You are on page 1of 42

2.3.

Estimating and
Forecasting Demand
In the previous part we have tried to look at
the concept of elasticity and their importance
for decision making.
Our analysis of elasticity was based on the
assumption that
“data already available and managers can easily
compute them”
However, it is not always the case rather
managers must develop a data set and use
statistical methods to estimate the demand
equation.
The estimated equation can be used to predict
demand for their product and the coefficients
of elasticity. 1
Estimating Demand
Statistical Methods is commonly used to
estimate and forecast demand and other
important economic variables
Statistical method is more analytical and uses
empirical data in basic statistical models to
generate forecast about economic variables.
The statistical methods are,
Econometrics model
Time Serious
2
Estimation of demand (con…)
A) Econometrics model
This method estimate on the basis of cause and
effect relationship between different variables
Consider the simple demand function
Q = B0 + B1P
 The law of demand implies that the coefficient-
B1 be negative.
An estimation of the value of B0 and B1 is
required for decision making purpose
The most widely used technique to estimate
these coefficients is the Least – Squares
Method (LSM) or (OLS). 3
Estimation of demand (con…)
 Consider two variables, X and Y, and the
value of Y depends up on the value of X.
 If we know what type of relationship exist
between X, Y we can estimate the value of
Y given the value of X.
Suppose we have collected data on these
variables for a number of periods
 and we plot them on the graph
 we can see the probable relationship
between X and Y 4
Concepts of demand (Cont…)
The problem is to estimate the
Y = B0 + B 1 X line that best fit the scattered
Y
data (estimate Coefficients)
The best estimation of the
coefficients B0 and B1 is to fit
the line through the data points
so that the sum of squared
vertical difference from each
point to the line is minimized.

X  This technique is called Least-


Square Regression Method (OLS)
5
Steps in Estimating Demand
Function
Using regression analysis estimating
coefficient of demand has four steps
1. Developing Theoretical Model
2. Data Collection
3. Choice of Functional Form
4. Estimation,
 Interpretation, and
 Testing of the validity of the
estimates.
6
1. Developing Theoretical Model
Demand Estimation requires a proper
formulation of a theoretical model of the
economic relationship among variables
Model expected to be based on sound
economic theory
Model formulation process involves
Determining which variables should be
included in the analysis
Predicting the nature and magnitude of
the relationship between variables.
7
Developing Theoretical Frame-work
(Cont…)
Example
Consumers’ demand for goods and services
is affected by factors like,
 its own price,
 consumers’ income,
 advertisement,
 Prices of related goods, and the
like.
 Socio-demographic factors
The general form could be developed as,
Qd = f ( P, I, Ad, Po …)
8
Developing Theoretical Frame-work (Cont…)

Prior knowledge about the relationship


between variables can be used to assess
empirical results of regression analysis.
 If the sign of the estimated coefficients are
inconsistent with the economic theory, then
the process should be re- examined
E.g. If the estimated coefficient of price is
not negative, it is possible that
 the function is not correctly formulated
 There was error in data collection or entry
 The relevant variables were omitted
 There is some unique about the good or
service 9
2. Data Collection
Once relevant variables are identified, data
for each of the variable must be collected
Data may be collected from
 survey

 market experiment

Time series or Cross section data may be used


Time series data:- Twenty months of data on
quantity demanded, price, income, price of
other goods from a single market would be
an example 10
Data Collection (cont…)
Cross – section:- Data can be collected
from a number of markets at a single
point in time.
Example
Quantity demanded, income, price, data
can be collected from 100 different
markets during a particular period.

11
3, Choice of Functional Form

In the first step a general relationship has


been formulated among variables involved.
That is , Q = f (P, I , P0, Ad)
Estimation using regression analysis requires,
the choice of specific functional form for the
above relationship
A linear model is the simplest possible form.
This can be expressed as,
Q = B0 + B1P + B2I + B3P0+ B4Ad
12
Choice of Functional Form (cont…)

The liner form has the following


advantage

It can be estimated without


modification
Coefficient of the valuables have a
simple interpretation
We can also easily calculate and
interpret elasticity for each variable
13
Choice of Functional Form (cont…)
Various functional form can be used for
regression analysis other than the
linear Model
The most common form is the
multiplicative (exponential) functional
form.
This can be expressed as
Qd = B0PB1IB2P0B3AdB4
In this form the equation can not be
estimated using ordinary least square
because it is not linear.
14
Choice of Functional Form (cont…)
With a simple transformation of the
function it will be a liner function.
Like using logarithmic transformation
Log (Qd) = log (B0PB1IB2P0B3TB4 )
The above form can be modified as,
log(Q)=log B0+B1 log P +B2logI +B3log P0+B4log
Ad
Since the equation is now linear the
coefficient can be estimated using the
ordinary least square method
15
Choice of Functional Form (cont…)
The advantage of multiplicative form
 It directly estimate the value of
elasticity
The estimated coefficient represent
the value of elasticity which is
constant.
The choice of an appropriate functional
form depends on the underlying
theoretical model and the intended use of
the results.
16
4.Estimation, interpretation, and
testing of the validity of the
estimates
A) Estimation and interpretation:
This can be done either manually or using the
statistical software packages,
There are many software packages
The output provided is relatively easy to
interoperate
 Suppose demand is a function of price,
income, and price of other goods and is given
in the following form.
Qd = Bo + B1P +B2I +B3Po
Income and prices are measured in birr.
The software output is presented in the following
table. 17
Estimation and interpretation (Cont …)
Price of
Constant Price Income other goods
(P) (Po)

Coefficient 50.78 -4.989 0.0034 -1.2801


Standard 10.2189 1.346 0.0045 0.5890
Error
T-statistics (4.97) (-3.71) (0.76) (-2.17)

No of observations = 182 R2 = 0.6837


18
Estimation and interpretation (Cont …)

The resulting demand function is,


Q = 50.78 - 4.989P + 0.0034I-1.2801P0

Constant (B0) (50.78):


It suggest the quantity demanded if
other variables are zero
It indicates the cumulative impact of
other determinants of demand
Its is significant 19
Estimation and interpretation (Cont …)

Other Coefficients:
Indicate change in quantity demanded
associated with a unit change in each
variable
i. Price Coefficient- P (-4.989P):
Its negative value is in line with the
economic theory
The negative value indicates the inverse
relationship between price and quantity
demanded
It is also significant
20
Estimation and interpretation (Cont …)

ii.Coefficient of price of other good-


P0 (-1.2801):

It indicates the relationship


between the two commodities
Its value is negative meaning that the
two commodities are complements
It is also significant
21
Estimation and interpretation (Cont …)

iii. Income Coefficient (0.0034):


Its value is positive meaning that the
commodity is a normal good
But it is not significant ????

The value of coefficient of determination


(R2):
Indicates the overall explanatory power of the
model
Indicate that about 68% of the variation in
quantity can be explained by the independent
variables 22
B. Testing the Regression Estimates
( Significance test)

Once the parameters are estimated


the strength of the relationship
between the dependent and
independent variables need to be
tested in two ways.
i. Coefficients of determination (R2):
ii. The use of T-statistics to test the
strength of the relationship between
variables
23
i. Coefficients of Determination (R2):

i. Coefficients of determination (R2)


The vale of (R2) ranges between 0 and 1
If the value of R2 = 0, it means that there
is no relationship between the independent
variables and dependent variable
If the equation explains all the variation,
the value of R2 = 1
The higher the value of R2 the equation will
fit the data well
24
Coefficients of Determination (R2):
However, how high must be the value of R2 to
be a best fit is not yet precisely defined.
Income and Consumption at least R2 = 0.95
Out put and cost the value of R2 = 0.4-0.5
might be regarded as a fit
ii. Testing the validity of explanatory
variable
The test called t-test (statistics) is used to
determine if the relationship between
dependent variable and independent variables is
significant or not
25
ii. Testing the validity of explanatory
variable

This test requires the standard deviation


(error) of each coefficient be computed
The estimated coefficient will be divided by its
respective standard error and this ratio is
called T-statistics.
Consider the above example
Constant 50.78/10.2189= (4.97)
Price -4.989/1.3458 = (-3.71)
Income 0.0034/0.0045 = (0.76)
P of other -1.2801/0.5890 = (-2.17)
26
Testing the validity of explanatory variable
(Cont…)

 These value (T-values) will be used for


hypothesis testing
The hypothesis is that,
Ho : B0, B1,B2,B3 =0
“ Coefficients are not significantly different
from zero”
 Accepting (failing to reject) the hypothesis
(Ho) means, the coefficient is not different
from zero,
 hence no relation exists between the
dependent and the independent variable. 27
Testing the validity of explanatory variable
(Cont…)

How it is tested ?
If the absolute value of the computed t-
value is equal or greater than the table
value (for n-k-1 degree of freedom) then
the hypothesis is rejected.
Meaning the coefficient is significant or
different from zero
k= number of independent variables
n = number of observation 28
Testing the validity of explanatory
variable (Cont…)

In the above example, the t-distribution


(table value) for more than 120
observation is plus or minus 1.98
Thus, two coefficients P and P0 are
significantly different from zero as a
result
 The hypothesis is rejected at only 5%
probability of error
 The coefficient of price even at only 1%
of probability of error
29
Other tests
Normality: Nnormal distribution of error
term with zero mean and constant variance
Homoscedasticity: Constant variance
Overall explanatory power of the model
Multicolliniarity: there is no strong
relationship between the chosen independent
variables.
In other words no multi-co- linearity was
detected. :
30
Example using hypothetical data
Time Series Model: Uses only the time-series
history of the variable to predict the future
value
The focus of time series analysis is to identify
the components of change in the historical
data.
Traditionally these components are,
Trend: It is a long term increase or decrease in
the variables
 The time series of human population of
Ethiopia exhibited an upward trend 31
32
33
34
Demand forecasting (Cont…)
Seasonality Component:
Represents changes that occur at regular
interval.
 A large increase in meat sales in the
month of April-May and fall in the month
of February-March would be an example
of seasonality.
Cyclical Component: refers to sustained
periodic of high values followed by low values.
 Business cycle is a good example
35
Demand forecasting (Cont…)
Random fluctuation: Variables that do not
follow any pattern
Various methods can be used to determine
trend, seasonality, and cyclical pattern
using time-series data.
However, by definition changes in the
variable due to random factor are not
predictable.
To forecast using time-series analysis it is
necessary to specify a mathematical model
that represents the generation process. 36
Demand forecasting (Cont…)
There are a number of forecasting method.
Some of them are,
Linear Trend projection method: very popular
in practice, the trend projection method
involves extrapolating the past trend on the
future.
Exponential smoothing: in this method,
forecasts are modified in the light of observed
errors.
Moving average method: in this method, the
forecast for the next period represents a
simple arithmetic average or a weighted
arithmetic average of the last few observations.
37
Demand forecasting (Cont…)

Linear Trend Model :


Using linear trend model we assume that
the sales or price increase or decrease
linearly over time
For example the analyst would like to
estimate sales for the period say, 2000-
2011

38
Demand forecasting (Cont…)
Sales Year Time
X1 2000 1
X2 2001 2
X3 2002 3
X4 2003 4
X5 2004 5
.

X11 2011 11 39
Demand forecasting (Cont…)
The straight line that best fit the data
scatter is estimated using simple
Regression Analysis
The fitted line (Solid) indicates a positive
trend in sales
The broken line shows the forecast for the
future demand say 2015
If the linear relationship is assumed
Q = a + bt
40
Demand forecasting (Cont…)
The value of b is estimated and statistically
tested
If b is +Ve and significant there is a positive
trend
If b is –Ve and significant there is a negative
trend
If b is not significant sales are constant over
time
Correcting for seasonal Effect
When there is seasonal effect (variation), we
have to correct for such variation using
dummy variable. 41
Demand forecasting (Cont…)
Suppose the sales of a particular
product is higher in the fourth
quarter than any other quarter this
may suggest a seasonal variation and
can be corrected using dummy
variables for the seasonal effect.
Dummy Variable is variable that takes a
only the value of zero or one
42

You might also like