You are on page 1of 52

Demand Estimation and Demand

Forecasting
Dr. A. N. Sah

Introduction

Prediction is very difficult, especially if it's


about the future. Nils Bohr
The successful business manager is a
forecaster first; purchasing, producing,
marketing, pricing, and organizing all
followanonymous.

What is forecasting?

Forecasting is a process used for predicting


future demand based on past demand
information.

Demand for products and services is usually uncertain.

Forecasting can be used for

Strategic planning (long range planning)

Finance and accounting (budgets and cost controls)

Marketing (future sales, new products)

Production and operations

Types of Forecasts

Short term
Medium Term
Long term

Short Term Forecasting

It normally relates to a period not exceeding a


year
Benefits of Short term forecasting

Evolving a Sales Policy


Determining Price Policy
Fixation of Sales Target

Long Term Forecasting

It refers to the forecasts prepared for long


period during which the firms scale of
operations or the production capacity may be
expanded or reduced

Benefits of Long term forecasting


Business Planning
Manpower Planning
Long-Term Financial Planning

Factors involved in Demand Forecasting


Undertaken at three levels:
a. Macro-level
b. Industry level eg., trade associations
c. Firm level
Should the forecast be general or specific (product-wise)?
Problems or methods of forecasting for new vis--vis
well established products.
Classification of products producer goods, consumer
durables, consumer goods, services.
Special factors peculiar to the product and the market
risk and uncertainty.

Criteria of a good forecasting


method
1.

2.
3.
4.
5.

Accuracy measured by (a) degree of


deviations between forecasts and
actuals, and (b) the extent of success in
forecasting directional changes.
Simplicity and ease of comprehension.
Economy.
Availability.
Maintenance of timeliness.

Presentation of a forecast to the


Management
1.

2.
3.

4.
5.
6.

Make the forecast as easy for the management to


understand as possible.
Avoid using vague generalities.
Always pin-point the major assumptions and
sources.
Give the possible margin of error.
Omit details about methodology and calculations.
Make use of charts and graphs as much as
possible for easy comprehension.

Various macro parameters found useful for


demand forecasting
1.
2.
3.
4.
5.
6.
7.

National income and per capita income.


Savings.
Investment.
Population growth.
Government expenditure.
Taxation.
Credit policy.

Significance of Demand Forecasting

Production Planning
Sales Forecasting
Control of Business
Inventory Control
Growth and Long Term Investment Program
Economic Planning and Policy Making

Sources of Data

Primary: which are collected for first time for


purpose of analysis
Secondary : are those which are obtained
from someones else records

Techniques of Demand Forecasting


Statistical Methods

Consumption
level
method

Time series analysis

Smoothing techniques

Least Square Method

Econometric
Methods

Regression Method

Simple

Multivariate

Consumer Survey Methods

Complete enumeration Method: All potential users of


product are contacted and are asked about their future
plan of purchasing the product in question
Limitations

Very expensive in case of widely dispersed market


Consumers may not know their actual demand and may
be unable to answer query
Their plans may change with a change in factors not
included in questionnaire

Contd

Sample Survey: Only a few potential


consumers and users selected from relevant
market are surveyed
Method is simpler, less costly and less time
consuming.
Surveys are done to understand market
demand, tastes ad preferences, Consumer
expectations etc

Opinion Poll Method

Aim at collecting opinions of those who are


supposed to possess the knowledge of the
market e.g sales representatives, sales
executives, consultants and professional
marketing experts
This method includes
Expert opinion
Delphi method

Expert opinion

Under this method each expert is asked independently to


provide a confidential estimate and results could be averaged.

Experts may include executives directly involved in the


market such as suppliers, distributors or dealers or marketing
consultants, officers of trade association etc.

Advantage is that there is no danger that group of experts


develop a group- think mentality. Moreover, forecasting is
done quickly and easily without need of elaborate need of
statistics.

Delphi Method

This method is an attempt to arrive at a consensus on


some issues by questioning a group of experts
repeatedly until the responses appear to converge
along a single line or the issues causing
disagreement are clearly defined.
Generally a panel consisting 9 to 12 experts
A coordinator is required for the process

Market Experimentation

Test marketing

A test area is selected, which should be a representative of the whole


market in which the new product is to be launched.

A test area may include several cities having similar features i.e.
population, income levels, cultural and social background, choice and
preferences of consumers

Market experiments are carried out by changing prices, advertisement


expenditure and other controllable variables influencing demand

After such changes are introduced in the market, consequent changes


in demand over a period of time are recorded.

Contd

Experiments in laboratory or consumer clinic


method

Under this method consumers are given some money to


buy in a stipulated store goods with varying prices,
packages, displays etc.

They are also requested to fill a questionnaire asking


reasons for the choices they have made

The experiment reveals the consumers responsiveness to


the changes made in prices, packages and displays.

Limitations of market experiment


methods

Very expensive
Being costly, carried out on a scale too small to permit
generalization with a high degree of reliability
Based on short term and controlled conditions which
may not exist in an uncontrolled market
Tinkering with price increases may cause a permanent
loss of customers to competitive brands

Types of data used in Statistical


methods

Time series data refer to data collected over a


period of time recording historical changes in price
, income and other relevant variables influencing
demand for a commodity

Cross sectional analysis is undertaken to


determine the effects of changes like price, income
etc on demand for a commodity at a point in time

Types of Statistical Methods

Consumption level Method


Time series Analysis (Trend Projection)
Smoothing Techniques

Moving Averages
Least Squares Method
Exponential Smoothing Technique

Econometric Method
Barometric Method

Consumption Level Method

Under this method consumption level method may


be estimated on basis of co-efficient of Income
elasticity and price elasticity of Demand
D* = D(1+M*.e)
D* =Projected per capita demand
D= Actual Per capita Demand
M*= Percentage change in per capita income/price
E=elasticity of demand

Illustration
Suppose Income elasticity of demand for
chocolates is 3. In year 1995 per capita income is
$500 and per capita annual demand for
chocolates is 10 million in a city. It is expected
that in year 2000 per capita income will increase
by 20 % . Then projected per capita demand for
chocolates in 2000 will be?

Time Series Analysis

It attempts to forecast future values of time series by examining


past observations of data
The time series relating to sales represent the past pattern of
effective demand for a particular product. Such data can be
presented either in a tabular form or graphically for further
analysis.
The most popular method of analysis of the time series is to
project the trend of the time series.a trend line can be fitted
through a series either visually or by means of statistical
techniques.
The analyst chooses a plausible algebraic relation (linear,
quadratic, logarithmic, etc.) between sales and the independent
variable, time. The trend line is then projected into the future by
extrapolation.

Time Series Analysis


Popular because: simple, inexpensive, time series
data often exhibit a persistent growth trend.
Disadvantage: this technique yields acceptable
results so long as the time series shows a persistent
tendency to move in the same direction. Whenever a
turning point occurs, however, the trend projection
breaks down.
The real challenge of forecasting is in the prediction of
turning points rather than in the projection of trends.

Time Series Analysis

Reasons for fluctuations in time series data

Secular Trend : value of a variable tends to increase or decrease


over a period of time
Cyclical Fluctuations are major expansions and contractions that
seem to recur every several years
Seasonal variation refers to regularly recurring fluctuation in
economic activity during each year
Irregular influences are variations in data series resulting from
wars, natural disasters or other unique events

Four sets of factors: secular trend (T), seasonal variation


(S), cyclical fluctuations (C ), irregular or random forces
(I).
O (observations) = TSCI

Trend Projection

Simplest form of time series analysis is projecting


trend based on assumption that factors responsible
for past trends in variable to be projected will
remain same in future.

Trends refer to long term persistent movement of


data in one direction-increase or decrease

Trend component of time series is the overall


direction of the movement of the variable over a
long period.

Reasons for studying Trends

Studying secular trends permits us to project past


patterns, or trends, into the future
In many situations studying the secular trend of a time
series allows us to eliminate the trend component from
the series.
Methods for trend Projections:
Least squares method
Smoothing Techniques
Moving Average
Exponential smoothing

Moving average Method

This method assumes that demand in future year


equals the average of demand in past years
Under this method 3 yearly,4 or 5 yearly etc moving
average is calculated by moving total of values in
group of years(3,4,5)is calculated, each time by
ignoring first entry and incorporating last one
For Three period Moving average the forecasted
value of time series for next period is average value
of previous three periods in time series

Moving average Method

In order to decide which of these moving averages


forecasts is better ie closer to actual data root-meansquare-error (RMSE) is calculated for each
forecast and using moving average that results in
smaller RMSE
The greater the number of periods used in moving
average the greater is the smoothing effect because
each new observation receives less weight. Useful
when time series data is more erratic.

Three-quarter Moving Average forecasts


Quarter

Firms Actual
Market Share (A)

Three Quarter
Moving Average
Forecast (F)

A-F

(A-F)2

20

22

23

24

21.67

2.33.

5.4289

18

23.00

-5.00

25

23

21.67

1.33

1.7689

19

21.67

-2.67

7.1289

17

20.00

-3.00

22

19.67

2.33

5.4289

10

23

19.33

3.67

13.4689

11

18

20.67

-2.67

7.1289

12

23

21.00

2.00

13

21.33

Total= 78.3534

Five Quarter Moving Average forecasts


Quarter

Firms Actual
Market Share (A)

Fiv3 Quarter
Moving Average
Forecast (F)

A-F

(A-F)2

20

22

23

24

18

23

21.4

1.6

2.56

19

22

-3

17

21.4

-4.4

19.36

22

20.2

1.8

3.24

10

23

19.8

3.2

10.24

11

18

20.8

-2.8

7.84

12

23

19.8

3.2

10.24

13

20.6

Total= 62.48

Three & Five year Moving Average


Comparison

RMSE= {(A-F)2 / n}1/2

RMSE = 78.3534/9 = 2.95


RMSE = 62.48/7 = 2.99
Thus Three Year Moving Average is marginally better than
corresponding Five year

Exponential Smoothing

A serous criticism of using moving averages in forecasting is that they give equal
weight to all observations in computing the average even though more recent
observations are more important
It uses a weighted average of past data as basis for a forecast by giving heaviest
weight to more recent information and smaller weights to observations in more
distant past on assumption that future is more dependent on recent past than on
distant past
The value of time series at period t (A t) is assigned a weight (w) between 0 and 1
both inclusive, and forecast for period t (F t) is assigned 1-w . The basic Equation :
Ft+1 = wAt + (1-w)Ft
Where Ft+1 = forecast for next period
At = Actual value of time t (most recent actual data)
Ft = forecast for present period
w = weight ie smoothing constant

Contd..

Rules of Thumb:
When magnitude of random variations is large, w is
taken as lower value so as to even out the effects of
random variation quickly
When magnitude of random variations is moderate, a
large value is assigned to w
It has been found appropriate to have w between 0.1 and
0.2 in many systems
To identify best forecast amongst many arrived from
different values of W,RMSE is used and forecast having
least RMSE is considered as best

Illustration : Exponential Smoothing


Time period

Actual Sales
(Rs 000)

Forecasted Sales

60

64

58

66

70

60

70

74

62

10

74

63

Contd..

Forecast sales of time period 8,9and 10


Take a smoothing constant w= 0.2

Econometric Methods

Combine statistical tools with economic


theories to estimate economic variables and to
forecast intended economic variables
An econometric model may be a single
equation regression model
Types of Econometric Method
Regression Method

Regression Method

It attempts to find out relationship between


dependent and independent variables
It is a statistical technique for obtaining the
line that best fits data points
It is obtained by minimizing sum of squared
vertical deviations of each point from
regression line and method used is called
Ordinary Least Squares method (OLS)

Contd

Linear Equation
Y= a +bX Where X and Y are averages
Objective of regression analysis is to estimate
linear relationship ie a and b
a = Y-bX
b = NXY (X) (Y)
N X2 - (X)2

Year t

Advertising
Xt (millionRs)

Sales
Yt (000
units)

X2

XY

45

25

225

50

64

400

10

55

100

550

12

58

144

696

10

58

100

580

15

72

225

1080

18

70

324

1260

20

85

400

1700

21

78

441

1638

10

25

85

625

2125

N = 10 X = 144 Y=656 X2=2448 XY=10254

Estimating Linear equation

b = 10(10254) (144)(656)
10(2448) (144)2
b = 2.15
a = Y bX where Y & X are averages
Y = 34.54 + 2.15X
It means that an increase of Rs 1 million in ad
expenditure will bring an increase of 2.15
thousand units in sales ie 2,15000 units

When a time series data reveals rising trend


for e.g. in sales then equation is:
S= a +bT where a and b are estimated using
following two equations
Estimating
Linear Trend-Least Squares
S= na + bT
Method
ST = a T + b T2

Year

Sales of Bread(000 in tonnes)

1992

10

1993

12

1994

11

1995

15

1996

18

Illustration: Suppose
1997 that a local bread manufacturer company
14 wants to assess
demand for its product for years 2002,2003 and 2004. for this purpose it uses
1998
20
time series data of its sales over past 10 years.
1999

18

2000

21

2001

25

Estimation of Trend Equation


Year

Sales

T2

ST

1992

10

10

1993

12

24

1994

11

33

1995

15

16

60

1996

18

25

90

1997

14

36

84

1998

20

49

140

1999

18

64

144

2000

21

81

189

2001

25

10

100

250

n=10

S=164

T=55

T2 = 385

ST= 1024

Contd.

164 = 10a + 55b


1024 = 55a + 385b
S = 8.26 + 1.48T
For 2002, S2 = 8.26 + 1.48(11) = 24,540
tonnes

Problems: Demand Forecasting


1. Using method of least
squares, fit straight line
trend and estimate the
annual sales of 1997.

Year

Sales( lacs
in Rs)

1991

45

1992

56

1993

78

1994

46

1995

75

Contd..
2. Suppose number of
refrigerators sold in past 7
years in a city is given in
table. Forecast demand
for refrigerator for year
2002 and 2003 by
calculating 3-yearly
moving average

Year

Sales

1995

11

1996

12

1997

12

1998

13

1999

13

2000

15

2001

15

Contd..

3. Estimate demand for


sugar in 2003-04 if
population in 2003-04
is projected to be 70
million by using
method of least
squares to estimate
regression equation of
form: Y= a+ bX
Data on Consumption
of Sugar:

Year

Population
(millions)

Sugar consumed
(000tonnes)

95-96

10

40

96-97

12

50

97-98

15

60

98-99

20

70

99-2000

25

80

2000-01

30

90

2001-02

40

100

Thank you

You might also like