Journal of Development Economics 80 (2006) 251 268 www.elsevier.

com/locate/econbase

A natural experiment for identifying the impact of dnaturalT trade barriers on exports
Chris Milner *, Evious Zgovu
School of Economics and CREDIT, University of Nottingham, Nottingham, NG7 2RD, UK Received 1 May 2003; accepted 1 February 2005

Abstract The paper investigates the relative importance of trade policy and dnaturalT sources of export taxation in Malawi, a landlocked African economy. These sources of export taxation are in turn used to explore how export supply would respond to trade liberalisation as opposed to measures which lower other international trade costs. The findings indicate that trade policy barriers are now only a limited source of dtrueT export taxation and that trade policy reform needs to be complemented with reforms to reduce international trade, including transport, costs. D 2005 Elsevier B.V. All rights reserved.
JEL classification: O24; F13A Keywords: Trade barriers; Exports

1. Introduction Despite significant liberalisation of trade policy across developing countries, there are still a significant number of countries (especially in sub-Saharan Africa) where there has been a weak or sluggish export response. A number of explanations for this muted export and investment response have been offered. Some commentators emphasise problems of limited credibility and fear of policy reversals (e.g., Collier, 1997). Rodrik (1999) adds to this a concern about the limited social and physical infrastructure of many African

* Corresponding author. Tel.: +44 115 9515624; fax: +44 115 9514159. E-mail address: chris.milner@nottingham.ac.uk (C. Milner). 0304-3878/$ - see front matter D 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.jdeveco.2005.02.008

252

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

economies to cope with adjustment induced by trade reforms. There are also a number of recent studies that draw attention to the extent of non-trade policy or dnaturalT barriers to trade, associated with for example a limited transport infrastructure, locational characteristics or port efficiency (Amjadi and Yeats, 1996; Milner, 1997; Clark et al., 2004). These studies typically measure the relative importance of trade policy and natural barriers and therefore the impact of liberalisation on total levels of protection, but do not seek to identify or quantify the impact of natural barriers on export performance. These studies, in line with most research on natural barriers, are unable to persuasively distinguish between the effects of natural barriers and other country-specific characteristics on trade performance. This paper fills this gap for a specific, landlocked economy in Africa (Malawi), by modelling export supply so as to capture the relative impact of policy and natural barriers on export growth. Malawi provides an opportunity to conduct a dnaturalT experiment on natural barriers since its natural barriers changed as a result of first civil war in Mozambique and the subsequent peace. Following the civil war in Mozambique, for example, Malawi had to import and export through South Africa. As a result the cif-fob differential on imports rose by a dramatic 45 percentage points. True export protection measures that incorporate both policy and natural sources of taxation are constructed for the period 19701998. These measures of internal relative price changes are then introduced into an econometrically estimated export supply model for Malawi, alongside other influences (external real exchange rate and productive capacity). This innovation allows us to simulate the in and out of sample impact of changes in natural and policy barriers on export supply. These simulations show that the contribution that trade liberalisation has made thus far to export growth is small, relative to what could be achieved by further liberalisation and/or lowering natural barriers. Indeed for non-traditional exports, the elimination of natural barriers is predicted to induce a significantly larger export response than that resulting from import liberalisation. The remainder of the paper is organised as follows. Section 2 reviews the trade policy regime in Malawi, providing some summary information also on natural barriers. Section 3 sets up the modelling framework, exploring how the relative price effects of these barriers can be represented within a dtrue protectionT framework and incorporated into the export supply function. Section 4 sets out the estimates of dtrueT export taxation from policy and natural barriers. These true protection rates of exportables (alternatively viewed as internal real exchange rates) are introduced, alongside other factors, into export supply functions for Malawi in Section 5. The estimated functions for traditional and nontraditional exports are used in Section 6 to decompose the total export response to lowering dtrueT export taxation into policy and natural barriers. The conclusions and implications of the study are set out in Section 7.

2. Trade barriers and regime changes in Malawi Malawi is a small landlocked economy in the southern African region that shares borders with Tanzania, Zambia and Mozambique. It is not natural resource abundant, but

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

80 70 60

253

Percent

50 40 30 20 10 0 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 Average tariff Import-weighted black market premium

Fig. 1. Import policy barriers (19701998).

its major wealth-producing activities draw upon its relative abundance of low-skilled labour and agricultural land. As a result of this and relatively high transaction costs facing manufacturing activities, its exports have traditionally been dominated by particular agricultural products (tobacco, tea and sugar). Indeed the share of tobacco in total exports has tended to grow post-1970. The UK and US are significant markets for Malawis exports, with South Africa and Zimbabwe still only receiving about one-sixth of the countrys exports. Fig. 1 includes information on average import tariff rates for the period 19701998.1 It captures the three distinct episodes for trade policy over this period; a relatively free trade regime (197079), a restrictive trade regime (198086) and a liberalising regime (198798). During 197079, there was limited intervention in the tradeable goods sector, with import tariffs used mainly for fiscal purposes and limited use of import licensing and foreign exchange licensing on selected competing consumer goods imports.2 These conditions changed sharply after 1979 in response to crisis conditions. Further pressure on foreign exchange reserves was also exerted by the escalation of the civil war in Mozambique. Import tariffs were increased sharply and the coverage of quantitative import and foreign exchange controls increased substantially. Fig. 1 plots also the changes in the extent of non-tariff barriers to trade, using the importweighted black market premium as a proxy measure. It shows a sharp rise in the black market premium in particular after 1978, with a fall back to below 40% after 1986. Tax reforms began in 1985, including tariff reduction and introduction of some export incentives were complemented by the abolition of import licensing and quantitative restrictions in 1988 and with the substantial easing of foreign exchange controls. By the mid-1990s, the average tariff was about 20% and the black market premium was below 10%. In addition to barriers against imports which indirectly discourage exports (implicit) direct taxation of exports has been used in Malawi either through periodic export surcharging or through the price-setting behaviour of the countrys agricultural

1 2

Information of duty collected on imports rather than scheduled tariffs was used. Like many other low-income developing countries, Malawi has a relatively high dependence on trade taxes.

254
90 80 70 60

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

Percent

50 40 30 20 10 0 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 International transport costs Implicit export tax

Fig. 2. Export taxation and natural barriers: 19701998.

commodity marketing board (Kydd and Christiansen, 1982). We can capture this in Fig. 2 with the ratio of the difference between the world and domestic price relative to the world price. The average implicit export tax rate has tended to rise over time, though with considerable fluctuation year-on-year. Natural barriers3 are often overlooked in policy analyses. For a landlocked economy such as Malawi, with its limited and underdeveloped local freight capacity, it is particularly important not to do so. When confronted by disruption of its traditional and shorter routes to the sea through Mozambique, Malawi was forced to shift most of its international freight through South Africa, using road transport through Zambia and Zimbabwe (a distance of about 2500 miles from its commercial centre of Blantyre).4 Not surprisingly, the route diversion increased markedly the countrys already substantial natural trade barriers. Fig. 2 plots the countrys average international transport costs (freight and insurance costs as a percentage of the f.o.b. import value). The normal barrier was about 15% of the import value up to 1982. This jumped sharply after then rising to over 60% for the remainder of the period.

3. Modelling framework: export supply and dtrueT export taxation We posit that besides externaldomestic relative prices and capacity variables, export supply is also affected by internal relative prices or the level of true export taxation arising

The term dnaturalT barrier is commonly used to capture the effects of factors such as geographical distance and location on international transactions costs. To the extent that it is such exogenous geographical factors involved the term is not misleading. To the extent, however, that it is endogenous factors such as the scale of international transactions or national policies affecting the competitiveness or efficiency of sectors such as international transport, then the term dnaturalT may be misleading. Rail transport through Tanzania was not used because of political differences, until the eventual complete closure of the Mozambique route. Dar es Salaam is however about 2600 miles from Blantyre.
4

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

255

from policy and natural barriers to trade.5 This extended function allows us to investigate external and internal relative price effects simultaneously. The latter are normally assumed constant in the standard export supply model. The externaldomestic relative prices do not capture any anti-export biases associated with internal relative price incentives (that is, inter-sectoral effects).6 Allowing for these considerations, the extended export supply function can be expressed as:
d 2 d 3 lt 1 Xt d0 ITd Xt Reert Kt e

where X t is real exports measured in period t , K represents the economys productive capacity to produce exports, ITXt is an index of the rate of true taxation of exports, Reer is real effective exchange rate (expressed in terms of units of domestic currency per unit of foreign currency), sometimes referred to as a PPP real exchange rate, and l is an error term. Rewriting in logarithmic form, we have: lnXt lnd0 d1 lnITXt d2 lnReer t d3 lnKt lt d1 b0; d2 N0; d3 N0 2

The choice of the measure of productive capacity is contentious. Goldstein and Khan (1978, 1985); Bond (1985) and Noland (1989) use trend output. The argument for this measure is that relative prices alone cannot fully explain the willingness and ability to supply exports. Beenstock et al. (1994) argue that supplying foreign markets is inversely related to increases in domestic demand pressures. Specifically, increasing domestic demand pressures for the export good makes it more profitable to supply the domestic market than the foreign market. Among other notable studies, Muscatelli et al. (1995) use the stock of fixed capital to capture the effects of increasing productive capacity and productivity on export supply for some Asian newly industrialised economies. While Haynes and Stone (1983) measure capacity utilisation as deviation-from-trend income. Our choice for the present work is in fact data-constrained, and we take agricultural GDP as a proxy of capacity.7

5 The use of such bsupply-quantityQ equations has been questioned, and instead the bsupply-priceQ equation in which price is the dependent variable is advocated (Goldstein and Khan, 1978; Haynes and Stone, 1983, among others). The assumption of dprice-settingT is more applicable to the analysis of supply behaviour of major exporting countries or manufacturing firms with appreciable world export market control. In the context of a small open developing economy, however, producers face an infinitely elastic demand for the export commodities in world markets and are dprice-receiversT. Bond (1985) and Ahmed (2000) observe that in this context, it is permissible to use single equation export supply functions. 6 In theoretical terms, it is quite possible for a PPP-type real exchange rate (RER) and domestic relative price (i.e., true protection measure) to move in opposite directions (see for example Milner and McKay, 1996). 7 Although a similar proxy is to be found in other studies, there is a potential problem of endogeneity in using agricultural GDP since it includes (traditional) exports. The later exogeneity testing shows that the problem is not in fact serious.

256

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

3.1. Taxation of exports * ) is defined as the change in the domestic The rate of true taxation of exports (T Xt relative price ( P ) of exportables (relative to the endogenously determined price of nontradeables) induced by trade policy and natural barriers, namely:   PNt 4 TXt D PXt 3

where X denotes exportables and N denotes non-tradeables. Rises in the relative price of non-tradeables increase true export taxation, and falls reduce the rate of taxation.8 Consider the small country case with perfect substitutability between traded and local goods, i.e., with full transmission of changes in the border prices of exports to locally consumed exportables (or similarly from import prices to locally produced import substitutes).9 The nominal, domestic price of exportables can be affected by explicit e i sources of taxation (t X ) or implicit export taxation (t X ) from policies of export marketing n agencies or from international transactions, in particular transport costs (t X ) borne by domestic producers of exports. Thus, for convenience, setting initial relative prices equal to unity, the rate of true export taxation in the absence of any export subsidies can be written as: 4 TXt P N P P Xt 1 Xt 4

P i n where Xt =t e Xt + t Xt + t Xt denotes the sum of explicit and implicit taxation of exports; and, N is the proportionate endogenous adjustment of the price of non-tradeables to changes in P trade policy and natural barriers on all tradeable goods. Assuming substitutability between tradeables and non-tradeables and requiring homogeneity, we can express the proportionate change in the price of non-tradeables as follows: N xP M 1 x P X P 0VxV1 5

X is the where x is the index of substitutability between non-tradeables and importables; P e i n proportionate fall in the nominal price of exportables following changes in t Xt , t Xt and t Xt ; M is the proportionate change in the nominal price of importables following explicit and P e i or implicit subsidisation of importables from tariff, t Mt , non-tariff, t Mt , and natural n barriers, t Mt .
8 The traditional literature on true protection (see for example Clements and Sjaastad, 1985; Greenaway and Milner, 1993 and Milner, 1995) expresses the measure as the inverse of Eq. (3). 9 The small country assumption is a reasonable one for Malawi. Perfect substitutability between traded and domestic goods, especially in the importables basket, is less credible. The empirical estimates of true protection allow therefore also for imperfect substitution.

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

257

M and P X gives us a measure of the overall true rate of Substituting (5) into (4) and for P export taxation arising from explicit and implicit sources of policy taxation and subsidisation from natural barriers, namely: P P x Xt * x Mt 2 P TXt 6 1 Xt P P n i e i n where Xt =t e Mt =t Mt + t Mt + t Mt .The rate of true export taxation depends Xt + t Xt + t Xt and therefore not only on the direct effects on the price of exportables (t Xt ), but also on the indirect effects of trade policy interventions on the price of non-tradeables (t Xt , t Mt ) and on the domestic substitution relationships (x ). True taxation of exports increases if any t Xt and t Mt increase (if x N 0). Similarly for given t Xt and t Mt , the rate of true export taxation increases as x increases.

4. Evidence on true export taxation for Malawi The measurement of true export taxation requires information on nominal rates of trade taxation and estimates of the substitution index (x ). Some information on the component elements of trade taxation in Malawi for the period 19701998 were described in Section 2. The detailed annual information and the sources of this information are given in Appendix A. Short and long-run indices of substitutability between exportables (aggregate or traditional and non-traditional separately) and non-tradeables were estimated using time series data on prices of categories of tradeable and non-tradeable goods. Time series models of relative price change, augmented for income and trade balance effects, were estimated for annual data for the period 19701998.10 The estimated indices indicate relatively high substitutability between non-tradeables and importables, and as expected higher substitutability in the long run (0.87 compared with 0.77 in the short run for aggregate exports). Also in line with the values found for other developing countries, the index is higher when traditional rather non-traditional exportables are used in the regression model. Applying the alternative measures of nominal protection and estimates of the incidence parameter to Eq. (6), we can derive a number of alternative estimates of true export taxation for Malawi. These are reported in Table 1 for the same periods or policy episodes discussed earlier. For each period, short and long-run measures are reported for exportables as a whole and separately for traditional and non-traditional exportables, both for the full and partial transmission cases.11

The indirect method of inferring the substitutional relationship from revealed adjustment over time in the relative prices of non-tradeables ( P N /P X ) to changes in the relative price of tradeables ( P M /P X ) is commonly used in the true protection or incidence literature (see Greenaway and Milner, 1993). The estimated time series equations are available from the authors on request. 11 Time series information on the prices of imports and import-competing goods and on the prices of exports and locally consumed exportables were used to estimate transmission coefficients. (Regression estimates are available from the authors). The short- and long-run estimates for exportables were 0.72 and 0.79, and for importables, the estimates were 0.65 and 0.82.

10

258

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

Table 1 Percentage rates of true export taxation for Malawi 197079 SRa Full transmission All exportables Traditional exportables Non-traditional exportables Partial transmission All exportables Traditional exportables Non-traditional exportables
a

198082 LRa 42 47 24 SR 62 68 46 LR 63 68 48

198386 SR 76 80 60 LR 78 83 62

198798 SR 70 74 47 LR 72 77 48

40 43 23

25 25 16

38 40 21

44 45 36

57 61 43

59 60 50

71 76 57

49 49 36

65 69 44

SRshort run, and LRlong run.

Firstly, it should be noted that positive true export taxation is consistently found for all measures and all time periods, with the highest values consistently found for the end of the restrictive trade policy episode (198386). Secondly, and as expected, true taxation rates are always higher for a particular category of measure for the full price transmission case than the partial transmission case, for the long run than the corresponding short-run measure and for traditional rather than non-traditional exportables. Thirdly, we find that the long-run rate of true export taxation is consistently higher than the corresponding nominal rate of export taxation. It is evident from Fig. 3 that the movements in the true rates of taxation for traditional and non-traditional exportables are closely (positively) correlated. It is evident also that the increase in true taxation after 1978 was sharp and sustained. By contrast, the easing of taxation after 1986 is less marked and more gradual, with some evidence of reversal after 1993/94. Before turning to investigate the impact of true export taxation on export supply, we can consider the relative importance of alternative source of export taxation on the dtrueT rate of taxation. In Table 2, we show what the long-run true rates would be if a particular

Percent

100 90 80 70 60 50 40 30 20 10 0 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 Traditional exportables Non-Traditional exportables

Fig. 3. Long-run rates of true export taxation: 19701998.

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

259

Table 2 Impact of eliminating sources of import and export taxation on long-run rate of true export taxation (19871998)a Traditional exports (%) Actual rate Hypothetical rates with removal of: Implicit export taxes Import tariffs (t ) Quantitative import restrictions Import transport costs
a

Non-traditional exports (%) 48 48 36 47 22

77 52 71 76 64

Corresponds to the full transmission information in Table 1.

source of export taxation were eliminated. It is evident that by the period 198798, import trade barriers, especially QRs, were a relatively small source of export taxation; eliminating import tariffs lowers the true tax rate from 77% to 71% and eliminating QRs from 77% to 76%. By contrast, eliminating implicit export taxes would substantially lower true taxation of traditional exports (77% to 52%), while eliminating only import transport costs on (the more intermediate import intensive) non-traditional export sector would lower true taxation from 48% to 22%.12

5. Estimation of the export supply function The export supply function (Eq. (1)) was estimated using 19701998 annual data for Malawi. Data definitions and sources for this stage of empirical work are set out in Appendix B. Econometric analysis of the export supply functions involved a number of now-standard procedures. These included examination of the order of integration of the data time series to inform appropriate modelling techniques. Augmented DickeyFuller unit root tests were used to test the order of integration and the results shown in Table A3.1 in Appendix C (from Microfit 4.0 ) indicate that the series are integrated order 1, I (1). The result that the data series contain unit roots necessitates analysing whether the trends driving the unit root series are linked, that is, whether the individual export supply models are cointegrated . To test for the presence of cointegrating relationships, we used the multivariate maximum likelihood regression methodology advanced by Johansen (1988) and Johansen and Juselius (1992). The Johansen methodology uses the VAR which is a representation of a large class of dynamic structural models (Hamilton, 1994). This being the case, it can be used to investigate the long-run and short-run dynamics (through an error correction model, the ECM). Here, we report the steps taken and results obtained in applying the methodology.

12 Note that our estimates of true export taxation do not include any direct transport costs on exports for data reasons. Although indirect effects via transaction costs on imported intermediates are captured, it may well mean that we are underestimating the effects of transport costs on true export taxation.

260

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

As a starting point, we established the lag length of the VAR using the standard model selection Akaike Information criterion (AIC), Schwarz Bayesian criterion (SBC) and the HannanQuinn criterion (HQC). AIC, SBC and HQC values estimated using Microfit 4.0 tend to point to a lag length (k ) of the VAR of between orders 1 and 2 as the best representation of the data. A higher VAR of order 2 (consistently favoured by AIC) is opted for to limit possible serial correlation of the residuals from a lower order VAR. The next task was to test for cointegration based on the maximal eigenvalue and trace statistics of the stochastic matrix from the Johansen reduced rank regressions; results (from PcFiml version 9 ) are reported in Table A3.2 in Appendix C. On the bases of both these statistics, the null of no cointegration (r = 0) is rejected but that of a single cointegrating vector (r = 1) cannot be rejected at 5% significance level in both models. Thus, there exists only one statistically significant cointegrating relation among the four variables in the traditional export supply and non-traditional export supply models. We also analysed weak exogeneity ; the test results are shown in Table A3.3. The null hypothesis of weak exogeneity is rejected for the dependent variable (export volume) but cannot be rejected for all the explanatory variables in both models at the 5% significance level. This shows that export volume is explained within the systems but that the explanatory variables are dweakly exogenousT to the determination of the long-run export supply relationship.13 On the bases of evidence of the existence of a single cointegrating vector and dweak exogeneityT of the right-hand side variables in the models, we estimated the single equation auto-regressive distributive functions of traditional and non-traditional export supply conditioned on the exogenous variables. The auto-regressive distributive function is particularly attractive as it avoids the finite sample bias suffered by static estimators and is more efficient than VAR methods in small samples like ours here (Inder, 1993). The long-run export supply models are reported in Table 3.14 The elasticity of export supply with respect to the various explanatory variables has the expected signs. The elasticities with respect to true export taxation are statistically significant with 1% increase in true export taxation rate decreasing the volume of traditional exports supplied by 0.65% and non-traditional by 0.54%. An error correction mechanism (ECM) from the equilibrium model was used to model the short-run dynamics. The dynamic model estimation results are reported in Table 4. The proportions of explained variation are reasonably high (nearly 70%). The models pass the

Other testing (Granger-causality and Granger block-causality tests) actually indicates dstrong exogeneity.T These results are available from the authors on request. 14 Estimates using VAR methods (i.e., the cointegrating vectors) are reported in Table A3.4 in Appendix C. The results lead to general conclusions (e.g., with respect to expected signs) that are consistent with those reached using results from the single equation methodology. The estimates of export supply elasticities with respect to production capacity and real effective exchange rate from the single equation method are more consistent with those found in the literature. We use these therefore for the subsequent analysis, though the qualitative implications of the analysis are not dependent on the use of the single equation methodology.

13

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268 Table 3 Solved long-run export supply models Traditional exports Coefficient (S.E.) Constant lnT jt lnK t lnReert 0.01 (1.95) 0.65 (0.17)*** 1.22 (0.24)*** 0.70 (0.28)** R 2 = 0.92; sample = 29 (19701998) ECM: (19711998) ADF(0), with intercept, tau stat. = 6.8349** ECM ADF: 95% = 4.9527 Non-traditional exports Coefficient (S.E.)

261

1.78 (3.53) 0.54 (0.14)*** 1.02 (0.35)*** 0.85 (0.40)** R 2 = 0.90; sample = 25 (19741998) ECM: (19711998) ADF(0) with intercept, tau stat = 6.7580** ECM ADF: 95% = 4.9527

T jt true export taxation of traditional exports ( j = TX ) and non-traditional exports ( j = NX ), K production capacity proxied by an index of agriculture output; Reerreal effective exchange rate as a proxy of external relative prices; ** and *** denote statistical significance at 5 and 1%, respectively; ECMs tau statistics are for unit root tests in the residuals. Statistical significance of tau stat. implies rejection of the null of no cointegration in the models.

usual diagnostic tests. The tests show that we cannot reject the null hypotheses of (a) no autocorrelation in the residuals using results from the Lagrange Multiplier (AR (lags 1 k )) F -statistic test for autocorrelated residuals; (b) no auto-regressive conditional heteroscedasticity (ARCH) in the disturbances using the ARCH F -statistic, White heteroscedasticity tests using squares (to test the null of unconditional homoscedasticity) and using cross-products (to test for heteroscedastic residuals); (c) normality (skewness and kurtosis) of the residuals on the basis of the JarqueBera Normality ChiSQ test results,

Table 4 Short-run (dynamic) export supply models Traditional exports Coefficient (S.E.) ECM(t 1) DlnTXj (t 1) DlnK (t 1) DlnReert DlnReer(t 1) 0.53 (0.17)*** 0.57 (0.22)*** 0.96 (0.35)*** 0.53 (0.25)*** R 2 = 0.68; sample = 19731998 AR 12 F (2,20) = 0.86 [0.54] ARCH 1 F (1,20) = 0.30 [0.58] Normality v 2 = 0.71 [0.70] X i2 F (8,13) = 0.65 [0.73] X i X j F (14,7) = 0.58 [0.82] RESET F (1,21) = 0.91 [0.35] Non-traditional exports Coefficient (S.E.) 0.92 (0.19)*** 0.46 (0.17)*** 0.92 (0.47)** 0.75 (0.31)*** R 2 = 0.69; sample = 19741998 AR 12 F (2,20) = 1.24 [0.41] ARCH 1 F (1,20) = 0.01 [0.96] Normality v 2 = 2.42 [0.30] X i2 F (8,11) = 0.20 [0.98] X i X j F (14,5) = 0.22 [0.99] RESET F (1,19) = 1.40 [0.25]

AR F -statistic from the Lagrange-multiplier (LM) test for autocorrelated residuals (AR). The null hypothesis is no autocorrelation; this is rejected if the test statistic is too high. ARCHautoregressive conditional heteroscedasticity (ARCH) is an LM test for autocorrelated squared residuals. D denotes dfirst differenceT operator; ** and *** denote statistical significance at 5 and 1%, respectively; values in [!] are p -values.

262

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

and (d) no misspecification of the models on the basis of the Ramsey RESET F -statistic test results. The estimated coefficients are qualitatively similar to those in Table 3, and again are consistent with the hypothesised relationships. As one might expect, short-run elasticities are lower than the corresponding long-run values, export supply responding less in the short term to given changes in export taxation, the real exchange rate and productive capacity. Note also that the coefficient on the error correction term (ECM) is higher in the case of non-traditional exports. One would expect non-traditional exports to adjust more quickly to long-run equilibrium.

6. Export supply response to lowering alternative trade costs Armed with estimated export supply functions and information about the relative importance of trade policy and other sources of export taxation, we can explore the in and out of sample period export supply responses to dassumedT changes in alternative elements of trade costs (for convenience adopting a dholding other things constantT methodology). 6.1. Common percentage reductions in trade costs (in-sample) The first experiment is to illustrate in turn the export response for the in-sample period for a 10% reduction only in each source of true export taxation (import tariffs, non-tariff import barriers, implicit export taxes and natural barriers). We use the coefficient on the index of true export taxation from the estimated long-run supply models (Table 3), holding the other variables constant. The results for this simulation are reported in Table 5, part (a) for traditional exports and part (b) for non-traditional exports. Note that a common percentage reduction implies differential absolute reductions in the particular element of

Table 5 Hypothetical export responses to 10% reduction in alternative trade costs Percentage increase with reduction in: Import tariffs only a) Traditional exports 197079 0.9 198082 0.9 198386 0.5 198798 0.4 b) Non-traditional exports 197079 1.9 198082 1.5 198386 1.1 198798 1.1 NTBs only 0.2 0.2 0.1 0.1 Natural barriers only 1.1 0.9 0.5 0.8 Export taxes 2.9 1.6 1.4 2.2 All trade costs 5.3 3.8 2.8 3.7

0.4 0.3 0.3 0.2

2.2 1.6 1.1 2.0

0.0 0.0 0.0 0.0

4.6 3.5 2.7 3.4

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

263

trade costs. The dsimulatedT percentage changes in export supply are reported for the policy episodes discussed earlier. By considering a common reduction, we are able to indicate where greater marginal effort might have been placed by the government of Malawi (and the multilateral agencies) in order to increase export supply. In the case of traditional exports, the largest simulated increases in export supply arise from a given lowering in implicit export taxation, while for non-traditional, it is natural barriers reductions which exert the greater leverage. Note also that the export supply response to the import tariff liberalisation does not exceed that of the reduction of natural barriers in any of the subperiods. Of course this does not mean that one can advocate one form of trade cost reduction over another, given that there may be differential implementation costs and differential welfare and other costs and benefits associated with the lowering of each source of trade cost. Indeed, one may well wish to use all forms of trade cost reduction to promote exports. One should note from Table 5 that the simultaneous reduction of all trade costs produces greater reduction in true taxation and increase in export supply than is evident from the simple addition of each of the component reductions. This arises because policy and natural barriers in this institutional setting have multiplicative effects on total trade costs. An alternative way of illustrating the relative importance of alternative sources of trade costs is to consider the following thought experiment: what would have happened to export supply if each of the policy sources of trade costs had been eliminated after the start of the liberalisation period in 1987 (assuming everything else constant), and how would this have matched up to a 75% reduction in natural barriers. The elimination of implicit export taxes would (hypothetically) have generated a 26% increase in traditional export supply, compared with a 5% increase from the elimination of import tariffs and an 8% increase associated with the partial removal of natural barriers. For non-traditional exports, the important (hypothetical) export supply responses are associated with import tariff elimination (about 15% increase) and natural barrier reduction (24% increase approximately). The story is clear, even for our incomplete capturing of natural barriers, it is evident that for a landlocked country, like Malawi, natural barriers have been overall a more important constraint on export supply than border import taxes.

Table 6 dPredictedT out of sample export supply effects of actual changes in trade costs (19982001) Trade Cost Change (%) Percentage changes in: Traditional exports Import tariffs NTBs Natural barriers Export taxation All trade costs Fall: 29.8 to 11.1 Increase: 0.9 to 1.8 Fall: 54.6 to 11.2 Fall: 58.5 to 19.8 5.2 1 8.5 24.3 55.5 Non-traditional exports 14 2.3 25.2 0 45.5

264

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

6.2. Out of sample simulations Finally, we consider the dpredictedT out of sample impact of the actual changes in trade costs that occurred between 1998 and 2001 on export supply (again holding other things constant). These are reported in Table 6, which also records the changes in the nominal rates of each source of trade cost. The predicted percentage change in export supply is shown for each trade cost component separately, and for the combined or overall change in trade costs. The changes in import policy barriers were more modest than for natural barriers and export taxation, and as a result, it is the change in the latter (58.5% to 19.8%) which is predicted to have the greatest impact on traditional export supply (+ 24%) and the fall in natural barriers (54.6% to 11.2%) which impacts most on non-traditional export supply (+ 25%). Overall, there was a large decline in nominal trade costs and in true export taxation rates, which dshouldT have brought about large increases in export supply; over 55% for traditional and over 45% for nontraditional. The predicted increases are substantially in excess of the actual increases in Malawis exports in this period, which were subject among other things to the effects of drought. Note, in any case, that actual exports will be influenced also by external demand conditions.

7. Conclusions This paper has integrated policy-induced and other trade costs into a single framework, in order to examine both the relative importance of policy and natural sources of true export taxation and the impact of both sources of export taxation on relative incentives domestically to supply (traditional and non-traditional) exports. It has applied the framework to Malawi over the period 197098; a period over which this landlocked, developing country in Africa experienced marked changes in both its trade policy regime and in the nature of trade costs associated with transporting goods in and out of the country. The evidence shows that, for non-traditional exports in particular, international transport costs have been a major source of true export taxation, and the export supply response is sensitive to trade cost-induced changes in true export taxation. It is clearly important in economies like Malawi to seek to increase the effectiveness of trade policy reform through the adoption of complementary measures to reduce other trade costs. It is also evident that it is important that both internal and external real exchange rate effects are incorporated into export supply functions, in particular, where policy and other trade costs cause marked variations between movements in the relative internal and external prices of tradeable goods.

Acknowledgement The authors are grateful to the editor and a referee for very helpful comments on earlier drafts of the paper.

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

265

Appendix A. Nominal rate of trade taxation and sources of information The values for the rates of taxationexplicit (t e ), implicit (t i ) and dnaturalT (t n ) for exportables (X ) and importables (M ) over the period 197098 taken from:
e tM i tM n tM

e tX i tX

Nominal tariff rate. Sources: Statement of External Trade (Malawi Government, 19701999c); International Financial Statistics (IMF, various issues). Nominal tariff equivalent rate. Source: African Development Indicators (World Bank, various issues). Nominal rate of import protection due to dnaturalT barriers proxied by unit import external transport costs. Sources: Statement of External Trade (Malawi Government, 19701999c); International Financial Statistics (IMF, various issues). Nominal export tax rate. Implicit export tax rate (the ratio of the difference between the world and domestic price relative to the world price of relevant commodities). Source: Financial and Economic Review (Reserve Bank of Malawi, 19701999), and International Financial Statistics (IMF, various).

Annual trade cost values

Year 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001
e tM i tM n tM e n tM tM

i X = tX

9.5 11.7 14.4 10.0 11.3 12.7 17.5 18.9 21.5 29.6 32.1 42.7 55.9 66.1 65.2 61.7 64.2 30.7 22.6 24.8 35.1 28.5 32.7 31.3 22.2 21.5 29.5 25.6 29.8 13.9 14.5 11.1

2.2 5.9 3.4 4.1 4.5 0.0 8.0 9.3 0.0 0.0 0.0 15.3 25.1 24.4 32.2 30.8 30.5 14.8 10.2 11.4 8.3 12.4 6.4 16.1 3.9 3.4 3.8 5.2 0.0 3.2 2.6 1.8

15.6 15.6 13.8 13.8 13.5 13.2 13.5 27.0 27.0 28.2 38.3 35.4 54.0 55.9 66.7 66.7 66.7 66.7 66.7 61.4 61.3 66.7 61.3 66.7 50.9 66.7 66.7 66.7 54.6 17.7 17.7 11.2

1.5 1.8 2.0 1.4 1.5 1.7 2.4 5.1 5.8 8.3 12.3 15.1 30.1 37.0 43.5 41.1 42.8 20.4 15.0 15.2 21.5 19.0 20.0 20.9 11.3 14.3 19.7 17.1 16.3 2.4 2.6 1.2

28.8 35.0 33.6 29.2 30.8 27.6 41.3 60.3 54.3 66.1 82.7 108.5 165.1 183.4 207.6 200.2 204.2 132.6 114.5 112.9 126.3 126.5 120.4 134.9 88.3 105.9 119.7 114.5 100.7 37.1 37.3 25.3

33.7 22.3 32.3 24.7 22.9 28.9 30.9 15.2 35.8 37.8 49.8 26.5 25.8 52.6 40.4 55.3 56.2 41.9 31.7 55.7 46.8 36.6 64.2 66.9 76.3 50.7 52.2 61.1 58.5 40.1 36.5 19.8

266

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

Appendix B. Variable definitions and data sources

Xt Real exports in period t , proxied by export volume indexes (1980 = 100) of traditional (agricultural) exports and non-traditional (semi-processed primary and manufactures) exports. The sources of data are Economic Report , and Monthly Statistical Bulletin , Malawi Government (19701999), and Financial and Economic Review, Reserve Bank of Malawi (19701999). Real effective exchange rate index (1980 = 100) defined as E.Pxw /P d . Source: International Financial Statistics Yearbook and International Financial Statistics Yearbook (IMF, various issues). The economys productive capacity to produce exports proxied by a real index of agricultural gross domestic output (1980 = 100). Data sources: Economic Report , and Monthly Statistical Bulletin , Malawi Government (19701999), and Financial and Economic Review, Reserve Bank of Malawi (19701999).

Reert K

Appendix C
Table A3.1 Unit root test results Variable lnX TX lnX NX lnT TX lnT NX lnK lnReer DlnX TX DlnX NX DlnT TX DlnT NX DlnK DlnReer Intercept/Trend included Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept and and and and and and trend trend trend trend trend trend Model (lag length) ADF(2) ADF(1) ADF(4) ADF(3) ADF(1) ADF(2) ADF(1) ADF(1) ADF(0) ADF(0) ADF(0) ADF(0) Test statistic 3.5577 3.2878 2.0283 1.3696 2.1423 1.7469 7.2550** 5.5267** 6.8378** 4.3740** 4.6642** 8.2523** ADF: 95% 3.5943 3.5867 3.6119 3.6027 3.5867 3.5943 2.9798 2.9798 2.9750 2.9750 2.9750 2.9750 Inference Random walk Random walk Random walk Random walk Random walk Random walk Stationary, I(1) Stationary, I(1) Stationary, I(1) Stationary, I(1) Stationary, I(1) Stationary, I(1)

DdFirst differenceT operator; ADFaugmented DickeyFuller; optimal lag length is the largest order for which the test statistic (tau ) is significant at 10%. Tests carried out using, and critical values provided by, Microfit 4.0 (Pesaran and Pesaran, 1997). **Indicates rejection of the null hypothesis of a unit for first difference series at 5%. The null is firmly rejected for all series in favour of difference stationarity. Table A3.2 Cointegration tests in the export supply systems Ho: rank = r Traditional exports r =0 rp1 r j2 r =0 r =1 rp2 rp3 Trace Statistic 23.9** 5.8 0.1 36.3*** 20.5 8.9 2.5 95% 21.0 14.1 3.8 27.1 21.0 14.1 3.8 Maximal Statistic 30.0** 5.9 0.1 68.2*** 27.8 11.4 2.5 95% 29.7 15.4 3.8 47.2 29.7 15.4 3.8

Non-traditional exports

*** and *** denote significance at 1 and 5%, respectively.

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268 Table A3.3 Tests of weak exogeneity Null: variable is weakly exogenous Export volume (lnX ) Productive capacity (lnK ) Real effective exchange rate (lnReer) Rate of true export taxation (lnT TX ; lnT NX ) Traditional export supply model v 2* 12.681 0.44845 0.80901 0.98758 p -value 0.0004 0.5031 0.3684 0.3203 Non-traditional export supply model v 2* 16.001 0.23531 1.603 0.199

267

p -value 0.0001 0.6276 0.2055 0.6555

*1 degree of freedom. Test statistics computed using PcFiml version 9 .

Table A3.4 Cointegrating vectors (a) Traditional export supply Order of VAR = 2; Chosen number of cointegrating vectors, r = 1 27 observations from 1972 to 1998; normalised on export volume (lnX TXt ) Vector 1 lnX TXt lnT TXt lnK t 1.0000 0.87469 -1.3493 (0.000) (0.27740) (0.27668) Thus, lnXT X t 0:87469 lnTTX t 1:3493 lnKt 0:72415 lnReert
0:27740 0:27668 0:11028

lnReert -0.72415 (0.11028)

(b) Non-traditional export supply Order of VAR = 2; Chosen number of cointegrating vectors, r = 1 27 observations from 1972 to 1998; normalised on export volume (lnX NXt ) lnT NXt lnK t Vector 1 lnX NXt 1.0000 0.20835 0.79376 (0.0000) (0.079163) (0.10565) Thus, lnXN X t 0:20835 lnTN X t 0:79376 lnKt 0:58619 lnReert
0:079163 0:10565 0:064865

lnReert 0.58619 (0.064865)

References
Ahmed, N., 2000. Export response to trade liberalisation in Bangladesh: a cointegration analysis. Applied Economics 32, 1077 1084. Amjadi, A., Yeats, A.J., 1996. Have transport costs contributed to the relative decline of African exports? Working Paper, World Bank: Washington. Beenstock, M., Lavi, Y., Ribon, S., 1994. The supply and demand for exports in Israel. Journal of Development Economics 44, 333 350. Bond, M.E., 1985. Export demand and supply for groups of non-oil developing countries. IMF Staff Papers 32, 56 77 (March). Clark, X., Dollar, D., Micco, A., 2004. Port efficiency, marine transport costs and bilateral trade. Journal of Development Economics 75, 417 450. Clements, K.W., Sjaastad, L.A., 1985. How protection taxes exporters. Thames Essay, vol. 39. Trade Policy Research Centre, London. Collier, P., 1997. Ugandan trade policy: liberalisation in an environment of limited credibility. World Economy 20, 649 665.

268

C. Milner, E. Zgovu / Journal of Development Economics 80 (2006) 251268

Goldstein, M., Khan, M.S., 1978. The supply and demand for exports: a simultaneous approach. Review of Economics and Statistics 59, 275 286. Goldstein, M., Khan, M.S., 1985. Income and price effects in foreign trade. In: Jones, R.W., Kenen, P.B. (Eds.), Handbook of International Economics, vol. 2, pp. 1041 1099. Greenaway, D., Milner, C.R., 1993. Trade and Industrial Policy in Developing Countries. Macmillan, London. Hamilton, J.D., 1994. Time Series Analysis. Princeton University Press, New Jersey. Haynes, S.E., Stone, J.A., 1983. Specification of supply behaviour in international trade. Review of Economics and Statistics 65, 626 632. Inder, B., 1993. Estimating long-run relationships in economics: a comparison of different approaches. Journal of Econometrics 57, 53 68. Johansen, S., 1988. Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control 12, 231 254. Johansen, S., Juselius, K., 1992. Testing structural hypotheses in a multivariate cointegration analysis of the PPP and UIP for UK. Journal of Econometrics 53, 211 244. Kydd, J., Christiansen, R., 1982. Structural change in Malawi since independence: consequences of a development strategy based on large-scale agriculture. World Development 10, 347 369. Milner, C.R., 1995. Relative incentives and trade strategies: typologies and possibilities. Economic Record 71, 230 239. Milner, C.R., 1997. On natural and policy-induced sources of protection and trade regime bias. Weltwirtschaftliches Archiv 132, 740 752. Milner, C.R., McKay, A., 1996. Real exchange rate measures of trade liberalisation: some evidence for Mauritius. Journal of African Economies 5, 69 91. Muscatelli, V.A., Stevenson, A.A., Montagna, C., 1995. Modeling aggregate exports for some Asian newly industrialized countries. Review of Economics and Statistics 77, 147 155. Noland, M., 1989. Japanese trade elasticities and the J-Curve. Review of Economics and Statistics 71, 175 179. Pesaran, M.H., Pesaran, B., 1997. Microfit 4.0: Windows Version. Oxford University Press, Oxford. Rodrik, D., 1999. The New Global Economy and Developing Countries: Making Openness Work. Overseas Development Council, Washington.