large variations in
the interest rates. For example, in Pakistan the default rates are usually
close to 1–2 per cent and rarely reach 10 per cent (Aleem, 1990).
Fact 5: It does not seem that lenders have high monopoly power. For
example, Aleem (1990) does not find high profits for moneylenders in his
sample.
10.4.2 Potential explanations

Now we try to come up with potential explanations for the five stylised
facts. The neoclassical model of capital markets under perfect competition
implies lenders need to cover the cost of deposits and default in order
to be able to continue lending money. Let us assume the deposit interest
rate is ρ (this is the cost of funds for the lender) and the lender charges
a uniform interest rate equal to i to all borrowers. With probability d
borrowers default and pay nothing to the lender. With perfect competition
and free entry in the credit market, lenders receive zero profits, therefore
122
Chapter 10: Credit
(1 − d)(1 + i) − (1 + ρ) = 0
1+ρ
⇔ 1+i =
1−d
The first line basically shows profits (revenue minus cost) for a loan of size
1. It shows the average profit margin on each loan. This shows that when
the probability of default goes up, the interest rate on loans increases. In
order to cover the cost of deposits, lenders need to charge a higher interest
on those repaying when more borrowers default.
Activity 10.4
Is this model consistent with the stylised facts about credit markets presented earlier?
How could this model explain a wide gap between deposit and borrowing interest rates
(i and ρ in the model)?
It is unlikely that the probability of default is the same for everyone.

Consider an extension of the model where an individual with assets w
chooses capital k > w to invest and produce f(k) units of output. Therefore
the borrower needs to borrow k – from a lender. Assume the interest rate
is i hence he has to repay (1 + i)(k – w) at the end of the period. Although
output is certain, the borrower can spend ηk units of capital and hide his
success to avoid repaying the loan (strategic default). Borrowers would
choose to repay if net profits with repayment is larger than net profits
when avoiding repayment:
f(k) − (1 + i)(k − w) ≥ f(k) − ηk
(1 + i)w
⇔ k ≤ (1 + i) − η
The right-hand side shows the maximum amount of capital that an

investor with wealth w can gather for production (from their own wealth
and credit). Let us focus on borrowers for whom the optimal level of
capital is greater than the right-hand side (i.e. they are credit constrained).
These borrowers can borrow up to the limit
(1 + i)w
k = (1 + i) − η (1)
The right-hand side of this equation is an increase in wealth (w) and

evasion cost (η). Therefore individuals with higher wealth and those with
higher evasion costs get larger loans. The size of the loan also decreases in
interest rate because the benefits of avoiding repayment increases.
Activity 10.5
Is the relation between loan size and wealth in this extension of the model consistent
with the stylised facts presented earlier?
How do lenders set the interest rate? Lenders need to incur fixed
monitoring costs equal to Φ for each loan and, as before, face a deposit
rate of ρ. The market is competitive and profits are driven to zero.
Therefore the market interest rate for a loan of size k – w is given as
(1 + i)(k − w) − (1 + ρ)(k − w) − Φ = 0
Φ
(2)
⇔ 1+i=1+ρ+ k−w
This equation shows that smaller loans are going to have higher interest
rates. This means (from (1)) that poorer individuals are going to
get smaller loans that are more expensive. Furthermore, there could
potentially be a high wedge between deposit and lending interest rates
due to the presence of monitoring costs. These results concur with the
123
stylised facts presented earlier. The model shows that monitoring costs are
critical and small changes in them could result in large efficiency gains.
Activity 10.6
You can directly show the claim made above by replacing loan size from (1) in (2) and
finding an expression for i that only depends on wealth and not k. How does the market
interest rate depend on the cost of default (η)?
10.4.3 Information flows

Why do formal banks not lend money to the poor in a similar way to
moneylenders? What makes moneylenders and other informal creditors
different from formal banks? The model we have seen above suggests that
in order to understand these questions we need to look at differences in
monitoring costs across formal and informal institutions.
Informal lenders usually have extensive interactions with borrowers which
makes monitoring far easier. Here we consider some examples of the
type of credit interactions made possible because of the perfect flow of
information.
The poor might borrow from their landlords working on nearby plots.
Once the harvest is cultivated, the landlord would be the first to know.
Others might borrow from business partners. A common example is a
vegetable vendor getting supplies from a wholesaler on credit at the
beginning of the day and repaying the loan the following day. Interlinked
transactions could work as credit channels. Wholesalers provide credit
to farmers for purchasing fertiliser or other expenditures and receive the
repayments as crops at harvest time.
Activity 10.7
If informal lenders have lower monitoring costs, why is it that they charge such high
interest rates compared to formal banks? Frame your answer in terms of the model
discussed in the previous section.
Sometimes informal creditors use some form of collateral that is not

available to formal lenders. Agricultural labourers get loans from landlords
on the implicit assumption that failure to repay results in some days of
unpaid work.
Formal credit institutions would fail in all these circumstances because
they lack the information to properly monitor individuals or use the forms
of collateral unavailable to the poor.
► Stop and read

Ray (1998) Chapter 14, pp.532–78 (sections14.2–14.4, and 14.5–14.5.2) and Banerjee
and Duflo (2010).
10.5 Microfinance
Could it be that the poor’s inability to escape poverty is because they do
not have access to cheap credit? Many poor households have little land
and work as agricultural labourers. Agricultural activities are seasonal and
depend on weather conditions. In good times, jobs are available but in bad
times the poorest households are unable to find jobs. Giving credit to these
households to cope with fluctuations in their incomes and possibly to start
a small business of their own could lift them out of poverty.
124
Chapter 10: Credit
Many government initiatives tried to help the poor by providing subsidised

credit usually through government banks. For example, in India banks
were forced to open four branches in areas where there were no banks
if they wanted to open one branch in an area where there were already
banks between 1969 and 1990 (Burgess and Pande, 2005). Although these
strategies could be successful in the short run, there are reasons to believe
they are not a sustainable solution. First, the fact that loans are subsidised
means credit demand is very high and often the funds are diverted to
unintended recipients with political connections. Second, high default
rates and accumulation of subsidies increase the cost of maintaining
subsidised credit. For example, the 4:1 rural branch expansion policy in
India was stopped partly because of default rates close to 40 per cent and
high costs.
Microfinance provides an alternative to a subsidised government loan
programmes. Microfinance institutions (MFIs) started on the premise of
serving the poor in a financially viable way. MFIs were started by NGOs
with a not-for-profit approach but currently many variations of MFIs exist,
including for-profit institutions (Cull et al., 2009).
MFIs are now active in several countries with several million customers.
Some MFIs are profit oriented (e.g. Banco Compartamos of Mexico), while
others promote a non-profit vision (e.g. Grameen Bank of Bangladesh).
Morduch (1999) and Cull et al. (2009) present interesting patterns in MFI
operations.
During the 2000s MFIs received much attention as a ‘magic bullet’ to
alleviate poverty. In fact, Muhammad Yunus and Grameen Bank won
the Nobel Peace Prize in 2006. In the next section we review the basic
principles of microfinance before moving to discuss one impact evaluation
of microfinance loans.
► Stop and watch

Now watch Muhammad Yunus speaking on history of microfinance:
www.tedxvienna.at/talks/#/4822.
10.5.1 Basics
MFIs provide small loans (or savings, insurance, etc.) to individuals unable
to access the formal channels of credit, such as banks. Lack of access
is usually due to unavailability of collateral and fear of default by poor
borrowers. So how do MFIs overcome the market failures in the credit
markets? How do they make sure that loans are repaid?
MFIs usually provide business loans to poor rural women. Instead of
collateral, MFIs use exclusion from future credit to promote repayment.
The reasoning behind this is simple. Borrowers can get a one-off benefit
by not repaying their current loan. But this will have a lifetime cost of not
being able to get credit in the future. If the net present value of access to
credit is greater than the one-off benefit of non-payment, the individual
will repay. The following activity asks you to show formally how this
works.
Activity 10.8
Assume an individual lives for N periods and in each period she can borrow money at
interest rate i to fund a production activity, but only if she has a good credit history. In
other words, she cannot get a loan if she has defaulted before. The project requires k
units of capital and will produce f(k). In the absence of credit the individual gets zero
output.
125
•• Find the profit of the business activity for one period if the individual borrows money
and repays.
•• Now write the condition under which the individual would repay her loan in the first
period. Assume the individual repays all future loans if she repays in the first period.
•• Now assume there is a second MFI in the village and that two MFIs do not share
information on the defaulting borrowers. What is the condition to ensure repayment
in the first period (continue to keep the assumption in (b)?
Group lending or joint liability is another mechanism for ensuring

repayment. Instead of lending to a single individual, MFIs lend to a group of
individuals (to each member separately) and if any of the group members
defaults, all members are excluded from future loans. These groups are
formed by individuals living in the same village, therefore members have
interactions beyond the MFI loan and are better able to monitor actions. If
a member is taking irresponsible action, others can impose social sanctions
(e.g. isolation). On the other hand, if one project fails despite its best
efforts, others have an incentive to help with repayments.
► Stop and read

Cull et al. (2009) and Ray (1998) Chapter 14, pp.578–86 (section 14.5.2 to end of
chapter).
10.5.2 Impact of microfinance

Despite all the excitement about microfinance, there has been little
reliable empirical work on estimating the impact of microfinance until
recently. Morduch (1999) section 6 provides a detailed discussion of issues
with the early studies. The main issue is that microfinance borrowers are
not a random sample of the population and therefore might have different
outcomes, even in the absence of borrowing. The following activity asks
you to think carefully about the nature of the selection bias.
Activity 10.9
A microfinance institution is active in a village in Bangladesh. In order to convince donors
that microfinance is changing the lives of the poor, the MFI has set up a meeting with a
team of researchers to help it estimate the causal effect of microfinance borrowing on
business profitability.
a. A researcher proposes to compare profitability of businesses borrowing from
microfinance to non-borrowers in the same village. Does this give us the causal effect of
borrowing from microfinance on profitability? Why? Could you mention a mechanism
that biases the estimate upward (i.e. the estimated effect is higher than true impact)?
b. Another researcher proposes to compare the profitability of businesses before getting
a microfinance loan and after they received the loan. Does this approach give the
causal effect of microfinance loans? Why?
A more reliable approach for estimating the impact of microfinance is to

randomly provide access to microfinance to some areas (e.g. villages) and
compare outcomes to the areas that did not have access. Randomisation
will ensure that, in the absence of microfinance loans, treatment and
control areas are on average similar on outcomes. Banerjee et al. (2013)
was the first randomised evaluation for this purpose. The authors
collaborated with Spandana, a large MFI active in South India, and
identified 104 suitable slums (where there was no pre-existing MFI and
where poor people lived) in Hyderabad, Andhra Pradesh, India. In 52
randomly selected neighbourhoods Spandana opened a branch in 2005.
126
Chapter 10: Credit
Authors collected detailed household surveys 15–18 months after the first
branch opened.
Although Spandana did not open a branch in control slums in 2005, other
MFIs started to operate both in control and treatment areas. Therefore,
control slums had access to microfinance to some extent. In treatment
areas 27.1 per cent of individuals borrowed money from MFIs, while in
control neighbourhoods, 18.3 per cent had microfinance loans. As the
treatment slums have significantly higher probability of borrowing from
an MFI, we can still compare outcomes between the two to identify the
impact of better access to microfinance.
Activity 10.10
What are the characteristics of the loans offered by Spandana? What are the criteria to
be eligible for a loan from Spandana? To answer this activity you need to read parts of
Banerjee et al. (2013).
Before going through the results let us think about the potential impacts
of a micro loan. First, individuals could borrow money from the (cheaper)
MFI and repay their debts to moneylenders. Second, MFI lending could
help households finance capital costs (e.g. buying a cart or table) and start
a new business. Third, households could get a loan to buy durable goods
like televisions or fridges. Fourth, the loan could help with human capital
investment and increase the education, health and skills of households.
The results suggest that on average the treatment group households
are no more likely to start a business but that profits are slightly higher
relative to the control group. Per capita monthly expenditure is the same
but treatment households increase durable consumption and reduce
consumption of temptation goods. The expenditure on education and
health seems unaffected. Therefore it seems MFI lending has only a
modest effect across several outcomes over a period of two to four years.
The results reported here compare average outcomes of treatment
households to control households regardless of whether they actually
borrowed money from an MFI. But only 27 per cent of households in
the treatment group received an MFI loan 15–18 months after Spandana
opened its branches. Therefore we are not estimating the effect of
getting a loan from an MFI. What is estimated here is the effect of having
increased access to MFI lending due to a Spandana branch opening in the
neighbourhood. This is known as the intent-to-treat (ITT) estimate. Since
assignment to treatment was random, we can argue that the ITT shows the
causal effect of increased access to MFI credit.
But what if we wanted to estimate the effect of MFI lending on those who
actually borrowed money? This is known as the treatment-on-the-treated
(TOT) estimate. Note, however, that randomisation does not ensure
that the individuals who decided to borrow from the MFI are a random
sub-sample of the treatment group. In the case of MFI borrowing, many
eligible households did not even apply for an MFI loan. Others chose to
form a group and applied for loans. It is unlikely that those who applied
for MFI loans are similar to those who did not. Therefore we do not have a
valid control group to estimate the TOT here.
One way to solve this endogeneity problem is to use assignment-to-
treatment as an instrument for receiving the treatment. We have seen
here that households assigned to the treatment slums are more likely to
borrow money from MFIs. This suggests that being assigned to a treatment
slum satisfies the relevance condition for a valid instrument. But is it
127
also the case that treatment assignment affects outcomes (e.g. per capita
consumption) only through increased probability of borrowing? In other
words, does this instrument satisfy the exclusion restriction? Probably not,
because borrowing households create an externality on non-borrowers
when they expand their business. Therefore even non-borrowers have
different outcomes due to externalities. The simple fact of being in the
treated slums affects outcomes and the exclusion restriction is violated.
Activity 10.11
This activity helps you better understand the difference between intent-to-treat (ITT)
estimate and treatment-on-the-treated (TOT) estimates. Suppose we ran the same
experiment and located MFI branches in a random neighbourhood. Assume anyone taking
an MFI loan sees an increase of 50 Rupees in expenditure (this is the true causal effect
of borrowing that we did not observe in the actual experiment). Furthermore, assume
there are 1,000 individuals in the treatment group and 1,000 in the control group. At the
baseline all individuals are similar and have an expenditure of 1,000 Rupees.
a. Consider a case where no one in the control group receives an MFI loan and everyone
in the treatment group receives the loan. Compare ITT and TOT estimates. Which one
is bigger and why?
b. Now consider a more realistic case where only 50 per cent of individuals in the
treatment group receive a loan. Still assume no one in the treatment group receives a
loan. Compare and contrast ITT and TOT estimates.
c. Finally, consider a case where 50 per cent of treatment group and 30 per cent
of control group receive a loan. Calculate ITT and TOT. How do they compare to
estimates derived in a and b above?
The distinction between ITT and TOT estimates is an important recurring

point in randomised controlled trials. From a policy point of view we
might be interested in one or the other. In the MFI experiment, if the
policy maker is interested in implementing a similar programme, probably
with similar take up, then the ITT estimate is informative. But ITT is not
very informative about the impact of borrowing money from an MFI. In
other words, the results here cannot rule out significant effects on the
subsample of borrowers.
► Stop and read

Banerjee et al. (2013). For similar randomised experiments in other countries, see
the following articles: Crépon et al. (2011) Morocco; Augsburg et al. (2012) Bosnia-
Herzegovina; Angelucci et al. (2013) Mexico; and Attanasio et al. (2011) Mongolia.
10.6 Summary
In this chapter we have reviewed stylised facts about credit markets in
developing countries. We emphasised the role of informal credit markets
and discussed microfinance as an innovation that has increased access to
credit for the poor. A recent randomised experiment, however, casts doubt
on the claims made about microfinance being a miracle.
► Stop and read

128
Chapter 10: Credit

• explain stylised facts about credit markets
• analyse and discuss models of credit markets and discuss their
empirical relevance
• outline the key features of microfinance
• critically evaluate microfinance impact studies.

1. Banerjee et al. (2013) found that many outcomes were not
significantly different across treatment and control groups. We,
however, know that only a small fraction of households received an
MFI loan. Could this modest result be driven by the fact that we are
looking at ITT estimates? Why?
2. Banerjee et al. (2013) find that 15–18 months after the Spandana
branches opened, around 88 per cent of households in the control
group had an outstanding loan. Most of these are from non-MFI
sources. A similar figure persists in the treatment group too. Why do
you think so many households are still borrowing money from other,
potentially more expensive, sources? Could any features of the MFI
loans explain low take up?

on the VLE.
129
Notes
130
Chapter 11: Insurance
11.1 Introduction
• introduce the concept of insurance and discuss limitations of insurance
markets
• discuss the role of informal insurance for rural households and
understand its limitations
• understand how informal credit could work as insurance.

• recall and relate the concept of risk aversion to demand for insurance
• outline the limitations of formal and informal insurance and analyse
their relevance in a given context
• discuss the case of informal insurance in northern Nigeria.

Townsend, R. ‘Consumption insurance: an evaluation of risk-bearing systems
in low-income economies’, Journal of Economic Perspectives 9(3) 1995,
pp.83–102.
Udry, C. ‘Credit markets in northern Nigeria: credit as insurance in a rural
economy’, World Bank Economic Review 4(3) 1990, pp.251–69.

Giné, X., R. Townsend and J. Vickery ‘Patterns of rainfall insurance
participation in rural India’, World Bank Economic Review 22(3) 2008,
pp.539–66.
Morduch, J. ‘Income smoothing and consumption smoothing’, Journal of
Udry, C. ‘Risk and insurance in a rural credit market: An empirical investigation
in northern Nigeria’, Review of Economic Studies 61(3) 1994, pp.495–526.
11.1.5 Works cited

Cole, S., X. Giné, J. Tobacman, P. Topalova, R. Townsend and J. Vickery
‘Barriers to household risk management: evidence from India’, American
Economic Journal: Applied Economics 5(1) 2013, pp.104–35.
11.1.6 Overview
The poor in developing countries face various shocks. We have seen that
a large share of the labour force works in agriculture. Weather shocks,
pests and price fluctuations are a few examples of factors that influence
agricultural productivity and income. In addition to income volatility,
uninsured individuals face expenditure shocks too. Unanticipated illnesses
are not uncommon and could result in a significant shock to household
expenditure, pushing the poor from malnutrition to starvation or worse.
So how do poor households cope with income and expenditure shocks?
131
In principle, households could use three instruments to buffer income

shocks and reduce consumption volatility. First, they could borrow from
various sources in times of hardship. In fact, one of the functions of
credit discussed in the previous chapter was consumption smoothing.
Second, they could use savings as a buffer for income shocks. In times
of abundance, households accumulate wealth (monetary savings,
purchase of livestock or other assets) and during shortages they run
down wealth to have less volatile consumption. Third, households
can engage in arrangements with other agents (households, private
companies or government) and insure themselves against negative
shocks. Insurance could be a formal contract where premiums are paid
to receive compensation when a pre-specified event happens. It could
also compromise informal arrangements between households in the same
village or family to help each other out.
An alternative approach is to reduce income volatility by diversifying
sources of income. For example, Banerjee and Duflo (2011) report in a
survey of 27 villages in West Bengal that the median household had three
working members engaged in seven occupations. Income shocks to one
occupation (e.g. farming) are less damaging if the household receives
money from other sources (e.g. construction). Adoption of more resilient
crops (sometimes with lower productivity) is another way of reducing
susceptibility to weather shocks (Morduch, 1995).
In this chapter we first discuss why individuals might prefer a smooth
consumption stream and therefore demand insurance. We then discuss one
article that deals with the extent of consumption smoothing in developing
countries. Evidence shows that consumption is much less volatile than
income but attempts at consumption smoothing fall short of perfect
smoothing (full insurance). To better understand why full insurance is
probably unachievable, we outline failures in insurance markets and
discuss why innovations like rainfall insurance have low take-up. Finally,
we discuss several articles that deal with the functioning of informal
insurance arrangements.
11.2 Basic model of insurance

Individuals are risk averse if their utility function is concave. To
understand how this relates to disliking risk, let us consider an example.
Asif owns a plot and in good years gets 40,000 Rupees from his harvest
but in bad years he gets 10,000 Rupees. From his experience he calculates
that bad years happen with 50 per cent probability. Consuming c gives Asif
utility u(c) = √c . Therefore Asif’s expected utility, if he consumes only his
farm income, is
E[u(c)] = 0.5 × √40000 + 0.5 × √10000
= 0.5 × 200 + 0.5 × 100 = 150
Now consider a different situation where Asif’s income is fixed at
the expected level of income in the previous scenario. Therefore his
consumption is equal to expected income as follows
E[c] = 0.5 × 40000 + 0.5 × 10000
= 25000 Rupees
Consuming the expected income will deliver utility equal to
u(E[c]) = √25000 ≈ 158
which is higher than the expected utility of consuming the risky income
stream. Therefore Asif is willing to sacrifice some expected income to
increase the certainty of the income stream. He basically dislikes risk.
132
Activity 11.1
What is the minimum certain income that gives Asif the same utility as the risky income
above? What fraction of expected income is Asif willing to give up to receive the secure
income?
Figure 11.1 shows the concept of risk aversion in a more general setting.
The utility function in the figure is concave (negative second derivative)
and therefore shows risk aversion. Asif would enjoy high consumption
with some probability and receive low consumption in other times. Low
consumption gives u(cl) and high consumption yields u(ch) utility. The
point on the straight line that connects low and high states shows the
expected utility of this risky bundle is E[u(c)]. On the other hand, we can
find expected income (consumption) on the horizontal axis and the utility
of consuming this fixed income stream on the vertical axis as u(E[c]). As
shown on the figure, concavity implies u(E[c]) > E[u(c)](this is Jensen’s
inequality.).
Utility
u(c)
u(ch)
B H
u(E[c])
E[u(c)]
A
u(cl)
L
Cl E[c] Ch Consumption
Figure 11.1 Expected utility of consumption versus utility of expected

consumption.
Let us consider an insurance product that costs q and fully insures Asif.
In other words, Asif would get his expected income minus the cost of
insurance. Does Asif want to purchase this scheme? To answer this
question we need to think about the utility that Asif gets with and without
the insurance. In the absence of insurance Asif gets an expected utility
which is an average of low and high consumption utility (point A in Figure
11.1). But with insurance, Asif gets a fixed consumption, which is equal to
expected consumption minus the cost of insurance. Therefore Asif would
buy the scheme if the following inequality is satisfied:
( )
u (1 − p)cl + pch − q ≥ (1 − p)u(cl) + pu(ch)
The left-hand side is utility with insurance and right-hand side is without.
If the price of insurance, q, is sufficiently small, Asif certainly buys
insurance. This is because we have seen he is risk averse and is willing to
sacrifice some expected income to reduce risk. The main implication of
this simple theory is that Asif’s income fluctuations should not translate
into consumption fluctuations when he can insure risk.
133
Activity 11.2
What is the maximum price Asif is willing to pay for insurance?
(Hint: Use Figure 11.1 and see how far you can lower Asif’s fixed consumption under
insurance. A graphical solution is enough here.)
Now assume Asif lives in a village with a large number of similar

individuals. The probability of getting low income is the same but differs
between individuals. Insurance products are not yet present in the village.
Could individuals in the village insure themselves against income shocks?
To answer this question, let us decompose the income of individuals in the
village as follows
Yij = A + θj + ∈ij (1)
Yij is the income of individual i living in village j and is decomposed to a
global average A, a village level shock θj, and an idiosyncratic shock to
each individual �ij. Consider an arrangement between the villagers by
which everyone pools their incomes together and consumes the average
income
Yj = A + θj
Note that individual shocks are independent and have an average of zero
in the village and therefore drop out when income is pooled but the village
level shock applies to everyone in the village. The individuals in the village
would prefer to get Yj instead of Yij because both have the same average
(expected value) but Yj is less risky. The following activity asks you to
investigate the implications of this arrangement.
Activity 11.3
Consider the above pooling arrangement. Do income fluctuations translate into
consumption fluctuations? In a regression of individual consumption on individual income
while controlling for average village income, what should be the coefficient estimates on
individual income and average village income based on the above theory? What does this
tell you about the ability of individuals to insure against aggregate shocks (village level)?
► Stop and read

Ray (1998) Chapter 15, pp.591–600 (sections 15.1–15.2).
11.3 Empirical evidence on consumption smoothing

Is there any evidence that households are fully insuring themselves against
idiosyncratic shocks? Townsend (1995) studied surveys from India and
Thailand to see if households are fully insured. In three Indian villages a
one dollar increase in individual income raised individual consumption by
14 cents. Furthermore, individual consumption is significantly affected by
village level income. These show households were able to insure against
idiosyncratic risks but that there was far from full insurance because
household income still impacts on household consumption. In the set of
Thai villages that Townsend (1995) analysed there is less insurance and a
dollar increase in household income increases consumption by 34 cents,
but the coefficient estimate on the average village income is double that
of household income. This shows the inability of the household to insure
against aggregate shocks.
In the next section we look at failures in the insurance markets before
discussing potential (informal) mechanisms through which households in
developing countries have achieved the sorts of consumption smoothing
measured by Townsend (1995).
134
► Stop and read

Townsend (1995).
11.4 Limits to insurance

Market failures are at the heart of insurance markets. Asymmetric
information and limited enforcement are two core issues that we discuss
here. In a last subsection we consider some new developments in rainfall
insurance and think about why they were mostly unsuccessful.
11.4.1 Limited information

Usually the insured parties have more information about their characters
and actions. The insurance company, on the other hand, has to rely on
observable outcomes such as an illness or a failed harvest. This creates
three potential issues:
1. We might have outright fraud in the sense that nothing has happened
to the insured parties yet they claim compensation.
2. The insured party might have a riskier-than-expected character
(adverse selection).
3. The insured party might fail to put effort into preventing damage
(moral hazards).
We will discuss each of these in turn.
If the insurance company is unable to fully observe the occurrence of the
insured event, the insured individuals have an incentive to lie about the
situation and receive unlawful compensation. Sophisticated contracts and
the possibility of legal action, however, reduce the importance of this issue
in formal insurance. Under informal arrangements, although the terms
of contracts are not written, close-knit relationships and geographical
proximity reduces monitoring costs. This makes it very hard to lie about
one’s harvest or illness.
Adverse selection is a situation where an insurance arrangement is
designed for a given set of customers with certain characteristics but
the insurance company is unable to observe all those characteristics
and therefore inadvertently draws in riskier customers. Suppose a car
insurance company studies a market and realises the average person
has a 10 per cent probability of having an accident with $1,000 dollars,
damage during a given year. It therefore offers insurance to car owners
at a premium of $100 to ensure viability at the lowest price (due to
competition). If a random sub sample of the market sign up for the offered
insurance scheme, then indeed 10 per cent of customers would have an
accident per year and this will cost $1,000 in compensation. Therefore the
insurance company would on average spend 0.1 × 1000 per customer and
receives 100 from premiums to cover these costs.
On second thoughts, it seems unlikely that a random sub-sample will
take up the scheme. Drivers with higher-than-average risk have greater
incentives to purchase insurance. The scheme is a better deal for them. By
contrast, safe drivers know they will not have an accident and do not see
any benefit in having insurance. This means that the average customer will
have a higher-than-expected risk and therefore the insurance company
makes a loss. This is a case of the market unravelling. For any given
premium, safer drivers are less likely to buy insurance which results in
higher-than-expected damages for the insurance company. To keep up with
this situation, the insurance company needs to increases the premium,
eventually leading to a breakdown of the market.
135
Activity 11.4
Based on adverse selection could you explain why car insurance is mandatory in many
countries? How do insurance companies try to reduce adverse selection? Does offering
multiple contracts play a role?
Once damages are insured, individuals have lower incentives to take

precautions to reduce the likelihood of shocks. This is called moral hazard.
There are many actions (choices) that affect the probability of shocks or
change the likely damage. Considering the classic case of car insurance it
is not difficult to see why insured drivers might become careless drivers.
As another example, think about a case of informal insurance where
individuals in a village agree to pool their incomes as suggested in the
theory of perfect insurance. This means individuals receive average village
income regardless of their own outputs. So why would anyone expend
effort to improve yields? Of course this logic suggests that no one will put
in any effort and most likely there will not be enough resources for pooling
incomes to get the pre-insurance average village income. Therefore the
insurance arrangement breaks down. In a village, however, most actions
are observable. Individuals go past each other every day and it is not
difficult to observe if someone is shirking from work.
Activity 11.5
Consider a health insurance product that offers free doctor visits. How could this change
precautionary actions taken by households? How could this change the pattern of visits
and the cost to the insurance company? Do you think the business of health insurance for
the poor is a viable activity?
One way of dealing with moral hazard is not to provide perfect insurance.
If individuals’ efforts matter for their own consumption at least to some
degree moral hazard is mitigated. This could explain why even in village
communities not all idiosyncratic risks are insured or why most health
insurance policies ask for co-payments from patients.
► Stop and read

11.4.2 Limited enforcement

In formal insurance contracts the insured parties pay premiums in advance
but in informal settings no written contract is signed and ex ante transfers
are not made. So could it be that informal insurance arrangements break
down because of inability to enforce ex post payments? In the earlier
example where households pooled all their earnings to maintain average
village income for everyone, the families with high outputs are helping
those with low outputs in the expectation of receiving help once they
themselves have low outputs.
At a given year, those with low outputs are pretty happy with the mutual
insurance but high output families have an incentive to deviate. The gains
from deviation depend on the difference between utility under high output
and average output. There is, however, a penalty for families deviating
from the village norms. They will be cut off from all future insurance
payments and might be shunned in the village. Depending on the costs
and benefits of deviation the mutual insurance agreement could be
sustained or repealed.
136
Activity 11.6
Suppose a migrant (or stranger) has recently moved to a village. Do you think they would
be able to enjoy mutual insurance with the existing village members? Why?
Perfect insurance may not be feasible precisely because of enforcement

issues. If the gap between high and average outputs is large and yet
everyone gets the same consumption (perfect insurance), the incentives
to deviate from mutual insurance are large for high output individuals.
Allowing for some retained output lowers deviation incentives and could
make mutual insurance more practical.
11.4.3 Aggregate risk

Let us revisit the model of income pooling among villages (equation (1)).
Even after perfect insurance individuals’ consumption depend on village
level shocks. To see this, note that consumption is equal to average village-
level income in this setting
consumptioni = Yj = A + θj
θj is village-level shock and directly impacts on average consumption. A
more general statement is that shocks to the whole network cannot be
insured within the network. Aggregate shocks shift everyone’s income
and so the average. You might be thinking that weather shocks are an
important determinant of agricultural productivity but they happen in a
village or even in several villages at the same time. Floods usually affect all
villages in a water basin and droughts affect regions rather than individual
households.
How could households cope with these shocks? Is there any significant
benefit to mutual insurances, given the extent of weather shocks on
agriculture? This is, in the end, a question about relative importance of
individual idiosyncratic shocks and aggregate village shocks. If everyone
in the village has the same vulnerability to weather shocks, the gains from
mutual insurance are dwarfed by aggregate fluctuations. Townsend (1995)
and others, however, argue that there is considerable variability within
villages. A plot on the slope of a hill might have different soil and moisture
compared to another one across the village on a plain. Furthermore, crops
vary in their resilience to weather shocks and yield. Although extreme
weather shocks could plunge a whole village into destitution, in normal
years there is still significant scope for mutual insurance.
Activity 11.7
How could insurance companies offer drought insurance? Do you think extreme weather
insurance is a profitable business? How might governments differ in providing disaster
relief?
► Stop and read

11.4.4 Other issues

Some forms of risk are easily observed. For example, rainfall is easily
measured at weather stations and greatly impacts on agricultural
productivity. Insurance companies could bring together a large number
of villages with uncorrelated levels of rainfall and offer them rainfall
insurance. Insured farmers would then receive a payout if rainfall falls
below a predetermined level. Giné et al. (2008) offered a similar rainfall
137
insurance to farmers in the district of Mahbubnagar in Andhra Pradesh,

India. The level of rainfall at three weather stations in the district was
used to determine insurance payouts during three phases of cultivation.
Although the insurance was marketed through an established microfinance
institution, only 4.6 per cent of farmers took it up.
Cole et al. (2013) returned to Andhra Pradesh and Gujarat and
implemented randomised experiments to test for barriers preventing
farmers from taking up rainfall insurance. In Gujarat they randomly gave
discount vouchers to farmers. Estimates of demand elasticity derived from
the effectively randomised prices showed a demand elasticity greater
than 1. A 10 per cent increase in premiums led to a more than 10 per
cent reduction in take up. A similar price sensitivity was confirmed in
Andhra Pradesh where a large unconditional cash grant (sufficient to buy
one policy but not conditional on buying one) increased take up by 40
percentage points.
Activity 11.8
Could the large effect of cash grants reflect the importance of liquidity constraints? What
other factors would imply a large effect of cash grants?
An interesting finding of Cole et al. (2013) is that endorsement by a

trusted agent increases take up. Insurance premiums are paid in advance
and payouts are conditional. Farmers may not trust the insurance company
to pay compensation in bad times. Consistent with this is that in the half of
home visits where a trusted local agent endorsed the insurance educator in
Andhra Pradesh, take up was higher by around six percentage points.
There are, however, other potential reasons for low insurance take up not
tested by Cole et al (2013). There may be doubts about the usefulness
of insurance policies. Good rainfall at the weather station does not
necessarily mean a good harvest for the farmers. Different plots within
the same village sometimes have different microclimates which determine
yield. For example, the Mahbubnagar district in Andhra Pradesh, one of
the districts in Cole et al. (2013), contains more than 18,000 sq km of land
with four million inhabitants. But there are only three weather stations.
Furthermore, the government provides disaster relief after extreme
weather events like droughts or floods which could reduce demand for
insurance as the worst cases are already insured against.
11.5 Informal insurance: credit as insurance

In this section we review Udry’s (1990) study of Nigerian villages as an
example of informal insurance arrangements among rural households.
This study also shows how informal lending and borrowing can be used as
an insurance mechanism to alleviate the impact of idiosyncratic shocks.
Udry (1990) carried out monthly surveys of four villages in northern
Nigeria over a period of one year. The survey asked detailed questions
about assets and debts held by the households. Only 10 per cent of
households did not lend and borrow, while 50 per cent both lent and
borrowed during the sample period. Despite the high prevalence of
lending and borrowing in the sample, only 7.5 per cent of loans originated
from formal financial institutions. Most loans were for short periods but
simultaneous lending and borrowing was prevalent even for high income
households. Most of the loans did not have a written record or witness
and did not explicitly mention a nominal interest rate. Although most
households had land, no collateral was used in these transactions.
138
There were certain enforcement mechanisms for repayment of loans.

First, those who ignored their responsibilities were excluded from future
loans. Second, lenders appealed to religious leaders or the village head to
retrieve unpaid loans. Ignoring community verdicts could result in being
left out from the community.
Activity 11.9
Could the threat of exclusion from future loans work if there is more than one lender
available to each borrower? Why?
Transacting parties in these loans knew each other very well: 97 per
cent of loans (by value) were between relatives or residents of the same
village and 65 per cent of loans occurred between partners who knew
and transacted with each other for more than three years. When asked
about an unexpected event experienced by their trading partners, 82 per
cent of the time individuals mentioned a correct instance. Therefore the
information asymmetries discussed earlier are by-and-large absent in this
context.
Activity 11.10
Why is there almost no lending and borrowing across villages? What are the
characteristics of the few inter village loans?
Udry (1990) shows that credit transactions in the four villages surveyed
actually take the form of state-contingent contracts between borrowers
and lenders. Table 4 in Udry shows that a lower fraction of borrowers who
had received a shock were perceived to be in default (i.e. excluded from
future loans) even when the realised interest rates were negative. In other
words, lenders were more lenient for borrowers who had an unexpected
shock (e.g. flooding and pest infestation). The interest rates were lowered
for borrowers receiving a shock. While 72 per cent of loans for borrowers
with a shock had zero or negative realised interest rate, only 39 per cent of
loans for borrowers without a shock had zero or negative realised interest
rates.
Activity 11.11
Could the pattern presented in Table 4 of Udry (1990) be a result of limited liability?
Which results (in Table 4 and elsewhere) are not consistent with an explanation based on
limited liability?
Table 5 in Udry (1990) shows that shocks to the lenders also matter
for realised interest rates and repayment periods. If a lender receives a
shock, the borrower pays the debt with higher interest and in a shorter
timeframe. These loans are not like usual credit. The terms of the loans
allow for greater flexibility to accommodate shocks. Almost perfect
information in the village is the reason why such transactions are possible.
It is impossible to uphold such transactions in an anonymous urban
environment. The fact that transactions were limited to households
in the same village implies the limited role of informal insurance in
accommodating aggregate shocks to the whole community.
► Stop and read

Udry (1990).
139
11.6 Summary
In this chapter we discussed why individuals demand insurance and briefly
looked at the evidence on consumption smoothing in developing countries.
Insurance markets are, however, susceptible to several failures.
Asymmetric information results in moral hazard and adverse selection in
insurance markets. The former stems from the fact that actions that affect
the occurrence probability of the insured event are not perfectly observed
and receiving insurance reduces incentives to take precautions. Adverse
selection, however, is a result of unobserved characteristics that matter for
the inherent risk of being insured
In informal settings the flow of information is close to perfect and most
actions and events are observed. This alleviates the problems of moral
hazard and adverse selection to some extent, but enforcement mechanisms
limit the practicality of informal insurance. When contracts are implicit
it is hard to uphold them in conventional courts. In informal settings,
however, there are other instruments like social sanctions and exclusion
from future credit and insurance that hold up ex ante promises.
Informal insurance is usually restricted to a village (to enjoy full
information) and therefore shocks that affect the village as a whole cannot
be insured. Unfortunately weather shocks usually have a regional nature
and affect several villages at the same time. It seems natural that farmers
have an unmet demand for weather insurance. We looked at rainfall
insurance in India and tried to make sense of low take up rates. Similar to
preventive health care we found that demand for rainfall insurance is price
elastic: a 10 per cent reduction in the price increases take up by more than
10 per cent. Trust in insurance providers and liquidity constraints seem to
be important factors in explaining low take up.
In the last part we reviewed informal insurance arrangements in four
Nigerian villages. In this setting lending and borrowing work as a state-
contingent contract that alleviates idiosyncratic shocks. Realised interest
rates and the repayment periods respond to both borrower and lender
shocks and therefore credit is working as a method of insurance.

• recall and relate the concept of risk aversion to demand for insurance
• outline the limitations of formal and informal insurance and analyse
their relevance in a given context
• discuss the case of informal insurance in northern Nigeria.

1. Consider two individuals Asif and Bita with the same utility functions:
(c) = √c . Asif and Bita face an uncertain income stream. Specifically,
half of the time they receive a high income equal to 3,600 Rupees and
at other times they receive a low income equal to 1,600 Rupees.
a. Assume the risks are independent for Asif and Bita. Could they
improve utility by pooling their incomes?
b. Now consider a situation where incomes are perfectly correlated. If
Asif receives a low income, so does Bita. Could they improve utility
by pooling their income?
140
2. Explain why informal insurance arrangements like pooling income and

consuming the average income work only in small communities.
3. How are insurance and credit related in northern Nigerian villages
studied by Udry (1990)? How does credit work as a means of
insurance?

on the VLE.
141
Notes
142
Part 3: The state and the process of development
Part 3: The state and the process

of development
143
Notes
144
Chapter 12: Infrastructure and development
Chapter 12: Infrastructure and

development
12.1 Introduction
• discuss the features of infrastructure that necessitate government
intervention
• discuss the impact of infrastructure on development
• explore the impacts of transport, telecommunication and water
infrastructure on development.

• explain why government intervention is required in the provision of
infrastructure
• outline the mechanisms for the economic impacts of infrastructure
• critically evaluate empirical work on estimating the effects of
infrastructure.

Ahuja, A., M. Kremer and A.P. Zwane ‘Providing safe water: evidence from
randomized evaluations’, Annual Review of Resource Economics 2(1) 2010,
pp.237–56. Available at http://goo.gl/O4I2Ol
Donaldson, D. ‘Railroads of the Raj: estimating the impact of transportation
infrastructure’, American Economic Review (forthcoming) http://economics.
mit.edu/files/6038.
Jensen, R. ‘The digital provide: information (technology), market performance,
and welfare in the South Indian fisheries sector’, Quarterly Journal of
Economics 122(3) 2007, pp.879–924.

Aker, J.C. and I.M. Mbiti ‘Mobile phones and economic development in Africa’,
Devoto, F., E. Duflo, P. Dupas, W. Pariente and V. Pons ‘Happiness on tap: piped
water adoption in urban Morocco’, American Economic Journal: Economic
Policy 4(4) 2012, pp.68–99.
Dinkelman, T. ‘The effects of rural electrification on employment: new evidence
from South Africa’, American Economic Review 101(7) 2011, pp.3078–108.
Duflo, E. and R. Pande ‘Dams’, Quarterly Journal of Economics 122(2) 2007,
pp.601–46.
Galiani, S., P. Gertler and E. Schargrodsky ‘Water for life: the impact of the
privatization of water services on child mortality’, Journal of Political
Economy 113(1) 2005, pp.83–120.
Goyal, A. ‘Information, direct access to farmers, and rural market performance
in Central India’, American Economic Journal: Applied Economics 2(3) 2010,
pp.22–45.
Kremer, M., J. Leino, E. Miguel and A.P. Zwane ‘Spring cleaning: rural water
impacts, valuation, and property rights institutions’, Quarterly Journal of
Economics 126(1) 2011, pp.145–205.
145
12.1.5 Works cited

Burgess, R. and D. Donaldson ‘Can openness mitigate the effects of weather
shocks? Evidence from India’s famine era’, American Economic Review
100(2) 2010, pp.449–53.
Burgess, R., R. Jedwab, E. Miguel, A. Morjaria and G. Padro-i-Miquel ‘The value
of democracy: evidence from road building in Kenya’, Working Paper, LSE
2013.
12.1.6 Overview
Around the world infrastructure investment comprises a large share
of national governments’ expenditure as well as that of donors and
international organisations like World Bank. Infrastructure covers a broad
range of facilities that are needed for the functioning of communities.
Piped water, roads, railways, power lines, dams, canals, schools, hospitals
and prisons are just a few examples.
Infrastructure plays a critical role in the development process. Historically
governments assumed the role of providing infrastructure. Given the large
fixed costs of building infrastructure and large externalities from such
developments, economic theory also supports government intervention. It
is, however, harder to support a given form of intervention. There is some
evidence that regulated private provision of infrastructure is feasible and
desirable, at least in some areas (e.g. telecommunication). You have also
learned about issues in the provision of education and health services by
government bodies in the earlier chapters. Government managers have
low powered incentives to enhance the quality of services, whereas private
managers have high powered incentives as their benefits are usually linked
to innovations that attract new customers.
We start this chapter by studying the important features of infrastructure
that necessitate government intervention. We then briefly discuss why
estimating the impact of infrastructure is a difficult task. In the remaining
part of the chapter we consider three areas: transport, telecommunications
and water infrastructures. In each area we look at recent studies that
deliver reliable estimates of the causal effect of infrastructure on
development.
12.2 Infrastructure and the role of government

Infrastructure features several characteristics that warrant government
intervention at various levels. First, implementing infrastructure projects
requires large fixed costs. Building a kilometre of rural road is very
expensive and unless there is a large enough network of roads, user
demand is low. For infrastructure to be a viable business, there has to be a
user price of greater than marginal cost so the upfront costs are covered.
This is not consistent with perfect competition and warrants a degree of
monopolistic power and hence government regulation of it.
Second, many infrastructure investments are complementary in nature.
They either complement other inputs in the production process or interact
with other infrastructure facilities. Once mobile phone infrastructure
is established, producers (and consumers) can get better information
on markets, including spot prices and availability of products. This
information feeds in to the production process by facilitating better
business planning for use of inputs and markets for outputs. Therefore
the productivity of other inputs depends on the availability of information
infrastructure. On the other hand, the usefulness of information technology
itself depends on availability of transport infrastructure. Having timely
146
information might not be of much use if there is no reliable transport link.

Producers with better information about input prices would know where
the inputs are the cheapest, but they wouldn’t benefit unless goods could
be transported to their production plants.
Activity 12.1
What are the complementarities arising from electricity infrastructure?
Third, infrastructure projects usually involve large externalities. The

impact of an additional link in a road network is not restricted to the
two cities at the ends of the link. Households in other cities also benefit
from increased connectivity. Expansion of piped water not only improves
the physical wellbeing of connected households but also reduces the
risk of contagious water-borne diseases in the whole community (e.g.
intestinal worms). The presence of externalities implies private investment
in infrastructure is less than optimal as individuals fail to incorporate
external benefits in their investment decisions.
Activity 12.2
Suppose a private company builds and owns all the roads in a country and charges road
users to access them. Does the company still ignore the ‘external’ benefits of adding a
link to the road network? Why? What benefits of the extra link might remain external to
the company?
Fourth, infrastructure is similar to public goods. A public good is non-

excludable and non-rival. The former means the supplier cannot exclude
individuals from consuming the product. The latter is about the fact
that a pure public good does not run out if an extra user is added. In
the case of infrastructure, it might be difficult to exclude non-paying
users. For example, it might be difficult to charge anything on a rural
road. Furthermore, at low traffic levels, additional users do not impose
a cost on existing ones (no rivalry and congestion). Even if some forms
of infrastructure are easily excludable, fairness arguments might make
exclusion politically unfeasible. Could a water company charge rural
households so as to recoup the cost of all piping? It might be easier and
more convincing to fund these projects with taxes.
The presence of large externalities and fixed installation costs necessitate
government intervention. One form of government intervention is
to monopolise the installation and supply of infrastructure facilities
through government-run companies. This approach could suffer from
government failures like favouritism and inefficient allocation of projects.
A second approach is to publicly fund infrastructure projects but delegate
construction and operation of facilities to private companies (similar
corruption issues might arise here as well). Finally, a third approach
is government regulation of privately funded infrastructure building
and operation. Depending on the type of infrastructure, either of these
methods might be superior. Our main focus in this chapter is, however,
not on the efficient mode of provision but on estimating the impact of
infrastructure on development outcomes.
12.3 Impact of infrastructure on development

Although there is great support for infrastructure projects, and
governments around the world spend huge sums of money on building
new roads, railways and other facilities, it is not always clear whether
147
these projects are having the advertised impacts. It is important to quantify

the gains from infrastructure in order to assess their cost effectiveness.
Furthermore, to be able to draw generalisable conclusions from specific
infrastructure projects it is crucial to understand the mechanisms through
which they improve outcomes.
Identification of the causal effect of infrastructure is challenging for
several reasons. A government’s placement of infrastructure is usually
driven by political and economic motives. Areas wielding more political
influence usually get more infrastructure projects. For example, Burgess
et al. (2013) studied Kenyan road building and found significant
ethnic favouritism when there were autocratic governments. Favoured
areas received government support along many dimensions (not just
infrastructure) which made it hard to identify the effect of infrastructure.
On other occasions, governments might have a developmental perspective
and expand infrastructure in the poorer regions. Again, since the poorer
areas receiving infrastructure investments are different along many
dimensions it is hard to find a valid counterfactual that informs us about
the outcomes in the absence of infrastructure.
Different types of infrastructure could work through different channels.
Transport infrastructure reduces travel time and cost, and allows freer
movement of goods and labour. Information technologies facilitate the
flow of information and allow better decision-making for producers and
consumers. We consider the example of electricity infrastructure below
and leave a discussion of various mechanisms to the relevant sections.
Expanding the electricity infrastructure could create new business
opportunities and new ways of enjoying life. An electricity connection
could relieve the households of chores like wood collection and cooking
with wood stoves. Households could use the freed-up time to engage in
market production or just enjoy the leisure. Electric stoves have health and
satisfaction effects. Having electricity at home makes business activities like
sewing a more feasible option and it could also extend the effective working
hours of the household into the hours of darkness.
Dinkelman (2011) used the rapid expansion of the electricity network
in South Africa to estimate the effect of electricity connections. She
used land gradient (hilliness) as an instrument for the likelihood of
being connected to the electricity grid. She found female labour force
participation increased by nine percentage points and female weekly work
hours increased by 8.9 hours between 1996 and 2001 (for the areas with
average increase in connections). Her findings support the idea that home
production has become less time consuming due to a switch from wood
cooking to electric cooking and lighting.
Activity 12.3
Why does comparing average household income for households in areas with access to
electricity with the unconnected areas unlikely to give the causal effect of electricity on
income?
Activity 12.4
What are the identification assumptions for the use of land gradient as an instrument
for connection to electricity in Dinkelman (2011)? Do you think these assumptions are
plausible?
148
12.4 Impact of transport infrastructure

The cost of transport is a major element of internal and international trade
costs. Investment in roads, railways, airports, ports and other forms of
transport could improve the access of producers and consumers to markets
by lowering trade costs. This could increase income through specialisation
in comparative advantage and lower price fluctuations through access to
more diverse supplies.
We will discuss comparative advantage and gains from trade more formally
in the next chapter and here rely on an intuitive argument. Different
regions have different productivities in various commodities. When trade
costs are lowered, firms involved in the production of commodities without
a comparative advantage face competition from outside producers. The
outsiders undercut the price of inside producers because they have a
lower marginal cost (they are more efficient). On the other hand, firms
involved in commodities with a comparative advantage can access a bigger
market because of lower trade costs. Therefore sectors with a comparative
advantage expand at the expense of other tradable sectors. Regions
specialise in their comparative advantage which improves aggregate
productivity and lowers consumer prices, leading to increased welfare.
As an example, consider a region that is good at producing cotton and
another that is good at production of livestock. In autarky each region
needs to diversify and produce both livestock and cotton. Opening up
to trade allows specialisation in the activity that each region is best at.
Cotton-efficient regions therefore specialise in cotton production and
livestock-efficient regions focus on livestock production. Both regions
engage in the export of their comparative advantage and import of the
other commodity. The increased average productivity of cotton and
livestock results in a reduction in the prices and more production leads to
improved welfare.
The possibility of trade reduces the vulnerability of communities to
adverse shocks. Shortages in one area could be covered by surplus
production in other areas (arbitrage) reducing price volatility. Burgess
and Donaldson (2010) show that being connected to the railway during
1875–1919 in colonial India reduced the vulnerability of communities to
rainfall shocks and decreased the severity of famines.
Donaldson (forthcoming) uses the expansion of railways in colonial India
to reliably estimate the effect of trade costs on income. He collected data
on the expansion of more than 67,000 km of railways between 1853 and
1930 in British India. Before railways the main mode of transport was
bullocks travelling at around 30 km/day. Population centres connected
to railways experienced a dramatic fall in travel time and cost. It was
then possible to travel 600km per day by train. Donaldson tries to answer
two critical questions. First, how much did the expansion of railways
reduce trade costs? Second, how much did the income of connected areas
improve as a result?
To answer the first question, Donaldson looked at salt price differences
across regions. Salt is a commodity that is produced in certain regions
(e.g. salt mines or from the sea) but is consumed everywhere. He collected
data on all salt varieties in all districts before and after expansion of
railways. The difference between salt prices in consumer districts and
producer district reflect the transport costs. Estimating the reduction in
salt price differentials after a location is connected to the railway would
therefore give an estimate of the relative reduction in transport costs.
149
Table 2 in Donaldson (forthcoming) shows that the effective distance

between salt-consuming and salt-producing districts increased the price of
salt at destination. Furthermore, the estimated costs for different modes
of transport revealed a much lower unit cost of railways relative to other
modes.
The trade cost estimates are then used to construct measures of the
shortest effective distance between all trading pairs in India. Table 3
in Donaldson’s article shows that the effective distance is a significant
determinant of the value of exports. Pairs closer in terms of effective
distance engage in more trade. The expansion of the railways reduced
effective distances and therefore significantly increased trade volumes for
connected areas. To see the impact of this in a reduced form, Donaldson
regressed agricultural income on a dummy variable that is equal to one
when a district is connected to the railway. Getting a railway connection
increased average per acre agricultural income by 16.4 per cent.
Activity 12.5
It is important to understand the basis for prioritising railway building in some areas
and delaying it in others. How does Donaldson (forthcoming) deal with the possibility
that railway construction was not random and the government targeted areas based on
economic potential? Are you satisfied that the 16.4 per cent increase in income shows
the causal effect of a railway connection?
In order to explore the mechanisms through which railways enhanced

income, Donaldson solves a general equilibrium trade model and finds
that ‘trade share’ – the share of district expenditure that is spent on
purchasing district’s own products – captures all model mechanisms. In
other words, the trade share variable (which is constructed using estimates
of trade costs and volumes) is a summary statistic for the effect of lower
trade costs that happens through the comparative advantage mechanism.
If the impact of railways on income is channelled through comparative
advantage we expect the trade share variable to be significant. Table 5
in Donaldson’s article shows 86 per cent of the original effect of railways
on income is captured by the trade share variable. Therefore mechanisms
other than comparative advantage account for only 14 per cent of the
increase in income.
► Stop and read

Donaldson (2013).
12.5 Impact of information technology

The expansion of the telecommunications infrastructure could enhance
information flows in the markets. Better information reveals arbitrage
opportunities for producers and consumers. For example, reliable price
information helps producers find the most profitable market. Mobile
phones and the internet could substantially improve the quality and
speed of price data. Places with higher prices attract more suppliers
and places with lower prices attract more buyers. Therefore the match
between supply and demand improves, which leads to a reduction in
price dispersion and potentially a reduction in average prices (see Aker
and Mbiti, 2010, for data on the fast growth of mobile subscriptions and
potential mechanisms of impact).
Goyal (2010) studied the impact of a policy by ITC Limited, a soy bean
processing firm, to eliminate intermediaries in the procurement of soy
150
beans. ITC introduced internet kiosks in some villages to provide daily

data on the price of soy beans offered by agricultural markets and ITC. It
further created hubs where farmers could test the quality of the produce
and sell it. The policy was gradually implemented in Madhya Pradesh,
India, from 2000 to 2004. Since the policy improves farmers’ information
about prices and increases their trading options (testing and warehousing)
intermediaries need to offer a better price to attract farmers. Goyal
finds that as soon as an internet kiosk is established in a district the
monthly price of soy beans (received by farmers) increases by 1–3 per
cent. Furthermore, price dispersion across markets decreased after the
intervention.
Activity 12.6
What are the potential mechanisms triggered by the introduction of internet kiosks and
hubs that could affect soy bean price? Could the 3 per cent increase in average price
be driven by a reduction of inermediaries’ bargaining power? Could it be due to other
mechanisms?
By easing the flow of information and reducing search costs, information

technologies enable market participants to gain more from transactions.
Jensen (2007) uses the expansion of mobile network coverage in Kerala,
India, to identify the causal effect of mobile phones on fisheries along the
coast. Over one million individuals are directly working in fisheries in
Kerala and 70 per cent of adults eat fish at least once a day. Fishermen,
however, do not observe the market prices while on the sea. Due to high
transport costs and limited time of market operation fishermen are unable
to visit more than one market per day. Due to poor road conditions and
the unavailability of storage facilities, arbitrage does not take place on
the coast (fish is perishable). Therefore supply of fish in a local market is
entirely from fishermen catching near that market.
Before the introduction of mobile phones in Kerala the price of fish in
the local markets showed great dispersion. For example, while the price
of sardines in Thaikadappuram was close to 10 Rupees, in Kasaba just
17 kilometres away from Thaikadappuram, sellers did not have enough
customers and hence faced a price of zero (survey by author reported
in Table I of Jensen, 2007). Lack of information about market prices,
unavailability of storage facilities and inadequate transport links, therefore
led to wastage in several markets at the same time as very high prices in
others.
Activity 12.7
This activity explains the law of one price. Assume there are only two fish markets and the
price of fish is P1 and P2 in the first and second markets respectively.
a. In a situation where fish could be transferred from one market to the other without
any cost (full arbitrage), what is the relation between P1 and P2?
b. Now assume fishermen need to pay a per unit transportation cost equal to θ. How
does the relationship between the two prices change?
c. Finally, consider a case where fish is perishable and carrying fish for more than an
hour on the road will result in a complete loss. Still assume a per unit transport cost
of θ regardless of the distance. How does the relationship between prices depend on
the time it takes to get to the other market?
Starting in 1997 mobile phone coverage was gradually expanded along the
coast in the three districts studied here. Surveys show that by 2001 more
than 60 per cent of fishing boats and most retailers were using mobile
151
phones to coordinate sales. Before discussing the results, let us think about
the potential theoretical channels for the impact of mobile phones.
Without mobile phones the fishing boats only rely on their own catch to
predict local market prices. A high catch would indicate that other boats
might have high catch too and therefore the local market price might be
low. Additionally a high catch increases the benefits of a given price gap
between markets. Therefore fishermen with a high catch are more likely
to incur the transport costs and divert their catch to nearby markets. An
increase in the uncertainty of price prediction and transport costs would
deter potential switchers from going to nearby markets, which creates
high price dispersion across markets. With the availability of mobile
phones, fishing boats have the option of buying a search technology (a
handset and subscription to use the network) to get more accurate price
information. Fishermen could now find out about arbitrage opportunities
in all accessible markets (from the sea) and supply their catch to the
market with a better deal. This reduces price dispersion across markets
and removes wastage.
The gradual expansion provides an ideal setting for estimating the effect
of mobile phones on fisheries. The three districts studied here sequentially
received mobile coverage in 1997, 1998 and 2000. Jensen (2007)
collected weekly data on all fishing markets in the three districts over
the entire period of mobile expansion (a huge data collection effort). He
employed a difference-in-differences identification strategy and compared
outcomes before and after mobile coverage was expanded to a given
district to the change in outcomes in uncovered districts. The identification
assumption is that, in the absence of mobile phones, there would not have
been any differential change in outcomes across regions.
Activity 12.8
Does a difference in average household income across the regions of study pose a threat
to the identification assumption? Why? Explain.
Overwhelming graphical evidence suggests that Jensen is most likely

to be estimating the causal effects of mobile phone. Figure III and IV in
his article reveal that right after coverage was expanded in a district,
a significant share of fisheries purchased phones and price dispersion
declined dramatically exactly at the same time. The same pattern was
repeated in three districts and it seems unlikely that a similar event had
happened exactly at the time of network expansion only in the district that
received the coverage.
The results show that on average the introduction of mobile phones
reduced the coefficient of variation in the sardine price by 38 percentage
points and eliminated wastage completely (reduced by 4.8 percentage
points). Therefore the quantity of sardines sold increased by 23 kg while
the average price received by fishermen insignificantly decreased by 0.05
Rupees/kg. Fisheries profits, however, increased significantly by 133
Rupees per day (about nine per cent increase on average profits). At the
same time, the average consumer price declined by 0.39 Rupees/kg (4 per
cent reduction over a baseline of 11 Rupees/kg). Therefore it seems that
mobile phones had a significant positive effect on producer and consumer
surplus.
152
Activity 12.9
How does the presence of fisheries using mobile phones impact on other boats not using
the search technology and therefore supplying to their local markets? What does the
evidence in Table VIII of Jensen (2007) suggest about your proposed channel?
What are the potential policy lessons from the expansion of the mobile
network in Kerala? The expansion of the mobile network in Kerala
provides a good example of a sustainable development strategy. Profit
maximising mobile operators provide a service that is valued by fisheries.
Fisheries happily pay for mobile phones to improve their profits and in
doing so reduce price dispersion and improve consumer welfare. Although
it might seem that government intervention is not needed in the case of
mobile expansion you should note that in the absence of government
regulation, operators might only expand in areas with high profit
potentials (like large cities). Poor rural areas with a low willingness to pay
for mobile services get the lowest priority. Fortunately in almost all cases of
network expansion the governments set coverage targets to reduce cherry
picking by operators.
Activity 12.10
What is the impact of fish perishability on generalising the results of Jensen (2007) to
other contexts?
► Stop and read

Jensen (2007).
12.6 The impact of water facilities

An important element of public health are facilities that provide access
to clean water. There could be large health gains from improved water
quality as many water-borne diseases like diarrhoea are a central cause
for infant and child mortality in developing countries. An increase in the
quantity of water used, due to better access, could potentially improve
health outcomes by more frequent washing and better sanitation. Despite
substantial health gains from abundant clean water the evidence from
developing countries shows highly elastic demand for simple quality
improving technologies like filters and chlorine (Ahuja et al., 2010). This
might increase the need for government intervention in providing access
to clean water.
Kremer et al. (2011) is the first randomised controlled trial that looks at
the health effects of improving source water quality. In collaboration with
a local NGO, the authors randomly chose 200 springs in rural western
Kenya and constructed a concrete casing to seal off the source from
contamination. Forty-three per cent of households in this area use springs
for drinking water. The majority of these naturally occurring springs are
located on private land, but everyone has free access to the source. This
means private owners do not have an incentive to improve the quality of
these springs. The water usually appears on the ground and the source is
accessible to both humans and animals leading to faecal contamination.
The protective casing made the water come through a standing pipe and
prevented source contamination.
Spring protection greatly improved source water quality (E. Coli
contamination was reduced by 66 per cent) but only moderately increased
the quality of water in the household (E. Coli reduced by 24 per cent).
153
This suggests some degree of water recontamination during transport and

storage. Incidence of diarrhoea among children under the age of three,
however, fell by 25 per cent which shows large health gains from water
quality improvements.
Activity 12.11
How does the Kremer et al. (2011) study show that the effects are not driven by an
increase in the quantity of water consumption?
Devoto et al. (2012) looked at the impact of piped water connections on

household welfare. The randomised experiment was done in the context
of urban Morocco, where households already had access to clean water
but needed to spend time and effort in fetching water (from public taps
in each neighbourhood). It was easy to connect these households and in
order to promote this the water company introduced an interest free loan
that paid for the unsubsidised connection fee. The experiment randomly
chose 434 households from the pool of households eligible for the credit
connection (total of 845) and carried out a door-to-door information
campaign in addition to helping with the application process. Sixty-nine
per cent of the treatment households received a connection while 10
per cent of the control households were connected (despite being in the
control group). The study uses exogenous variation in access to piped
water to identify the causal effect of easier access on physical health and
mental wellbeing.
The treatment led to more free time, which was used for leisure and
socialising and not production or schooling. Furthermore, households
increased their water consumption but this did not lead to an
improvement in health outcomes. You should note that these households
already had access to public taps for drinking water and therefore the tap
inside the house did not change the source of drinking water. Therefore
the connection only changed the quantity and not the quality of water
consumed.
Activity 12.12
Devoto et al. (2012) found that 27 per cent of the control households had piped
water inside the house 18 months after the treatment. Furthermore those closer to the
treatment households had a higher probability of being connected to the mains. What are
the potential reasons for this result?
Another aspect of water infrastructure is whether it is the government

or privately run companies that are responsible for the distribution of
water. Is it the case that government provision leads to inadequate quality
and coverage or are there large externalities that make private provision
inefficient? Galiani et al. (2005) decided to approach these questions using
the context of Argentina’s privatisation of water companies. During the
1990s Argentina privatised water companies in 28 per cent of the country’s
municipalities (covering 60 per cent of the population).
Activity 12.13
Why is comparing the infant mortality rate in privatised areas before and after
privatisation unlikely to yield the causal effect of privatisation (time series estimate)? Why
is comparing infant mortality in areas with a privatised water supply to areas with public
provision not going to give the causal effect of privatisation (cross sectional estimate)?
Galiani et al. (2005) used a difference-in-differences strategy to compare

municipalities with privatised water services to the ones with public
154
companies before and after privatisation. The identification assumption

is that in the absence of privatisation the change in infant mortality
would have been the same in privatised and public water municipalities.
A potential violation of this assumption might have come from targeted
privatisation by municipality governments. For example, it might be that
the central government pressurised municipalities experiencing a local
recession to privatise their water companies to boost economic activity.
Under this scenario the recession is a factor that is specific to privatised
municipalities and impacts on infant mortality. Therefore the difference-
in-differences estimation is going to confound the effects of recession and
privatisation. Galiani et al. (2005), however, tried to rule out scenarios like
this through various robustness checks and it seems they are getting the
causal effect of privatisation on infant mortality.
The results show that infant mortality fell by 8 per cent in privatised
municipalities. Furthermore, they find that poorer areas saw a larger
decline in infant mortality as a result of privatisation. What are the
channels through which privatisation impacted on infant mortality?
Galiani et al. (2005) show that privatised companies expanded the
number of connections radically. The difference-in-differences estimates
suggest connections in privatised municipalities increased between 2–4
percentage points between 1991 and 1997. Furthermore a case study of
the privatised water company in Buenos Aires shows large improvements
in efficiency of the privatised companies in terms of increased water and
sewage production, reduction in spillage and non-stop services during the
summer.
► Stop and read

Ahuja et al. (2010).
12.7 Summary
The presence of fixed costs, large externalities and complementarities
require some form of government intervention in infrastructure projects.
The debate about the form of intervention is still unsettled. In any
case it is important to understand both the magnitude of the effects of
infrastructure on outcomes and the mechanisms that are responsible.
Knowing the magnitude helps in cost benefit analysis for similar
infrastructure projects. Evidence on mechanisms, on the other hand, helps
us to understand what has caused the effects and inform us about the
design of future infrastructure projects.
Until recently, reliable estimates of the causal effect of infrastructure were
not available because political and economic motives greatly influence
the placement of infrastructure projects and result in biased estimates.
We reviewed several articles in the areas of transport, telecommunication
and water infrastructure that overcome this empirical challenge and yield
reliable estimates of the causal effect of infrastructure on development.
Investments in transport reduce trade costs and improve access to markets.
Therefore, regions specialise in their comparative advantage leading to
improvements in productivity and welfare. Evidence from the expansion of
railways in colonial India shows agricultural income in districts connected
to the railway increased by 16 per cent. Easier and faster transport
enables households to smooth out shocks more effectively. The evidence
from colonial India again suggests that being connected to the railway
significantly reduces the severity of famines.
155
Telecommunication infrastructure improves the speed and quality of

information. Better price information allows farmers to find a better
deal for their crops and results in less price dispersion across markets.
Evidence from internet kiosks in India suggests farmers received 1–3 per
cent higher prices after they were able to get more accurate information
on the price of soy beans. Further evidence from the expansion of mobile
phone networks shows welfare gains for both producers and consumers as
a result of better information flows.
Water treatment facilities are a critical component of public health
infrastructure. Clean water reduces the incidence of water-borne
diseases like diarrhoea that are the main cause of infant mortality in
developing countries. Furthermore, better access to clean water could
improve household welfare by reducing the time spent on fetching water.
Recent evidence on improving the quality of springs in Kenya suggests
large reductions in the incidence of diarrhoea. In this area we also saw
large improvements in the quality and access to safe water from the
privatisation of water companies in Argentina.
► Stop and watch

You should now listen to the following LSE public lecture on mobile phones for
development: http://goo.gl/GLnguC

• explain why government intervention is required in the provision of
infrastructure
• outline the mechanisms for the economic impacts of infrastructure
• critically evaluate empirical work on estimating the effect of the
infrastructure.

1. Consider Jensen’s (2007) study of fish markets in Kerala. He compares
outcomes in districts with mobile phone coverage to those without,
before and after coverage was expanded. State the identification
assumption. Outline a potential story that contradicts the identification
assumption.
2. What are the potential channels through which the introduction of
mobile phones in Kerala’s fisheries affects producer and consumer
welfare (surplus)?
3. Improving the quality of water consumed has large health benefits but
an increase in quantity does not impact on health outcomes. Discuss
from theoretical and empirical perspectives.

on the VLE.
156
Chapter 13: Trade policy
13.1 Introduction
• introduce trade theories and explain gains from trade
• discuss theoretical impacts of trade liberalisation and present empirical
evidence
• outline elements of trade policy and their welfare implications.

• analyse gains from trade using various trade theories
• identify distributional issues in trade liberalisation
• discuss the welfare implications of trade policy.

Ray (1998) Chapters 16, 17 and 18.
Bernard, A.B., J.B. Jensen, S.J. Redding and P.K. Schott ‘Firms in international
trade’, Journal of Economic Perspectives 21(3) 2007, pp.105–30.
Melitz, M.J. and D. Trefler ‘Gains from trade when firms matter’, Journal of

Goldberg, P.K. and N. Pavcnik ‘Distributional effects of globalization in
developing countries’, Journal of Economic Literature 45(1) 2007, pp.39–82.
Pavcnik, N. ‘Trade liberalization, exit, and productivity improvements: evidence
from Chilean plants’, Review of Economic Studies 69(1) 2002, pp.245–76.
Verhoogen, E.A. ‘Trade, quality upgrading, and wage Inequality in the Mexican
manufacturing sector’, Quarterly Journal of Economics 123(2) 2008,
pp.489–530.
13.1.5 Works cited

Baldwin, R.E. ‘The political economy of trade policy’, Journal of Economic
Grossman, G.M. and E. Helpman ‘Protection for sale’, American Economic
Review 84(4) 1994, pp.833–50.
Harrison, A. and A. Rodríguez-Clare ‘Trade, foreign investment, and industrial
policy for developing countries’ in Rodrik, D. and M. Rosenzweig
Handbook of development economics. (Oxford; Amsterdam: Elsevier, 2010)
[ISBN 9780444529442] Volume 5, Chapter 63, pp.4039–214.
Maggi, G. and P.K. Goldberg, ‘Protection for sale: an empirical investigation’,
13.1.6 Overview
Countries around the world are trading an increasingly larger number of
goods and services. In 2011, total commodity exports in the world stood
at around $18 trillion. Reductions in barriers to trade in recent decades
have resulted in greater integration of markets around the world, with
significant implications for developing countries. In this chapter the study
157
of trade theories enables us to understand mechanisms that generate

gains from trade liberalisation. Empirical evidence suggests that trade
liberalisations are associated with large reshuffling of resources towards
more productive firms and therefore result in high productivity gains.
We also consider the merits of the 'infant industry' argument and other
protectionist views in the final section. The breadth of material covered
in this chapter prevents us from engaging in a deep discussion of the
extensive literature on trade and development.
13.2 Theories of trade

Trade theories try to explain the observed trade patterns between
countries and study their welfare consequences. They offer insights for
analysing various trade policies like import tariffs and export subsidies.
Old trade theories emphasised the role of comparative advantage due to
factor endowments, technological differences or heterogeneous tastes in
explaining trade flows. The mixed success of old theories in explaining
inter-industry trade and the inability to account for large, observed,
within-industry trade paved the way for new trade theories. Economies of
scale and a set of horizontally differentiated firms generate intra-industry
trade. When combined with sectoral comparative advantage, new trade
theories provide a reasonable account of aggregate trade flows. New
theories, however, could not account for new stylised facts emerging
from firm level data. Trade is rare and only a few highly productive firms
account for most trade flows. Recent waves of trade theories relaxed the
representative firm assumption in previous theories and provide nice
explanations for these observations.
13.2.1 Comparative advantage

The simplest answer to why countries trade could be because they need
the goods that others possess. Oil is found in only a few countries, but
everyone needs it, leading to oil trade. Countries’ factor endowments
are one source of comparative advantage. Technological differences and
consumer preferences are two other factors that generate comparative
advantage.
To understand better the concept of comparative advantage, let us
consider a simple example of two countries and two products, both
produced solely using labour. For ease of exposition we assume constant
return to scale technologies (linear). Table 13.1 shows the required units
of labour for production of each commodity. Country A can produce 1kg
of wheat using 2 units of labour and 1 car using 3 units of labour. Country
B on the other hand can produce 1kg of wheat using 1 unit of labour and
1 car using 2 units of labour. Furthermore, both countries have 6 workers
that can allocate to production of cars and wheat.
One car One kg of wheat
Country A 3 2
Country B 2 1
Table 13.1: Units of labour required for production of each commodity.
Is there any gain from trade for these countries? To answer this question
we first consider the production possibilities of each country under no
trade (car production is in full units but wheat could be produced in
fractions). Let (c,w) represent production of c units of cars and w kg of
wheat. Country A can choose (1,1.5) by allocating labour equally to each
activity. It can also specialise in either activity and produce (2,0) or (0,3).
158
Similarly country B could choose from four possible allocations of labour

corresponding to (0,6),(1,4),(2,2),(3,0) production plans. These points show
the production possibility frontier (PPF) of countries A and B in autarky.
Given the preferences of individuals in these countries, one point from
PPF is picked for actual production and consumption. If both commodities
are produced (and consumed), in the competitive equilibrium the car
wheat price ratios are 32 and 2 in countries A and B (remember that in
competitive equilibrium the marginal rate of transformation, marginal rate
of substitution and price ratio were equalised). The car wheat price ratio
is smaller in country A and therefore it has a comparative advantage in
the production of cars. In other words, country A transforms a smaller kg
of wheat into one car and therefore it is comparatively more efficient in
the production of cars. On the other hand, country B has a comparative
disadvantage in the production of cars.
Activity 13.1
Which countries have an absolute cost advantage in the production of cars and wheat
(assuming wages are the same across countries)?
Opening up to trade in this simple model implies countries specialise in

their comparative advantage. Country A specialises in car manufacturing
and country B focuses on wheat production. This results in a world
production of 2 cars and 6 kg of wheat. Is this better or worse than the
autarkic outcome? How do countries decide on how many cars can be
traded for how much wheat? This is a question about the international car
wheat price ratio. Assuming both countries never choose zero consumption
of any commodity, we can show that the outcome with free trade is an
improvement over autarky and the international car–wheat price ratio
must be between 1.5 and 2.
Country B can exchange 1.8 kg of wheat with 1 car from country A. This
yields (1,4.2) consumption of car and wheat, which is better than the
autarkic allocation of (1,4) for Country B. Country A is also happy with this
trade as it would receive (1,1.8) which is an improvement over the autarkic
outcome of (1,1.5). Therefore, an international car–wheat price ratio of
1.8 leads to welfare gains for both countries. Notice that the argument
here is inconclusive about the exact equilibrium price ratio and 1.8 is only
an example that shows the working of the idea. To find the equilibrium
price ratio you need to incorporate consumer preferences and solve for the
world equilibrium prices, which is beyond the scope of this course.
Activity 13.2
Consider the same example as in Table 13.1 but now assume country B uses 2 units of
labour to produce 1 kg of wheat. Which country has a comparative advantage in the
production of cars? What are the potential production plans under autarky? What is the
pattern of specialisation with free trade? Do both countries gain from free trade? Why?
Would trade be welfare-improving if country A has the same productivity parameters as
country B?
It is worth noting that country B has an absolute advantage in production

of both commodities because it uses less labour than A to produce any of
the commodities. This, however, is irrelevant for trade patterns (as far as
labour is immobile). What determines trade flows in this simple example is
comparative advantage.
159
Comparative advantage in the example outlined above is due to

productivity differences. Trade models featuring productivity-based
comparative advantage are called Ricardian models. In Heckscher-Ohlin
models trade is based on comparative advantage due to the relative
abundance of factors of production. The simplest form of a Heckscher-
Ohlin model considers two countries producing two commodities using
two factors. Countries specialise in producing of commodities that use
their relatively abundant factor more intensively.
Let us modify the earlier example and add in capital as another factor
of production. Also assume that both countries have access to the same
technologies for the production of cars and wheat. Car production is,
however, capital intensive whereas wheat employs labour more intensively.
Also assume country A has a higher labour-to-capital ratio compared to
country B. In other words, labour is relatively more abundant in country
A. With free trade, country A specialises in production of wheat and trades
some of its wheat production with country B’s car manufacturing. Since
developing countries usually have lower capital labour ratios this theory
predicts they specialise in labour intensive production and import capital
intensive commodities from developed countries.
Activity 13.3
Could there be factor endowment comparative advantage when there is only one factor
of production (e.g. labour)?
13.2.2 New trade theories

Although theories based on comparative advantage are to some extent
successful in explaining inter-industry trade, they could not provide an
explanation for the large and growing intra-industry trade. Trade data
reveals countries engage in trade of closely substitutable commodities.
For example, Melitz and Trefler (2012) show more than 35 per cent of
all trade volumes in 2006 was within narrowly defined industries. For
example, one country exports Ford cars and, at the same time, imports
Toyota cars from a second country.
New trade theory appeared partly to explain intra-industry patterns of
trade. In these models consumers get utility from consuming a larger
variety of goods produced in the same industry. Production features
economies of scales, therefore serving a larger market reduces average
cost of production. In these models, intra-industry trade increases welfare
by increasing the set of varieties available to consumers. Furthermore,
access to larger markets reduces production costs and increases profits.
Activity 13.4
Consider countries A and B which can produce car varieties. Each new variety has a fixed
cost (e.g. design and assembly line) of 6 units of labour and once invented requires 1
unit of labour to produce 1 car. Consumers get more utility from consumption of more
varieties and higher quantities. Also assume both countries have 15 workers. Identify
potential production plans under autarky. What are the potential production plans under
free trade? Would consumers prefer free trade?
When new trade theories are combined with industry level comparative
advantage they provide a reasonable account of international trade flows
across countries. They, however, fail to account for large productivity
disparities across firms within narrowly defined industries. Recent
evidence suggests that high productivity firms self-select into exporting
160
because they are better able to overcome fixed export costs and compete
in international markets. Exporters, however, constitute a small fraction
of all firms within given industries. Bernard et al. (2007) show in US
Census for Manufacturers that the percentage of firms exporting varies
from 2 to 38 across sectors and, on average, 18 per cent of firms export.
In their sample, exporters are 11 per cent more productive than non-
exporters within manufacturing sectors. Similarly, exporters have higher
sales and employ more workers. Interestingly, exporters in developing
countries seem to be more capital-intensive and employ a larger number
of skilled workers. In contrast to this a simple comparative advantage story
predicts that developing countries that have abundant unskilled labour
should specialise in export of commodities using unskilled labour more
intensively. Models with heterogeneous firms introduce a new source
of gains from trade which we explore below (section 13.3). Table 1 in
Bernard et al. (2007) shows the range of trade models and their ability in
explaining stylised facts.
13.2.3 Gravity
The gravity model states that trade flows between two regions are
positively related to incomes of trading partners and negatively related
to the distance (trade costs) between the two. The gravity model outlines
factors that influence trade costs and size of the target market. Higher
income increases demand for imports and also raises the ability of
countries to engage in exporting. On the other hand, trade costs are higher
for countries further apart because of transport and other costs (such as
familiarity with the destination market).
Gravity is complementary to the theories outlined above. For example,
comparative advantage determines patterns of specialisation while trade
costs and size of the markets pin down volume of trade flows and the
choice of trading partners.
► Stop and read

13.3 Trade liberalisation

Let us think about a situation where barriers to trade are significantly
reduced. This could be a transition from high to lower tariffs or the
expansion of transport infrastructure like railways, ports, etc. In this
section we discuss gains and losses from such reductions in trade barriers.
The term ‘trade liberalisation’ is, however, normally used to describe the
removal of policy barriers to trade. The central questions that we ask are
whether trade liberalisation raises welfare on average and how such gains
are distributed.
13.3.1 Gains from trade liberalisation

In trade theories based on comparative advantage, welfare gains arise
from specialisation in comparative advantage. By putting resources into
the production of commodities that countries are relatively better at,
countries could achieve superior consumption baskets. Trade liberalisation
removes the barriers to using such gains and therefore improves welfare.
New trade theory predicts additional welfare gains from liberalisation,
in terms of an increased number of varieties available to consumers and
the utilisation of economies of scale and lower production costs. Trade
liberalisation increases the choices available to consumers as foreign
161
varieties become available; since consumers attach utility to more

varieties, welfare is increased. On the other hand, access to larger markets
allows producers to use economies of scale and achieve lower average
costs. In a dynamic setting, access to a larger market results in stronger
incentives to invest in innovations, due to higher profitability.
In the presence of firm heterogeneity trade liberalisation results in the
contraction and exit of low productivity firms and the expansion of high
productivity firms. Contraction of low productivity firms happens along
two channels:
1. First, the presence of foreign competitors in domestic markets drives
down prices and lowers profits for low productivity firms, leading to
contraction or exit.
2. Secondly, expansion of exporters due to better access to foreign
markets increases the domestic price of factors (labour and capital)
as well as the cost of production, leading to greater pressure on low
productivity firms.
Therefore, after liberalisation a greater percentage of resources are
employed by high productivity firms and the allocative efficiency of inputs
is improved. Average productivity would also improve.
► Stop and read

Melitz and Trefler (2012).
13.3.2 Distributional impacts

Trade liberalisation has distributional implications. In a Heckscher-
Ohlin model, opening to trade results in the expansion of sectors
with comparative advantage increasing the price of factors used more
intensively in these sectors. If factors of production are not distributed
equally among the individuals in society, trade liberalisation creates
gainers and losers. Owners of the factors used intensively in the
comparative advantage sector gain from higher compensation and owners
of other factors lose.
Activity 13.5
Let us think about a country with two sectors and two factors of production (labour and
capital). The country has a comparative advantage in the capital intensive sector. Why
might this country have a comparative advantage in this sector? Now assume there are
10 individuals in this country and each owns 1 unit of labour and 1 unit of capital. How
does trade liberalisation change the income distribution in this country? Now assume that
capital is concentrated in the hands of 5 individuals. How does trade liberalisation change
the income distribution?
Therefore, even if in the long-run, trade liberalisation increases total

welfare, it might involve large short-run adjustment costs and a rise in
inequality. In heterogeneous firm theories, the reallocation of resources
to high productivity firms has similar implications. Although aggregate
productivity is improved, the low productivity firms exit the market.
Depending on the bankruptcy laws and friction in the labour markets we
might have high adjustment costs borne mostly by workers in the low
productivity plants as they need to relocate and find a new job.
Distributional concerns and the presence of adjustment costs create a
significant role for government policy during trade liberalisation episodes.
Measures aimed at alleviating these problems should not necessarily involve
162
trade policy. For example, the government might use existing poverty
alleviation schemes to help out the workers who have lost their jobs.
For an elaborated discussion of the links between trade and inequality and
the empirical evidence, see Goldberg and Pavcnik (2007).
13.3.3 Empirical evidence

Pavcnik (2002) uses Chile’s trade liberalisation during the 1970s to
estimate the productivity gains from trade. During 1974 to 1979 Chile
eliminated all non-tariff barriers and reduced tariff rates to a unified 10
per cent ad valorem rate across all industries. Pavcnik uses a census of all
manufacturing plants in Chile with 10 or more employees during 1979–86
to investigate the impact of liberalisation.
The identification strategy compares firms exposed to trade (import
competing and export oriented) to firms in non-traded sectors over time.
This is similar to a difference-in-differences strategy except for the fact
that she does not have data from before the liberalisation period. Pavcnik
(2002) reports that the productivity of import competing plants increased
by 3–10 per cent more than firms in non-traded sectors after liberalisation.
This confirms that firms facing greater competition after liberalisation
became more efficient. On the other hand, export-oriented plants did not
see productivity gains after liberalisation, potentially because they already
had high productivity. She also finds exiting plants are on average 8 per
cent less productive than firms continuing to produce post liberalisation.
Overall, two-thirds of trade liberalisation gains were due to the exit of
low productivity firms and expansion of high productivity firms. One-
third of the productivity gain came from within-plant productivity gains,
potentially due to reallocation of resources between activities within
plants.
Bernard et al. (2007) review several other articles and conclude that
productivity gains from within-industry reallocation of resources dominate
cross-industry reshuffling emphasised by comparative advantage theories.
► Stop and read

Bernard et al. (2007).
13.4 Trade policy

In the previous section we have seen several arguments for gains from
trade but, in some circumstances, governments might want to impose
restrictions on trade flows. Two lines of argument could justify restricted
trade as an optimal strategy. First, the government might be concerned
about inequality. We have seen that trade liberalisation involves large
reshuffling of resources and there is some evidence that the recent waves
of globalisation have led to higher inequality (Goldberg and Pavcnik,
2007). The government might therefore restrict trade in sectors that harm
the poor (e.g. protect low skilled sectors). Second, free trade gives Pareto
efficient outcomes when all markets are working. The presence of market
failures and externalities could justify protection as a second-best policy.
We explain this dea more when we discuss the infant industry argument
below.
A large literature on the political economy of trade policy proposes a
different explanation for observed trade restrictions. Trade liberalisation
increases competition and erodes monopolistic profits. Monopolies are
likely to engage in activities to fence off their monopolistic rents and
therefore seek government protection from foreign competition. Other
163
sectors might want a different set of protective policies. For example,

one sector might produce an intermediate product. Other sectors using
this input would like trade to be liberalised for the intermediate good
they are using to reduce their costs but prefer restrictions on their own
product to maintain monopoly rents. In equilibrium, sectors with stronger
lobbies receive more protection at the expense of others who remain
unprotected. In this framework, trade barriers are not optimal from a
social perspective but nevertheless we observe them as an equilibrium
outcome (see Baldwin, 1989, for a review of early literature, Grossman
and Helpman,1994, on the theoretical modelling of the issue, and Maggi
and Goldberg, 1999, for empirical evidence).
In the remaining part of this section we look at import tariffs and their
likely impact on welfare in a world without market failures. We then
introduce the infant industry argument and potential market failures
justifying it. Finally, we touch on the role of trade agreements.
13.4.1 Import tariffs

Setting a tax on the import (or export) of goods and services will help
the government to control trade flows and raise tax revenue at the
same time. To better understand how tariffs work, let us consider a very
simple demand and supply framework for garments. Figure 13.1 shows
a downward sloping domestic demand and an upward sloping domestic
supply. A higher price results in lower demand and higher supply because
for consumers the good is more expensive and for producers it is more
profitable. The autarkic equilibrium price and quantity are shown on the
figure as Pa and Qa. Let us assume that the domestic market is small and
the international price of garments is lower than the autarkic domestic
price (P*<Pa). With free trade the price of garments will fall to the
international price (P*) and the country imports Qd – Qs units of garments
to accommodate the additional demand.
Activity 13.6
Why does opening its domestic markets to trade imply this country imports garments
from the rest of the world?
Consumer and producer surpluses under autarky are given by the areas C
and A+B+D. Total surplus (welfare) is the sum of these. With free trade
the price falls to P* and consumer surplus increases to C+D+E+B+F+R+G.
Producer surplus (profits of firms in the garment industry) is, however,
reduced to A. As you can see, some of the producer surplus is transferred
to consumers through lower prices (i.e. area B+D). This is a distributional
effect. But total surplus with free trade is larger than autarkic welfare by
F+E+R+G. This is the gain from trade.
Now suppose the government introduces an ad valorem tariff equal to t.
The effective price of imported garment is now P*(1 + t). This eases off
competitive pressure on domestic producers and allows them to expand
production to Q's increasing their surplus to A+B. Consumers lower their
demand because of higher prices and garment imports is given by
Q'd – Q's. Consumer surplus with tariffs is given by the area C+D+E. The
government collects tax revenue from setting tariffs, which is equal to
(P*t × (Q'd – Q's). This is area R in Figure 13.1. Adding up consumer and
producer surpluses and government tariff revenue, you can calculate the
total surplus. Total surplus is reduced by the sum of F+G compared to the
case of free trade. This is the deadweight loss arising from tariffs (taxes).
164
Price
Supply
C
Pa
D E
P*(1+t)
B F R G
P*
A Demand
Qs Q’s Qa Q’d Qd Quantity

Figure 1: Impact of an ad valorem tariff on welfare.
13.4.2 Import substitution and infant industry argument

Given the arguments in the previous section is it really the case that
imposing a tariff always reduces social welfare? After all, many countries
historically had high tariffs and used other instruments to manipulate
trade flows. The argument on the sub-optimality of import tariffs rests on
the assumption of no market failures. Furthermore, it ignores dynamic
considerations that might arise due to foreign competition or lack thereof.
Here we consider the impact of relaxing these assumptions.
One popular argument for trade restrictions relies on economies of scale.
Nascent industries with increasing returns to scale are unable to operate
at a profitable scale in the presence of foreign competition. The incumbent
producer has an average cost advantage that makes it impossible for
the domestic producer to undercut prices. The argument follows that
government needs to fence off nascent industries from foreign competition
until they are big enough (thereby enjoying economies of scale) to be able
to compete in the international markets.
Activity 13.7
Think about an industry with large fixed set up costs equal to F and constant marginal
cost equal to μ << F. Derive the average cost of production for this industry. Now assume
there is an incumbent producer in this market. Is it possible for a new entrant to start
production in the presence of the incumbent?
The infant industry argument implicitly assumes that credit markets

are imperfect and therefore the new entrant cannot get credit for an
extended number of periods to incur losses while it is catching up with
the incumbent. This is quite plausible but there are two inherent problems
with the infant industry argument. First, the government needs to
identify industries with increasing returns that need protection. Second,
the government needs to decide when to remove its support. Both of
these decisions are highly contentious and induce rent-seeking activities.
Industry lobbies try to bribe policy makers in order to receive favourable
treatments and once under protection they have an interest in remaining
protected.
165
Activity 13.8
Car manufacturing started off in country A as a nascent industry with increasing returns
and received protection against foreign competition in the form of high import tariffs. The
industry has evolved to a large set of inefficient factories employing one per cent of the
labour force in the economy. What are the potential arguments that car producers might
put forward to keep the high tariffs?
► Stop and read

13.4.3 Trade agreements

In the recent decades import tariffs have been falling dramatically. Harrison
and Rodríguez-Clare (2010), for example, show average statutory tariffs
were significantly reduced between 1980 and 2004 for many developing
countries. For example, the average statutory tariff rate in India was reduced
from 74 per cent in 1980 to 28 per cent in 2004. Most of these tariff
reductions are due to expansion of the World Trade Organization (WTO)
and the proliferation of regional and bilateral trade agreements. In trade
agreements the participating countries mutually agree to ease off trade
barriers to the transfer of goods, services or workers between themselves.
The European Union is an example of an agreement that encompasses all
three areas and allows free trade within the member countries.
► Stop and read

13.5 Summary
The failure of old trade theories based on comparative advantage to
explain intra-industry trade led to the emergence of new trade theories
that featured economies of scale and a preference for more varieties.
When combined with industry-level comparative advantage, new trade
theories provide a reasonable account of international trade flows. New
data on the nature of firms engaged in trade necessitated the need to
relax the representative firm assumption in old and new trade theories.
Fixed export costs in these models imply that only firms with high enough
productivities enter foreign markets.
The evolution of trade theories sheds light on potential gains from trade.
In old theories, specialisation in comparative advantage resulted in mutual
welfare gains for trading partners. Liberalisation in new trade theories
leads to the expansion of available consumption varieties and use of
economies of scale by producers. Adding in firm heterogeneity brings in
the additional reshuffling of resources between firms involved in trade and
those serving domestic customers only. Greater access to foreign markets
and higher competition mean low productivity firms contract and exit the
market, leading to higher average within-industry productivity. Empirical
evidence suggests that this final channel might dominate the inter-industry
specialisation emphasised in comparative advantage theories.
The reshuffling of resources post-liberalisation episodes has significant
implications for the success of trade policy. Ignoring the distributional
effects of trade policy could significantly undermine the implementation of
liberalisations and erode the steady state gains identified above. Existing
institutions such as labour market regulations and bankruptcy laws have a
key role in shaping the speed of adjustment and likely costs.
166
Distributional concerns and market failures provide grounds for

protectionist policies. Benevolent governments intending to reduce
the distributional impacts of trade might set tariffs or quotas to restrict
flows of goods and services. Credit market failures and presence of scale
economies could justify protectionist policies, because new entrants would
be unprofitable in the short run. Protectionist policies, however, induce
rent-seeking activities and, once in place, are hard to get rid of. It seems
that the new wave of multilateral trade agreements under the auspices of
the WTO has led to a lowering of tariff barriers to trade.

Having completed this chapter, as well as the Essential reading and
• analyse gains from trade using various trade theories
• identify distributional issues in trade liberalisation
• discuss the welfare implications of trade policy.

1. Consider the markets for garments, as in the text, but now assume that
the international price is higher than the autarkic equilibrium price
(p* > pa).
a. Find consumer and producer welfare under free trade and
compare it to autarkic surpluses. Is free trade welfare improving?
Draw an appropriate diagram and show surpluses as areas. Is the
country still a garment importer?
b. Now consider an ad valorem export subsidy on garments equal to
s. What is the equilibrium domestic price after this subsidy? Find
the consumer and producer surplus. Is the sum of the consumer
and producer surplus greater than the total surplus under free
trade? Is total surplus with a subsidy greater than the total surplus
under free trade?
3. Another argument for protecting industries is based on learning by
doing. Unless you start building ships, you do not learn an efficient
way of producing them. Under free trade, established shipbuilders
who already have high productivity capture the domestic market and
reduce the viability of any domestic shipbuilding industry. Discuss the
merits and deficiencies of this argument.
4. Provision of subsidised credit to exporting firms is welfare reducing
because it pushes firms to export beyond the optimal level. Discuss.

on the VLE.
167
Notes
168
Chapter 14: Environment and development
Chapter 14: Environment and

development
14.1 Introduction
• understand the importance of the environment for the process of
development
• discuss the impacts of the environment on economic outcomes
• explain some of the mitigation strategies for environmental issues.

• outline how environmental issues affect developing countries
• explain the impacts of climate change, pollution and natural disasters
in developing countries
• critically evaluate mitigation strategies for controlling deforestation
and pollution.

Burgess, R., O. Deschenes, D. Donaldson and M. Greenstone ‘Weather and
death in India’, Working Paper, 2011. Available at: http://goo.gl/dAXALt
Burgess, R., M. Hansen, B. Olken, P. Potapov and S. Sieber ‘The political
127(4) 2012, pp.1707–754.
Strömberg, D. ‘Natural disasters, economic development, and humanitarian
aid’, Journal of Economic Perspectives 21(3) 2007, pp.199–222.

Dell, M., B. Jones and B. Olken ‘Temperature shocks and economic growth:
evidence from the last half century’, American Economic Journal:
Macroeconomics 4(3) 2012, pp.66–95.
Duflo, E., M. Greenstone, R. Pande and N. Ryan ‘Truth-telling by third-party
auditors and the response of polluting firms: experimental evidence from
Ebenstein, A. ‘The consequences of industrialization: evidence from water
pollution and digestive cancers in China’, Review of Economics and Statistics
94(1) 2012, pp.186–201.
Graff Zivin, J. and M. Neidell ‘Environment, health, and human capital’,
Greenstone, M. ‘Estimating regulation-induced substitution: the effect of the
Clean Air Act on water and ground pollution’, American Economic Review
93(2) 2003, pp.442–48.
Jayachandran, S. ‘Liquidity constraints and deforestation: the limitations of
payments for ecosystem services’, American Economic Review 103(3) 2013,
pp.309–13.
Maccini, S. and D. Yang ‘Under the weather: health, schooling, and economic
consequences of early-life rainfall’, American Economic Review 99(3) 2009,
pp.1006–026.
169
Newell, R.G., W.A. Pizer and D. Raimi ‘Carbon markets 15 years after Kyoto:
lessons learned, new challenges’, Journal of Economic Perspectives 27(1) 2013,
pp.123–46.
Stern, N., ‘The economics of climate change’, American Economic Review 98(2)
2008, pp.1–37.
14.1.5 Works cited

Acemoglu, D., S. Johnson and J.A. Robinson ‘Reversal of fortune: geography and
institutions in the making of the modern world income distribution’, Quarterly
Besley, T. and R. Burgess ‘The political economy of government responsiveness:
theory and evidence from India’, Quarterly Journal of Economics 117(4) 2002,
pp.1415–451.
Intergovernmental Panel on Climate Change, ‘Climate change 2013: the physical
science basis’. Available at:
www.ipcc.ch/report/ar5/wg1/#.UnDk7PlSh8E.
Jayachandran, S. ‘Air quality and early-life mortality: evidence from Indonesia’s
wildfires’, Journal of Human Resources, 44(4) 2009, pp.916–54.
Sachs, J.D. ‘Institutions do not rule: direct effects of geography on per capita
income’, National Bureau of Economic Research Working Paper 9490, 2003.
Zheng, S. and M.E. Kahn, ‘Understanding China’s urban pollution dynamics’,
14.1.6 Overview
The process of economic development might create environmental
externalities that inflict irreversible damages to individuals and the
environment. In this chapter we consider three types of environmental issues
that might have large impacts in developing countries: climate change,
environmental pollution and natural disasters. We review recent evidence on
the impact of these phenomena on economic outcomes (section 14.2) and
try to evaluate some of the mitigation strategies (section 14.3). We will only
cover part of these topics due to the vast range of issues involved and the
diversity of impacts.
Climate change is a result of global warming due to a rise in the stock of CO2
and other greenhouse gases. While developing countries’ contribution to the
current stock of CO2 is small and developed countries’ emissions in the past
account for most of the stock, continuing with the current level of emissions
could have wide-ranging consequences across the globe. It is interesting to
note that developed countries are still major CO2 emitters: 44 per cent of all
CO2 emissions come from high income countries, while low income countries
account for less than one per cent of emissions in 2010 (The World Bank:
World Development Indicators). We review the existing evidence on the
impact of temperature and rainfall fluctuations on individuals in developing
countries. We consider the determinants of deforestation in the tropics, as
a major contributor to emissions, and potential strategies for controlling it.
We also touch on the carbon trading schemes as a common policy tool for
controlling emissions in the European Union and other rich countries.
Environmental pollutions can be defined as damaging externalities from
industrial production. In contrast to climate change, most of the damages
here are concentrated in the vicinity of the polluting factory. The biological
knowledge of harmful substances and methods of curbing pollutions are also
well established. Most countries use various types of regulation and restrict
pollution below a critical threshold. There is, however, increasing evidence
that developing countries have difficulties in enforcing such regulations. We
discuss one randomised experiment that tried to fix the regulatory structure
for polluting factories in Gujarat, India.
170
Natural disasters, in their different forms, are a fixture of news every

day. Floods, droughts, hurricanes, volcanic eruptions and so forth
happen across the world. Developing countries, however, seem much
more vulnerable to disasters. There is also a chance that climate change
contributes to more devastating natural disasters. Similar disasters that
happen in rich and poor countries have drastically different deaths
rates. Furthermore, developing country governments are often unable to
organise appropriate relief and prevention strategies. We present stylised
facts about the impact of disasters across developing and developed
countries and discuss potential issues in disaster relief policies.
14.2 The impact of the environment on outcomes

We have seen that many individuals in developing countries rely on
agriculture. Agricultural activities rely heavily on the weather. Most
agriculture in developing countries is rain fed and subject to rainfall
fluctuations. Even more resilient irrigated plantations greatly depend
on temperature fluctuations and climatic events like storms. Loss of
agricultural income could have devastating impacts on the livelihoods
of poor rural households. In this section we review some of the evidence
on the impact of climate change, pollution and natural disasters on the
livelihoods of individuals in developing countries.
14.2.1 Climate change

Climate change could impact on humans in several ways. Increased
temperatures could change agricultural productivities. Rising sea
levels could flood islands and undefended coastal areas. Higher ocean
temperatures and increased acidity due to higher CO2 content could affect
marine life significantly and reduce fish stocks. Although climate change is
an externality, four features distinguish it from conventional externalities:
1. Climate change arises due to the actions of all countries of the world
and it impacts on everyone in a heterogeneous way.
2. As climate change happens through the increased stock of greenhouse
gases, the negative consequences materialise over very long periods.
3. The consequences of climate change are uncertain as events happen
over long periods and are felt across the world.
4. The impacts of climate change are potentially very large and often
irreversible, increasing the stakes in getting the policies right.
See Stern (2008) for a great review of these issues and their implications
for policy making.
Activity 14.1
Outline the key differences between the externality arising from a factory that pollutes a
nearby lake and climate change externality.
Separating short-term weather fluctuations from long-run trends in

climate change is a daunting task. However, the evidence increasingly
supports the idea that the climate is changing in response to emissions,
pollutions, and other actions of human beings. While in its previous
reports the Intergovernmental Panel on Climate Change (IPCC) suggested
there might be a link between human actions and global warming, in its
Fifth Assessment Report, the IPCC firmly concluded that human actions
are responsible for observed changes in the climate:
171
Warming of the climate system is unequivocal, and since the 1950s, many
of the observed changes are unprecedented over decades to millennia. The
atmosphere and ocean have warmed, the amounts of snow and ice have
diminished, sea level has risen, and the concentrations of greenhouse gases
have increased. (IPCC, 2013)
As many of the changes in the climate are unprecedented, we lack the

ability to forecast changes in the climate, let alone predict its impacts on
our lives. This, however, does not mean we should give up hope. There are
useful lessons to be learned from the impact of weather fluctuations. In
this section we discuss three studies that look at the impact of temperature
and rainfall variations on economic outcomes.
Burgess et al. (2011) investigate the relationship between weather
fluctuations and mortality in India. They combined daily temperature
and rainfall data at district level with mortality and other variables.
Theoretically, high temperatures could affect mortality directly
and indirectly. Being exposed to high temperatures might result in
cardiovascular failure when the body tries to stabilise core temperature
(heat stress). This effect could be stronger if individuals are engaged in
physical labour in the outdoors. On the other hand, high temperatures
and rainfall result in a worse disease environment. For example, malaria
mosquitoes thrive in such weather conditions and the chances of intestinal
infections increase. Finally, high temperatures could directly impact on
mortality by causing damage to agricultural crops and lowering individual
incomes.
Burgess et al. (2011) adopted a flexible specification to estimate the
impact of temperature variations on mortality. They split temperature
observations into 15 ‘bins’ ranging from below 10°, 10–12, 12–14, …,
34–36, and above 36°C. They regress mortality rate on 15 variables
showing the number of days with mean daily temperatures in each of the
15 temperature bins. The results show the significant impacts of days in
temperature bins above 30°C only for rural areas and in growing season
(see Figures 1, 3, 4, 6 in Burgess et al.). Urban areas seem to be unaffected
by temperature variations. High temperatures also have a negative impact
on mortality if they happen during the growing season.
Overall Burgess et al. show high temperatures have striking effects on
rural mortality in India. The results are consistent with the dependence of
rural households on agricultural incomes. Unless new resistant crops are
introduced, it is likely that rural mortality increases as a result of more
frequent hot days and higher average temperatures.
Activity 14.2
Explain why observing a significant effect of temperature during the growing season and
for rural areas only is supporting the indirect channel of temperature impact on mortality
through income.
Maccini and Yang (2009) find there is a significant effect of rainfall shock
during a child’s birth year on their health and education in rural Indonesia.
They find that 20 per cent higher-than-average rainfall during a child’s
birth year results in a 3.8 percentage point reduction in the likelihood
of their reporting poor health and 0.22 more years of schooling. This is
consistent with the findings of Burgess et al. (2011) because they also
saw a significant impact of deficient rainfall on mortality and agricultural
productivity.
172
The two studies we discussed so far estimate the impact of weather shocks
on individuals. There might be aggregate mechanisms that affect GDP
growth over and above its effect on individuals. Dell et al. (2012) use a
panel of around 125 countries during 1950–2003 and combine yearly
temperature and precipitation data for each country with GDP figures
to estimate the aggregate impacts of temperature fluctuations. The fact
that countries with higher temperatures are poorer is well established
(northern Europe versus Sub-Saharan Africa). But whether one can draw
causal conclusions from cross country correlations is hotly debated in
the literature. For example, Sachs (2003) assigns a causal role to high
temperatures in reducing incomes, while Acemoglu et al. (2002) argue for
other factors being responsible for the observed correlations.
Activity 14.3
Why might the observed cross country negative correlation between temperature and
income not show the causal effect of temperature on income?
Dell et al. (2012) use country fixed effects together with region and poor
country specific time effects to identify the causal effect of temperature
on income. Including country fixed effects (one dummy variable that is
equal to 1 for observations relating to that country and zero otherwise)
would capture any observed or unobserved factor that matters for growth
but is fixed over time. For example, the income effect of sparse population
density for hotter countries (e.g. because of large deserts) will be captured
by the country fixed effect because having a sparse population is a fixed
factor over time. Furthermore, including year dummies interacted with
region and poor country dummies allow for yearly changes with specific
effects on different regions and poor versus rich countries. To the extent
that we expect yearly temperature and precipitation changes within
countries to be exogenous with respect to other (time varying) factors
influencing economic growth within the same country, this identification
allows causal interpretation of the results.
The results show that a 1°C rise in temperature leads to a 1.3 percentage
points reduction in contemporaneous growth in poor countries but has no
effect in rich ones. Interestingly, the contemporaneous effect persists over
time, suggesting that temperature shocks have growth effects rather than
just level effects. Dell et al. (2012) try to shed light on the mechanisms
driving temperature effects. They find temperature fluctuations affect both
agricultural and industrial value-added growth rates in poor countries.
This is in contrast to Burgess et al. (2011) finding that urban mortality,
productivity and wages were not affected. Dell et al. (2012) also do not
find a significant effect for rainfall in poor countries.
► Stop and read

Burgess et al. (2011).
14.2.2 Pollution
Detrimental impacts of various types of pollution on health are widely
recognised, but estimating the health and economic impacts of pollution
is not straightforward. For example, residential sorting could result in a
concentration of poorer households in high pollution areas. Poor households
might have worse health and economic outcomes, due to factors other than
pollution leading to an overestimation of the negative impacts. Alternatively,
cities offer greater employment opporunities but are also a source of
pollution. Individuals living in cities might be more productive and take
173
more health investment to compensate for the harmful impacts of pollution,

leading to an underestimate of pollution effects. Environmental pollution
could both have contemporaneous and long-run effects. Prenatal and early
life exposure to pollution could have damaging lifetime effects. Developing
countries might be particularly vulnerable to environmental pollution due to
a lack of enforcement powers by the government. In a later section we will
discuss some of the issues around environmental regulation.
Activity 14.4
Mercury is a poisonous element which is banned in many industrial processes. Small-
scale gold mining in developing countries, however, uses a method of extraction which
exposes workers to mercury. To see the devastating impact of exposure to mercury, see
the article by BBC News on the Indonesian gold mining industry: www.bbc.co.uk/news/
magazine-24127661. Why do you think individuals might underestimate the damages
caused by mercury in developing countries?
Until recently most of the studies on the impact of pollution were done
in developed countries. Here we review two recent studies on developing
countries. Jayachandran (2009) studied the impact of Indonesian wildfires
in 1997. Wildfires resulted in high density of particulate matters in the
air over a short period in specific regions. Mothers in the third trimester
of pregnancy exposed to this event experienced a reduction in survival
rates of their infants. Jayachandran combined daily satelite imagery (for
expansion of the smoke) and census data (to measure the size of cohorts)
and arrived at a large and significant estimate of a 17 per cent increase in
mortality for children under the age of two.
In a second study, Ebenstein (2012) used the rise in water pollution levels
due to economic development in Chinese regions in the past decades to
estimate the impact of pollution on the likelihood of cancer. The evolution
of industries and hence pollution varied greatly across different regions.
The author used regional variation in river basin pollutions to identify the
effect of water pollution on digestive cancers. He estimated the effect of
water pollution on the likelihood of digestive cancer using ordinary least
squares while he controlled for average education in the area, share of
farming, whether the area is urban, level of air pollution and other factors.
He found that deterioration of water pollution by one grade in a six-grade
scale results in a 9.7 per cent increase in digestive cancers.
Activity 14.5
What is a potential problem for causal interpretation of Ebenstein’s (2012) results? Does
the fact that China has strict migration controls help you alleviate some of the concerns?
Graff Zivin and Neidell (2013) provide a review of the literature on the
impact of pollution on health and other outcomes from both developing
and developed countries.
14.2.3 Natural disasters

While low and high income countries have similar exposure to disasters,
low income countries suffer much more from natural disasters. Strömberg
(2007) reports that during 1980–2004 high income countries received
1,476 disasters which affected 440 million individuals and resulted in
75,425 deaths. However, during the same period 1,533 disasters happened
in low income countries and affected 496 million individuals and resulted
in 907,810 deaths. This means high income countries had 51 deaths per
disaster while low income countries had 592 deaths per disaster (more
than 10 times larger, see Table 3 in Strömberg’s article).
174
While some of the difference in death figures across high and low income
countries can be attributed to differences in magnitude of the disasters
striking low income countries, a regression framework confirms the idea
that poor countries are more vulnerable to disasters. Table 4 in Strömberg
(2007) reveals that the same type of disaster would have on average 70
per cent lower deaths in high income countries (controlling for year and
continent fixed effects and several other factors).
Activity 14.6
Why might poor countries have higher death tolls in the face of disasters?
Governments of poor countries have fewer resources to engage in effective

disaster relief or prevention. Humanitarian aid plays a role in reducing
suffering, but it is not clear whether aid could go beyond providing
blankets and temporary food supplies. In the chapters on insurance
and credit we have seen the difficulties of consumption smoothing in
developing countries, especially in the face of aggregate shocks like
disasters that affect large areas. In a later section we consider some
further evidence on how political incentives might shape government
responsiveness to disasters.
► Stop and read

Strömberg (2007).
14.3 Mitigation strategies

In this section we look at some strategies to reduce the negative
consequences of environmental phenomenon discussed above. We
try to deal with the following questions. How can developing country
governments and individuals mitigate the consequences of environmental
shocks? What are potential policies for reducing CO2 emissions and
slowing climate change? How can governments effectively enforce
environmental regulations? What shapes the incentives in disaster relief?
14.3.1 Deforestation
Deforestation is responsbile for 20 per cent of the CO2 generated by
humans. The IPCC (2013) puts CO2 emissions as the main contributing
factor to rising surface temperatures. In addition to CO2 emissions, the
disappearance of forests results in the mass extinction of plant and animal
species. Biodiversity has both an intrinsic and functional (e.g. for the
stability of ecosystem) importance.
Deforestation is happening mostly in the tropical forests of the Amazon,
the Congo basin and South East Asia. The main motives for cutting down
trees is logging and freeing up land for agriculture. National governments
often have de jure forest utilisation standards but local entities are
responsible for their enforcement. Due to imperfect monitoring, local
officials can allow logging beyond mandated quotas. Evidence from
Indonesia suggests illegal logging is responsible for 60 to 80 per cent of
deforestation (Burgess et al., 2012).
Activity 14.7
Explain why private individuals residing in forest areas might rationally choose to clear
out the forest for agriculture despite its negative consequences. What is the bribe
individuals are willing to give to the local authority to illegally clear the forest?
175
Burgess et al. (2012) use a novel dataset and identification strategy to

measure deforestation and better understand its drivers in Indonesia.
They use data from Moderate Resolution Imaging Spectroradiometer
(MODIS) satellite imagery to construct a measure of deforestation for each
250 metre by 250 metre block of forest during 2001–2008. This captures
both legal and illegal logging and is ideal for detecting local variation in
deforestation activity. Furthermore, by mapping the MODIS observations
to land use classifications they are able to observe if deforestation is
happening in production and conversion zones where the quota system
applies or in conservation and protection zones where no logging should
take place.
Activity 14.8
Why can Burgess et al. (2012) not rely on official production statistics in their study?
What type of deforestation is not captured by satellite data?
The scale of logging captured in the data is not very small and it is unlikely
the local officials do not notice the activity. Therefore firms engaged in
logging need to acquire permission from local officials either legally or
illegally through bribes. Burgess et al. (2012) propose a theory that firms
search to find the most profitable place for logging. Therefore district
officials face a downward sloping demand curve for permits. Asking a high
price (or for a bribe) will reduce the number of firms applying for permits.
Effectively, districts compete to attract loggers to increase rents for local
officials. As the wood markets in Indonesia are defined at province level
based on this theory, provinces with more districts are likely to have more
intense competition, leading to higher levels of deforestation and lower
wood prices.
Activity 14.9
Explain why comparing deforestation rates across provinces with high and low numbers
of districts is unlikely to give the causal effect of number of districts on deforestation.
To test this prediction Burgess et al. (2012) use the creation of new
districts in provinces as an exogenous variation to the number of districts.
Between 1998 and 2009 the number of districts more than doubled in
the forest islands of Indonesia. Presumably district splitting is unrelated
to deforestation and therefore it is exogenous with respect to the logging
activity.
The results suggest that adding a district to a province on average
increases the deforestation rate by 8.2 per cent and reduces local wood
prices by 3.4 per cent. This is consistent with the competition story from
their model. A larger number of districts intensifies competition between
districts in attracting loggers, which increases overall logging activity and
reduces the price of wood. The impact of an additional district on logging
in conservation areas is larger than production areas. Since all logging
in conservation areas is illegal, this provides direct evidence of increased
illegal logging within the province after district splitting. This study clearly
shows that incentives faced by local officials and the enforcement regime
matter greatly for forest conservation and that policies ignoring such
incentives are likely to fail.
176
Activity 14.10
Burgess et al. (2012) find that an additional district in the province results in 3.8 per cent
more deforestation contemporaneously. After three years the effect reaches 8.2 per cent.
Why might you expect an increase in deforestation since the initial district creation date?
One potential way of reducing deforestation is to offer payments

for ecosystem services (PES) to individuals to preserve the forest.
Jayachandran (2013) analyses the take up of one PES programme
offered by a local NGO in western Uganda to households owning forests.
Jayachandran, together with the NGO, ran an experiment offering PES
to 65 randomly chosen villages over the course of two years. Individuals
signed a contract to preserve trees on their own land and received
payments that were, on average, equivalent to around 10–15 per cent of
their annual expenditure.
The experiment is still ongoing, but preliminary take up shows that only
33 per cent of households took up the programme. Correlations in the
take up among the control group suggest that owners cutting trees to sell
are less likely to take up PES, while owners clearing forest for agriculture
are more likely to sign up for PES. This could be because of the timing
of PES payments and liquidity constraints faced by owners. Selling trees
provides cash right now but preserving them will deliver cash in one year’s
time. On the other hand, the timing of agricultural income as a result of
forest clearing is similar to PES payments. Owners in need of immediate
cash or facing liquidity constraints are therefore less likely to take up the
programme.
► Stop and read

Burgess et al. (2012).
14.3.2 CO2 trading schemes

One of the most important contributors to global warming is CO2
emissions. Combustion of fossil fuels, constituting 80 per cent of global
energy consumption, produces CO2 which remains in the atmosphere for
centuries and reflects surface radiations, resulting in higher temperatures.
Therefore, CO2 creates an externality that is global with uncertain impacts.
One method of correcting for this externality is to impose a tax equal
to the marginal cost of CO2 emissions on agents creating the externality
(Pigouvian taxation). Another method is to set permits for emissions and
allow firms to buy and sell permits on a CO2 market. A multitude of such
trading schemes is now in place around the developed world. See Newell
et al. (2013) for an overview of design issues and performance of existing
markets.
Activity 14.11
Given the distinct features of climate change externalities, why might Pigouvian taxation
not be appropriate for correcting them?
14.3.3 Environmental regulation

Environmental regulation is a natural and widely practised solution
for the prevention of pollution damages. Many countries in fact have
regulations for industries in terms of allowed chemicals and levels of
toxic emissions. However, it seems that developing countries are facing
higher levels of pollution with potentially devastating impacts (for a
177
review of China’s urban pollutions see Zheng and Khan, 2013). Higher
pollution levels in developing countries could be a result of higher needs
for income generation relative to pollution reduction. This suggests that as
countries get richer they tend to introduce more stringent regulations to
curtail environmental pollutions. On the other hand, it might be that poor
country governments lack the capacity to enforce existing environmental
regulations. The deforestation evidence in Burgess et al. (2012), discussed
above, supports the latter conclusion. In this section we look at a
randomised experiment that tries to overcome the enforcement problem.
Several countries use third party auditors to control compliance with
national environmental regulation. Usually polluting firms choose and pay
auditing firms. This creates a conflict of interest. Auditing firms have an
incentive to report lower pollution levels to receive payments and to be
chosen in future. Duflo et al. (2013) conducted a randomised experiment
in collaboration with the environmental agency in Gujarat to alter the
auditing market structure and study its impact on pollution.
They randomly assigned half of 473 audit-eligible factories in two
populous regions of Gujarat to a treatment group. The treatment changed
the auditing market structure in four ways. First, treatment plants had to
use a predetermined auditing firm. Second, payments to auditing firms
were made from a central fund at a flat rate. Third, 20 per cent of plants
randomly received a re-audit by a technical agency to verify the accuracy
of the earlier measurements (everyone knew the probability of a back
check visit was 20 per cent but verification visits were unannounced).
Fourth, second year payments to audit firms depended on the accuracy of
readings in the first year.
Duflo et al. (2013) report high corruption in the existing audit
arrangements with polluting firms reporting measurements just below
maximum standards. Back checks in the control plants revealed 29 per
cent of reports were false. The treatment significantly reduced false reports
by 23 percentage points (this is an 80 per cent reduction over the 29 per
cent false reports in the control). Furthermore, treatment plants reduced
pollution levels by an average of 0.21 standard deviations, suggesting
adjustment in polluting behaviour in response to stricter enforcement. This
suggests an important role for alternative market structures in enforcing
environmental regulation.
Activity 14.12
Could we identify which of the four changes in the regulation market is responsible for
the reduction in false reports? Design a hypothetical experiment to identify the effect
of an increase in the probability of back checks. Why might it be better to keep the four
treatment elements together rather than separately identifying each?
Another important issue in environmental regulation is substitution towards

other forms of pollution. Clean air regulation under good enforcement
could result in better air quality through installation of scrubbers and
electrostatic precipitators. These devices remove pollutants from the air,
but residuals are not eliminated and they are often dumped in water bodies
or injected into the ground. This could cause water and soil pollution and,
while the clean air policy seems to be working, a broader picture shows its
failures. In the context of the US Clean Air Act Amendments, Greenstone
(2003) assesses the relevance of this argument for iron and steel industries.
He, however, finds no support for cross contamination in this context as
measured in terms of lead, particulate matter and ground level ozone
pollutants released in to air and other media.
178
► Stop and read

Duflo et al. (2013).
14.3.4 Disaster relief

Effective policies to prevent or reduce damages from natural disasters
could reduce income volatility for individuals and help them in
consumption smoothing. There is, however, evidence that high income
countries can better afford effective relief policies for several reasons:
1. Rich countries have stronger and more durable materials and
structures leading to less destruction.
2. Most agriculture in rich countries is irrigated and not exposed to
droughts.
3. Early warning systems for hurricanes and weather forecasting are
more widely available.
4. Medical care and food are available in greater volumes after disasters.
5. Since disaster response is handled by governments, more efficient and
accountable governments do better.
Besley and Burgess (2002) present evidence from Indian states on the final
mechanism above. They show that states with higher levels of newspaper
circulation are more responsive to food shortages as a result of droughts
and floods. In states with higher newspaper circulation, elected state
governments spend more on public food distribution programmes (in
response to shortages).
Activity 14.13
Why would you expect states with higher newspaper circulation to be better at providing
calamity relief?
If developing countries lack the means to provide calamity relief reliably,

then it is likely that international humanitarian aid could help. Strömberg
(2007) analyses country level data and finds three stylised correlations in
the data:
1. Donors tend to provide greater relief for high profile emergencies that
receive good news coverage at the expense of more invisible ones.
2. Donors tend to favour historically important allies.
3. Disasters in countries with high importance in the foreign policy of
the donor or those with robust economic ties tend to receive higher
amounts of aid.
► Stop and read

Strömberg (2007).
14.4 Summary
The environment impacts on the livelihood of individuals in developing
countries to a great extent. This is both because of their higher
vulnerability and inability to use avoidance strategies or relief policies
effectively. Climate change is likely to cause unknown fluctuations in
temperature and rainfall, resulting in greater risk to agricultural income.
The articles we reviewed on the impact of weather fluctuations on
mortality and economic growth allude to potentially large and devastating
impacts.
179
While climate change is a global externality demanding global action

and coordination, there are other environmental issues that have serious
consequences for developing countries. We investigated environmental
regulation for pollution control. The evidence suggests that lack of proper
enforcement powers, potentially due to audit market structures, could be
responsible for inability of regulation to fix polluting behaviour.
Finally, we considered the differences in vulnerabilities in the face of
natural disasters. Developing countries have much greater death rates
for similar disasters and domestic relief policies seem to be inadequate,
potentially due to state inefficiencies and political motivations.
International aid could play a role here but it seems geopolitical interests
shape aid flows to some extent.

• outline how environmental issues affect developing countries
• explain the impacts of climate change, pollution and natural disasters
in developing countries
• critically evaluate mitigation strategies for controlling deforestation
and pollution.

1. Explain how residential sorting might invalidate causal inference from
a cross sectional comparison of health outcomes between polluted and
clean areas in a city or country.
2. Consider the payment for ecosystem services programme (PES) in
Uganda and assume a scenario where the results of the experiment
show that those who took up the scheme preserved their forests
during the period of the contract. What are the potential limitations
of generalising the results of this scheme to other contexts like
Brazil or Indonesia? How does the fact that forests are owned by the
government qualify your policy recommendations for Indonesia?

on the VLE.
180
Chapter 15: Taxation and development
15.1 Introduction
• discuss the structure of tax systems in developing countries
• explain one model of tax evasion and the role of information in tax
collection
• discuss incentives for investment in fiscal capacity.

• describe the differences in tax systems across developing and
developed countries
• critically evaluate the role of information in tax collection
• discuss factors shaping investment in fiscal capacity.

De Mel, S., D. McKenzie and C. Woodruff ‘The demand for, and donsequences
of, formalization among informal firms in Sri Lanka’, American Economic
Journal: Applied Economics 5(2) 2013, pp.122–50.
Gordon, R. and W. Li ‘Tax structures in developing countries: many puzzles
and a possible explanation’, Journal of Public Economics 93(7–8) 2009,
pp.855–66
Piketty, T. and N. Qian ‘Income inequality and progressive income taxation in
China and India 1986–2010’, American Economic Journal: Applied Economics
1(2) 2009, pp.53–63.

Besley, T. and T. Persson ‘Taxation and development’ in Auerbach, A.J.,
R. Chetty, M. Feldstein and E. Saez (eds) Handbook of public economics.
(London; Amsterdam: Elsevier, 2013) Volume 5, Chapter 2, pp.51–110.
Available at: http://goo.gl/pjTPb5
Kleven, H.J., M.B. Knudsen, C.T. Kreiner, S. Pedersen and E. Saez ‘Unwilling
or unable to cheat? Evidence from a tax audit experiment in Denmark’,
Econometrica 79(3) 2011, pp.651–92.
Olken, B.A. and M. Singhal ‘Informal taxation’, American Economic Journal:
Applied Economics 3(4) 2011, pp.1–28.
15.1.5 Works cited

Kleven, H.J., C.T. Kreiner and E. Saez ‘Why can modern governments tax so
much? An agency model of firms as fiscal intermediaries’, National Bureau
of Economic Research, Working Paper No. 15218, 2009.
15.1.6 Overview
We have seen developing countries face challenges like poor infrastructure,
malfunctioning health facilities and teacher absence. Provision of public
goods and correction of market failures are central tasks of the states that
require expenditure of resources. Governments could finance expenditures
from three sources:
181
1. taxation of economic activity

2. use of natural resources
3. reliance on international aid.
(In this chapter we look at tax revenue and in the next chapter we
consider aid.)
Apart from providing revenue, taxation is as an instrument for reducing
inequality and correcting externalities. A progressive tax system imposes
higher taxes on the rich resulting in a more equal after-tax (net) income
distribution. Higher taxes could help in redistribution to the poor in the
form of direct transfers, infrastructure programmes or other pro-poor
expenditures. By raising the marginal costs, Pigouvian taxation could
fill the wedge between private and social marginal returns and correct
for externalities. Many countries impose taxes on petrol, cigarettes and
alcohol based on this argument.
Therefore, the power to tax is a central determinant of state effectiveness
in performing its classical duties. Tax collection is, however, not a trivial
task and requires costly investments in state capacity to monitor economic
activity and appropriate resources from profit-seeking economic agents. A
large literature on public finance establishes that individuals change their
behaviour in response to taxes to maximise after-tax incomes and lower the
burden of taxes. This could involve real responses where individuals work
less and earn less in response to the imposition of income tax or corporations
invest less as a result of a hike in profit taxes. Tax evasion is another form
of behavioural response to taxes where real income is unchanged yet
individuals choose to report a lower level of income. The state may not be
able to collect much revenue when behavioural responses are substantial.
Tax administrators (agents responsible for the collection of taxes on behalf
of government) are the other side of the story. They deal with the crucial
task of inspecting economic activity and assessing whether the right
amount of tax is paid. In the absence of the right incentive structures, tax
administrators may not put in enough effort or accept bribes from evading
taxpayers. This is a classical principal agent problem with the government
as the principal and the administrators as agents. Despites its absolute
importance we, however, do not cover this side of the story.
In the next section we review the differences in the structure of tax systems
across developing and developed countries. Although statutory tax rates
do not seem radically different, effective tax rates vary substantially.
Developing countries collect much lower taxes than developed ones
and rely disproportionately on trade and corporation taxes. To better
understand these differences, in section 15.3 we consider a simple model
of tax evasion and discuss the role of third party reporting and withholding
in improving tax collection. We also look at the issue of informality and
its impact on tax collection, emphasising why businesses might optimally
choose to remain informal in developing countries. In section 15.4 we
extend the analysis and think about the state’s decision to invest in and
expand fiscal capacity. Finally, we review some evidence on the significance
of informal taxation, an often ignored form of local financing.
15.2 Tax structures in developing countries

Tax structures in developing countries differ dramatically from the
patterns observed in developed countries. The first marked difference
is that developing countries collect lower tax revenues compared to
developed countries. For example, Figure 6 in Besley and Persson (2013)
182
shows a positive correlation between tax revenue as a fraction of GDP and

the level of GDP per capita. Higher income countries collect more taxes as
a fraction of GDP.
The second important difference is the composition of tax revenue.
Developing countries tend to rely more on corporate income taxes,
consumption and production taxes, border taxes (e.g. tariffs) and
seignorage, whereas developed countries rely much more on income taxes.
Gordon and Li (2009) present evidence from a sample of countries that
confirms this pattern. In their sample an average rich country raises 25
per cent of GDP as tax as opposed to 14 per cent in poor countries. In rich
countries 54 per cent of tax revenue is from income taxes, while only 31
per cent comes from this source in developing countries. Similarly, the
average border tax contribution to revenue is 9 per cent and 1 per cent in
poor and rich countries respectively. Figure 8 in Besley and Persson (2013)
also shows a significant positive correlation between the share of income
tax in revenue and GDP per capita.
Despite large differences in the share of tax revenue and its composition,
the statutory tax rates are rather similar across developing and developed
countries. Figure 9 in Besley and Persson (2013) shows that countries with
similar top statutory income tax rates have strikingly different income tax
revenue as a share of GDP. Gordon and Li (2009) further show the top
statutory corporate and VAT rates are also rather similar across developing
and developed countries. Therefore it does not seem the difference in tax
collection (effective tax rates) stems from different statutory tax rates.
Activity 15.1
Explain how it is possible to see a divergence between statutory and effective tax rates.
Note that effective tax rates are calculated by dividing collected tax revenue by the tax
base.
Piketty and Qian (2009) shed some light on the puzzle of similar statutory
but markedly different effective tax rates by focusing on the cases of China
and India. They propose that exemption thresholds play a critical role in
restricting a government’s ability to raise taxes. Income tax was established
in China in 1980 but the exemption threshold (the level below which
individuals pay no tax) was set so high that most of the population was
exempt. The exemption threshold was, however, fairly fixed in nominal
terms, which resulted in an increase in the percentage of the population
paying taxes over time. The percentage of the population paying income
tax increased from 0.1 per cent in 1986 to 20 per cent in 2008. Therefore,
the income tax revenue as a share of GDP rose from 0.1 per cent in 1986
to 1.5 per cent in 2005.
Piketty and Qian (2009) observe an opposite pattern in India. Income tax
exemptions were annually adjusted in line with nominal income growth.
Therefore, a constant proportion of the population paid income taxes over
time. Over a similar period the fraction of the population paying income
tax in India was about 2–3 per cent and consequently income tax revenue
stayed at about 0.5 per cent of GDP.
Activity 15.2
Consider a simple income tax schedule with an exemption threshold set at z and a
marginal tax rate of τ applicable to income beyond that threshold. Find the average tax
rate for an individual earning income y. How does this average tax rate vary with the
level of income? Based on this formula, explain why high exemption thresholds could be
responsible for low effective tax rates in developing countries.
183
► Stop and read

Piketty and Qian (2009).
15.3 Why tax structures are so different?

Why are tax revenue and composition so different across developed and
developing countries? One potential answer might be that developing
countries have a different preference for income redistribution. Developing
countries might prefer to have lower levels of redistribution compared to
rich countries and therefore need lower revenues. Although we cannot rule
out preference heterogeneity, it is hard to use this story to justify differences
in tax composition (e.g. differences in use of personal income tax).
From theories of optimal taxation we know that an optimal tax structure
should maintain production efficiency and minimise the distortionary
effect of taxes. For example, taxing capital income discourages investment
and results in lower than optimal investment. As another example, we
have seen in Chapter 13 that tariffs reduce welfare by reducing gains from
trade. Tax structures in developing countries go against these principles.
Tariffs and corporation taxes constitute a significant share of tax revenue
in developing countries. Therefore from a theoretical perspective it is
important to understand whether developing countries are choosing
suboptimal tax structures.
Activity 15.3
Consider a simple supply and demand framework for a single commodity. The government
imposes a per unit tax on consumers. Does social welfare increase or decrease?
Another possible explanation that could reconcile differences in tax

structures with optimal tax theory relies on information and enforcement
constraints that are present in developing countries. Individuals could
respond to the imposition of taxes by evasion and informality. Formal
taxpayers (e.g. registered for government services) could lower tax liability
on the intensive margin by reporting less than actual income (evasion).
On the extensive margin taxpayers could opt out of registration and fall
through the government tax net. The inability of governments to detect
cheating could result in low tax revenues, large informal sectors and
widespread evasion. In what follows we first consider a simple model of
evasion and discuss differences in information structures of tax collection,
then we investigate the potential role of informality in explaining the
patterns of taxation.
15.3.1 Enforcement and tax evasion

Lack of enforcement powers by the tax authorities and high tax evasion
could be reasons why effective tax rates differ drastically between poor
and rich countries.
In this section we review a simple model of tax evasion to better
understand factors that influence taxpayers’ evasion decision. Consider a
single taxpayer with true pre-tax income equal to y*. There is a linear tax
system in place with a single tax rate equal to τ. The government does
not observe true income unless it audits the individual in which case true
income is revealed. Taxpayers receive a fine proportional to the evaded
tax on top of the true tax payment if found cheating. Due to resource
constraints, however, the government cannot audit everyone and randomly
audits p fraction of individuals. Assuming linear utility, the taxpayer
chooses the level of tax evasion to maximise expected income as follows:
184
max (1 – p) ((1 – τ)y* + τe) + p((1 – τ) y* – θτe)

e
where y* is true income, y is reported income, e = y* – y shows the amount
of evaded income, θ < 1 shows the fine rate, and τ is the statutory tax
rate. The first term in the objective function is income when the taxpayer
is not audited, multiplied by the probability of not being audited. The
second term is audit probability multiplied by income in case of audit.
Upon audit, the full amount of tax plus the penalty are paid (τy* + θτe)
and therefore subtracted from pre-tax income. To find the optimal level
of evasion we should look at the first order condition but this problem is
linear in e and therefore the taxpayer either chooses full evasion or no
evasion (technically some parameter values imply indifference between
any level of evasion). If the probability of evasion and the penalty rate
are sufficiently small, the chances of getting caught are low and there is a
positive marginal expected gain from evasion. Therefore taxpayers evade
to the full extent. On the other hand, if the probability of audit and the
penalty rate are large the marginal expected benefit of evasion is negative
and no evasion happens. In summary, evasion is determined as follows:
e* = { y*
0
if
if
p (1 + θ) < 1
p (1 + θ) > 1
This model does not look very consistent with observations from
developed countries. Audit rates are not particularly high and penalties are
modest but, contrary to the prediction of this model, evasion is very small
in developed countries.
Activity 15.4
In this model we have assumed an audit is equivalent to detecting the full amount of
evasion. How do the predictions of the model change if only µ fraction of audits results in
detection of evaded tax?
Third party reporting and withholding are important features of developed

country tax systems which are absent in the above model. Under third
party reporting employers, banks and investment funds report taxable
income on behalf of individuals to the tax authority. With withholding they
also withhold the tax and reimburse it to the government directly. These
features of the tax system imply actual audit rates are not a good proxy for
detection probabilities. In fact, tax authorities could combine third party
reports and individual tax returns to uncover discrepancies and make
targeted audits. Therefore, there is little scope for evasion on third party
reported income.
Kleven et al. (2011) designed a randomised experiment to estimate the
response of taxpayers to changes in the probability of audit in Denmark.
In the baseline they made an unannounced comprehensive audit of half of
their sample. They found substantial evasion on the self-reported income
but zero evasion on third party reported income. In the following year
they randomly sent out letters to two groups, stating an audit probability
of either 100 or 50 per cent. A third group was not audited. The findings
here substantiate the baseline evidence. Self-reported income was
responsive to the threat of audit and higher probability of audit resulted
in less evasion, but third party reported income was not responsive. This
study clearly shows the impact of the information environment on evasion
behaviour.
In developing countries the share of self-reported income is much higher
than in developed countries. If evasion on self-reported income is easier
185
we would expect lower effective tax rates in poor countries. Furthermore,

Kleven et al. (2009) show that the success of third party reporting relies
on firm size. In small firms, collusion between employers and employees
for tax evasion is feasible and keeping transactions off the books is less
costly. Bigger firms with higher turnover and more employees, however,
need to keep accurate accounts and collusion is harder to sustain. The
process of development and economic growth could lead to the emergence
of larger firms and a shift from self employment to salary employment in
big firms, leading to a better enforcement environment for tax collection.
Activity 15.5
Could you explain why the enforcement environment might be different for trade taxes
(tariffs) compared to personal income taxes? How does this justify higher border taxes in
developing countries?
15.3.2 Informality
One important feature of developing countries is the presence of a large
informal sector. Basically, firms not registered with the government are
harder to tax. If businesses use cash transactions and avoid formal banks
it is much harder for the government to observe business activity and levy
taxes. In Gordon and Li (2009) the average estimated size of the informal
economy for developing countries is 30 per cent of GDP as opposed to 14
per cent in developed countries.
Activity 15.6
Is it possible to estimate the causal effect of informality on tax collection by regressing
tax revenue as a share of GDP on the informal economy as a share of GDP? What are the
potential problems with causal interpretation of this estimate?
Gordon and Li (2009) offer one theoretical explanation for why so many
businesses decide to remain informal in developing countries. The key
benefit of formality in their model is access to formal credit. The cost of
formality is being taxed by the government. When the benefits of access
to credit outweigh the cost of tax payments, businesses choose formality.
Therefore a higher tax rate would imply a smaller fraction of businesses
choosing to be formal. There might also be heterogeneity in patterns of
formality across industries. Sectors that have higher benefits from access
to credit remain formal in the presence of higher taxes. Therefore the
government might optimally choose to tax these sectors heavily.
Based on this model, improvements in financial institutions increase the
benefits of getting formal credit and hence induce more businesses to
choose formality, leading to higher effective tax rates. If the process of
development is associated with enhancement of finance, then firms in
developed countries attach higher values to financial access and a lower
fraction remain informal.
Capital intensive businesses receive larger benefits from access to finance.
Therefore, the benefits of formality are larger for capital intensive firms,
meaning they would remain formal even with high tax rates. This justifies
heavier reliance on corporate taxes in developing countries.
Activity 15.7
Why might the benefits of access to credit be larger for larger firms? What does this imply
about patterns of formality and firm size?
186
► Stop and read

Gordon and Li (2009).
Our empirical knowledge of why firms remain informal is very limited.

Here we discuss a randomised experiment in Sri Lanka that investigates
this question.
De Mel et al. (2013) try to test between two views on why firms remain
informal. The exit view is what we discussed above: firms remain informal
because the cost of formality is higher than the benefits. The other view is
the exclusion view and argues that burdensome entry regulations prevent
businesses from reaping the benefits of formality. In the experiment they
randomly allocate small businesses (1–14 employees) to four treatment
groups and a control group to test between these hypotheses. The first
treatment provided information on costs and benefits of registration and
reimbursed the direct registration costs (around 1,000 Rupees). On top of
the information treatment the second, third and fourth treatment groups
were offered 10,000, 20,000 and 40,000 Rupees to become registered.
At the baseline, a majority of businesses had no paid workers and only 23
per cent of these were registered at local level but as Figure 1 in De Mel
et al. (2013) shows, registration rates are increasing in firm size. Very
few informal businesses reported paying fines or bribes because of not
being registered and only a small fraction received visits from government
officials. Furthermore, the baseline survey showed that very few informal
businesses had good knowledge of the registration procedures and costs.
Business owners also revealed willingness to register if someone else paid
the cost (61 per cent did so!).
Activity 15.8
How do De Mel et al. (2013) make sure their survey is representative of both formal
and informal businesses? Would it be an issue if they relied on existing surveys from
government sources?
The results show no extra registration for firms provided with information
and reimbursement of costs relative to the control group. Offering a
bonus of 10,000 or 20,000 Rupees resulted in a 17–22 per cent increase
in registration and raising the bonus to 40,000 increased registration to
48 per cent (20,000 Rupees is about a month of profits for the median
firm). Many of the unregistered businesses in the 40,000 Rupees treatment
mentioned that they could not get formal permission from landowners to
prove they are legally allowed to run a business on premises.
Activity 15.9
The fact that the first treatment group (information and reimbursement) did not impact
on registration provides evidence against the exclusion view. Discuss.
What are the impacts of formalisation on business profitability and

survival? De Mel et al. (2013) followed up businesses over a period of
15–31 months after treatment to see whether formality improves business
outcomes. The increase in business profits is only marginally significant
for registered firms and survival rates are not different during follow ups.
This is despite the fact that a significant number of registered businesses
reported having receipt books and advertising in the last six months.
Interestingly there was no significant difference in terms of taxes paid by
registered and unregistered businesses.
187
► Stop and read

De Mel et al (2013).
15.4 Fiscal capacity

In previous sections we have seen that information constraints and a large
informal sector could potentially explain why developing countries collect
lower taxes than developed countries. The question you might ask at this
stage is why governments in poor countries did not invest in reducing
these constraints. A new line of enquiry pioneered by Besley and Persson
provide some answers to these questions (e.g. see Besley and Persson,
2013). They argue that governments make purposeful investments in
state capacity to tax (fiscal capacity) and to secure property rights (legal
capacity). In this framework policy makers’ incentives to improve the tax
system is at the core of the analysis.
Several economic and political factors could affect government
investment decisions in fiscal capacity. First, investments in the tax
system are complementary to investments in legal capacity in the sense
that once the legal environment is improved, more business activity
takes place and therefore there are greater returns to fiscal capacity
investments (higher additional tax revenue). This creates a link between
economic development and improvements in the tax system. Higher
levels of economic development imply higher returns on fiscal capacity
improvements and increase the likelihood that governments will decide
to incur the fixed cost of implementing third party reporting, withholding
and other measures to improve tax collection.
Activity 15.10
Informal firms could not benefit from the legal system to resolve contractual disputes
(e.g. in labour contracts). Explain how this could generate a second complementary link
between legal and fiscal capacity.
Second, the existence of a common interest or an urgent need for public

spending is inducive to investments in fiscal capacity. Faced with the threat
of war, citizens support government intititives to raise taxes to prevent
foreign invasion. Many developed countries have expanded tax bases and
invested in fiscal capacity during the First and Second World Wars (see
Figure 5 in Besley and Persson, 2013).
Third, the political constraints on redistributive politics and stability of
governments (political turnover) have great implications for the fiscal
regime. Besley and Persson (2013) identify three types of governments.
In the case of a common interest state, where public goods have much
value relative to redistribution, politics does not matter and incumbent
governments and opposition both agree on the expansion of fiscal capacity
(e.g. case of war above). When redistributive incentives are high and
political turnover is low, the state invests in fiscal capacity, but most of the
tax revenue is spent on transfers to the incumbent group. Finally, there
might be a case of weak states where redistributive motives are strong but
political turnover is high. In these states, building fiscal capacity backfires.
When the opposition takes office it can use the fiscal capacity to extract tax
revenue and transfer resources to its own group. Therefore, the incentives
to invest in fiscal capacity are low.
188
Activity 15.11
Explain how ethnically diverse countries might end up in the position of weak or
redistributive states.
Fourth, the presence of non-tax revenue such as aid and natural resources
impacts on incentives. Receiving non-tax revenue enhances the ability
of governments to finance fixed costs and invest in fiscal capacity (and
other social infrastructure) when the common interest motive is strong.
But when redistributive motives dominate, additional resources will be
channelled to transfers and not investments in fiscal capacity.
Activity 15.12
How could the expectation of receiving aid or natural resources in the future impact on
incentives to invest in fiscal capacity?
► Stop and read

Besley and Persson (2013) pp.51–63 (sections 1 to 3).
15.5 Informal taxation

One often ignored feature of developing country tax systems is the
presence of informal taxation. To finance local public goods, communities
either need to raise formal taxes, get grants from central government
or rely on informal arrangement between themselves. As we have seen
before, formal taxation is limited in developing countries and if there
are no other informal arrangements, the provision of local public goods
will be particularly deficient. Olken and Singhal (2011), however,
present evidence from 10 countries showing the importance of informal
arrangements in communities for the provision of local public goods.
Individuals from rural communities often contribute to the provision of
public goods like the construction of roads or maintenance of facilities.
Contributions are in the form of both monetary payments and free labour.
Arrangements between individuals determine the level of contributions
from each household and penalties for free riders. The system of informal
taxation might be coordinated by local officials, but enforcement is
through social sanctions rather than formal legal action.
The evidence from Olken and Singhal (2011) shows that a significant
share of households in the 10 countries studied pay informal taxes. In
several countries 50 per cent of household made some contributions.
These informal payments constitute 0.85 per cent of household
expenditure in the median country but 15.7 per cent of total taxes paid.
These numbers are not particularly high, but they represent a much larger
share of local spending. Informal tax payments are regressive in the sense
that poor households face a larger average rate than rich families and
ignoring them in distributional analysis of formal tax system understates
the regressiveness of the tax system. Finally, they find that labour
payments are more commonly used among poorer households.
Activity 15.13
Do you think informal taxation could be a response to a lack of information on individual
incomes by formal tax collectors? What other explanations could result in the presence of
informal taxation?
189
15.6 Summary
The ability to tax is a key defining feature of government effectiveness in
correcting market failures and pursuing equity objectives. We have seen
developing countries not only collect lower shares of GDP as taxes but also
use tax instruments that differ from developed countries.
Ignoring information and enforcement issues, it seems that tax structures
in developing countries are not in line with predictions of optimal tax
literature. Further analysis, however, shows the importance of third party
reporting and the role of medium and large businesses in facilitating
detection of fraudulent activities. Third party reporting is well instituted in
developed economies and empirical evidence shows there is little evasion
on income that is subject to third party reporting. But in developing
countries small businesses are responsible for a larger share of the
economy and enforcement of an independent reporting structure is much
harder. Evidence from Denmark confirmed the intuition that self-reported
income is subject to substantial evasion.
The natural question is why some governments were able to implement
information and enforcement environments conducive to greater tax
capacity while others failed to do so. This view treats improvements in
tax collection as purposive investments by governments to enhance fiscal
capacity. With agreement on the spending of public funds (common
interest) statesmen have greater incentives to invest in fiscal capacity
expansions, while in times of political instability and strong redistributive
motives, incumbents fear irreversible investments in tax collection could
result in more tax capacity for the opposition group. These arguments put
the interplay of political and economic factors at centre stage in shaping
investment in fiscal capacity.

• describe the differences in tax systems across developing and
developed countries
• critically evaluate the role of information in tax collection
• discuss factors shaping investments in fiscal capacity.

1. Based on the theoretical model of Gordon and Li (2009) explain how a
large informal sector could lead to a poverty trap.
(Hint: Where does the government get the revenue to enhance its
services?)
2. What are the potential policy implications of De Mel et al.’s (2013)
study? Specifically, what are the barriers to formality? What are the
policy options to induce higher levels of formality?

on the VLE.
190
Chapter 16: International aid
16.1 Introduction
• explain different types of development assistance
• discuss the impact of aid on growth
• outline some of the unintended consequences of aid.

• compare and contrast different types of development assistance
• critically evaluate the impact of aid on growth
• outline and discuss the unintended consequences of aid.

Easterly, W. ‘Can the West save Africa?’, Journal of Economic Literature 47(2)
2009, pp.373–447.
Burnside, C. and D. Dollar ‘Aid, policies, and growth’, American Economic
Review 90(4) 2000, pp.847–68.

Burnside, C. and D. Dollar ‘Aid, policies, and growth: reply’, American Economic
Review 94(3) 2004, pp.781–84.
Easterly, W., R. Levine, and D. Roodman ‘Aid, policies, and growth: comment’,
Frazer, G. ‘Used-clothing donations and apparel production in Africa’, Economic
Journal 118(532) 2008, pp.1764–784.
Kuziemko, I. and E. Werker ‘How much is a seat on the Security Council worth?
Foreign aid and bribery at the United Nations’, Journal of Political Economy
114(5) 2006, pp.905–30.
Nunn, N. and N. Qian ‘US food aid and civil conflict’, American Economic Review
(forthcoming).
16.1.5 Works cited

Berger, D., W. Easterly, N. Nunn and S. Satyanath ‘Commercial imperialism?
Political influence and trade during the Cold War’, American Economic
Review 103(2) 2013, pp.863–96.
Besley, T. and T. Persson Pillars of prosperity: the political economics of
development clusters. (Princeton: Princeton University Press, 2011).
Collier, P. The bottom billion. (Oxford: Oxford University Press, 2007).
Easterly, W. ‘Can foreign aid buy growth?’, Journal of Economic Perspectives
17(3) 2003, pp.23–48.
Easterly, W. The white man’s burden: why the West’s efforts to aid the rest have
done so much ill and so little good. (New York: Penguin Books, 2006).
Sachs, J. The end of poverty: economic possibilities for our time. (New York:
Penguin Books, 2005).
191
16.1.6 Overview
International aid is a much-discussed source of revenue for developing
countries’ governments. Official Development Assistance (ODA) is rising
in real terms and reached a peak of $123 billion (US) in 2009. The largest
recipient region is Sub-Saharan Africa with a 33 per cent share of total
ODA, with the Middle East and North Africa, and Central and South Asia
lagging behind with 21 and 15 per cent shares respectively (Besley and
Persson, 2011, Chapter 6).
International organisations like IMF, World Bank, other affiliates of United
Nations, governments of rich countries, non-governmental organisations
(NGOs) and private donors are the various types of donors in the field
of development assistance. Aid is controversial, because it is difficult to
rigorously assess its impact on recipient countries. There are three stylised
views on aid. Aid proponents argue that aid helps countries escape poverty
if enough resources are spent on building infrastructure, public institutions
and enhancing physical and human capital. In this view, insufficient aid
is to blame for the disappointing performance of most recipient countries
(e.g. see Sachs, 2005).
In contrast to this optimistic view, aid pessimists claim that most aid is
wasted and does not contribute to improving the situation of the poor
in developing countries. Some would argue that aid actually harms the
process of development by increasing the likelihood of civil conflict and
crowding out local production. Cutting off all aid is, however, an extreme
conclusion to this argument. Essentially the question of what would have
happened in the absence of aid is hard to answer. Proponents would argue
that in the absence of aid, African growth performance would have been
worse. The pessimists argue this is an implausible counterfactual since it is
hard to imagine that African growth would have been very negative in the
absence of aid (e.g. see Easterly, 2006).
A middle stance focuses on conditionality and identifying the type of
aid that might work in different circumstances (see Collier, 2007). The
current wave of donor-funded randomised experiments is in line with
this view. Having failed to find robust evidence of the impact of aid on
growth performance, some donors and academics started to focus on
implementing and evaluating small-scale projects with an eye towards
shedding light on mechanisms that could lift individuals out of poverty.
Many of the randomised experiments we saw in the previous chapters are
run through NGOs funded by local or global philanthropists. Increasingly,
multinational organisations like the World Bank and donor countries are
endorsing this approach of one-experiment-at-a-time as a fruitful method
of improving the lives of the poor.
In this chapter we start by a discussion of different types of development
assistance. We then review some of the evidence on the link between
growth and aid. In the end, this literature is inconclusive due to
identification problems and a lack of robust correlations between aid
and growth. In the last section we review some articles on the negative
unintended impacts of aid.
16.2 Forms of development assistance

Aid to poor countries can take several forms. In its most basic – and
probably most usual form – donors provide money to the government or
certain groups in a poor country without specifying any condition, leaving
full discretion to the recipient. This form of aid could be used for the
192
private consumption of individuals in charge of expending aid money or it

could be put into activities not originally intended by donors. For example,
it might be diverted to the purchase of weapons rather than medical
supplies. On the other hand, recipients might know their needs better than
donors and allocate the funds more efficiently.
In most cases, however, there are conditions attached to imbursement of
aid. Conditional aid itself can have many variants. First, aid can take the
form of budgetary support to the recipient government or organisation
(e.g. microfinance banks), conditional on the implementation of certain
policies or improvements in specified outcomes. The contract design
literature suggests that conditions should be based on verifiable and
enforceable decisions made by recipient countries. But it is not clear how
donors can hold recipient countries accountable to the use of aid, apart
from the threat of stopping future aid. For example, during the period
of the Washington Consensus on institutional and policy reforms, the
World Bank and International Monetary Fund made extensive use of such
conditions in promoting market-oriented reforms which led to some policy
changes but failed to produce significant changes in aggregate incomes
(Easterly, 2009).
Activity 16.1
Money is fungible in the sense that once revenue from different sources is combined
(e.g. tax and aid) then it is hard to distinguish where revenue from each source is spent.
Explain how this might pose a challenge to upholding donor conditions.
Second, donors and NGOs could organise the building of infrastructure

projects or provision of other public goods (e.g. health facilities, disaster
relief, etc.) in locations of interest. This is development support and
requires organisational presence in the poor country and might involve
large overhead costs, but ensures the project of interest is implemented
and provides closer interaction between donors and actual recipients.
Activity 16.2
How might a divergence between donors’ priorities and recipients’ needs render
conditional aid ineffective?
Third, aid might be given in the form of technical assistance to recipient

governments to help improve their performance or help them identify
better projects. For example, tax officers from developing countries
could help developing governments enhance their tax systems. This
form of assistance invests resources in transmitting knowledge of
how to govern the bureaucracy and how to run public institutions to
developing countries. If guidance is modified to fit the given context of
specific countries, this type of assistance could reduce costs of policy and
institutional reform.
Fourth, donor governments might give military support to the recipient
government in the form of outright intervention, training or supply of
weapons. This form of assistance is usually justified on the basis of helping
a fragile state to overcome internal or external militant groups but could
be linked to strategic interactions as we discuss below.
The final form of aid is post-conflict assistance in war-torn countries
to help sustain peace, reconstruct the infrastructure and, sometimes,
design rules for new states. This might be the most controversial type of
intervention and it is not clear whether the advertised outcomes (peace
and prosperity) will follow. In a recent study, Berger et al. (2013) use
193
declassified Cold War CIA documents to show in countries where the USA
government had a role in installing or supporting political leaders, imports
from USA surged after the intervention but exports did not change.
Kuziemko and Werker (2006) further provide evidence that in fact aid
flows greatly impact on political motives. They use the rotation of 10 out
of 15 positions on the United Nations Security Council between countries
to see if becoming a temporary member of the Council has an effect on aid
flows. On average, becoming a member of the Security Council increases
US aid to that country by 59 per cent and UN aid by 8 per cent. This effect
is larger in times of important decision-making. Additionally, the timing of
the increase in aid flows closely tracks entry and exit of the country in and
out of the Security Council.
16.3 Does aid promote growth?

Until the randomised evaluation revolution, most of the aid literature was
focused on estimating the impact of aid on aggregate outcomes, especially
growth. As you will see below, theoretically it is not clear whether aid
should have a long-lasting effect on growth. Empirical investigation of the
impact of aid is also hindered by endogenous allocation of aid.
16.3.1 Theoretical impact of aid

How might aid impact on economic growth? To answer this question you
need to adopt a framework for thinking about the determinants of growth.
You have seen several models in the macro part of the course that could
have different implications for the effect of aid on growth. Growth effect of
aid in the Solow model is likely to be transitory unless it changes the deep
parameters like savings rates or productivity. The next activity asks you to
analyse these effects.
Activity 16.3
Consider an economy that functions under Solow assumptions away from its steady state.
What is the impact of providing cash aid to this country if all the money is invested in the
capital stock? What will be the effect of aid if the country started in a steady state? Now
consider aid in the form of technical assistance or investment in infrastructure that raises
productivity parameter in the Solow model. How does aid affect growth in and out of the
steady state here?
Poverty trap models provide a better justification for aid. Based on these
models, countries might be stuck in a poor equilibrium, and providing
enough aid could push them on track to move to a favourable one. This
argument provides an almost non-falsifiable justification for proponents
of aid. In the absence of positive effects, practitioners could argue that the
amount of aid was not enough and therefore did not allow the recipient to
escape the poverty trap. Testing for presence of poverty traps is a daunting
task, as you have seen in Chapter 4. Therefore it is not clear whether this
view can be supported, based on empirical evidence.
There are other potential mechanisms triggered by aid that go beyond the
growth models. For example, the resource curse literature shows that there
might be a negative growth effect from windfall revenue from extraction
of natural resources, mainly due to political economy factors such as rent
seeking, corruption and power struggles. Aid is not very different from
resource windfalls and provides a source of rents that might trigger such
mechanisms, as in resource curse models. We look at two possibilities
under this heading in section 16.4 below.
194
16.3.2 Empirical studies

There is a large literature trying to link aid to macro development
outcomes like economic growth. The usual approach is to use a regression
of growth (or other outcomes) on aid and other covariates as follows
git = α + βait + Z΄γ + єit (1)
where git shows growth in country i at time t, a captures aid, and Z is a set
of control variables that affect growth and might be correlated with aid.
The problem with putting a causal interpretation on the estimate of β from
(1) is that aid is not given at random. In principle donors choose countries
with dismal economic performance precisely because they care about the
lives of individuals living in those countries and want to prevent worse
outcomes. We have also seen above that strategic motives might be behind
donor governments’ decisions. Therefore the coefficient estimate of β from
(1) will not capture the true effect of aid on growth.
Activity 16.4
Why does the non-random allocation of aid lead to a biased estimate of β in (1)?
Suppose donors pick underperforming economies for their donations. Provide a rationale
for why this type of selection might lead to an underestimation of the positive effects of
aid.
Most of the studies in the literature are unable to correct for the
endogeneity of aid and therefore are not able to identify the true impact
of aid on growth. Burnside and Dollar (2000) was an extremely influential
article that tried to use an instrumental variable approach to identify
the causal effect of aid on economic growth. They used a sample of 56
countries over the period of 1970–93 to estimate β above. The variable
that captures aid in their study is the ratio of aid over GDP and they
tried to include several additional variables that might be important in
explaining growth performance. Specifically they include a policy index,
which shows whether the recipient has adopted good policies (index
constructed from trade openness, inflation and fiscal surplus, which is a
limited measure of good policies) and also an interaction of aid and policy
index.
OLS and 2SLS results (in the absence of policy aid interaction term) show
that the aid-GDP ratio has no effect on growth. Once Burnside and Dollar
(2000) controlled for the interaction term, it seems aid is effective in
enhancing growth performance in countries that adopted good policies.
The 2SLS specification, however, delivers insignificant coefficients even for
the interaction terms (see Table 4 in the article). The authors concluded
that aid is good for growth when policies are favourable. This seems to
provide a rationale for conditional aid but how credible are the results?
A later article by Easterly et al. (2004) questioned the robustness of these
results and showed that once they included additional observations, even
a similar methodology to that used by Burnside and Dollar (2000) yields
no significant results (see Burnside and Dollar, 2004, for a reply to this
critique).
Let us think about the validity of the instrumental variable approach used
here. The first assumption required for a valid instrument is relevance.
The instruments should have a (strong) correlation with the endogenous
variable. Here Burnside and Dollar use population, arms imports, Egypt
dummy, Franc zone dummy, and Central America dummy as instruments
for the endogenous aid-GDP ratio. It seems that relevance is satisfied.
195
Evidence suggests smaller countries get more aid as a ratio of GDP

(probably because donors are not willing to increase aid as much to
compensate for an increase in population), countries importing arms from
the USA and other rich countries are more likely to receive assistance,
Egypt has a special position in USA aid policy, Franc zone countries are
favoured by France, and Central America is also important to the USA.
The first stage results (the middle panel in Table 4 of Burnside and Dollar)
show these variables are indeed a significant determinant of aid–GDP
ratio.
The second and more difficult to defend assumption for the validity
of instruments is exclusion restriction. Instruments should not affect
growth other than through determining aid. In other words, all the five
instruments used here should not affect growth directly or indirectly
except for their effect on aid. Do you think this is a reasonable
assumption?
Burnside and Dollar (2000) rely solely on a statistical test known as the
‘over-identification’ test. This test uses predicted values of the error term
and sees whether they are correlated with the instruments. Burnside and
Dollar cannot reject the null hypothesis of no relation between instruments
and the error in their regressions. But the econometrics literature has
shown the over-identification test has low power and may not be able to
pick up endogenous instruments. Therefore in order to convince academics
that instruments are exogenous, you need to reject potential theoretical
mechanisms that provide a direct or indirect link between instruments
and aid. The following activity asks you to outline a story that results in
violation of exclusion restriction here.
Activity 16.5
One of the instruments used in Burnside and Dollar (2000) is population. Do you think
that the only impact of population on economic growth is its impact on aid? Try to outline
a plausible story and link population to growth directly or through a third variable which
is not included in the original regression. Arms importation is the second instrument used
here. How might arms imports relate to growth?
► Stop and read

Burnside and Dollar (2000). Easterly (2003) also provides a critical overview of the aid
growth literature.
16.4 Unintended consequences

Aid could trigger mechanisms that are not conducive to development. In
this section we review two studies that identify situations where aid is the
most likely cause of negative unintended consequences.
16.4.1 Food aid

Food aid is usually motivated by humanitarian objectives and many would
think it is both morally and economically right to help out malnourished
individuals in regions hit by droughts and famines. Yet food aid could
have negative impacts on development by eroding incentives for local
production and, potentially, through conflict. Free provision of food items
could hurt local agriculture at times when farmers need the money more
than ever due to shortages. In addition to this, in areas hit by civil conflict
some food aid might be diverted to armed groups. This is because often
aid convoys have to travel long distances in areas weakly governed by the
recipient state.
196
Activity 16.6
Why might allocation of food aid be endogenous to incidence of conflict? Clearly outline
a coherent story.
Nunn and Qian (forthcoming) study the impact of US food aid on conflict.
They used an instrumental variable approach to overcome the endogeneity
problem in a regression of conflict on food aid. They used US weather
conditions as a source of exogenous variation to US wheat production,
together with the average likelihood of a country receiving food aid from
the USA to construct a plausibly exogenous instrument for US food aid.
The US government fixes the price of wheat and buys any surplus from
farmers. In times of good harvest the government accumulates wheat
and in the following year sends more food aid to recipient countries.
(The authors have a detailed discussion of the exclusion restriction for
this instrument. You should at least read their introduction if you are
interested.)
The results from Nunn and Qian (forthcoming) show that an increase in
US food aid increases the duration of existing internal conflicts. Although
this result confirms the presence of negative consequences for aid,
you should bear in mind several interpretation issues. The instrument
basically focuses on the variation in US food aid that comes from surplus
production in the last year. The impact of surplus aid might be different
from other forms of humanitarian aid. We might worry about the outcome
of the conflict in the absence of humanitarian aid. This is essentially a
question about counterfactual scenario. Does the instrumentation allow us
to conclude that ‘stopping US food aid to countries with a history of civil
conflict reduces the likelihood of conflict’? Since the identifying variation
is coming from changes in surplus wheat stocks, we probably will not be
able to extend the argument to food aid from other sources and countries.
16.4.2 Used clothing donations

Frazer (2008) tested for the impact of used clothing donations on the
African apparel industry. He used an instrumental variable approach to
overcome the endogeneity of clothing donation supplies with respect to
the status of the apparel industry in the recipient country. Specifically,
he used the geographical proximity of the recipient to OECD donors
and the supply of charity used clothing in OECD countries to test for
the availability of second hand clothes in the recipient market. The
exclusion of instrument is partly founded on the idea that individuals
in rich countries think they are donating to local needy individuals but
in fact most clothing donations end up in poor countries. The results of
instrumental variable estimation shows that a one per cent increase in
used clothing imports results in 0.6 per cent decrease in the size of the
apparel industry in the recipient country. This is a large effect and could
account for 40 per cent of Africa’s decline in production of apparel.
Activity 16.7
Explain why supplying free used clothing to African countries might be detrimental to the
textile industry in those countries.
16.5 Summary
Development assistance could take several forms and might be
dispensed by many types of donor organisations. It seems that the macro
literature on the impact of aid on aggregate outcomes such as growth is
197
inconclusive, due to identification issues. Some recent studies, however,

cast doubt on the positive impacts of even the most innocuous forms of
aid, such as food aid and distribution of used clothing.
This seems a very disappointing conclusion: we do not know if aid
positively impacts on growth and in some cases with better identification,
aid has even negative effects on conflict and local production. It is,
however, intuitive to think that aid might work under some circumstances
and not others. Our goal as economists might be to identify types of
aid that are most effective under certain conditions. This has led to an
expansion of randomised evaluations supported by donors in developing
countries to find out what forms of interventions are most beneficial. The
aim is to focus on micro changes with identifiable outcomes. Giving aid to
a government and conditioning it on a set of policy reforms at best would
result in simultaneous changes that make it hard to learn what has worked
and what has not. On the other hand, the randomised experiments we
have seen try to test ideas one at a time and, instead of being concerned
about aggregate outcome, they focus on closely observed individual
outcomes.
Stop and read

Easterly (2009).
Now listen to the following LSE Public Lecture by William Easterly on ‘We do not know
how to solve global poverty and that’s a good thing’: http://goo.gl/Rggvew

• compare and contrast different types of development assistance
• critically evaluate the impact of aid on growth
• outline and discuss the unintended consequences of aid.

1. Explain conditional and unconditional aid. What are the potential
drawbacks and benefits of conditional aid over unconditional aid?
2. Assuming the exclusion restriction is satisfied in the regression of
growth on aid in Burnside and Dollar (2000), what are other potential
problems with the identification strategy? Be specific and try to
outline a clear story as to why there might be other problems with the
identification.
(Hint: You might want to use the fact that policy variables are
endogenous and yet not dealt with in the instrumentation process.)

on the VLE.
198
Appendix 1: Sample examination paper
Important note
This Sample examination paper reflects the examination and
assessment arrangements for this course in the academic year 2013–
2014. The format and structure of the examination may have changed
since the publication of this subject guide. You can find the most recent
examination papers on the VLE where all changes to the format of the
examination are posted.
Candidates should answer FOUR of the following TWELVE questions. All

Time allowed: 3 hours
Candidates should answer three questions: Question 1 from Section A, and
two questions from Section B. Question 1 carries 40% of the total mark.
Each of other questions carries 30% of the total mark.
Calculators are not allowed in this examination.
Section A
Answer all parts of Question 1.
Question 1
State whether the following statements are true or false and explain your
answers briefly. You should answer all eight parts. Each part carries 5
marks. Concise and clear answers will be rewarded. Answers with no
explanation receive no credit.
a. Empirical evidence suggests conditional aid increases economic growth
if recipient countries adopt good policies.
b. The evidence from West Bengal, India, suggests that land
redistribution involves large equity–efficiency trade-offs.
c. Rural farmers can perfectly insure against climate change by pooling
their incomes together in each village. Therefore there is no need for
government intervention.
d. PPP converted GDP figures give a more reliable estimate of the income
gap between rich and poor countries compared to exchange rate
converted GDP figures.
e. Statutory tax structures are very similar across developing and
developed countries but tax revenue as a share of GDP varies
significantly.
f. Empirical evidence supports Solow’s prediction of absolute
convergence.
g. Increasing returns to scale on its own can generate a poverty trap
and provide a potential explanation for large income gaps between
developed and developing countries.
h. The fact that the average income of households borrowing from a
microfinance institution is lower than that of non-borrowers shows
that microfinance loans on average reduce income.
199
Section B
Answer two questions from this section. Sub-parts of each question carry
equal marks.
Question 2
The following questions relate to Easterly, W., ‘Inequality does cause under
development: insights from a new instrument’, Journal of Development
Economics 84(2) 2007, pp.755–76.
a. Outline three theoretical arguments for why inequality might be
harmful for economic growth.
b. Explain why an OLS regression of growth on inequality does not
identify the causal effect of inequality on growth.
c. Why might suitability of land for sugar relative to its suitability for
wheat be related to current time inequality? How does this relate to
the identification strategy used in this article?
d. How much does inequality affect growth? Interpret the coefficient
estimate of inequality from the second column in Table A2 (below).
e. Compare the results in the first two columns of Table A2 (below). Does
the author find evidence that endogeneity biases the OLS coefficient
upwards? What can explain the difference between the
OLS and IV estimates?
f. State the identification assumptions of the instrumental variable
approach here. Do you think the assumptions are plausible in this
context?
Ordinary least squares Instrumental variables

(1) (2)
GINI coefficient –0.040 –0.121
(4.27) (4.45)
Observations 107 97
F-statistics from first stage NA 21.2
Table A2. Adapted from W. Easterly ‘Inequality does cause under

development: Insights from a new instrument’, Journal of Development
Economics 84(2) 2007, pp.755–76.
Notes: The table shows the coefficient estimate of regressions of log per
capita income in 2002 on average GINI coefficient during 1960–98 for the
sample of countries in Easterly (2007). Column (1) uses ordinary least
squares while column (2) uses the instrument studied in Easterly (2007).
t-statistics are reported in parenthesis below the coefficients.
Question 3
The following questions refer to E. Duflo ‘Schooling and labor market
consequences of school construction in Indonesia: evidence from an
unusual policy experiment’, American Economic Review 91(4) 2001,
pp.795–813.
a. Describe the INPRES programme studied by Duflo (2001). What were
the programme’s objectives?
b. Discuss how the programme might have affected an individual
household’s education decisions.
200
c. With reference to columns (1) to (3) of Panel A in Table A3 (below),

explain the results regarding educational attainment.
d. What is the identification assumption for interpreting the estimated
effect from Panel A as causal?
e. How does the information in Panel B of Table A3 (below) reassure you
that Duflo is estimating the causal effect of the programme?
Level of school construction in region of birth

High Low Difference
(1) (2) (3)
Panel A
Age 2–6 in 1974 8.49 9.76 –1.27
(0.043) (0.037) (0.057)
Age 12–17 in 1974 8.02 9.40 –1.39
(0.053) (0.042) (0.067)
Difference 0.47 0.36 0.12
(0.070) (0.038) (0.089)
Panel B
Age 12–17 in 1974 8.02 9.40 –1.39
(0.053) (0.042) (0.067)
Age 18–24 in 1974 7.70 9.12 –1.42
(0.059) (0.044) (0.072)
Difference 0.32 0.28 0.034
(0.080) (0.061) (0.098)
Table A3. Average years of schooling for various groups. Adapted from
E. Duflo ‘Schooling and labor market consequences of school construction
in Indonesia: evidence from an unusual policy experiment’, American
Notes: Numbers show average years of schooling for each group.
Standard errors are reported in parenthesis below the averages. The
rows and columns labelled ‘Difference’ show the differences in years of
schooling. Individuals aged 2–6 in 1974 started primary school when
INPRES was done, while older cohorts (aged 12–17 or 18–24 in 1974) are
past primary school age at this point.
Question 4
a. Assume a landlord is deciding to get someone to work on his land.
What are the three types of tenancy contracts he could offer? How is
the risk of agricultural activity spread in each of these contracts?
b. Now assume the farm output could be either high, H, or zero. The
probability of getting a high output depends on the effort expended by
the farmer. Specifically, e∈[0,1] shows both the level of effort and the
probability of getting the high output. The landlord does not observe
effort and therefore could condition payments only on observed
output. How would the three types of contracts you identified above
look like in this world? Write a parametric representation of the three
contracts.
c. Assume cost of effort is ½ ce2, where c is a positive constant. Further
assume that both landlord and tenant are risk neutral. Find the first
best (efficient) level of effort.
201
d. Now consider the three contracts you wrote a parametric

representation for in (b) and find the farmer’s effort level under each
of the contracts. Which contract delivers the same level of effort as the
first best?
e. Now assume the landlord offers the contract that delivers the first best
level of effort. Find the terms (payments) of the contract that maximise
the landlord’s payoff. Assume the outside option of the tenant is 0.
Question 5
The following questions are partly based on R. Burgess, M. Hansen, B.
Olken, P. Potapov and S. Sieber ‘The political economy of deforestation in
the tropics’, Quarterly Journal of Economics 127(4) 2012, pp.1707–54.
a. Why should tropical deforestation be an important topic in climate
change debates?
b. Think about a situation where the national government sets logging
quotas that are distributed and enforced by district officials. Further
assume that each province works as a unified wood market where
loggers (legal or illegal) could choose to operate in any district within
the same province. Now intuitively explain why an increase in the
number of districts in a province might lead to more deforestation.
c. With reference to Table A5 (below) interpret the impact of number of
districts on deforestation.
d. Does Table A5 (below) provide evidence of illegal logging?
e. With reference to Table A5 (below) explain how the effect of number
of districts on deforestation changes over time.
All forest (1) Production / Conservation /

Conversion Protection
(2) (3)
Panel A: number of districts 0.038 0.044 0.047
in province (0.016) (0.018) (0.033)
Panel B: number of districts 0.082 0.081 0.101
in province (some of L0–L3) (0.020) (0.019) (0.043)
Observations 608 296 312
Table A5: Impact of the number of districts in the province on

deforestation as measured with satellite data. Table adapted from
R. Burgess, M. Hansen, B. Olken, P. Potapov and S. Sieber ‘The political
127(4) 2012, pp.1707–54.
Notes: Panel A and B show coefficient estimates of regressions of
deforestation on a number of districts. In Panel B three lags of the number
of districts are included and the reported coefficient is the sum of the
contemporaneous effect of number of districts and its three lags. Standard
errors are reported in parenthesis.
202
Appendix 2: Examiners’ commentary
Section A
Question 1
State whether the following statements are true or false and explain your
answers briefly. You should answer all eight parts. Each part carries 5 marks.
Concise and clear answers will be rewarded. Answers with no explanation
receive no credit.
a. Empirical evidence suggests conditional aid increases economic growth if
recipient countries adopt good policies.
Reading for this question
Subject guide, Chapter 16; Easterly (2009); Burnside and Dollar (2000);
and Easterly (2003).
Approaching the question
This question basically requires a summary of the empirical findings
of Burnside and Dollar (2000). You first need to discuss their findings
and confirm that they indeed concluded the statement in the question.
You would then need to discuss various problems with that study, using
insights you learned from Easterly (2003, 2009). Overall, this statement is
false and, while some articles conclude aid helps growth, their empirical
strategies are not convincing.
b. The evidence from West Bengal, India, suggests land redistribution involves
large equity-efficiency trade-offs.
Subject guide, Chapter 9; Ray (1998) Chapter 12; Banerjee et al. (2002).
To answer this question you would first need to explain the nature of
the equity-efficiency trade-off. This is when land is split into smaller
pieces and farming becomes less efficient because of lack of access to
technologies and inability to overcome fixed technology adoption costs. In
other words, efficiency is reduced but equity is improved. You should then
very briefly discuss the West Bengal tenancy reform and bring in the result
from Banerjee et al. (2002) on improvements in rice yield after tenancy
reforms that allowed sharecroppers to keep at least 75 per cent of the
harvest. Therefore, the statement is false and the West Bengal evidence, in
fact, suggests equity and efficiency could go hand in hand.
c. Rural farmers can perfectly insure against climate change by pooling their
incomes together in each village. Therefore there is no need for government
intervention.
Subject guide, Chapter 11; Ray (1998) Chapter 15.
This is more of a theoretical question and there is no need to provide
knowledge of empirical studies. You need to decompose individual
incomes to Yij=A+θj+∈ij and explain what each term captures. Then
you should clearly explain why a pooling mechanism within the village
will remove the idiosyncratic shock but is unable to insure against the
203
aggregate
_ shock. Therefore, consumption in the presence of pooling is
Cij=Y j=A+θj. The next step is to argue that climate shocks are likely to
affect the village as a whole, and therefore should be captured by θjYin
ij=A+θj+ϵij
the model. This suggests that the pooling mechanism is unable to insure
against climate shocks and hence the statement is false.
d. PPP converted GDP figures give a more reliable estimate of the income gap
between rich and poor countries compared to exchange rate converted GDP
figures.
This is true. You are expected to discuss the difference between PPP
and exchange rate conversions of GDP and argue why the former gives
a more comparable GDP figure. An intuitive account of the PPP idea is
fine. Briefly, exchange rates are based on price of traded goods, therefore,
non-traded goods could be cheaper in poorer countries. We need to find a
proper measure of the purchasing ability of income in given countries. PPP
collects price data for both traded and non-traded goods in all countries
and hence can construct a more reliable estimate of price differences.
e. Statutory tax structures are very similar across developing and developed
countries but tax revenue as a share of GDP varies significantly.
Subject guide, Chapter 15; Gordon and Li (2009).
The statement is true. Gordon and Li (2009) show the statutory tax rates
are very similar across developing and developed countries. Yet, developed
countries collect a much larger fraction of GDP as taxes.
f. Empirical evidence supports Solow’s prediction of absolute convergence.
Subject guide, Chapter 3; Ray (1998) Chapter 3; and Mankiw et al.
(1992).
The statement is false. You are expected to outline the absolute
convergence hypothesis and its implications and discuss whether the
empirical evidence is supporting it. Absolute convergence states that
countries have the same steady state level of capital per worker (and
hence output per worker) and, therefore, underdeveloped countries
(low capital per worker) should have higher growth rates because they
are further away from the steady state. This suggests underdeveloped
countries catch up with the developed ones and eventually all countries
will have the same level of output per worker and the same output per
worker growth rate. You have investigated the empirical evidence in
Activity 9 of Chapter 3 in the subject guide. Ray (1998) Chapter 4 also
provide an overview. You are expected to talk about lack of convergence
in both growth rates and income levels across developed and developing
countries.
204
g. Increasing returns to scale on its own can generate a poverty trap and
provide a potential explanation for large income gaps between developed
and developing countries.
Subject guide, Chapter 4; Ray (1998) Chapters 4 and 5; and Murphy, et al.
(1989).
The statement is false. You are expected to define increasing returns
to scale and also show understanding of the concept of poverty traps
(multiple equilibria). Based on insights from Murphy et al. (1989) you
should argue that we need to have some sort of market failures and
increasing returns to scale in order to get multiple equilibria. See Chapter
4 (section 4.3.2) of the subject guide.
h. The fact that the average income of households borrowing from a
microfinance institution is lower than that of non-borrowers shows
microfinance loans on average reduce income.
Subject guide, Chapters 2 and 10; Banerjee and Duflo (2011) Chapter
7; and Banerjee and Duflo (2010). Also see references on programme
evaluation in Ravallion (2001).
The statement is false. This question is testing your understanding of the
fact that correlation does not imply causation (see Activity 6 in Chapter
2 of subject guide). Households targeted by the microfinance institution
are a selected sample of the population. For example, it might be that
microfinance has a charitable mission and targets the poorer households.
Comparing incomes of the borrowers with non-borrowers will give
misleading results because we do not expect the income of borrowers and
non-borrowers to be the same in the absence of microfinance lending (by
construction the poor have lower incomes to begin with).
Section B
Answer two questions from this section. Sub-parts of each question carry equal
marks.
Question 2
The following questions relate to W. Easterly, ‘Inequality does cause
underdevelopment: insights from a new instrument’, Journal of
Subject guide, Chapter 5; Ray (1998) Chapter 7; and Easterly (2007).
Approaching the questions
a. Outline three theoretical arguments for why inequality might be harmful for
economic growth.
Any three plausible and well-articulated channels are acceptable, whether
mentioned in the references or not. First, inequality could harm growth
by raising demands for redistribution and hence increasing distortionary
taxes that reduce investment incentives and hence stunts growth. Second,
inequality could harm growth by fuelling civil conflict and violence. A
society torn by violence is unable to protect investments and hence will see
205
insufficient investment and poor growth. Third, inequality could hamper

growth by eroding the level of trust in the society and raising discontent
between employees and employers. If people cannot work together and
cooperate, factories and other production teams will be less productive.
b. Explain why an OLS regression of growth on inequality does not identify the
causal effect of inequality on growth.
OLS regression of growth on inequality is unlikely to isolate the causal
effect of interest because of the presence of reverse causality, omitted
variable bias, and measurement error. You should mention the three issues
briefly, then elaborate on at least one. We pick on reverse causality here.
Growth could affect inequality. For example, growth might be because
of rapid industrialisation. This could mean the incomes of the industrial
workers are increasing rapidly while those of agricultural workers remain
unchanged (in the short run). This will enlarge the gap between these
sectors and increase inequality. A simple regression will confound this link
with the causal effect of inequality on growth.
c. Why might suitability of land for sugar relative to its suitability for wheat be
related to current time inequality? How does this relate to the identification
strategy used in this article?
This relates to the arguments discussed in Easterly (2007). The gist of
the argument is that sugar plantations require large farms in order to be
profitable, whereas wheat production could be done efficiently on small
farms. Large farms historically relied on slave labour, whereas wheat
production was carried out by small households. Slavery is conducive to
higher degrees of inequality. Therefore, countries with a higher fraction
of sugar-suitable land were likely to have higher historical inequality.
Once established, inequality is likely to persist over time, and therefore
we expect these countries to have higher current inequality as well. The
author uses an index of land suitability for sugar relative to wheat to test
for inequality in a regression of growth on inequality.
d. How much does inequality affect growth? Interpret the coefficient estimate
of inequality from the second column in Table A2.
When you are asked to read a result, you should report the coefficient
and its statistical significance. One percentage point increase in the GINI
coefficient reduces GDP per capita by 0.121 log points (approximately
12.1 per cent). Furthermore the t-statistics reported below the coefficient
shows it is significant at 5 per cent.
e. Compare the results in the first two columns of Table A2. Does the author
find evidence that endogeneity biases the OLS coefficient upwards? What can
explain the difference between the OLS and IV estimates?
From Table A2 we see that the OLS estimate is larger than the IV estimate
(the OLS estimate is smaller than the IV in absolute value). As we know,
OLS confounds any endogeneity issues with the causal effect of interest
and IV removes the endogeneity issues (if its assumptions are satisfied).
We therefore conclude that endogeneity issues biased the estimate
upwards. In other words, we estimated a larger coefficient than the
true coefficient (given by the IV estimate). Once the bias is removed the
estimate becomes smaller. Note the magnitude of the OLS coefficient is
smaller but upward or downward bias takes the sign of coefficients into
account.
After discussing this you would need to explain potential factors that could
result in such a bias. For example, you need to find a positive, confounding
correlation between inequality and GDP per capita. For example, it could
206
be that higher GDP per capita results in higher inequality (higher GINI is
higher inequality) because of the reverse causality story mentioned above.
This creates a positive correlation between inequality and growth which
could bias OLS estimates upwards.
f. State the identification assumptions of the instrumental variable approach
here. Do you think the assumptions are plausible in this context?
IV requires two identification assumptions: relevance and exclusion
restriction. You should mention that relevance is satisfied from the large
F-statistics reported in Table A2 and also the plausible theoretical link
between sugar-wheat suitability and inequality. You should then carefully
state what exclusion restriction is and discuss whether you can think of
potential factors that result in violation of it (see activities in Chapter 5 for
more discussion).
Question 3
The following questions refer to E. Duflo ‘Schooling and labor market
consequences of school construction in Indonesia: evidence from an
unusual policy experiment’, American Economic Review 91(4) 2001,
pp.795–813.
Subject guide, Chapter 7; Banerjee and Duflo (2011) Chapter 4; and Duflo
(2001).
a. Describe the INPRES programme studied by Duflo (2001). What were the
programme’s objectives?
INPRES was a large school construction programme in Indonesia. The
main goal was to increase equality of access to schools. Therefore it
targeted areas with initial low enrolment rates and built more schools in
those areas.
b. Discuss how the programme might have affected an individual household’s
education decisions.
By building more schools (and maintaining the quality of schools, for
example, the student–teacher ratios), the travel cost of going to school will
be cut for the affected areas. This could also reduce the opportunity cost of
attending school (students could work on the farm for some hours because
they travel for shorter distances). The reduction in cost of education could
therefore increase the optimal years of education.
c. With reference to columns (1) to (3) of Panel A in Table A3, explain the
article’s results regarding educational attainment.
The young cohort in high intensity areas are affected more than other
groups by the school construction programme. The table shows average
years of education for this group is 8.49. While this is higher than the
old cohort in the high intensity region (8.02), we cannot conclude the
difference (0.47) is solely due to the school construction programme. On
the other hand, the difference between the young cohort in high and low
intensity areas is negative (–1.27) which shows a negative impact of the
school construction. Neither of the above estimates is reliable. Therefore,
Duflo relies on difference-in-differences estimation. This is shown in the
third row, third column of Panel A. Being exposed to the programme
increased years of education by 0.12 years. This estimate is, however,
insignificant because the t-statistics is 1.34 = 00.089
.12
less than 2.
207
d. What is the identification assumption for interpreting the estimated effect

from Panel A as causal?
The identification assumption here is: had it not been for the school
construction programme, the difference between average years of
schooling for young and old cohorts would have been the same for high
and low intensity regions. This is the usual parallel trends assumption
required for the working of difference-in-differences. Notice you need to
carefully learn how to articulate this identification assumption. Saying it
is a parallel trends assumption, without tailoring it to the given context,
would not give any mark.
e. How does the information in Panel B of Table A3 reassure you that Duflo is
estimating the causal effect of the programme?
You need to explain what Panel B is first and how it helps to reassure that
the parallel trends assumption is satisfied. In Panel B, Duflo compares
average years of schooling for two control groups. Both of the groups
were past the age of primary school when INPRES was done, therefore
the difference-in-differences estimation using these two groups should
yield zero. If we observe a number that is statistically different from zero,
we get worried that maybe the parallel trends assumption is violated.
The table shows the cohort aged 12–17 in high intensity regions has
0.034 more years of schooling. This number is much smaller than the DID
estimate in Panel A and it is also not significant. Therefore, it alleviates
concerns about validity of parallel trends assumption.
Question 4
a. Assume a landlord is deciding to get someone to work on his land. What
are the three types of tenancy contracts he could offer? How is risk of
agricultural activity spread in each of these contracts?
The landlord could offer a fixed wage, rent out the land or share the
output with the tenant (sharecrop). You need to briefly explain each
of these so we know you understand their meanings. In terms of risk
distribution of each of these contracts, a fixed wage puts all the risk on
the landlord because the worker receives a constant wage regardless of
output, rental contract puts all the risk on the tenant because the landlord
receives a fixed rent and the tenant receives whatever output that is
produced, sharecropping shares the risk because each party gets a share of
the output and if output falls, they both lose out.
b. Now assume the farm output could be either high, H or zero. The probability
of getting a high output depends on the effort expended by the farmer.
Specifically, e∈[0,1] both show the level of effort and probability of getting
the high output. The landlord does not observe effort and therefore could
condition payments only on observed output. How would the three types of
contracts you identified above look like in this world? Write a parametric
representation of the three contracts.
Any type of land contract in this world would feature, at most, two
different payments because output can take only two values. Therefore
any contract is of the form (l,h), where l shows payoff of the farmer
(tenant) when output is observed to be low and h shows the payoff when
output is high.
208
A wage contract is (w,w) (i.e. the payment made is fixed regardless of

realised output). A rental contract is (–R,H – R), that is to say the tenant
pays out of pocket when output is zero and then retains the rest of output
when it is high, rental payment to the landlord is R. A sharecropping
contract is a sharing rule and therefore, (0,μH) (i.e. tenant receives zero
when nothing is produced and gets μ per cent of output when it is high).
1 2
c. Assume cost of effort is 2 ce , where c is a positive constant. Further assume
that both landlord and tenant are risk neutral. Find the first best (efficient)
level of effort.
Under risk neutrality we can look at expected utility. The first best level of
effort is found by maximising social welfare, that is the sum of landlord
and tenant’s welfare:
W = (1 – e) × 0 + e × H – 1 ce2 = eH – 1 ce2
2 2
Notice the payments from landlord to the tenant are unimportant when
we consider social welfare because they are only transfers between parties
and under any transfer aggregate welfare remains the same. Maximising
W with respect to e by taking its derivative gives the following first order
condition
∂W H
= H – ce = 0 ⇒ e*
∂e c
To check that this is indeed the first best level of effort, we check the
second order condition by looking at the second derivative and check it is
negative
∂2W
= –c <0
∂e2
Therefore, e* above is the first best level of effort.
d. Now consider the three contracts you wrote a parametric representation for
in (b) and find the farmer’s effort level under each of the contracts. Which
contract delivers the same level of effort as the first best?
The tenant would maximise its own payoff under the given contract
structure, therefore
1
max Wtenant = (1 – e) × l + e × h – ce2
e 2
h–l
Again the first order condition is e = c . This is the incentive
~
compatibility constraint. This is farmer’s optimal level of effort given

payments (l, h) under any contract. Replace for (l, h) from the contract
{
parametrisation in (b) to find
0 if fixed wage
H rent
e~ = c
if
μ H if sharecrop
c
Therefore it is clear that only under rental contract will the farmer exert
the efficient level of effort. You need to briefly discuss the intuition for
this result. In a rental contract the tenant is receiving any extra output
produced above the fixed rent (i.e. he is the residual claimant of any
additional output). Therefore, it is as if he owns the land and works on his
farm (social planner in this questions: who maximises social welfare?).
209
e. Now assume the landlord offers the contract that delivers the first best
level of effort. Find the terms (payments) of the contract that maximise the
landlord’s payoff. Assume the outside option of the tenant is 0.
The landlord would offer a fixed rent contract, given the assumption
in question. With this contract his payoff is R, regardless of the level of
output. Therefore, he would maximise R. But he needs to set R in a way
that the tenant is happy to accept the contract (participation constraint).
Therefore, the landlord would solve the following problem
max R
R
1
s.t. (1 − e*) (−R) + e* (H − R) − 2 e*2 ≥ 0
Obviously the landlord would like to set R as high as possible, therefore his
optimal R is when the tenant is indifferent between accepting the contract
and going with his outside option. In other words, the participation
constraint is satisfied with equality. Therefore
1
(1 − e*) (−R) + e* (H − R) − 2 e*2 = 0
H H2
Simplifying and using e* = c we get R* = 2c .
Question 5
The following questions are partly based on R. Burgess, M. Hansen, B.
Olken, P. Potapov and S. Sieber, ‘The political economy of deforestation in
the tropics’, Quarterly Journal of Economics 127(4) 2012, pp.1707–54.
Subject guide, Chapter 14; and Burgess et al. (2012).
a. Why should tropical deforestation be an important topic in the climate
change debates?
Tropical deforestation is responsible for 20 per cent of CO2 emissions.
Increased stock of CO2 in the atmosphere is the main source of global
warming. Furthermore, loss of forest (and biodiversity) could change
ecosystems and the climate balance.
b. Think about a situation where the national government sets logging quotas
that are distributed and enforced by district officials. Further assume that
each province works as a unified wood market where loggers (legal or
illegal) could choose to operate in any district within the same province. Now
intuitively explain why an increase in the number of districts in a province
might lead to more deforestation
Loggers try to acquire permits from district officials by bribing them. If
local officials care about money, then they might try to maximise their
bribe collection by selling permits (allowing illegal logging). Given the
fact that loggers could go to any district within the same province, it is
as if district governments are competing to attract loggers. An additional
district in the province introduces an additional official looking for loggers
to sell permits to and get bribes. The new district could undercut the
bribes asked by other districts and attract more loggers. This means the
price of (illegal) permits, and hence the price of wood, should fall and the
intensity of logging activity should increase, leading to more deforestation.
210
c. With reference to Table A5 interpret the impact of number of districts on

deforestation.
Column 1 shows that adding one more district to the province increases
the level of deforestation by 3.8 percentage points. This number is
statistically significant at 5 per cent because t-statistics is 2.38 = 0.038
0.016
.
d. Does Table A5 provide evidence of illegal logging?
Columns 2 and 3 show deforestation in forest area designated for
production/conversion and for conservation/protection. Obviously there
should not be any logging activity in conservation areas and any amount
of deforestation is illegal here. An additional district in the province results
in 4.7 percentage points more deforestation in conservation areas. This is
obviously illegal. While the estimate is insignificant (t-statistics less than
2), the magnitude is comparable to the original coefficient.
e. With reference to Table A5, explain how the effect of number of districts on
deforestation changes over time.
Panel B includes three lags of number of districts. This will capture
whether the effect of an increase in the number of districts takes time to
develop or fades away over time. If, after the initial establishment of a
new district, the government is in chaos and lacks control over logging,
then we would expect an initial increase in the level of deforestation that
eventually dies out. This hypothesis implies that the sum of the three
lags and the contemporaneous effect should be similar in magnitude
(i.e. coefficients on lag terms are small). Alternatively, it might be that
it takes time for district officials to learn the best ways of extracting
bribes and establishing connections with loggers. In that case, the effect
of an increase in the number of districts should increase over time. This
would imply that Panel B estimates are much larger than Panel A because
coefficients on lag terms are large. Panel B provides support for the latter
story. While the contemporaneous effect is 3.8 percentage points, after
four years there is an increase of 8.2 percentage points in deforested areas.
211
Notes
212

Desarrollo Economico - London University

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Desarrollo Economico - London University

Uploaded by

Copyright:

Available Formats

Economics of development

University of London International Programmes

Published by: University of London

Chapter 6: The role of institutions........................................................................ 69

11.5 Informal insurance: credit as insurance................................................................ 138

1.1 Route map to the guide

1.2 Introduction to development economics

points for thinking about issues in developing countries, the situation

1.4 Aims of the course

1.5 Learning outcomes for the course

1.6 Overview of learning resources

1.6.2 Essential reading

Banerjee, A. and E. Duflo, ‘Addressing absence’, Journal of Economic Perspectives

1.6.3 Further reading

Besley, T. and R. Burgess, ‘Land reform, poverty reduction, and growth:

Duflo, E., M. Greenstone, R. Pande and N. Ryan ‘Truth-telling by third-party

Morduch, J. ‘Income smoothing and consumption smoothing’, Journal of

1.6.4 Online library

Making use of the Online Library

1.7 The examination

should be advised of any forthcoming changes. You should also carefully

Part 1: Cross-country differences

Chapter 2: Overview of development

2.1.2 Learning outcomes

2.1.3 Essential reading

2.1.4 Further reading

Rosenzweig, M.R. and K.I. Wolpin, ‘Natural “natural experiments”’, Journal of

2.1.5 Works cited

2.2 Concept of development

2.3 Measuring development

2.3.1 Income and growth

Table 1: Quantities produced and prices in 2013.

2.3.2 How large are the income gaps?

1980 1990 2000 2010

How reliable are PPP comparisons?

► Stop and read

2.3.3 Poverty and inequality

► Stop and read

There is a close relationship between poverty and growth. If the fruits of

► Stop and read

2.3.4 Measuring well-being

Measuring subjective well-being is much more difficult, but the general

► Stop and read

2.4.1 Role of theory

2.4.2 Empirical methods

► Stop and read

Often in development economics, we are interested in knowing the

y and x to the variance of x.

2.4.3 Programme evaluation problem

D = E[YiT | T ] – E[YiC | T ] + E[YiC | T ] – E[YiC | C ] (2)

where Yi is test scores for household i, and Ti is a dummy variable that is

► Stop and read

2.4.4 Randomised controlled trials (RCTs)

5. Since experiments are often small, it might be difficult to generalise

► Stop and read

2.4.5 Methods using observational data

► Stop and read

2.5 Overview of the chapter

2.6 Reminder of learning outcomes

2.7 Test your knowledge and understanding

Average income ($) Village A Village B

Discuss whether it is possible to estimate the causal effect of borrowing

Reminder: Feedback to activities in this chapter are available

k0 k'0 k1 k2 k