Demand Estimation

Chapter 5
 Demand Estimation
 Managers need to know the quantitative
relationships based on real-life information
about demand and factors that affect it.
 Methods for obtaining information are:
 Consumer surveys

 Focus groups

 Electronic data collection

 Regression Analysis
 Method of estimating the quantitative relationship
among variables.
 Steps of regression analysis are:
1. Specify the regression model
2. Obtain data on variables
3. Estimate the quantitative effect of each
independent variable on the dependent variable
4. Test the statistical significance
5. Interpret the results for business decisions
 The Simple Regression Model
 Y = a + bX + u
where Y = dependent variable
X = independent variable
a = intercept
b = slope
u = random factor (error term)
 Regression analysis seeks the best
linear relationship between Y and X.
 The Simple Regression Model
 In model specification, it is possible that we may
not be aware of some independent variables that
have an impact on the dependent variable under
analysis.
 The variable “u” is included in the regression
model to account for the effects of other
independent variables that may have been left out
of the model.
 The error term is used to test the statistical
significance of the regression estimation.
 Data Used in Regression Analysis
 Two types of data:
1. Cross-sectional
Information on a variable at a given point in time
obtained from a cross-section of observations from a
group (e.g. city, university, country, gender, age)
2. Time Series
Information on one variable over time (e.g. annual
income over 10 years, weekly electricity consumption
over one year)
 Estimating the Regression
Equation
 The equation Y = a + bX is estimated to find the
best linear relationship between Y and X.
 a and b are the coefficient estimates (parameters)
of the regression equation.
 The coefficients are estimated using the method of
ordinary least squares (OLS).
 OLS entails finding the linear relationship such that
the sum of the squared distances between the
regression line and the actual data points is
minimized.
 OLS Estimation
Y

• • • •
• •
• • • • •
• • •
• • •

X
OLS estimation tries to fit a line through the scatter of
points such that the squared distance between the
line and the points is minimized.
 OLS Estimation
 When the regression equation is estimated by
OLS using sample data taken from a population,
the regression coefficients and the dependent
variable become estimates.
 Estimates are shown with a circumflex placed
over their symbols:

ˆY  aˆ  bˆ X
 OLS Estimation
 OLS estimation’s objective is to explain as
much of the variation in Y as possible by
using the information contained in the
(X,Y) pairs collected as data.
 For evaluating the explanatory power of
the regression equation, the actual values
of the dependent variable (Y) are
compared against the estimated values of
the dependent variable ( Ŷ).
 Explanatory Power
 OLS estimation tries to fit a line through
the scatter of points by using the
information about the sample means of Y
and X.
 OLS estimation tries to explain by how
much a given Y value deviates from the
sample mean of Y.
Explanatory Power
A Yˆ  aˆ  bˆ X
Unexplained •
deviation B
Total
•
deviation
• • •
•
Y •
• •ExplainedC
•
• deviation
• •

X
 Explanatory Power

AC  Yi - Y  total deviation of the ith observatio n from the sample mean Y

ˆ - Y  explained deviation of estimated Y from the sample mean Y
BC  Y i i

ˆ  unexplaine d deviation of the ith observatio n from the estimated Y

AB  Yi - Yi i
 Explanatory Power

TSS   (Yi  Y)2  Total sum of squares

RSS   (Yˆ i  Y)2  Regression sum of squares
ESS   (Yi  Yˆ i )2  Error sum of squares

RSS ESS
R 2
 1
TSS TSS
 Explanatory Power
 R2 is the coefficient of determination.
 R2 should be as high as possible.
 If R2 is 100%, this means that the total
deviation of Y from its sample mean is
explained perfectly by the regression
equation.
 The Regression Coefficients
 In most cases, a sample of data from
the population is used.
 The researcher must assess the degree to
which the results of the sample reflect the
population.
 The researcher needs to make
inferences about the population based
on what is known about the sample.
 Testing the Statistical
Significance
 Statistical significance of the regression
coefficients (bi) is tested by the t-tests.
 Hypotheses are formed in order to conduct
the test.
 In economics,
Null Hypothesis (H0) states that there is no
relationship between Y and X and thus the
coefficient is equal to zero
H0 : b = 0
Testing the Statistical
Significance

Alternative Hypothesis (HA) states

that there is a relationship between Y
and X and thus the coefficient is not
equal to zero
HA : b  0
 Conducting the t-Test
 The t-test is conducted by comparing the t-value
calculated from the regression estimation against
the t-critical value obtained from the Student’s t
Table.
 Calculated t:

ˆb  E(bˆ )
t
SEbˆ
 Conducting the t-Test
 The formula compares the estimated value
of the coefficient against the hypothesized
value of the coefficient and divides by the
standard error of estimation for that
coefficient.
 Test:
Reject H0 if |t-calculated| > t-critical
Fail to reject H0 if |t-calculated| < t-critical
 Conducting the t-Test
 Reading the Student’s t Table:
1. Determine the degrees of freedom:
d.f. = n - k - 1
where n = number of observations
k = number of parameters
2. Determine the level of significance ():
E.g. for a 95% confidence level,  = 0.05
3. Read from Column /2 or  and Row d.f.
 Forming the Hypotheses
 The null and alternative hypotheses can
be formed in three different ways:
Two-Tailed Test:
H0 : b = 0
HA : b  0
One-Tailed Test:
H0 : b  0 OR H0 : b  0
HA : b < 0 HA : b > 0
 Conducting the t-Test
 The t-critical value for a one-tailed test will
be read as:
Column  and Row d.f.

 The test will be conducted as before:

Reject H0 if |t-calculated| > t-critical
Fail to reject H0 if |t-calculated| < t-critical
 The Multiple Regression Model

 Y = a + b1X1 + b2X2 + b3X3 + b4X4

where Y = dependent variable
Xi = independent variables
a = intercept
bi = slope coefficients for each
independent variable
 Testing the Model Significance
 For multiple regression models, we need
to test the statistical significance of the
entire regression equation as well as each
individual coefficient.
 The F-Test is used to test the significance
of the model:
Explained Variation / k
F
Unexplaine d Variation / (n - k - 1)
 The F-Test
Null Hypothesis (H0) for the F-test states
that there is no relationship between Y and the
Xis and thus all the coefficients are equal to zero
H0 : bi = 0 (or, b1 = b2 = b3 = b4 = 0)
Alternative Hypothesis (HA) for the F-test
states that at least one of the coefficients is not
equal to zero and that there is a relationship
between Y and all Xis whose coefficients are
different from zero
HA : at least one bi  0
Conducting the F-Test
Reading the F-Distribution Table:
1. Determine the degrees of freedom:
d.f. for numerator = k
d.f. for denominator = n - k - 1
where n = number of observations
k = number of parameters
2. Determine the level of significance ():
E.g. for a 95% confidence level,  = 0.05
Table B.3a:  = 0.05, Table B.3b:  = 0.01
3. Read from Column k and Row n - k – 1
4. Reject H0 if F-calculated > F-critical
Fail to reject H0 if F-calculated < F-critical
Using Regression Analysis for
Forecasting
 Once the coefficients are estimated, we can
assign values to the independent variable(s)
and find a forecast value for the dependent
variable.
 The forecasted Y value will have some error
in it and thus we need to assign some level of
confidence to the forecasted value that we
produce:

ˆY  (t
n  k 1 )(SEE)
 Using Regression Analysis for
Forecasting
 In this formula,
tn-k-1 is t-critical chosen for building a
confidence interval around the
forecasted Y value, and,
SEE is the standard error of estimation
which measures the sampling error (the
error associated with not working with
the population).
 Example 1

 Estimate the multiple regression model in which

Y = consumption of soft drinks per capita in each
city
X1 = per capita income in each city (in
hundreds of TL)
X2 = mean annual temperature in each city (in C)
Multiple R 0.695820689
R Square 0.484166432
Adjusted R Square 0.461240495
Standard Error 49.44768027
Observations 48

ANOVA
df SS MS F Significance F
Regression 2 103273.6279 51636.81 21.1187 3.4E-07
Residual 45 110028.2888 2445.073
Total 47 213301.9167

Coefficients Standard Error t Stat P-value Lower 95% Upper 95%

Intercept -56.61441484 63.11655018 -0.897 0.374502 -183.738 70.50884
X Variable 1 -2.054389677 1.815498449 -1.1316 0.263804 -5.71099 1.602212
X Variable 2 4.695033589 0.823816926 5.69912 8.71E-07 3.035781 6.354286
 Questions
1. Write out the estimated regression model.

2. Comment on the explanatory power of the

regression model.

3. Test the significance of each coefficient

estimate by writing out the individual
hypotheses (t-critical = 2.010).
 Questions
4. Test the significance of the entire
regression model by writing out the
hypothesis (F-critical = 2.81).

5. When X1 = 17 and X2 = 30, what is the

95% confidence interval for the
forecasted value of Y? (t-critical =
2.010)
 Example 2
 For Electronic Data Processing, Inc.
(EDS), the following gives the
regression equation estimated for the
relationship between units demanded
(Y) and price (X1), advertising
expenditures (X2), and personal selling
expenditures (X3).
Yt = -117.513 - 0.296 Pt + 0.036 ADt +
0.066 PSEt

t (Pt) = -2.91
t (ADt) = 2.56
t(PSEt) = 4.61

SEE = 123.9

R2 = 95.8%

F = 85.4
 Evaluate the statistical significance of
the coefficients:
t-critical (α = 0.05, d.f. = 8) = 2.306

t (Pt) = -|2.91| > 2.306  significant

t (ADt) = 2.56 > 2.306  significant
t(PSEt) = 4.61 > 2.306  significant

It looks like all three independent

variables have a statistically significant
impact on units demanded.
 Calculate the price elasticity of demand
when
Pt = 3,200; ADt = 28,200; PSEt =
40,000

Y = -117.513 - 0.296 (3,200) + 0.036

(28,200) + 0.066 (40,000)
= 2590.487 units

EP = (dQ / dP) (P / Q)
= (-0.296) (3,200 / 2590.487)
= -0.37
 Calculate the advertising elasticity of
demand when
Pt = 3,200; ADt = 28,200; PSEt =
40,000

Y = -117.513 - 0.296 (3,200) + 0.036

(28,200) + 0.066 (40,000)
= 2590.487 units

EAD = (dQ / dAD) (AD / Q)

= (0.036) (28,200 / 2590.487)
= 0.39
 Construct a 95% confidence interval for the
estimated value of Y when
Pt = 3,200; ADt = 28,200; PSEt = 40,000

Y = -117.513 - 0.296 (3,200) + 0.036

(28,200) + 0.066 (40,000)
= 2590.487 units

C.I.95%
ˆ
=  (t n k 1 )(SEE)
Y
lower limit = 2590.487 - (2.306*123.9)
upper limit = 2590.487 + (2.306*123.9)
C.I.95% = (2304.77 , 2876.20)