Econometrics for Finance
5. The Multiple Regression Model
Siraj M. September, 2023
Learning Outcomes
Understand the extension of simple regression to multiple regression, with
more than one explanatory variable.
Interpret the effect of each explanatory variable on the dependent variable.
Understand and explain the meanings of the assumptions for the multiple
regression model.
Find interval estimates for single coefficients and linear combinations of
coefficients, and interpret the interval estimates.
2
Introduction
The model in Lectures 1 - 4 is called a simple regression model because the
dependent variable y is related to only one explanatory variable x.
Although this model is useful for a range of situations, in most practical cases
there are two or more explanatory variables that influence the dependent
variable y.
When we turn a model with more than one explanatory variable into its
corresponding econometric model, we refer to it as a multiple regression model.
3
Introduction: Notation
The regression relationship now becomes:
𝑌𝑖 = 𝛽1 + 𝛽2 𝑋2𝑖 + 𝛽3 𝑋3𝑖 + 𝜇𝑖
Y is the dependent variable, 𝑋2 and 𝑋3 the explanatory variables, 𝜇 the
stochastic disturbance term, and i the ith observation.
𝛽1 is the intercept term.
The coefficients 𝛽2 and 𝛽3 are called the partial regression coefficients, and their
meaning will be explained shortly.
4
Illustration
We will set up a model for a Juice Company that we call 3S juice.
To assess the effect of different price structures and different
levels of advertising expenditure, 3S juice sets different prices,
and spends varying amounts on advertising, in different regions.
Of particular interest to management is how sales revenue
changes as the level of advertising expenditure changes.
5
Illustration
Essential questions:
Does an increase in advertising expenditure lead to an increase
in sales?
If so, is the increase in sales sufficient to justify the increased
advertising expenditure?
Will reducing prices lead to an increase or decrease in sales
revenue?
This economic information is essential for effective
management.
6
Illustration
The econometric model is:
𝑆𝐴𝐿𝐸𝑆 = 𝐸 𝑆𝐴𝐿𝐸𝑆 + 𝜇𝑖 = 𝛽1 + 𝛽2 𝑃𝑅𝐼𝐶𝐸 + 𝛽3 𝐴𝐷𝑉𝐸𝑅𝑇 + 𝜇𝑖
SALES represents monthly sales revenue
PRICE represents price
ADVERT is monthly advertising expenditure, and
𝜇 is stochastic disturbance term
7
Illustration
Interpretation of parameters:
Mathematically, the intercept parameter 𝛽1 is the value of the
dependent variable when each of the independent variables takes
the value zero.
The other parameters in the model measure the change in the
value of the dependent variable given a unit change in an
explanatory variable, all other variables held constant.
8
Illustration
Interpretation of parameters:
𝛽2 = the change in monthly SALES when price PRICE is increased by
one unit (ETB1) and advertising expenditure ADVERT is held constant.
∆𝑆𝐴𝐿𝐸𝑆 𝜕𝑆𝐴𝐿𝐸𝑆
=
∆𝑃𝑅𝐼𝐶𝐸(𝐴𝐷𝑉𝐸𝑅𝑇 ℎ𝑒𝑙𝑑 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡) 𝜕𝑃𝑅𝐼𝐶𝐸
9
Illustration
Interpretation of parameters:
𝛽3 = the change in monthly SALES when advertising expenditure
ADVERT is increased by one unit (ETB1) and price PRICE is held
constant.
∆𝑆𝐴𝐿𝐸𝑆 𝜕𝑆𝐴𝐿𝐸𝑆
=
∆𝐴𝐷𝑉𝐸𝑅𝑇 𝑃𝑅𝐼𝐶𝐸 ℎ𝑒𝑙𝑑 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 𝜕𝐴𝐷𝑉𝐸𝑅𝑇
10
Illustration: OLS Estimators
To find the OLS estimators, let us first write the sample
regression function (SRF) as follow:
𝑌𝑖 = 𝛽1 + 𝛽2 𝑋2𝑖 + 𝛽3 𝑋3𝑖 + 𝜇𝑖
𝜇𝑖 is the residual term, the sample counterpart of the stochastic
disturbance term ui.
11
Illustration: OLS Estimators
To find the OLS As noted in previous lectures, the OLS
procedure consists of choosing the values of the unknown
parameters so that the residual sum of squares (RSS) 𝜇2 𝑖 is as
small as possible. Symbolically,
2
𝑚𝑖𝑛 𝜇2 𝑖 = 𝑌𝑖 − 𝛽1 − 𝛽2 𝑋2𝑖 − 𝛽3 𝑋3𝑖
12
Illustration: OLS Estimators
Finally after certain derivation, one can find the following formulas.
13
Illustration: Estimating the parameters
This table contains the least squares results for the sales equation for
the 3S Juice Company.
14
Illustration: Interpretation
What can we say about the coefficient estimates in the above table?
Coefficient on PRICE
The negative coefficient on PRICE suggests that demand is price
elastic; we estimate that, with advertising held constant, an increase in
price of ETB1 will lead to a fall in monthly revenue of ETB7,908.
Or, expressed differently, a reduction in price of ETB1 will lead to an
increase in revenue of ETB7,908.
15
Illustration: Interpretation
Coefficient on ADVERT
The coefficient on advertising is positive; we estimate that with price
held constant, an increase in advertising expenditure of ETB1,000 will
lead to an increase in sales revenue of ETB1,863.
We can use this information, along with the costs of producing the
additional juice, to determine whether an increase in advertising
expenditures will increase profit.
16
Illustration: Interpretation
Coefficient on intercept
The estimated intercept implies that if both price and advertising
expenditure were zero the sales revenue would be ETB118,914.
Clearly, this outcome is not possible; a zero price implies zero
sales revenue.
17
Illustration: Interpretation
Prediction
Suppose 3S is interested in predicting sales revenue for a price of ETB5.50
and an advertising expenditure of ETB1,200. This prediction is:
The predicted value of sales revenue for PRICE = 5.5 and ADVERT =
1.2 is ETB77,656.
18
Illustration: Interpretation
Estimation of the error variance 𝝈𝟐
There is one remaining parameter to estimate—the variance of the error term.
𝑁 2
𝑖=1 𝑒𝑖
𝜎2 =
𝑁−𝐾
𝑁 2
𝑖=1 𝑒
1718.943
𝑖
𝜎2 = = = 23.874
𝑁−𝐾 75 − 3
𝜎= 𝜎 2 = 23.874 = 4.8861
Sometimes 𝜎 is referred to as the standard error of the regression. Sometimes it is
called the root mse (short for mean squared error). 19
Illustration: Interval Estimation
Interval estimation for a single coefficient
Suppose we are interested in finding a 95% interval estimate for 𝛽2 , the
response of sales revenue to a change in price at 3S Juice Company.
Following the procedures described in lecture 4, and noting that we
have N-K = 75-3=72 degrees of freedom, the first step is to find a
value from the t(72)-distribution, call it tc, and use the formula:
𝛽2 − 𝑡𝑐 ∗ 𝑠𝑒 𝛽2 , 𝛽2 + 𝑡𝑐 ∗ 𝑠𝑒 𝛽2
20
Illustration: Interval Estimation
Interval estimation for a single coefficient
A 95% interval estimate for 𝛽2 based on our particular sample is obtained by
replacing 𝛽2 , se(𝛽2 ) and 𝑡𝑐 by their values 𝛽2 = -7.908, se(𝛽2 ) = 1.096, and
𝑡𝑐 =1.9335. Thus, our 95% interval estimate for 𝛽2 is given by:
−7.908 − 1.9335 ∗ 1.096,7.908 + 1.9335 ∗ 1.096 = (10.093, −5.723)
This interval estimate suggests that decreasing price by ETB1 will lead to an
increase in revenue somewhere between ETB5,723 and ETB10,093.
21
Illustration: Hypothesis Testing
Testing for Elastic Demand
With respect to demand elasticity, we wish to know whether:
𝛽2 ≥ 0: a decrease in price leads to a change in sales revenue
that is zero or negative (demand is price-inelastic).
𝛽2 < 0 : a decrease in price leads to an increase in sales revenue
(demand is price elastic).
22
Illustration: Hypothesis Testing
Testing for Elastic Demand
Using t test, since -7.215 < -1.666, we reject 𝐻0 : 𝛽2 ≥ 0 and
conclude that 𝐻1 : 𝛽2 < 0 (demand is elastic) is more compatible
with the data.
The sample evidence supports the proposition that a reduction in price
will bring about an increase in sales revenue.
Since 0.000 < 0.05, the same conclusion is reached using the p-value.
23
Illustration: Hypothesis Testing
Testing Advertising Effectiveness
The other hypothesis of interest is whether an increase in
advertising expenditure will bring an increase in sales revenue
that is sufficient to cover the increased cost of advertising.
Since such an increase will be achieved if 𝛽3 > 1 , we set up the
hypotheses:
24
Illustration: Hypothesis Testing
Testing Advertising Effectiveness
𝐻0 : 𝛽2 ≤ 1 and 𝐻1 : 𝛽2 > 1
Using t test, since 1.263 < 1.666, we do not reject 𝐻0 .
There is insufficient evidence in our sample to conclude that
advertising will be cost effective.
Using the p-value to perform the test, we again conclude that 𝐻0
cannot be rejected, because 0.105 > 0.05.
25
Interaction Variables
What if we wanted the marginal effect of one variable to depend on
the level of another variable?
How do we model this effect?
To illustrate this idea we will consider a life-cycle model for pizza
consumption.
Suppose that we are economists for 3S Pizza, and that we wish to study
the effect of income and age on an individual’s expenditure on pizza.
26
Interaction Variables
For that purpose we take a random sample of 40 individuals, age
18 and older, and record their annual expenditure on pizza
(PIZZA), their income in thousands of dollars (INCOME) and
age (AGE).
To account for interaction between AGE and INCOME e, we
model:
𝑃𝐼𝑍𝑍𝐴 = 𝛽1 + 𝛽2 𝐴𝐺𝐸 + 𝛽3 𝐼𝑁𝐶𝑂𝑀𝐸 + 𝛽4 (𝐴𝐺𝐸 + 𝐼𝑁𝐶𝑂𝑀𝐸) + 𝑒
27
Interaction Variables
In this model, the effects of INCOME and AGE are:
𝜕𝐸 𝑃𝐼𝑍𝑍𝐴
= 𝛽2 + 𝛽4 𝐼𝑁𝐶𝑂𝑀𝐸: The effect of AGE now depends
𝜕𝐴𝐺𝐸
on income.
𝜕𝐸 𝑃𝐼𝑍𝑍𝐴
= 𝛽3 + 𝛽4 𝐴𝐺𝐸:The effect of a change in income on
𝜕𝐼𝑁𝐶𝑂𝑀𝐸
expected pizza expenditure, which is the marginal propensity to
spend on pizza, now depends on AGE.
28
@@@@@@@@@@@@@@@@@@@@@ end of lecture 5 @@@@@@@@@@@@@@@@@@@@
29
