Ambo University Woliso Campus

School of Business and Economics

Department of Economics
Econometrics –II
Course code: Econ 2062
By: Tesfaye Etensa (Ass.Prof.)
Email:
[email protected]/tesfayeeten
[email protected]
Phone No:+251917054131
@ All rights are reserved
 Ambo University Woliso Campus
School of Business and Economics
Economics Department
Econometrics –II (Econ 2062)
 (This ppt was prepared from econometrics books of “Tesfaye
Etensa (2017): “Introduction to Econometrics: Theory and
Practice with Stata”
Prepared by:
Tesfaye Etensa
E-mail: [email protected]
[email protected]

@ All rights are reserved

May, 2020
Woliso, Ethiopia
 CHAPTER ONE

Regression with Qualitative

Information
Dummy Variables Regression
 Outline of the Chapter
1.1 Introduction
1.2 Dummy as independent variables
1.2.1 Cautions in the use of dummy variables
1.2.2 Regression with qualitative variables: Analysis of
Covariance (ANOVA)
1.2.3 Regression with a mixture of qualitative and
quantitative variables: Analysis of Covariance
(ANCOVA)
1.2.4 Interactions between qualitative variables
1.2.5 Interactions between qualitative & quantitative variable
(Slope indicators variables)
1.3 Structural stability
1.4 Dummy as dependent variable: Qualitative Response Model
1.4.1 Linear probability Model
1.4.2 Logit model
1.4.3 Probit model
1.4.4Tobit model
1.1. “Introduction”
 1.1 Introduction
 In regression analysis the dependent variable is
not only frequently affected by quantitative (ratio
scale variables like price, income, output, etc) but
also qualitative variables (nominal scale variables
like sex, race, nationality, etc)
 Such variables should be included in the model
 Dummy variables are commonly used as proxies
for qualitative factors such as sex, religion, etc
 Dummy variable is synonymously with non
measurable, qualitative in nature, nominal scale,
non numeric variable.
 1.1 Introduction
Qualitative information?

 It is a non measurable information that we obtain

or gather for a given variable.
 It is an indicator variable that is non measurable
or non quantify in nature.
 Indicator variable, binary variable, categorical
and dichotomous variable are use
interchangeable.
 1.1 Introduction
 Not all information can easily be quantified.
 So, need to incorporate qualitative information.
Example: 1. Effect of belonging to a certain group:
 Gender, location, marital status, occupation
 Beneficiary of a program/policy
2. Ordinal variables:
 Answers to yes/no (or scaled) questions...
 Effect of some quantitative variable may differ
between groups/categories:
 Returns to education may differ between
sexes or between ethnic groups …
 1.1 Introduction
Example 2: Suppose the firm utilized two types of
production process to obtain its output.
Yi = α + β D + u i
Where Y is output obtained
D is the dummy variable
0 if the output is obtained from machine A
Di = 
1 if the output is obtained from machine B
Example 3: Does sex makes any difference in a college
teachers salary, assuming that all other variables such
as age, education level, and experience etc are held
constant.
Yi = β 0 + β 1D + u i
 1.1 Introduction
Interest in determinants of belonging to a group
 Determinants of being poor …
Dummy dependent variable (logit, probit…)
Dummy Variable: a variable devised to use
qualitative information in regression analysis.
 A dummy variable takes 2 values: usually 0/1.
e.g. Yi=β0+β1*D+u; for ∀i ϵ group 1, and
for ∀i ∉ group 1.
If D = 0, E(Y) = E(Y|D = 0) = β0
If D = 1, E(Y) = E(Y|D = 1) = β0 + β1
Thus, the difference between the two groups (in
mean values of Y) is: E(Y|D=1) – E(Y|D=0) = β1.
The significance of this difference is tested by a t-
test of β1 = 0.
 Lab session

Use “lecture_1.xls” data to

practice what we learnt in
previous sections
1.2 “Dummy as
Independent
Variables”
 1.2 Dummy as Independent Variables
I. How we include dummy variable as explanatory
variable?
 Constructing artificial variable, which take on
values of 1 or 0,
0 indicating the absence of the attributes
1 indicating the presence of that attributes
 1.2 Dummy as Independent Variables

II. ANOVA & ANCOVA?

 ANOVA: A regression model which contains
regressors (explanatory variables) that are all
exclusively dummy variables.
 ANCOVA: Regression model which contains
quantitatively explanatory variables in addition to
dummy variables.
 A regression model which contains the mixed of both qualitative
and quantitative variable.
 1.2 Dummy as Independent Variables

III. Purpose of dummy variable

It allows for difference in intercept
It allow for difference in slopes
It help us to estimate equations with cross equation
restrictions
Test for the stability of regression coefficients
 1.2.1 Dummy variable Trap
 It is the Cautions in the use of dummy variables,
what we called it as “Dummy variable Trap”
 We should include (introduce) j-1 where j is
number of variables
 The general rule is that” If a qualitative variable
has m categories, introduce only (m-1) dummy
variables.”
 Example: Sex has two categories and hence we
introduced only a single dummy variable.
 If this is not fulfilled, we faced the problem of
perfect multicollinearity (perfect collinearity),
which is called “dummy variable trap.”
 1.2.1 Dummy variable Trap
The category for which no dummy variable is
assigned is known as the base, bench mark, control,
comparison, reference, or omitted category.
Hence, all comparisons are made in relation to the
bench mark category (we assigned 0 values)
If there is a constant term in the regression
equation the number of dummies defined should
always be one less than the number of groupings by
that category.
 b/c the constant term is the intercept for the base
group and the coefficients of the dummy variables
measures differences in intercept (the mean
difference)
 1.2.1 Dummy variable Trap
 If the coefficients β attached to the dummy
i

variables D, are called differential intercept

coefficients.
Reasons: It tell by how much the value of the intercept
of the category that receives that value of 1 differs from
the intercept coefficients of the base category.
 The intercept value ( α ) represents the mean value
of the benchmark category.
 If we don’t have constant term, we can’t used j-1
or m-1 because we don’t have dummy variable
trap.
 1.2.2 ANOVA Analysis

It is regression with qualitative variables.

A. A single dummy independent variable
 The model specification will be:
Yi = α + β D + u i β It doesn’t measure the slope
Yi = β 0 + β 1D + ε i rather it measures the mean
difference b/n the category
1 if male
Di = 
0 if female (otherwise )
Y is annual salary of a college teacher
α It measure the mean value
of the bench category
 1.2.2 ANOVA Analysis

Example: Suppose a researcher wants to find

out whether sex makes any difference in a
college teachers’ salary, assuming other
variables, education level, experiences, and age
being constant.
Assuming the disturbance term satisfy the
usual assumptions, of the classical linear
regression model, we obtain:
 1.2.2 ANOVA Analysis
Interpretation
 The mean (average) annual salary of female college
teacher: E(Yi/Di=0)= β 0
 The mean (average) annual salary of male college
teacher: E(Yi/Di=1)= β 0 + β 1
 The slope coefficients β 1 “tells by how much the
mean salary of a male college teacher differs from
the mean salary of his female counter part or
 the value of β 1 is on average measures the
difference between intercept.”
 β 0 + β 1 reflecting the mean salary of the male
college professors or the average (mean) salary of
non base group (actual value).
 1.2.2 ANOVA Analysis
The implication of slope coefficients ( β 1 ) :
The coefficient determines whether there is
discriminating on against female.
 If then, for the same level of other factors, women
earn less than men on average.
A test of the null hypothesis that there is no sex
discrimination (H0: β 1 = 0 ) can be easily made by
running regression by OLS.
Suppose we have the following regression result:
 1.2.2 ANOVA Analysis
Example
Observation Salary (Y) Sex (D)
1 27000 1
2 17500 0
3 42500 1
4 29000 1
5 23000 0
6 32000 1
7 18500 1
8 22000 0
9 24000 1
10 18500 0
 1.2.2 ANOVA Analysis
Brainstorm:
Based on the above table:
1. Estimate the coefficients of the variables
2. Square of correlation coefficient
3. Standard error and t-statistics
4. TSS, ESS and RSS
5. What makes this regression different from
simple regression you have learnt under
Econometrics-I?
6. Interpret the result
 1.2.2 ANOVA Analysis
Regression Result of above example

Yˆi = 19125 + 9708 . 33 D i

se (4463.514) (3457.423)
t (5.53) (2.18)
 1.2.2 ANOVA Analysis
Interpretation :
The estimated mean (average) salary of female
college teacher is birr 19,125(= βˆ ).
0

The mean salary of male college teachers is birr (

(βˆ + βˆ ) =28,833.
0 1

β =9708 is the mean difference between male and

1

female college teachers.

Since β is statistically significant, the results
1

indicate that the mean salary of two categories

are different, actually, the female teacher’s
average salary is lower than that of her
counterpart.
 1.2.2 ANOVA Analysis
 Interpretation:
If all other variables are held constant, there is sex
discrimination in the salaries of the two sexes or the
salary of the female is less than male by 9708, on
average.
B. A multiple dummy independent variable
 It is when more than two distinct values are involved.
 Always when there are N variables we develop N-1
dummy variables.
 Let, a given output is produced using three methods of
production says: Machine A, Machine B , and Machine
C. Yi = β 0 + β 1D1 + β 2 D 2 + u i
 1.2.2 ANOVA Analysis

1 if output is obtained from machine A

D1 = 
0 otherwise (if it comes from B & C)
1 if output is obtained from machine B
D2 = 
0 otherwise (if it comes from A & C)
Interpretation
 β represents the mean value of output obtained
0

from machine C.
 β is the mean difference in output associated
1

with a change from machine C to machine A.

 1.2.2 ANOVA Analysis
(β + β ) it is the mean value of output obtained from
0 1

machine A.
 β is the mean difference in output associated with
2

a change from machine C to machine B.

 (β + β ) is the mean value of output obtained from
0 2

machine B.
Exercise: Interpret the following model

Yi = β 0 + β 1 Gender + β 2 Instituion + ui

1 male 1 government
Gender =  ; Institutio ns = 
0 female 0 private
 1.2.2 ANOVA Analysis
Example2.: Wage differential between male and female
Two possible ways: a male or a female dummy.
1. Define a male dummy (male = 1 & female = 0).
 reg wage male
 Result: Yi = 9.45 + 172.84*D + ûi
p-value: (0.000) (0.000)
Interpretation the monthly wage of a male worker is, on
average, $172.84 higher than that of a female worker.
This difference is significant at 1% level.
2. Define a female dummy (female = 1 & male = 0)
 reg wage female
 Result: Yi = 182.29 – 172.84*D + ûi
p-value: (0.000) (0.000) Interpretation ??
 1.2.3 Analysis of Covariance (ANCOVA)
Unlike ANOVA, a regression model may contain
regressors that are all exclusively dummy, or
qualitative, in nature; ANCOVA, is regression with a
mixture of qualitative and quantitative
independent variables.
It is regression on both qualitative and quantitative
independent variables.
A. Single dummy independent variable
 Example: Suppose we identified two variables that affect the salary of a given
employee.
Wage = β 0 + β 1 gend + β 2 educ + u i
 1 female Gender & level of
gend = 
 0 male education is the only
wage = wage rate of individual observed factors
educ= level of education
affect wage.
 1.2.3 Analysis of Covariance (ANCOVA)
Interpretation
 β2
measures the slope.
β1
 is the difference in hourly wage between females
and males, given the amount of education. Hence,
the coefficient determines whether there is
discrimination against women.
If β < 01 , then for the same level of education,
women earns less than men, on average.
If we assume the zero conditional mean
assumptions E(U)=0, then:
β 1 = E ( wage gend = 1, educ ) − E ( wage gend = 0, educ )
Key: the level of education is the same in both
individuals; the difference, β is due to gender only.
1
 1.2.3 Analysis of Covariance (ANCOVA)
Interpretation
The intercept for male is β 0

The intercept for female is β0 + β1

Since there are just two groups, we only need two
different intercepts. This means that, in addition to β0,
we need to use only one dummy variable; we have
chosen to include the dummy variable for females.
Using two dummy variables would introduce perfect
collinearity because female + male = 1 , which means that
male is a perfect linear function of female.
 Including dummy variables for both genders is the
simplest example of the so-called dummy variable trap,
 1.2.3 Analysis of Covariance (ANCOVA)
Interpretation:
 1.2.3 Analysis of Covariance (ANCOVA)

Interpretation
Mean salary of female college professor:
E (Y / D = 1, educ ) = ( β 0 + β 1 ) + β 2 educ

Mean salary of male college professor:

E (Y / D = 0, educ ) = β 0 + β 2 educ
After we run OLS regression, if the t test shows
that it is statistically significant, we reject the null
hypothesis that the male and female college
professors’ levels of mean annual salary are the
same, and we accept the alternative hypothesis.
 1.2.3 Analysis of Covariance (ANCOVA)
Features of the dummy variable regression model
Introducing dummy variable
The general rule: If a qualitative variable has ‘m’ categories,
introduce only ‘m-1’ dummy variables. In our example, sex
has two categories, and hence we introduced only a single
dummy variable. If this rule is not followed, we shall fall into
what might be called the dummy variable trap, that is, the
situation of perfect multicollinearity.
The assignment of 1 and 0 values to two
categories, such as male and female, is arbitrary.
 1.2.3 Analysis of Covariance (ANCOVA)
Features of the dummy variable regression model
The group, category, or classification that is assigned
the value of 0 is often referred to as the base,
benchmark, control, comparison, reference, or omitted
category.
It is the base in the sense that comparisons are made with
that category.
The coefficient attached to the dummy variable D can
be called the differential intercept coefficient.
b/c it tells by how much the value of the intercept term of the
category that receives the value of 1 differs from the intercept
coefficient of the base category.
 1.2.3 Analysis of Covariance (ANCOVA)
B. Dummy Variables for Multiple Categories
Example: Suppose that, on the basis of the cross-
sectional data, we want to regress the annual
expenditure on health care by an individual on
the income and education of the individual.
Since the variable education is qualitative in
nature, suppose we consider three mutually
exclusive levels of education: less than high
school, high school, and college.
Now, unlike the previous case, we have more
than two categories of the qualitative variable
education.
1.2.3 Analysis of Covariance (ANCOVA)
Thus, following the rule that the number of
dummies be one less than the number of
categories of the variable, we should introduce
two dummies to take care of the three levels of
education.
Assuming that the three educational groups
have a common slope but different intercepts in
the regression of annual expenditure on health
care on annual income, we can use the following
model: Yi = β 0 + β1D1i + β 2 D2i + β 3 X i + ui
Where Yi= annual expenditure on heath care
Xi= annual income
 1.2.3 Analysis of Covariance (ANCOVA)
1 if high school education 1 if colleg e educatio n
D =  D2 = 
 1
0 otherwise
; 0 otherwise
We arbitrarily treating the “less than high school
education” category as the base category.
 Therefore, the intercept β 0 will reflect the intercept
for this category.
The differential intercepts β 1 and β 2 tell by how much
the intercepts of the other two categories differ from
the intercept of the base category, which can be readily
checked as follows:
 Assuming , we obtain from the above specification
 1.2.3 Analysis of Covariance (ANCOVA)
E (Yi | D1 = 0, D 2 = 0, X i ) = β 0 + β 3 X i
the mean health care expenditure functions for less
than high school
E(Yi | D1 =1, D2 = 0, Xi ) = (β0 + β1) + β3 Xi
the mean health care expenditure functions for the
high school
E(Yi | D1 = 0, D2 = 1, Xi ) = (β0 + β2 ) + β3 X i
the mean health care expenditure functions for the
college.
1.2.3 Analysis of Covariance (ANCOVA)

β 2 > β 1
 1.2.3 Analysis of Covariance (ANCOVA)
If a qualitative variable has more than one category,
the choice of the bench mark category is strictly up to
the researcher.
There is a way suppressed this trap by introducing as
many dummy variable as the number of categorical of
that variable provide we do not introduce the intercept
(constant term) in such a model.

Yi = β 1 D1 + β 2 D 2 + β 3 X + u i
 When we run regression, we use the non intercept option in your
regression packages (suppressed intercept)
 1.2.3 Analysis of Covariance (ANCOVA)
i. Dummy Variables for Multiple Categories :No Intercept Case
If there is no intercept, we have no comparison,
base group and we did not omitted one category.
Yˆi = 13 ,124 D 1 + 12 , 244 D 2 + 10 , 453 D 3
se (546) (425) (462)
t (13) (14) (12)
R 2
= 0 . 2546
Where Y is salary of teachers
 1 west  1 north
D1 =  ;D 2 = 
 0 otherwise  0 otherwise
 1 south
D 3 = 
 0 otherwise
 1.2.3 Analysis of Covariance (ANCOVA)
Interpretation
 B1= the mean salary of teachers in west=13124
 B2= the mean salary of teachers in north=12244
 B3= the mean salary of teachers in south=10453
ii. Dummy Variables for Multiple Categories :Case
when constant term is present
 Redo the above example, now assume we take west
as base category.

Yˆi = 13 , 456 . 63 − 845 . 23 D 2 − 1245 . 14 D 3

se (234.56) (125.98) (262.45)
t (21.26 ) (14.84) (13.68)
 1.2.3 Analysis of Covariance (ANCOVA)
Interpretation
 The mean salary of teachers in the west is about
13,456.63.
 The mean salary of teachers in the north is lower by
845.23 and that is the teachers in the south is lower
than 1245.15; the mean salary of teachers in the north
13,456.63- 845.23=12611.4
 -845.23 tell us that the mean salary of teachers in the
North is smaller by about 845.23 than the mean salary
of about 13456 for the bench mark category, west.

 N.B: Model with intercept is more appropriate than no

constant term b/c it facilitates comparisons.
 1.2.3 Analysis of Covariance (ANCOVA)
 Intercept indicators variables:
 The above examples we have seen under ANCOVA
analysis an example of intercept indicator variables, a
regression mixture of qualitative and quantitative
variables.
 It affects only intercept.
 It interact with dummy variables and qualitative
variables. This is why it affects only intercepts rather
than slope. Wage = β 0 + β 1 gend + β 1 educ + u i
 β 0 + β 1 + β 2 educ − − − when D = 1
E (Y ) = 
 β 0 + β 2 educ − − − − − − when D = 0
 1.2.3 Analysis of Covariance (ANCOVA)
Then the difference b/n them is:

β 1 = ( β 0 + β 1 + β 2 educ ) − ( β 0 + β 2 educ )

+ve: it is greater than other

-ve: it is less than other
 1.2.4 Interactions among Dummy Variables

Y i = α 1 + α 2 D 2i + α 3 D 3i + β X i + ui
Yi where annual expenditure on clothing
 Xi income
1 if female
D2 = 
 0 if male
1 if college graduate
D3 = 
 0 otherwise
• In many applications there may be interaction between the two
qualitative variables D and D 3 therefore their effect on mean Y
2

may not be simply additive but multiplicative as well

• Hence, we re specify the above model as follows:
1.2.4 Interactions among Dummy Variables

Yi = α 1 + α 2 D 2 i + α 3 D 3 i + α 4 ( D 2 i D 3 i ) + β X i + u i
E (Y i | D 2 = 1, D 3 = 1, X i ) = (α 1 + α 2 + α 3 + α 4 ) + β X i
α 2 = differential effect of being a female
α 3 = differential effect of being a college graduate
α 4 = differential effect of being a female graduate
* The above equation shows that the mean clothing expenditure
of graduate females is different by α 4 from the mean clothing
expenditure of females or college graduates.
* If α 2 , α 3 and α 4 are all + ve, the average clothing expenditure
of females is higher (than the base category, which here is
male non graduate), but it is much more so if the females also
happen to be graduates.
 1.2.4 Interactions among Dummy Variables

Similarly, the average expenditure on clothing by a

college graduate tends to be higher than the base
category but much more so if the graduate happens to be
a female.
 This shows how the interaction dummy modifies the
effect of the two attributes considered individually.
Whether the coefficient of the interaction dummy is
statistically significant can be tested by the usual t test.
If it turns out to be significant, the simultaneous presence
of the two attributes will attenuate or reinforce the
individual effects of these attributes.
 Omitting a significant interaction term incorrectly will lead to a specification
bias.
 1.2.4 Interactions among Dummy Variables
The importance of interactions among dummy
variables
 help us to get influential variables
to avoid misspecification bias
 1.2.5 Slope indicator variables
The interaction between dummy variables and
quantitative variables. They affect only slope,
i.e, it does not affect intercept.
It help us to captures the interaction effect of
dummy and quantitative variables on dependent
variables
Look at the following example
The price of condominium house can be explained as
a function of its characteristics such as its size,
location, number of bedrooms, age, floor and so on.
 1.2.5 Slope indicator variables
For our discussion, let us assume that the
number bed room of the house of the measured
in numbers, nbdr, is the only relevant variable in
determining house price.
prhou = β 0 + β1nbdr + ui
 β1 is the value of an additional number of bed
rooms.
 β 0 is the value of land alone
We can use dummy variable and indicator
variable interchangeable.
 1.2.5 Slope indicator variables

prhou = β 0 + ψneib + β1nbdr + ui

1 if desirable neibourhood
neib = 
0 if not desirable neibourhood
We make the reference group, non desirable
group.
Instead of assuming that the effect of location on
house price causes a change in the intercept.
Let us assume that the change is in the slope of
the relationship.
 1.2.5 Slope indicator variables
We can allow for a change in a slope by
including in the model an additional explanatory
variable that is equal to the product of an
indicator variable and continuous variable.
In our model, the slope of the relationship is the
value of an additional number of bed rooms.
If we assume 1 value for homes in desirable
neibourhood, and 0 other wise; we can specify
our model as follows:
prhou = β 0 + β1nbdr + ω ( nbdr * neib ) + ui
 1.2.5 Slope indicator variables
The new variable (nbdr*neib) is the product number
of bedroom and the indicator variables, is called an
interaction variable as it captures the interaction of
location and number of bedroom on condominium
house prices.
Or it is called a slope –indicator variable or a slope
dummy variable, b/c it allows for the change in the
slope of the relationship.
The slope indicator variable takes a value equal to
nbdr for houses in the desirable neibourhood, when
neib=1, and it is 0 for homes in other
neighbourhoods.
 1.2.5 Slope indicator variables
A slope indicator variable is treated as just like
any other explanatory variable in a regression
model.
 β 0 + β 1 nbdr + ω nbdr − − − when D = 1
E ( prhou ) = 
 β 0 + β 1 nbdr − − − − − − − − when D = 0

In the desirable neighbourhood, the price per

additional number of bedrooms of a house is β + ω 1

In the non desirable neighbourhood, the price

per additional number of bedrooms of a house is
β .1

If ω > 0 price per additional number of bedrooms

is higher in the more desirable neighbourhood.
 1.2.5 Slope indicator variables
The effect of including a slope indicator variable
also can be see by using calculus.
The partial derivatives of expected house price
with respect to number of bed rooms
∂E ( prhou )  β1 + ω when D = 1
=
∂ ( nbdr )  β1 w hen D = 0
If ω > 0 E ( prhou ) = β 0 + ( β1 + ω )nbdr
slope = β1 + ω
prhou ω E ( prhou ) = β 0 + β1nbdr

 β0 slope = β1

nbdr
 1.2.5 Slope indicator variables
If we assume that house location affects both the
intercept and the slope, then both affects can be
incorporated into a single model.
The model specification will be:
prhou = β 0 + ψ neib + β1nbdr + ω ( nbdr * neib ) + ui

 ( β 0 + ψ ) + ( β 1 + ω ) nbdr − − − when D = 1
E ( prhou ) = 
 β 0 + β 1 nbdr − − − − − − − − when D = 0

 Look numerical example from Principle of Econometrics

1.3 “Structural
Stability”
 1.3 Structural Stability
Testing for structural stability is help us to find
out whether two or more regressions are different,
where the difference may be in the intercepts or
the slopes or both.
Suppose we are interested in estimating a
simple saving function that relates domestic
household savings (S) with gross domestic
product (Y) for Ethiopia.
Suppose further that, at a certain point of
time, a series of economic reforms have been
introduced.
 1.3 Structural Stability
So far we assumed that the intercept and
all the slope coefficients (βj's) are the
same/stable for the whole set of
observations. Y = Xβ + e
But, structural shifts and/or group
differences are common in the real world.
May be:
the intercept differs/changes, or
the (partial) slope differs/changes, or
both differ/change across categories or
time period.
 1.3 Structural Stability
 The hypothesis here is that such reforms might have
considerably influenced the savings- income
relationship, that is, the relationship between savings
and income might be different in the post reform
period as compared to that in the pre-reform period.
 If this hypothesis is true, then we say a structural
change has happened.
 H0: Economic reforms might not have influenced the
savings and national income relationship
 H1: Economic reforms might have influenced the
savings and national income relationship

 How do we check if this is so?

 1.3 Structural Stability
We can test structural stability of testing parameter
by using two methods.

1. Using Dummy variables

2. Chow’s test

1. Using dummy variables

* Write the savings function as:
S t = β 0 + β 1 Dt + β 2Yt + β 3 (Yt Dt ) + u t
where S t is household saving at time t , Yt is GDP at time t and

0 if pre − reform (< 1991)

Dt = 
1 if post − reform (> 1991)
 1.3 Structural Stability
 β Is the differential slope coefficient
3

indicating how much the slope coefficient of the

pre-reform period savings function differs from
the slope coefficient of the savings function in
the post reform period.
Decision rule:
If β and β are both statistically significant as
1 3

judged by the t-test, the pre-reform and post-

reform regressions differ in both the intercept
and the slope.
 1.4 Structural Stability
If only β is statistically significant, then the
1

pre-reform and post-reform regressions differ

only in the intercept (meaning the marginal
propensity to save (MPS) is the same for pre-
reform and post-reform periods).
If only β 3 is statistically significant, then the
two regressions differ only in the slope (MPS).
Check structural stability for the f/wing
regression result:
Sˆt = −20.76005 + 5.9991 Dˆ t + 2.616285 Yˆt − 0.5298177 (Yˆt Dˆ t )
 1.4 Structural Stability
Example 2: Using the DVR to Test for Structural
Break:
 Recall the example of consumption function:
period 1: consi = α1+ β1*inci+ui vs.
period 2: consi = α2+ β2*inci+ui
 Let’s define a dummy variable D1, where:
for the period 1974-1991, and
for the period 1992-2006
 Then, consi = α0+α1*D1+β0*inci+β1(D1*inci)+ui
For period 1: consi = (α0+α1)+(β0+β1)inci+ui
For period 2 (base category): consi= α0+ β0*inci+ui
 Regressing cons on inc, D1 and (D1*inc) gives:
cons = 1.95 + 152D1 + 0.806*inc – 0.056(D1*inc)
p-value: (0.968) (0.010) (0.000) (0.002)
 1.4 Structural Stability
D1=1 for i ϵ period-1 & D1=0 for i ϵ period-2:
period 1 (1974-1991): cons = 153.95 +
0.75*inc
period 2 (1992-2005): cons = 1.95 + 0.806*inc
The Chow test is equivalent to testing α1=β1=0
in:
cons=1.95+152D1+0.806*inc – 0.056(D1*inc)
This gives: F(2, 29) = 6.76; p-value = 0.0039.
Then, reject H0! There is a structural break!
 1.4 Structural Stability
 For a total of m categories, use m–1 dummies!
 Including m dummies (1 for each group) results in
perfect multicollinearity (dummy variable trap). e.g.: 2
groups & 2 dummies:
 constant = D1 + D2 !!!

X = [constant D1 D2 ]
1 X 11 1 0

X = 1 X 12 1 0 
1 X 13 0 1 
 1.4 Structural Stability
2. Chow’s test
 One approach for testing the presence of structural change
(structural instability) is by means of Chow’s test. The steps
involved in this procedure:
 Step 1: Estimate the regression equation for the whole period
(pre-reform plus post-reform periods) and find the error sum of
squares ( ESSR ) or RRSS.
 Step 2: Estimate equation (model) using the available data in the pre-reform
period (say, of size n 1), and find the error sum of squares (ESS1) or RSS1
 Step 3: Estimate equation (model) using the available data in the pre-reform
period (say, of size n 2), and find the error sum of squares (ESS2) or RSS2.
 Step 4: Calculate RSSUR= RSS1+RSS2.
 Step 5: Calculate the Chow test statistic
( RSS R − RSSU ) / k
Fc =
RSSU /(n1 + n2 − 2k )
 Where k is number of estimated regression coefficients
 1.4 Structural Stability
α
 F ( k ,n1 +n2 −2k )
is the critical value from the t-
distribution with k (in our case k=2) and n1+n2-2k
degrees of freedom from a given significance level, α
 Decision rule: Reject the null hypothesis of
identical intercepts and slopes for the pre-reform
and post reform periods, that is
β0 = β3
H0 =  if Fc > Fb.
β2 = β4
 i.e, Rejecting H0 means there is a structural
change.
 1.4 Structural Stability
Example: RSS1=64499436.865 (Error sum of
squares in the pre-reform period); n1=12;
RSS2=2,726,652,790.434 (Error sum of squares in
the post-reform period); n2=11;
RSSR=13,937,337,067.461 (Error sum of squares
for the whole period)
RSSU=RSS1+RSS2=2,791,152,227.299
The test statistics is:
( RSS R − RSSU ) / k (13,937,337,067.461 − 2,791,152,227.2) / 2
Fc = = ≈ 190
RSSU /( n1 + n2 − 2k ) (2,791,152,227.299) /(12 + 11 − 2(2))
The tabulated value from the F-distribution with 2
and 19 degrees of freedom at the 5% level of
significance is 3.52.
 1.4 Structural Stability
Decision: Since the calculated value of F exceeds
the tabulated value, we reject the null hypothesis of
identical intercepts and slopes for the pre-reform
and post reform periods at the 5% level of
significance.
Hence, we can conclude that there is a structural
break.
 1.4 Structural Stability

Draw backs:
Chow’s test does not tell us whether the difference
(change) in the slope only, in the intercept only or
in both the intercept and the slope.

The Chow Tests

Using an F-test to determine whether a single
regression is more efficient than two/more separate
regressions on sub-samples.
 1.4 Structural Stability

The stages in running the Chow test are:

1. Run 2 separate regressions (say, before & after war
or policy reform, …) & save RSS's: RSS1 & RSS2.
RSS1 has n1–(K+1) df & RSS2 has n2–(K+1) df.
RSS1 + RSS2 = URSS with n1+n2–2(K+1) df.
2. Estimate pooled model (under H0: β's are stable).
RSS from this model is RRSS with n–(K+1) df
where n = n1+n2.
3. The test-statistic (under H0): [RRSS − URSS]
(K + 1)
Fcal =
URSS
[n − 2(K + 1)]
4. Find the critical value: FK+1,n-2(K+1) from table.
5. If Fcal>Ftab, reject H0 of stable parameters (and
favour Ha: there is structural break).
 1.4 Structural Stability

e.g.: we have the ff results from estimation of real

consumption from real disposable income:
i. For the period 1974-1991: consi = α1+β1*inci+ui
Consumption = 153.95 + 0.75*Income
p-value: (0.000) (0.000)
RSS = 4340.26114; R2 = 0.9982
ii. For the period 1992-2006: consi = α2+ β2*inci+ui
Consumption = 1.95 + 0.806*Income
p-value: (0.975) (0.000)
RSS = 10706.2127; R2 = 0.9949
iii. For the period 1974-2006: consi = α+ β*inci+ui
Consumption = 77.64 + 0.79*Income
t-ratio: (4.96) (155.56)
RSS = 22064.6663; R2 = 0.9987
 1.4 Structural Stability

1. URSS = RSS1 + RSS2 = 15064.474

2. RRSS = 22064.6663
K = 1 and K + 1 = 2; n1 = 18, n2 = 15, n = 33.
3. Thus, [22064.6663 − 15064.474]
Fcal = 2 = 6.7632981
15064.474
29
4. p-value = Prob(F-tab > 6.7632981) = 0.003883
5. Reject H0 at α=1%. Thus, there is structural break.
The pooled consumption model is an inadequate
specification; we should run separate regressions.
The above method of calculating the Chow test
breaks down if either n1 < K+1 or n2 < K+1.
Solution: use Chow’s second (predictive) test!
 1.4 Structural Stability

If, for instance, n2 < K+1, then the F-statistic will be

altered as follows: [RRSS − RSS ]
1
n2
Fcal =
RSS 1
n1 − (K + 1)
 The Chow test tells if the parameters differ on
average, but not which parameters differ.
 Also, it requires that all groups have the same σ2.
This assumption is questionable: if parameters can
be different, then so can the variances be.
One way of correcting for unequal σ2 is to use
dummy variable regression with robust standard
errors.
 1.4 Structural Stability
Using Dummy variables vs Chow’s test
Comparing the two methods, it is
preferable to use the method of
dummy variables regression.
This is because with the method of
DVR:
1. We run only one regression.
2. We can test whether the change is in
the intercept only, in the slope only, or
in both.
 Lab session

Use “Chowtest.xls” data to

practice what we learnt in
previous sections
 END OF CHAPTER ONE

THANK YOU FOR BEING WITH ME

BEING @ COMMITTED
Stay Safe!
2.2.“Dummy dependent
variable”:
Qualitative Response
Model
 2.2.1 Introduction
Qualitative Response Model shows situations in
which the dependent variable in a regression
equation simply represents a discrete choice
assuming only a limited number of values
Such a model is called
 Limited dependent variable
 Discrete dependent variable

 Qualitative response

Categories of Qualitative Response Models

there are two broad categories of QRM
 2.2.1 Introduction
1. Binomial Model: it shows the choice between
two alternatives
e.g: Decision to participate in labor force or not
2. Multinomial models: the choice between more
than two alternatives
e.g: Y= 1, occupation is farming
=2, occupation is carpentry
=0, government employee
Important terminologies
Binary variables: variables that have two categories and used to
an event that has occurred or some characteristics present.
 2.2.1 Introduction
Ordinal variables: variables that have
categories that can be ranked.
Example: Rank according to education
attainment (Y)
0 if primary education

Y = 1 if secondary education
 2 if university education

Nominal variables: variables occur when
there are multiple outcomes that cannot be
ordered.
 2.2.1 Introduction
Example: Occupation can be grouped as
farming, fishing, carpentry etc.
0 if farming N.B: Numbers are
1 if fishermen
 assigned arbitrarily
Y =
 2 if carpentry
 3 if government employee
Count variables: indicate the number of times
some event has occurred.
Example: How many years of education you
have attend?
In all of the above situations, the variables
are discrete valued.
 2.2.2 Qualitative Choice Analysis
In such cases instead of standard regression
models, we apply different methods of modeling
and analyzing discrete data.
Qualitative choice models may be used when a
decision maker faces a choice among:
The number of choices if finite
The choices are mutually exclusive (the
person chooses only one of the alternatives)
The choices are exhaustive (all possible
alternatives are included)
 Qualitative choice analysis
Throughout our discussion we shall restrict
ourselves to cases of qualitative choice where the
set of alternatives is binary.
For the sake of convenience the dependent
variable is given a value of 0 or 1.
Example: Suppose the choice is whether to
work or not. The discrete dependent variable
we are working with will assume only two
values 0 and 1: Y = 1 if i individual is working
th

i
0 if i th individual is notworking
where i = 1, 2, …, n.
 Qualitative choice analysis
 The independent variables (called factors) that are expected to
affect an individual’s choice may be X1 = age, X2 = marital
status, X3 = gender, X4 = education, and the like.
 These are represented by a matrix X.
Regression Approach
 The economic interpretation of discrete choice models is
typically based on the principle of utility maximization leading to
the choice of, say, A over B if the utility of A exceeds that of B.
 Let U1 be the utility from working/seeking work and let U0 be
the utility form not working. Then an individual will choose to
be part of the labour force if U1 -U0 > 0 , and this decision
depends on a number of factors X.
 Qualitative choice analysis
The probability that the ith individual chooses
alternative 1th (i.e. works) given his/her individual
characteristics, Xi is:
Ρ i = pr (Y i = 1 / X i ) = Pr[( U 1
− U o ) i > 0] = G ( X i , β )

The vector of parameters β = (β , β ,....,β ) ( measures the

1 2 k

impact of changes in X (say, age , marital status,

gender, education, occupation, and the like) on the
probability of labor force participation.
 the probability that the ith individual chooses
alternative 0 (i.e. not to work) is given by:
pr(Yi = 0 / X i ) = 1 − Ρi = 1 − Pr[(U 1 − U o ) i > 0] = 1 − G( X i , β )
 Qualitative choice analysis
Here Pi is called the response probability and
(1-Pi ) is called the non-response probability.
The mean response of the ith individual given
his/her individual characteristics Xi is:
E ( Y i / Xi = 1 * {G ( Xi , β ) }+ 0 * {1 − G ( X i , β ) }= G(X i,β )

The problem is thus to choose the appropriate

form of G ( X , β ).
i

 There are several methods to analyze regression models where

the dependent variable is binary.
 the four most commonly used approaches to estimating binary
response models (Type of binomial models). These are:
 Qualitative choice analysis
Linear probability models
The logit model
The probit model
The tobit (censored regression) model
1. The Linear Probability Models (LPM)
The term linear probability model is used to
denote a regression model in which the
dependent variable y is a dichotomous variable
taking the value 1 or 0.
 Linear Probability Models
In the 1960’s and early 1970’s the linear
probability model was widely used mainly
because it is a model that can be easily estimated
using multiple regression analysis.
A “limited dependent variable" y is one which
takes a limited set of values. The most common
cases are: Binary: yε {0,1 } ; Multinomial: yε{0,1,2,...,k} ;
;Integer: Integer: yε{0,1,2,...} y ε R .
+

The traditional approach to the estimation of limited

dependent variable models is parametric maximum
likelihood.
 Linear Probability Models
A parametric model is constructed, allowing the
construction of the likelihood function.
A more modern approach is semi-parametric,
eliminating the dependence on a parametric
distributional assumption. We will limit our
discuss to parametric approach.
When we use a linear regression model to
estimate probabilities, we call the model the
linear probability model.
The linear probability model is the regression
model applied to a binary dependent variable.
 Linear Probability Models
The linear probability model defines G( X , β ) = X β i i

The regression model when Y is a binary

variable is thus, Y = β 0 + β1 X 1 + β 2 X 2 + ... + β k X k + ε = Xβ + ε
Where y takes only two value: 0 & 1, and
β cannot be interpreted as the change in Y given a
j

one-unit increase in Xj,, holding all other factors

constant rather Y changes either from 0 to 1 or
from 1 to 0.
If we assume that the zero conditional mean
assumption holds, that is, E(ε) =0 , then we have, as
always,
 Linear Probability Models
E(Y ) = β0 + β1 X 1 + β 2 X 2 + ... + β k X k = Xβ
 pr(Y =1/ X ) = β + β X + β X + ... + β X = Xβ  it says that the
0 1 1 2 2 k k

probability of success is a linear function of the

Xj; it is called the response probability.
 Pr(Y = 0 / X ) = 1 − pr (Y = 1 / X ) it is the non-response
probability, and it is also a linear function of the
Xj .
Interpretation of LPM
The response probability is linear in the
parameters of Xj .
 Linear Probability Models
Example: Suppose that from hypothetical data
of house ownership and income and thus, the
LPM estimated by OLS (on home ownership) is
given as follows:

 The above regression is interpreted as follows

 The intercept of –0.9457 gives the “probability” that a family with zero
income will own a house. Since this value is negative, and since probability
cannot be negative, we treat this value as zero.
 The slope value of 0.1021 means that for a unit change in income, on the
average the probability of owning a house increases by 0.1021 or about 10
percent. This is so whether the income level is increased or not. This seems
patently unrealistic. In reality one would expect that Pi is non-linearly related
to Xi.
 Linear Probability Models
β j measures the change in the probability of
success when Xj changes, holding other factors
fixed.
Advantages of LPM: Easy to estimate and
interpret; it is too simple.
Drawbacks of LPM:
1. The dependent variable is discrete while the independent
variable is the combination of discrete and continuous
variables.
2. Usually we arbitrarily (or for convenience) use 0 and 1 for Yi .
If we use other values for Yi , say 3 and 4, will also change even
if the vector of factors Xi remains unchanged.
 Linear Probability Models
3. Error term assumes only two values.
If Yi=1 then ε = 1 − X β with the Probability, Pi;
i i

 If Yi=0 then ε = − X β with Probability, 1-Pi;

i i

 The variance of the disturbance terms depends

on the X’s and is thus not constant.; i.e., error
term is not normally distributed.
Now by definition Var(ε ) = [E(ε − E(ε )] = E(ε ) since E(ε = 0 by
i i i
2
i
2
i)

assumption. Therefore, using the preceding

probability distribution of we obtain:
 Linear Probability Models
This shows that the variance of Ui is
heteroscedastic because it depends on the
conditional expectation of Y, which, of course,
depends on the value taken by X.
 Thus, the OLS estimator of β is inefficient and
the standard errors are biased, resulting in
incorrect test.
4. The expectation (mean) of conditional on the
exogenous variables Xi is non sense.
Setting this mean to zero as in
E(ε / Xi ) = (1− Xi β)Pi + (−Xi β)(1− Pi ) = Pi − Xi β the classical regression
analysis
 Linear Probability Models
E(ε i / X i ) = 0  Pi = X i β

Yi = Pi +εi ε = Yi − Pi

 That is, the binary (discrete) disturbance term is

equal to the difference between a binary variable
Yi and a continuous response probability Pi.
Clearly this does not make sense.
the probability of an event is always a number
between 0 and 1 (inclusive). But here we can see
that: Pi = pr (Yi = 1 X i ) = X i β , i.e., Pi can take on
any value (even negative numbers) leading to
nonsense probabilities, the fitting probabilities
 Linear Probability Models
5. Non-Sensical Predictions
The LPM produces predicted values outside the
normal range of probabilities (0,1). It predicts
value of Y that are negative and greater than 1.
This is the real problem with the OLS
estimation of the LPM.
6. Non-normality of Ui
 Although OLS does not require the disturbance (U’s) to be normally
distributed, we assumed them to be so distributed for the purpose of
statistical inference, that is, hypothesis testing, etc. But the assumption of
normality for Ui is no longer tenable for the LPMs because like Yi, Ui takes on
only two values.
 Linear Probability Models
7. Functional Form
Since the model is linear, a unit increase in X
results in a constant change of in the probability
of an event, holding all other variables constant.
The increase is the same regardless of the
current value of X.
8. Questionable Value of R2 as a Measure of
Goodness of Fit (lower R2 values )
 In sum, non-normality of ui; possibility of Yi lying outside the 0-
1 range; hetroscedasticity of Ui; lower R2 values; the basic
problem is not logically attractive.
 Linear Probability Models
b/c the above mentioned problems, the LPM
model is not recommend for empirical works.
Therefore, what we need is a (probability)
model that has the following two features:
As Xi increases, Pi = E(Y = 1/X) increases but
never steps outside the 0-1 interval.
The relationship between Pi and Xi is non-
linear, that is, “ one which approaches zero at
slower and slower rates as Xi gets small and
approaches one at slower and slower rates as
X gets very large”
 Linear Probability Models

The above S-shaped curve is very much similar with

the cumulative distribution function (CDF) of a
random variable.
the CDF of a random variable X is simply the
probability that it takes a value less than or equal to x0,
were x0 is some specified numerical value of X.
 In short, F(X), the CDF of X, is F(X = x0) = P(X ≤ x0).
 Linear Probability Models
 Therefore, one can easily use the CDF to model regressions
where the response variable is dichotomous, taking 0-1 values.
 The CDFs commonly chosen to represent the 0-1 response
models are.
 the logistic – which gives rise to the logit model

 the normal – which gives rise to the probit (or normit)

model
2. Logit model
• Although LPM is simple to estimate and use, but the two most
important disadvantages are:
• the fitted probabilities can be less than zero or greater than
one and
• the partial effect of any explanatory variable is constant.
 Logit model
These limitations of the LPM can be overcome
by using more sophisticated binary response
models.
P (Y = 1 / X ) = P (Y = 1 / X 1 , X 2 ,...., X k )

 In a binary response model, interest lies primarily in the

response probability.
 where we use X to denote the full set of explanatory variables.
 For example, when Y is an employment indicator, X might
contain various individual characteristics such as education, age,
marital status, and other factors that affect employment status,
including a binary indicator variable for participation in a
recent job training program.
 Logit model
In the LPM, we assume that the response
probability is linear in a set of parameters, β . j

 To avoid the LPM limitations, consider a class

of binary response models of the form.
P(Y = 1 / X ) = G( β 0 + β1 X 1 + β 2 X 2 + .... + β k X k ) = G( Xβ )

Where G is a function taking on values strictly

between 0 & 1: 0 < G ( z ) < 1, for all real numbers z.
This ensures that the estimated response
probabilities are strictly between zero and one.
 Logit model
Various nonlinear functions have been
suggested for the function G in order to make
sure that the probabilities are between zero and
one.
In the logit model, G is the logistic function:
exp( z ) e Xβ
G (z) = = Λ(z) =
[1 + exp( z )] 1 + e Xβ

which is between zero and one for all real

numbers z. This is the cumulative distribution
function (cdf) for a standard logistic random
variable.
 Logit model
the response probability P(Y =1/X) is evaluated
as: P = P (Y = 1 X ) =
e Xβ

Xβ
1+ e
the non response probability P(Y =0/X) is
evaluated as:
e Xβ 1
1 − P = P(Y = 0 X ) = 1 − Xβ
=
1+ e 1 + e Xβ
Note that: both response and non- response
probabilities lie in the interval [0 , 1] , and
hence, are interpretable.
Odd ratio: the ratio of the response probabilities
(Pi) to the non response probabilities (1-Pi).
 Logit model
For the logit model, the odds ratio is given by:
e Xβ
P P(Y = 1 X ) 1 + e Xβ
= = = e Xβ = e β0 +β1 + e β2 X 2 + e β3 X 3 + ...e βk X k
1 − P P(Y = 0 X ) 1
1 + e Xβ

The natural logarithm of the odds ratio(log-

odds ratio) is:
Li = ln(
P
1 − P
) = β0 + β1X1 + β2 X2 +...+ βk Xk

L(the log of the odds ratio) is linear in X as well

as β (the parameters). L is called the logit and
hence the name logit model is given to it.
 Logit model
Thus, the log-odds ratio is a linear function of
the explanatory variables.
For the LPM it is Pi, which is assumed to be a
linear function of the explanatory variables.
Features of logit model
 As P goes from 0 to 1 (i.e., as Z varies from −∞ to +∞), the logit L
goes from −∞ to +∞. That is, although the probabilities (of
necessity) lie between 0 and 1, the logits are not so bounded.
 Although L is linear in X, the probabilities themselves are not.
This property is in contrast with the LPM model where the
probabilities increase linearly with X.
 Logit model
If L, the logit becomes negative and increasingly
large in magnitude as the odds ratio decreases
from 1 to 0 and becomes increasingly large and
positive as the odds ratio increases from 1 to
infinity.
LPM assumes that Pi is linearly related to Xi,
the logit model assumes that the log of the odds
ratio is linearly related to Xi .
Interpretation: Be remind that we doesnot directly
interpreted the coefficients of the variables
rather we interpreted their marginal effects and
 Logit model
β2, the slope, measures the change in L for a
unit change in X, that is, it tells how the log-odds
in favor of owning a house change as income
changes by a unit, say, birr 1000.
The intercept, β1 is the value of the logodds in
favor of owning a house if income is zero.
 Probit model
The estimating model that emerges from the
normal CDF is popularly known as the probit
model.
In the probit model, G is the standard normal
cumulative distribution function (cdf ), which is
expressed as an integral:
 In the probit model, G is the standard normal
cumulative distribution function
z
G ( z ) = Φ ( z ) =  φ (v)dv
−∞

Where Φ (z) is the standard normal density

 Probit model

Φ(z) = (2π )−1/ 2 exp(−z 2 / 2)

The estimating model that emerges from the
normal CDF is popularly known as the probit
model.
Here the observed dependent variable Y, takes
on one of the values 0 and 1 using the following
criteria.
 Define a latent variable Y* such that Y = X β + ε∗
1i i

1 if Y i ∗ > 0
Y = 
 0 if Y i ∗ ≤ 0
 Probit model
The latent variable Y* is continuous (-∞ < Y* <
∞).
It generates the observed binary variable Y.
An observed variable, Y can be observed in two
states:
if an event occurs it takes a value of 1
if an event does not occur it takes a value of 0
The latent variable is assumed to be a linear
function of the observed X’s through the
structural model.
 Probit model
However, since the latent dependent variable is
unobserved the model cannot be estimated using
OLS.
Maximization of the likelihood function for
either the probit or the logit model is
accomplished by nonlinear estimation methods.
Maximum likelihood can be used instead.
Most often, the choice is between normal errors
and logistic errors, resulting in the probit
(normit) and logit models, respectively.
 Probit model
The coefficients derived from the maximum
likelihood (ML) function will be the coefficients
for the probit model, if we assume a normal
distribution.
If we assume that the appropriate distribution
of the error term is a logistic distribution, the
coefficients that we get from the ML function
will be the coefficient of the logit model.
 In both cases, as with the LPM, it is assumed
that E[∈i/Xi] = 0
 Probit model
In the probit model, it is assumed that Var
(∈i/Xi) = 1; In the logit model, it is assumed that
Var (∈i/Xi) = π2 3 .
Hence, the estimates of the parameters (β ’s)
from the two models are not directly
comparable.
But as Amemiya suggests, a logit estimate of a
parameter multiplied by 0.625 gives a fairly
good approximation of the probit estimate of the
same parameter.
 Probit model
Similarly the coefficients of LPM and logit
models are related as follows:
 β LPM = 0.25 Logit, except for intercept
 β LPM = 0.25 Logit + 0.5 for intercept
The standard normal cdf has a shape very
similar to that of the logistic cdf.
 Probit model
The estimating model that emerges from the
normal CDF is popularly known as the probit
model, although sometimes it is also known as
the normit model.
Note that both the probit and the logit models
are estimated by Maximum Likelihood
Estimation.
 Probit model
Interpreting the Probit and Logit Model Estimates
The coefficients give the signs of the partial effects of
each Xj on the response probability, and the statistical
significance of Xj is determined by whether we can
reject H0: Bj=0 at a sufficiently small significance level.
However, the magnitude of the estimated parameters (
dZ/dX) has no particular interpretation. We care about
the magnitude of dProb(Y)/dX.
From the computer output for a probit or logit
estimation, you can interpret the statistical significance
and sign of each coefficient directly.
 Probit model
In the linear regression model, the slope
coefficient measures the change in the average
value of the regressand for a unit change in the
value of a regressor, with all other variables
held constant.
In the LPM, the slope coefficient measures
directly the change in the probability of an event
occurring as the result of a unit change in the
value of a regressor, with the effect of all other
variables held constant.
 Probit model
In the logit model the slope coefficient of a
variable gives the change in the log of the odds
associated with a unit change in that variable,
again holding all other variables constant.
But as noted previously, for the logit model the
rate of change in the probability of an event
happening is given by βj Pi(1 − Pi ), where βj is
the (partial regression) coefficient of the jth
regressor. But in evaluating Pi , all the variables
included in the analysis are involved.
 Probit model
In the probit model, as we saw earlier, the rate of
change in the probability is somewhat complicated
and is given by βj f (Zi ), where f (Zi) is the density
function of the standard normal variable and Zi =
β1 + β2X2i + · · · +βkXki , that is, the regression
model used in the analysis.
Thus, in both the logit and probit models all the
regressors are involved in computing the changes in
probability, whereas in the LPM only the jth
regressor is involved. This difference may be one
reason for the early popularity of the LPM model.
 Probit vs logit model
Is logit or probit model is preferable?
In most applications the models are quite similar, the main
difference being that the logistic distribution has slightly
fatter tails.
 That is to say, the conditional probability Pi approaches
zero or one at a slower rate in logit than in probit.
 Therefore, there is no compelling reason to choose one over
the other.
In practice many researchers choose the logit model
because of its comparative mathematical simplicity.
The standard normal cdf has a shape very similar to that of
the logistic cdf.
 Probit vs logit model
The probit and logit models differ in the
specification of the distribution of the error term
u.
 The difference between the specification and
the linear probability model is that in the linaer
probability model we analyses the dichotomous
variables as they are, where as we assume the
existence of an underlying latent variable for
which we observe a dichotomous realization.
 Probit vs logit model
The probit model and the logit model are not
directly comparable. The reason is that,
although the standard logistic (the basis of logit)
and the standard normal distributions (the basis
of probit) both have a mean value of zero, their
variances are different; 1 for the standard
normal (as we already know) and π2/3 for the
logistic distribution, where π ≈ 22/7.
Therefore, if you multiply the probit coefficient
by about 1.81 (which is approximately = π/√ 3),
you will get approximately the logit coefficient.
 Probit vs logit model
The R2’s for the linear probability model are
significantly lower than those for the logit and
probit models. Alternative ways of comparing
the models would be:
To calculate the sum of squared deviations
from predicted probabilities
To compare the percentages correctly
predicted
To look at the derivatives of the probabilities
with respect to a particular independent
variable.
 Tobit Model
An extension of the probit model is the tobit model
developed by James Tobin.
Let us consider the home ownership example.
Suppose we want to find out the amount of money the
consumer spends in buying a house in relation to his or
her income and other economic variables.
If a consumer does not purchase a house, obviously we
have no data on housing expenditure for such
consumers; we have such data only on consumers who
actually purchase a house.
 Tobit Model
Thus, consumers are divided into two groups, one
consisting of say, N1 consumers about whom we have
information on the regressors (say income, interest rate
etc) as well as the regresand ( amount of expenditure
on housing) and another consisting of say, N2
consumers about whom we have information only on
the regressors but on not the regressand.
 A sample in which information on regressand is
available only for some observations is known as a
censored sample. Therefore, the tobit model is also
known as a censored regression model.
 Tobit Model
Mathematically, we can express the tobit model
as
β 0 + β1 X 1i + ui if RHS > 0
Yi = 
0, otherwise
Where RHS= right hand side
The method of maximum likelihood can be used
to estimate the parameters of such models.
Measuring goodness of fit
The conventional measure of goodness of fit,R2 ,
is not particularly meaningful in binary
regressand models. Measures similar to R2,
 Measuring goodness of fit
Measures based on likelihood ratios: The
conventional measure of goodness of fit,R2 , is
not particularly meaningful in binary
regressand models.
 Measures similar to R2, called pseudo R2, are
available, and there are a variety of them.
Measures based on likelihood ratios
Let LURbe the maximum likelihood function when
maximized with respect to all the parameters and LR be
the maximum likelihood function when maximized
with restrictions βi =0 .
 Measuring goodness of fit
2
LR n
R 2
=1− ( )
L UR
the qualitative dependent variable model, the
likelihood function attains an absolute
maximum of 1. This means that, L ≤ L ≤ 1 R UR

Cragg and Uhler (1970) suggested a pseudo R2 2 2

that lies between 0 and 1. R = L − L 2 UR

n
R
n
2 2
(1 − L R
n ) L UR n

Mc Fadden (1974) defined R2 as

log L UR
R 2
=1−
log L R
 Measuring goodness of fit
Another goodness-of-fit measure that is usually
reported is the so-called percent correctly
predicted, which is computed as follows. For
each i, we compute the estimated probabilit y
that Yi takes on the value one, Yˆi .
ˆi
 If >0.5 the prediction of Yi is unity, and if <0
Yˆ i Y

Yi is predicted to be zero. The percentage of

times the predicted Yˆ i matches the actual Yi
(which we know to be zero or
1 if Ŷ ≥ 0.5
one) is the percent
∗
Yˆi = 
i

correctly predicted.  0 if Ŷ < 0.5

number of correct prediction s
i

CountR =
2

total number of observatio ns

 Brainstorm questions
1. Why LPM is not recommendable for empirical
analysis?
2. Drive logodds ratio.
3. For logit and probit model we can use OLS
estimators.
4. For logit and probit model we can use
π2

2
maximum likelihood estimators.
5. The variance that logit model assume is
where as probit model assumes 1.
 Lab session

Use “lecture_2.xls” data to

practice what we learnt in
previous sections
 END OF CHAPTER TWO

THANK YOU FOR BEING WITH ME

BEING @ COMMITTED
stay Safe!
 CHAPTER THREE

Introduction to Basic Regression Analysis with

Time Series Data
3.1 Nature of the Time Series data
3.2 Stationary & non stationary stochastic
Processs
3.3 Trend & Difference stationary stochastic
process
3.4 Integrated Stochastic Process
3.5 Tests of Stationary
141
 3.1 Nature of the Time Series data
Time series data have become so frequently and
intensively used in empirical research and thus,
econometricians have recently begun to pay very
careful attention to such data.
 Time series data are data collected for a single
entity (person, firm, and country) collected
(observed) at multiple time periods.
A time series is a sequence of numerical data in
which each item is associated with a particular
instant in time.
 Example: Monthly unemployment, weekly measures of money
supply, M1 and M2, daily closing prices of stock142
indices, and so on
 3.1 Nature of the Time Series data
Thus, the way we collect time series data can be
characterised as daily (stock prices, weather
report), weekly (gasoline supplied in thousands of
barrels), monthly (unemployment rate, CPI),
quarterly (GDP), & annual (GDP, Budget).
Quinquennially: Every 5 years (e.g: the census of
manufactures)
Decennially (e.g: Census of population)
Exchange rates daily date for 2 years=730
observation
Inflation rate for Ethiopia, quarterly data
143 for 30
 3.1 Nature of the Time Series data
Gross domestic investment in Ethiopia of annual
data for 40 yrs; 40*1= 40 observations.
Time Series data Vs Cross sectional data
Time Series Cross sectional
• It coming with temporal ordering over
period of time on a single entity. • At a given point of time
• the past can affect the future, but not • a different sample has drawn from
vice versa the population will generally yield
• We have d/t data for d/t samples different values of the independent
• Also viewed as random varaibles; we and dependent variables
do not know what the annual growth in • We have d/t data for d/t year
output will be in Ethiopia during the • The OLS estimates computed from
coming year. i.e, the outcomes of these d/t random samples will generally
random variables are not foreknown. differ and this why we consider OLS
estimators to be random variables.

144
 3.1 Nature of the Time Series data
 A sequence of random variables indexed by time is
called a stochastic process or a time series process.
(“Stochastic” is a synonym for random.)
 When we collect a time series data set, we obtain
one possible outcome, or realization, of the
stochastic process.
 We can only see a single realization, because we
cannot go back in time and start the process over
again.
 This is analogous to cross-sectional analysis where
we can collect only one random sample.)
145
 3.1 Nature of the Time Series data

Important terminology :
 Univariate analysis examines a single data
series.
 Bivariate analysis examines a pair of series.
 The term vector indicates that we are
considering a number of series: two, three, or
more.
 The term ‘‘vector’’ is a generalization of the
univariate and bivariate cases.
146
 3.2 Stationary and non-stationary Stochastic
Processes

 A random or stochastic process is a collection of

random variables ordered in time.
If we let Y denote a random variable, and if it is
continuous, we denote it as Y(t), but if it is discrete,
we denoted it as Yt.
Example of Yt is GDP, CPI, PDI; since most
economic data are collected at discrete points in time,
we use Yt notation.
If we let Y represent GDS, for our data we have
Y1,Y2,Y3,...,Y21,Y22,Y23, where the subscript 1
denotes the first observation (i.e., GDS of 1991/1992)
147
and the subscript 23 denotes the last observation (i.e.,
 3.2 Stationary and non-stationary Stochastic
Processes
Example of Y(t) is electro cardiogram, record
of heart activity.
N.B: Each of these Y’s is a random variable.
A. Stationary Stochastic Processes
 A stochastic process is said to be stationary
“if its mean and variance are constant over
time and the value of the covariance between
the two time periods depends only on the
distance or gap or lag between the two time
periods and not the actual time at which the
covariance is computed. 148
 3.2 Stationary and non-stationary Stochastic Processes

 In the time series literature, such a stochastic

process is known as a weakly stationary, or
covariance stationary, or second-order
stationary, or wide sense, stochastic process.
 To explain weak stationarity, let Yt be a
stochastic time series with these properties:
Mean : E (Y t ) = µ
Variance : Var (Yt ) = E (Yt − µ ) 2 = σ 2

Covariance: γ k = E[(Yt − µ)(Yt +k − µ)]

149
 3.2 Stationary and non-stationary Stochastic
Processes
Where γk, the covariance (or autocovariance) at
lag k, is the covariance between the values of Yt
and Yt+k, that is, between two Y values k periods
apart.
If k=0, we obtain γ0, which is simply the variance
of Y(=σ2); if k=1,γ1 is the covariance between two
adjacent values of Y.
γ 1 = E[(Y1 − µ )(Yt +1 − µ )]

Suppose we shift the origin of Y from Yt to Yt+m

(say, from 1997 to 2002 for our GDS data). Now if
Yt is to be stationary, the mean, variance, and
autocovariances of Yt+m must be the same as
those of Yt. 150
 3.2 Stationary and non-stationary Stochastic
Processes
In short, if a time series is stationary, its mean, variance,
and autocovariance (at various lags) remain the same no
matter at what point we measure them; that is, they are
time invariant.
 Such a time series will tend to return to its mean
(called mean reversion) and fluctuations around this
mean (measured by its variance) will have a broadly
constant amplitude.
If a time series is not stationary in the sense just defined,
it is called a nonstationary time series (keep in mind we
are talking only about weak stationarity).
In other words, a nonstationary time series will have a
time-varying mean or a time-varying variance 151 or both.
 3.2 Stationary and non-stationary Stochastic Processes
Why Stationary time series are important?
 Because if a time series is nonstationary, we can
study its behavior only for the time period under
consideration.
 Each set of time series data will therefore be for
a particular episode.
 As a consequence, it is not possible to generalize
it to other time periods. Therefore, for the
purpose of forecasting, such (nonstationary) time
series may be of little practical value.
A. Non stationary Stochastic Processes
 A special type of stochastic process (or time
series), is called, a purely random, or white noise,
process. 152
 3.2 Stationary and non-stationary Stochastic Processes
 We call a stochastic process purely random if it
has zero mean, constant variance σ2, and is
serially uncorrelated.
 One of the classical example of non stationary
time series is the random walk model (RWM).
 It is often said that asset prices, such as stock
prices or exchange rates, follow a random walk;
that is, they are nonstationary.
 We have two types of random walks
(1) random walk without drift (i.e., no
constant or intercept term) and
(2) random walk with drift (i.e.,
153a constant
 3.2 Stationary and non-stationary Stochastic Processes
1. Random walk without drift
 Suppose ut is a white noise error term with
mean 0 and variance σ2
.
 Then the series Yt is said to be a random walk if
Y t = Y t −1 + u t  shows, the value of Y at time t is
equal to its value at time (t−1) plus a random
shock; thus it is an AR (1) model.
 We can think of Y t = Y t −1 + u t as a regression of
Y at time t on its value lagged one period.
Y1 = Y 0 + u t
Y 2 = Y1 + u 2 = Y o + u 1 + u 2
Y3 = Y 2 + u 3 = Y 0 + u 1 + u 2 + u 3
154
 3.2 Stationary and non-stationary Stochastic Processes
 If the process started at some time 0 with a
Y = Y +  u
value of Y0, we have 1 0 t

E(Y ) = E (Y +  u ) = Y
1 0 t 0

Var (Yt ) = tσ 2

 the mean of Y is equal to its initial, or starting,

value, which is constant, but as t increases, its
variance increases indefinitely, thus violating a
condition of stationarity.
 In short, the RWM without drift is a
nonstationary stochastic process. In practice Y0
is often set at zero, in which case E (Yt) =0.
155
 3.2 Stationary and non-stationary Stochastic Processes
 An interesting feature of RWM is the persistence
of random shocks (i.e., random errors), which is
clear from Y = Y +  u : Yt is the sum of initial
1 0 t

Y0 plus the sum of random shocks.

 As a result, the impact of a particular shock does
not die away.
 For example, if U2=2 rather than U2=0, then all
Yt’s from Y2 onward will be 2 units higher and
the effect of this shock never dies out.
 That is why random walk is said to have an
infinite memory. The implication is that, random
walk remembers the shock forever; that is, it has
infinite memory. 156
 3.2 Stationary and non-stationary Stochastic Processes
☞We can rewrite the above equation as: Y 1 = Y0 +  ut

(Yt − Yt −1 ) = ∆ Yt = u t
☞Where ∆ is the first difference operator. It is
easy to show that, while Yt is nonstationary, its
first difference is stationary. In other words, the
first differences of a random walk time series are
stationary.
2. Random Walk with Drift
Let us modify Y = Y +  u as follows:
1 0 i

157
 3.2 Stationary and non-stationary Stochastic Processes
Where δ is known as the drift parameter. The
name drift comes from the fact that if we write
the preceding equation as .
it shows that Yt drifts upward or downward,
depending on δ being positive or negative. It is
also an AR(1) model.
Following the procedure discussed for random
walk without drift, it can be shown that for the
random walk with drift model :

158
 3.2 Stationary and non-stationary Stochastic Processes
As you can see for RWM with drift, the mean
as well as the variance increases over time.
Thus, it violating the conditions of (weak)
stationary. In short, RWM, with or without
drift, is a nonstationary stochastic process.
The random walk model is an example of what
is known in the literature as a unit root
process.
Unit Root Stochastic Process
Let us write the RWM as:
This model resembles the Markov first-order autoregressive
model that we discussed on autocorrelation.
159
 3.2 Stationary and non-stationary Stochastic Processes
 If ρ=1, becomes a RWM (without drift). If ρ is in
fact 1, we face what is known as the unit root
problem, that is, a situation of nonstationary; we
already know that in this case the variance of Yt
is not stationary.
 The name unit root is due to the fact that ρ=1.
Thus the terms nonstationary, random walk, and
unit root can be treated as synonymous. If,
however, |ρ|<1, that is if the absolute value of ρ
is less than one, then it can be shown that the
time series Yt is stationary in the sense we have
defined it. 160
3.3 Trend Stationary and Difference Stationary Stochastic Processes

If the trend in a time series is completely predictable and not variable, we call it a
deterministic trend, whereas if it is not predictable, we call it a stochastic trend.
 To make the definition more formal, consider the following model of the time series
Yt.
 --------- -------(a)
 Where ut is a white noise error term and where t is time measured chronologically.
 Now we have the following possibilities:
RWM without drift:
Pure random walk : If in (a), β 0 = 0, β1 = 0, β 2 = 1, we get --------(b)=non stationary

……………………….. --------(c)=stationary, Hence, a

RWM without drift is a difference stationary process (DSP).

161
3.3 Trend Stationary and Difference Stationary Stochastic Processes
RWM with drift: Pure random walk withdrift : If in (a), β ≠ 0, β = 0, β = 1, we get
0 1 2

----------(d)-non stationary
-----------(e)—stationary, this means
Yt will exhibit a positive (β1>0) or negative
(β1<0) trend. Such a trend is called a
stochastic trend. Equation (e) is a DSP process
because the nonstationarity in Yt can be
eliminated by taking first differences of the time
series.
Deterministic trend: Purerandomwalk withdrift: If in (a),β ≠ 0, β ≠ 0, β =1, weget
0 1 2

Y t = β 0 + β 1 t + u t , which is called a trend stationary

162 process (TSP).
 3.3 Trend Stationary and Difference Stationary
Stochastic Processes
Although the mean of Yt is β0+β1t, which is
not constant, its variance (=σ2) is constant.
Once the values of β0 and β1 are known, the
mean can be forecasted perfectly. Therefore, if
we subtract the mean of Yt from Yt, the
resulting series will be stationary, hence the
name trend stationary.
This procedure of removing the (deterministic)
trend is called detrending.

163
3.3 Trend Stationary and Difference Stationary
Stochastic Processes
Random walk with drift & deterministic
trend: If in (a) β 0 ≠ 0, β1 ≠ 0, β 2 = 1, we get
-----(f) non stationary
Deterministic trend with stationary AR (1)
component:
If in (a) β 0 ≠ 0, β1 ≠ 0, β 2 < 1, we get
Yt = β 0 + β1t + β 2Yt −1 + U t , which is
stationary around the deterministic trend.

164
 3.4 Integrated Stochastic Process

The random walk model is a specific case of a

more general class of stochastic processes
known as integrated processes.
the RWM without drift is nonstationary, but
its first difference is stationary.
the RWM without drift integrated of order 1,
denoted as I(1).
Similarly, if a time series has to be differenced
twice (i.e., take the first difference of the first
differences) to make it stationary, we call such
a time series integrated of order 2. 165
 3.4 Integrated Stochastic Process
In general, if a (nonstationary) time series has
to be differenced d times to make it stationary,
that time series is said to be integrated of order
d. A time series Yt integrated of order d is
denoted as Yt ∼I(d).
If a time series Yt is stationary to begin with
(i.e., it does not require any differencing), it is
said to be integrated of order zero, denoted by
Yt ∼I(0).
Thus, we will use the terms “stationary time
series” and “time series integrated of order
166
zero” to mean the same thing.
 3.4 Integrated Stochastic Process
Most economic time series are generally I(1); that
is, they generally become stationary only after
taking their first differences.
Properties of Integrated Series
Let Xt, Yt and Zt be three time series
i. If Xt ∼I(0) and Yt ∼I(1),then Zt =(Xt +Yt)=I(1);
that is, a linear combination or sum of stationary
and nonstationary time series is nonstationary.
ii. If Xt ∼I(d), then Zt =(a+bXt)=I(d), where a and
b are constants. That is, a linear combination of
an I(d) series is also I(d). Thus, if Xt ∼I(0), then
167
Zt =(a+bXt)∼I(0).
 3.4 Integrated Stochastic Process
iii. If Xt ∼I(d1) and Yt ∼I(d2), then Zt =(aXt
+bYt)∼I(d2), where d1<d2.
iv. If Xt ∼I(d) and Yt ∼I(d), then Zt =(aXt
+bYt)∼I(d*); d* is generally equal to d, but in
some cases d*<d.

168
 3.5 Tests of Stationarity: The Unit Root Test

A test of stationarity (or nonstationarity)

that has become widely popular over the
past several years is the unit root test.
 ------------------(i)
where ut is a white noise error term
We know that if ρ =1, that is, in the case of
the unit root, (i) becomes a random walk
model without drift, which we know is a
nonstationary stochastic process.
 169
 3.5 Tests of Stationarity: The Unit Root Test

Therefore, why not simply regress Yt on its (one

period) lagged value Yt −1 and find out if the
estimated ρ is statistically equal to 1?
 If it is, then Yt is nonstationary. This is the
general idea behind the unit root test of
stationarit y.
For theoretical reasons, we manipulate (i) as
follows: Subtract Yt −1 from both sides of (i) to
obtain: Yt − Yt −1 = ρYt −1 −Yt −1 + ut
, which can be written
alternatively 170
----------------------------(ii)
 3.5 Tests of Stationarity: The Unit Root Test

Where δ = (ρ−1) and , as usual, is the first-

difference operator.
In practice, therefore, instead of estimating
(i), we estimate (ii) and test:
 the (null) hypothesis that δ = 0, then ρ = 1,
that is we have a unit root (nonstationary).
the t value of the estimated coefficient of Yt
−1 does not follow the t distribution even in
large samples; that is, it does not have an
asymptotic normal distribution. Thus, we
use τ(tau) statistic. 171
 3.5 Tests of Stationarity: The Unit Root Test
1. Dickey–Fuller (DF) test
 Dickey and Fuller have shown that under the
null hypothesis that δ = 0, the estimated t value
of the coefficient of Yt −1 follows the τ(tau)
statistic.
 In the literature the tau statistic or test is known
as the Dickey–Fuller (DF) test
 In conducting the DF test, it was assumed that
the error term ut was uncorrelated.
 the DF test is estimated in three different forms,
that is, under three different null hypotheses.
172
 3.5 Tests of Stationarity: The Unit Root Test

Where t is the time or trend variable.

 Null hypothesis, δ = 0; that is, there is a unit
root—the time series is nonstationary.
 Alternative hypothesis, δ is less than zero; that is,
the time series is stationary.
Decision rule: If /tau statistics/> /tau critical
value/; we rejected the null hypothesis, it means
that Yt is a stationary time series. 173
 3.5 Tests of Stationarity: The Unit Root Test

2. Augmented Dickey–Fuller (ADF) Test

 In this case the ut are correlated, Dickey and
Fuller have developed a test, known as the
Augmented Dickey–Fuller (ADF) test.
 This test is conducted by “augmenting” the
preceding three equations by adding the lagged
values of the dependent variable ΔYt -i.
The ADF test here consists of estimating the
following regression:

174
 3.5 Tests of Stationarity: The Unit Root Test
 In ADF we still test whether δ = 0 and the ADF
test follows the same asymptotic distribution as
the DF statistic, so the same critical values can be
used.
Hypothesis we use under this test is:

Example: the GDP series using one lagged

difference of natural log of GDP of Ethiopia; the
results were as follows: 175
 3.5 Tests of Stationarity: The Unit Root Test

Decision rule: The t(=τ) value of the Yt −1 coefficient (

=δ) is 0.34, but this value in absolute terms is much l ess
than even the 1% and 5% critical τ value of −4.242 and
-3.540 respectively, again suggesting that even after
taking care of possible autocorrelation in the error
term, the Y series is not stationary.
3. The Phillips–Perron (PP) Unit Root Tests
 An important assumption of the DF test is that the error terms ut are independently
and identically distributed. The ADF test adjusts the DF test to take care of possible
serial correlation in the error terms by adding the lagged difference terms of the
regressand.

176
 3.5 Tests of Stationarity: The Unit Root Test
Phillips and Perron use nonparametric statistical methods
to take care of the serial correlation in the error terms
without adding lagged difference terms.
The Phillips-Perron test involves fitting the following
regression: Y t = β 0 + β 1 t + ρ Y t − 1 + u t
Under the null hypothesis that ρ = 0, the PP Z(t) andρ Z(
) statistics have the same asymptotic distributions as the
ADF t-statistic and normalized bias statistics.
 One advantage of the PP tests over the ADF tests is that
the PP tests are robust to general forms of
heteroscedasticity in the error term ut. Another
advantage is that the user does not have to specify a lag
length for the test regression. 177
 3.5 Tests of Stationarity: The Unit Root Test

In some situation, lack of power in both the ADF

and PPtests is widely acknowledged,
Usually ADF yields superior results than PP test,
if the data set has no missing observations and
structural breaks whilst PP test also yields
superior results than ADF test, if the dataset have
some missing observations and have structural
breaks
Decision rule: since tests statistic, Z(t) value is
greater that critical values we reject the null
hypotheses of non stationary. 178
Next assignment

179
 Next assignment
The series steps we should followed to do
with time series analysis:
Unit root stationary if all are
stationary at a level optimum leg
length we run directly VAR model.
Unit root stationary if all stationary
at 1st difference  optimum length
Johanson co integration
VECMIRF& VDF Granger
causality
180
 Lab Session

Use “lecture_3.xls” data to

practice what we learnt in
previous sections
 END OF CHAPTER THREE

THANK YOU VERY MUCH FOR BEING WITH

ME

BEING@COMMITTED!
STAY SAFE!

182
 CHAPTER FOUR

INTRODUCTION TO SIMULTANEOUS
EQUATION MODELS
4.1 Nature of Simultaneous Equation models
4.2 Simultaneity Bias & Inconsistency of OLS
estimators
4.3 Solution to Simultaneous Equations
4.4 Identification problem
4.5 Formal Rules (Conditions) for Identification
4.6 Estimation of Simultaneous Equations
Models183

183
 4.1 Nature of Simultaneous Equation models
 So far we have been discussed by focusing exclusively
on the problems and estimations of a single equation
regression models. In such models, a dependent
variable is expressed as a linear function of one or
more explanatory variables.
 i.e, there was a single dependent variable Y and one or
more explanatory variables, X’s.
 The cause-and-effect relationship in single equation
models between the dependent and independent
variable is unidirectional.
 That is, the explanatory variables are the cause and
the independent variable is the effect.
 But there are situations where such one-way or
unidirectional causation in the function is not
1 meaningful.
 4.1 Nature of Simultaneous Equation models
 This occurs if, for instance, Y (dependent
variable) is not only function of X’s (explanatory
variables) but also all or some of the X’s are, in
turn, determined by Y.
 There is, therefore, a two-way flow of influence
between Y and (some of) the X’s which in turn
makes the distinction between dependent and
independent variables a little doubtful.
 In simultaneous model there is more than one
equation –one for each of the mutually, or
jointly, dependent or endogenous variables.
 The number of equations in such models is equal
to the number of jointly dependent or
endogenous variables involved in the
phenomenon under analysis.
1
 4.1 Nature of Simultaneous Equation models
 Unlike the single equation models, in
simultaneous equation models it is not usually
possible (possible only under specific
assumptions) to estimate a single equation of
the model without taking into account the
information provided by other equation of the
system.
 If one applies OLS to estimate the parameters
of each equation disregarding other equations
of the model, the estimates so obtained are not
only biased but also inconsistent, i.e. even if the
sample size increases indefinitely, the
1 estimators do not converge to their true values.
 4.1 Nature of Simultaneous Equation models
 Example: the classic example of simultaneous
causality in economics is supply and demand.
 Both Prices and quantities adjust until supply
and demand are in equilibrium.
 A shock of demand or supply cause both
prices and quantities to move.
 As well known, the prices P of a commodity
and quantity Q sold are determined by the
intersection of the demand and supply curves
for that commodity.
 Look at the graph of dd and ss from class
discussion(???)
1
 4.2 Model specification

Thus, assuming for simplicity that the demand and

supply curves are linear and adding the stochastic
disturbance term U1 and U2, we may write the
empirical dd and ss function as:
Demand function :
 Qt = β 0 + β1 Pt + β 2Yt + U 1t − − − β1 < 0
d

 Supply function :
Qt = β 0 + β1 Pt + β 2Yt + U 2 t − − − β1 > 0
s

Equilibrium condition: Q t = Q t s d

 Where Qtd= quantity demand

 Qts=quantity supplied
 t=time; β0 , and β1 are the parameters 188
 4.2 Model specification
 B/C of simultaneous dependence between Q and P, then U1t
and Pt, and U2t and pt cannot be independent.
 If u1t in above equation changes b/c changes in other
variables affecting Qtd such as income, wealth and tastes), the
demand curve will shift upward if u1t is +ve and downward if
u1t is –ve.
 Thus, shift in demand curve changes both P and Q.
 Similarly, a change in U2t b/c of weather, import or export
restrictions, etc; will shift the ss curves, again affect both P
and Q.
 B/c of this simultaneous dependence b/c Q and P, u1t and Pt
and u2t and pt cannot be independent. Thus, a regression of
Q and P as in above equation would violate an important
assumptions of the classical linear regression model; namely
the assumption of no correlation b/n the explanatory
variable(s) and the disturbance term.
189
 ,

In simultaneous equation models variables are

classified as endogenous and exogenous.
 Endogenous variables: are variables that are
determined by the economic model (within the
system) and
Exogenous variables: are those determined from
outside of the system.
Exogenous variables are also called predetermined.
Since the exogenous variables are predetermined,
they are supposed to be independent of the error
terms in the model/ non stochastic.
are exogenous variables, lagged exogenous variables and
lagged endogenous variables. Predetermined variables are
non-stochastic and hence independent of the disturbance
terms. , X t , X t −1 and Yt −1are regarded as predetermined
(exogenous) variables.
1
 4.3Definitions of Some Concepts

Structural models: A structural model

describes the complete structure of the
relationships among the economic variables.
 Structural equations of the model =
endogenous variables+ exogenous variables +
disturbances (random variables).
The parameters of structural model express
the direct effect of each explanatory variable
on the dependent variable.
The Variables not appearing in any function
explicitly may have an indirect effect and is taken into
191 account by the simultaneous solution of the system.
 4.3Definitions of Some Concepts

Reduced form of the model: The reduced form

of a structural model is the model in which the
endogenous variables are expressed a function
of the predetermined variables and the error
term only.
Example: We may write the empirical demand-
and-supply functions as

192
 4.3Definitions of Some Concepts
Example: The following simple Keynesian model
of income determination can be considered as a
structural model.
 =------------(1)
 ------------(2)
where: C=consumption expenditure; Z=non-
consumption expenditure ; Y=national income; C
and Y are endogenous variables while Z is
exogenous variable.
Find the reduced form of the above structural
model. Since C and Y are endogenous variables
and only Z is the exogenous variables, we have to
express C and Y in terms of Z.
193
 4.3Definitions of Some Concepts

To do this substitute Y=C+Z into equation (1).

 -------(3)
Substituting again (3) into (2) we get;
α  1  U
 Y= +  Z + ----------(4)
1− β 1− β  1− β

Equation (3) and (4) are called the reduced

form of the structural model of the above. We
194
can write this more formally as:
 4.3 Definitions of Some Concepts

Parameters of the reduced form measure the total

effect (direct and indirect) of a change in
exogenous variables on the endogenous variable.
For instance, β in the above reduced form
equation(1), 1− β  measures the total effect of a
unit change in the non-consumption expenditure
on consumption. This total effect is β , the
direct effect, times  1  ,the indirect effect.
195 1− β 
 4.4 Inconsistency and Simultaneity Bias of OLS Estimators

 Biasedness:
 The two-way causation in a relationship leads to violation
of the important assumption of linear regression model,
i.e. one variable can be dependent variable in one of the
equation but becomes also explanatory variable in the
other equations of the simultaneous-equation model.
 In this case E[XiUi] may be different from zero. To
show simultaneity bias, let’s consider the following simple
simultaneous equation model.
Y = α 0 + α1X + U 

X = β0 + β 1Y + β 2 Z + V 
X = f (Y )
 Y = this shows that the 2 way causation in a
f (X )
relationship leads to violations of the important
196
assumptions linear regression model
  Suppose that the following assumptions hold,
Ε (U ) = 0 , Ε (V ) = 0
Ε (U 2 ) = σ 2
u , Ε (V 2 ) = σ 2
u

Ε (U i U j ) = 0 , Ε (V i V j ) = 0 , also Ε (UiVi ) = 0 ;

where X and Y are endogenous variables and

Z is an exogenous variable.
The reduced form of X of the above model is
obtained by substituting Y in the equation of
X.
X = β0 + β1 (α0 + α1 X + U) + β2 Z + V
β 0 + α 0β1  β 2   β 1U + V 
X = +   Z +  
197 1 − α 1β1  1 − α 1β1   1 − α 1β1 
 4.4 Inconsistency and Simultaneity Bias of OLS Estimators

 Applying OLS to the first equation of the

above structural model will result in biased
estimator because cov(X U ) = Ε( X U ) ≠ 0 . Now, let’s
i i i j

proof whether this expression.

 β1  β 1σ u2
=  Ε (U ) =
2
≠0
 1 − α 1β1  1 − α1 β1
 That is, covariance between X and U is not
zero. As a consequence, if OLS is applied to
each equation of the model separately the
coefficients will turn out to be biased. Now,
let’s examine how the non-zero co-variance of
the error term and the explanatory variable
will lead to biasness in OLS estimates of the
198
parameters.
 Estimators
 Consistency Problems: An estimator is said to
be consistent if its probability limit is equal to
its population value.
 Inconsistent estimates
β 2 σ 2U
1 − β1β 2
p lim( βˆ 1 ) = β 1 +
σu 2

β2  σv 2 
β1 = β1 +
ˆ  
1 − β 1β 2  σ X 2


199
 4.5 Solution to the Simultaneous Equations
 The obvious solution is to apply other methods of
estimation w/c gives better estimates of parameters.
 1. the reduced form method or indirect least
squares (ISLS)
 2. the method of instrumental variables
 3. two stage least squares (2SLS)
 4. limited information maximum likelihood (LIML)
 5. the mixed estimation
 6. Three stage least squares
 7. Full information maximum likelihood (FIML)

200
 4.5 Solution to the Simultaneous Equations
 N.B: 1-5---we can applied to one equation at a time,
and 6-7----the systems methods b/c they are applied
to all equations of the system simultaneously.

How to estimate the reduced form parameters?

 The estimates of the reduced from coefficients (π’s )
may be obtained in two ways.
1. Direct estimation of the reduced coefficients
by applying OLS.
2. Indirect estimation of the reduced form
coefficients.

2
 4.6 Direct estimation of the reduced form coefficients
 Direct Method: Express the three endogenous
variables(Ct , It , and Yt ) as functions of the
two predetermined variables (Gt, andYt-1)
directly using π’s as the parameters of the
reduced form model as follows.
Ct = π11Yt-1 + π12Gt + V1
It , =π21Yt-1 + π22Gt +V2
Yt =π31Yt-1 + π32Gt + V3
Note: π11 , π12 , π21 , π22 , π31 , and π32 are
reduced from parameters.

2
4.6 Direct estimation of the reduced form coefficients
 The reduced form π ’s may be estimated by
the method of least- squares –no restriction
(LSNR).
 This means we can apply OLS to reduced
form equation because we express all the
endogenous variables in terms of
exogenous variables.

203
 4.6 Direct estimation of the reduced form coefficients

 This method of obtaining the π 's is called least

squares no restriction (LSNR) because it doesn't take
into consideration any information on the structural
parameters.

 In this method what required is knowledge of the

predetermined variables appearing in the system not
about the coefficients of structural questions.

204
 4.7 Indirect estimation of the reduced form
coefficients
 It is known that there is a relationship
between the reduced form coefficients &
the structural parameters (explained in the
table).
 Therefore, to obtain values of coefficients
estimate the structural parameters by any
appropriate econometric techniques and
then substitutes these estimates in to the
system of parameters relationships to
obtain indirectly.
 This indirect method involved three steps.

205
 4.8 Indirect estimation of the reduced form coefficients

 1st step: Solve the system of endogenous variables so

that each equation contains only predetermined
explanatory variables.
 2nd step: Obtain the estimates of the structural
parameters by any appropriate econometric method.
 3rd step: Substitute the estimates of β's and γ's in to
the system of parameters relations to find the
estimates of the reduced form coefficients.
Advantage of indirect estimation of the reduced-form
coefficients
 Though it is complicated, it has a very good
importance.
a) The derivation of parameters like, π's, β's & α's is
2 more efficient because in this way we take in to
 4.9 Recursive models

 A model is called recursive if its structural

equations can be ordered in such a way that:
 the first equation includes only the predetermined
variables in the right hand side.
 the second equation contains predetermined variables and
the first endogenous variable (of the first equation) in the
right hand side and so on.
 The special feature of recursive model is that
its equations may be estimated, one at a time,
by OLS without simultaneous equations bias.

207
 4.9 Recursive models
 OLS is not applicable if there is interdependence
between the explanatory variables and the error term.
 In the simultaneous equation models, the endogenous
variables may depend on the error terms of the
model.
 Hence, the OLS technique is not appropriate for
estimation of an equation in a simulations equations
model.
 However, in a special type of simultaneous equations
model called Recursive, Triangular or Causal model,
the use of OLS procedure of estimation is appropriate.
 Consider the following three equation system to
208
understand the nature of such models:
 4.9 Recursive models

 Note that:

 In the above illustration, the X’s and Y’s are exogenous and
endogenous variables respectively.
 The disturbance terms follow the following assumptions.

 The above assumption is the most crucial

assumption that defines the recursive model.

2
 4.9 Recursive models
 If this does not hold, the above system is no
longer recursive and OLS is also no longer
valid.
 The first equation of the above system
contains only the exogenous variables on the
right hand side.
 Since by assumption, the exogenous variable is
independent of U1 , the first equation satisfies
the critical assumption of the OLS procedure.
 Hence, OLS can be applied straight forwardly
to this equation.
210
 4.9 Recursive models
 Let us build a hypothetical recursive model for
an agricultural commodity, say wheat.
 The production of wheat =Y1; , may be assumed
to depend on exogenous factors: X2 = climatic
conditions; and X3=last season’s price. The retail
price =Y2 may be assumed to be the function of
production level Y1= and exogenous factor X4=
disposable income.
 Finally, the price obtained by the producer = Y3
can be expressed in terms of the retail price; Y2
and exogenous factor; Xj= the cost of marketing
the producer.
 The relevant equations of the model may be
211 described as under:
 4.9 Recursive models

Y1 = α 1 + α 2 X 2 + α 3 X 3 + U 1

Y2 = α 4 + β 1Y 1 + α 5 X 4 + U 2

Y3 = α 6 + β 2Y 2 + α 7 X 5 + U 3

 In the first equation, there are only exogenous

variables and are assumed to be independent of U1.
 In the second equation, the causal relation between
Y1and Y2 is in one direction.
 Also Y1 is independent of U2 and can be treated just
like exogenous variable.
 Similarly since Y2 is independent of U3 , OLS can be
applied to the third equation.
212
 Thus, we can rewrite the above equations as follows:
 4.9 Recursive models

Y1 − α 1 − α 2 X 2 − α 3 X 3 = U 1
− β 1Y1 + Y2 − α 4 − α 5 X 4 = U 2
− β 2Y2 + Y3 − α 6 − α 7 X 5 = U 3
 We can again rewrite this in matrix form as
follows:

 The coefficient matrix of endogenous variables is

thus a triangular one; hence recursive models
213
are also called as triangular models.
 4.9 Problems of simultaneous equation models

 Simultaneous equation models create three distinct

problems. These are:
 Identification of each equation of the model
 Mathematical completeness of the model
 any model is said to be (mathematically)
complete only when it possesses as many
independent equations as endogenous variables.
 In other words if we happen to know values of
disturbance terms, exogenous variables and
structural parameters, then all the endogenous
variables are uniquely determined.
 Statistical estimation of each equation of the model
214
 4.10 The identification problem
Of the three problems, we are going to discuss the first
problem (the identification problem) in the following
section.
The identification problem
 In simultaneous equation models, the Problem of
identification is a problem of model formulation; it
does not concern with the estimation of the model.
 The estimation of the model depends up on the
empirical data and the form of the model.
 If the model is not in the proper statistical form, it may
turn out that the parameters may not uniquely
estimated even though adequate and relevant data are
available.
 In a language of econometrics, a model is said to be
identified only when it is in unique statistical form to
enable us to obtain unique estimates of its parameters
from the sample data.
2
 4.10 Identification Problem
The identical concept concerns with whether the
numerical estimates of structural equations can be
obtained from the estimated reduced form coefficients.
Look at a simple Keynesian model, to illustrate the
problem of identification (look at “Introduction to
Econometrics: theory and practice with Stata” by
Tesfaye E,)
An identification may be either exactly (fully or just
identified) or over identified or under identification.
A. Under identification (SEP>REP)
• It occurs when the parameters of structural equation is
higher than reduced form parameters.
• If the coefficients of the structural equations are
greater than the coefficients of the reduced form, then
we can say that the equation is under identified.
2
 Under identification (SEP>REP)
Example: look at f/wing demand and supply equations

2
 Under identification (SEP>REP)

218
 Under identification (SEP>REP)
 Equation number 10.37 and 10.39 were the two reduced form
equations derived from the structural equations number 10.32
& 10.33.
 Now if you compare the number of structural equation
coefficients (α0, α1, β0 and β1) are four where as from the
structural equations we have only two coefficients (π0 and π1).
 The coefficients of reduced form contain the coefficients of the
structural equations i.e α0, α1, β1 and β2 are found in π0 and
π1.
 But how we can find the values of α0, α1, β1 and β2 from π0
and π1. It is an ambiguous question??
 Since it is not possible to find these values from π0andπ1 or
the coefficients of the structural equations are greater than
the coefficients of the reduced form then we can say that the
equation is under identified and we can not compute four
structured coefficients from two reduced coefficients.
219
 Why under identification is happened?

 The reason to have under identified function in

the previous demand and supply function was
that:
 The same variables P and Q are appearing in
both functions (only endogenous variables in
both equation)
 There is no additional information.

220
 Exact /Just/ Identification (SEP=REP)
 It occurs when structural coefficients are equal
to reduced form coefficients.
 Now let’s incorporate additional variable in the
demand function in order to solve the above
problem.

221
 Exact /Just/ Identification (SEP=REP)

222
 Exact /Just/ Identification (SEP=REP)

223
 Exact /Just/ Identification (SEP=REP)

224
 Exact /Just/ Identification (SEP=REP)
 But in case of the demand function α0, α1, and α2, is 3
structural coefficients but in reduced form of equation
the coefficients are two.
 Since in the demand function the coefficient of the
reduced form (10.45) is less than the coefficients of the
structural equation (10.40).
 We can concluded that the demand function is under
identified (π2,π3) are less than α0,α1,and α2).
 But in case of supply function π2,π3 are equal to β0 , β1
then it is just identified.
 In conclusion, we can say that the supply function is
identified but the demand function is not identified on the
basis of this one can say that the system as a whole is not
identified.
225
 Over identification (SEP<REP)
 It occurs when the coefficients (parameters) of
structural equation is less than the coefficients
(parameters) of reduced forms.
 Let’s modify the demand function by
incorporating wealth (R) and supply function by
incorporating the lagged price we will have the
following equation.

226
 Over identification (SEP<REP)

227
 Over identification (SEP<REP)
 From equation number 10.47 and 10.48 we have
seven structural coefficients but in equation 10.49
and 10.45 we have eight reduced form coefficients.
 Since the coefficients of reduced form coefficients
are greater than the reduced form coefficients we
can say that the system as a whole is over identified.
 A function (an equation) belonging to a system of
simultaneous equations is identified if it has a
unique statistical form, i.e. if there is no other
equation in the system, or formed by algebraic
manipulations of the other equations of the system,
contains the same variables as the
function(equation) in question.
228
 4.11 Formal Rules (Conditions) for Identification
 Identification problems do not just arise only on
two equation-models.
 Using the above procedure, we can check
identification problems easily if we have two or
three equations in a given simultaneous equation
model.
 However, for ‘n’ equations simultaneous equation
model, such a procedure is very cumbersome.
 In general, for any number of equations in a given
simultaneous equation, we have two conditions
that need to be satisfied to say that the model is in
general identified or not.
 In the following section we will see the formal
conditions for identification.
229
 Formal Rules (Conditions) for Identification
 Actually the term ‘identification’ was originally used to
denote the possibility (or impossibility) of deducing the
values of the parameters of the structural relations from
a knowledge of the reduced form parameters.
 However, we think that the reduced form approach is
conceptually confusing and computationally more
difficult than the structural model approach, because it
requires the derivation of the reduced from first and then
examination of the values of the determinant formed
form some of the reduced form coefficients.
 The reduced form equation is time consuming process.
 The structural form approach is simpler and more useful.
 Thus, the so called order and rank conditions of
identification lighten the task by providing a systematic
230way.
 Formal Rules (Conditions) for Identification
 There are two conditions which must be fulfilled for
an equation to be identified. These are:
 1. the order condition for identification
 2. the rank condition for identification
 The identification of a system means the
identification of each question.
 The parameters identification in any equations
means there is unique value for each parameter in
equations.
 Equation is under identified when its statistical
form is not unique/ When one or more of its
equation of the model are identified we can say that
231the system as a whole is under identified.
 Formal Rules (Conditions) for Identification
 Equation identified: in this case a system is identified when all
the equations are identified.
 In identified system we can have two options:
 if an equation is under identified it is impossible to
estimate all its parameters using any econometric
techniques. However, if the equation is identified its
coefficients (parameters) can be statistically estimated.
 If the equation is exactly identified appropriate method
for estimation is the method of Indirect Least Square
(ILSM).
 If the equation is over identified, ILS will not give unique
estimates of the parameters b/c it will not yield unique
estimates of structural parameters.
 In this case we use various methods. These are:
 2SLS (Two Stages Least Squares) or
232  MLM(Maximum Likely hood methods)
 A. The order condition for identification
 This condition is based on a counting rule of the variables
included and excluded from the particular equation.
 It is a necessary but not sufficient condition for the
identification of an equation.
 The order condition may be stated as follows.
 For an equation to be identified the total number of variables (endogenous
and exogenous) excluded from it must be equal to or greater than the
number of endogenous variables in the model less one.
Let, G = total number of equations (= total number of
endogenous variables)
K= number of total variables in the model (endogenous
and predetermined)
M= number of variables, endogenous and exogenous,
included in a particular equation/ in a specific equation.
233
 A. The order condition for identification
 Then the order condition for identification may
be symbolically expressed as:

 The guidelines is that:

 If (K-M)> (G-1); the equation is identified.
 If (K-M)= (G-1); the equation is just/exactly
identified.
 If (K-M)< (G-1); the equation is under identified.
 If (K-M)>(G-1); the equation is over identified.
234
 A. The order condition for identification
Example 1:
Q d = α + α 1 P1 + α 2 I + U 1 − − − − ( 1 )
Q s = β 0 + β 1 P1 + U 2 − − − − − − − (2)
• Take the dd equation
• G= total number of equations/ total number of
endogenous variables=2
• K=total number of exogenous and endogenous
variables in equation (1), i.e., in demand equation=3
• The solution is that: (K-M)________(G-1)
• (3-3)____(2-1)=0<1, we conclude that the demand
equation is under identified.
235
 A. The order condition for identification
Take the ss equation
 Given: G=2; K=3; M=2;
 Solution:
 K-M-------------G-1
 (3-2)--------------(2-1)
 1=1 from these we can conclude that the
supply function is exactly identified.
Example 2: Given the structural model and
determine whether the equation are identified or
236 under identified.
 A. The order condition for identification
y1 = 3 y 2 − 2 x1 + x 2 + u 1 − − − − 1
y 2 = y3 + x3 + u 2 − − − − − − − −2
y 3 = y1 − y 2 − 2 x3 + u 3 − − − − − 3

• Take equation (1);

• Given; M (endogenous and exogenous variables) in this
specified equation is 4 (y1, y2, x1 and x2); K=6; G=3;
• (K-M)----------(G-1)
• 6-4-------------(3-1)
• 2=2-- this equation is identified and it is exactly
identified.
237
 A. The order condition for identification
• Take equation (2);
• Given; M (endogenous and exogenous variables)
in this specified equation is 3 (y2, y3, & x3);
K=6; G=3;
• (K-M)----------(G-1)
• 6-3-------------(3-1)
• 3>2-- this equation is identified and it is over
identified.

238
 A. The order condition for identification
• Take equation (3);
• Given; M (endogenous and exogenous variables) in
this specified equation is 4 (y3, y1, y2 and x3); K=6;
G=3;
• (K-M)----------(G-1)
• 6-4-------------(3-1)
• 2=2-- this equation is identified and it is exactly
identified.
 Example 3: if a system contains 10 equations with
15 variables, ten endogenous and five exogenous, an
equation containing 11 variables is not identified,
while another containing 5 variables is identified.
239
 A. The order condition for identification
 For 1st equation we have:
G=10; K=15; M=11;
Order condition:
 K-M> G-1
 15-11> 10-1
 4<9 that is the order condition is not
satisfied.
 For the 2nd equation we have:
 G=10; K=15; M=5
 Order condition:
 (K-M)> (G-1); 10>9-----the order conditions
240satisfied.
 B. The rank condition for identification
 The rank condition states that: in a system of G
equations any particular equation is identified if
and only if it is possible to construct at least one
nonzero determinant of order (G-1) from the
coefficients of the variables excluded from that
particular equation but contained in the other
equations of the model.
 The practical steps for tracing the identifiablity of
an equation of a structural model may be outlined
as follows.
 Firstly, write the parameters of all the equations of
the model in a separate table, noting that the
parameter of a variable excluded from an equation
241is equal to zero.
 B. The rank condition for identification

 Where y’s are the endogenous variables and

 x’s are the exogenous variables

 Ignoring the random disturbance the table of the

242parameters of the model is as follows:
 B. The rank condition for identification

Secondly, Strike out the row of coefficients of the

equation which is being examined for identification.
For example, if we want to examine the
identifiability of the second equation of the model
we strike out the second row of the table of
coefficients.
243
 B. The rank condition for identification
 Thirdly, Strike out the columns in which a non-zero coefficient of the
equation being examined appears.
 Table of structural parameter
Equatio Y1 Y2 Y3 X1 X2 X3
ons

1st equ. -1 3 0 -2 1 0

2nd equ. 0 1 1 0 0 1

3rd equ. 1 1 1 0 0 2

 By deleting the relevant row and columns we are left with the
coefficients of variables not included in the particular equation, but
contained in the other equations of the model.
 For example, if we are examining for identification the second
equation of the system, we will strike out the second, third and the
sixth columns of the above table, thus obtaining the following tables.
244
 B. The rank condition for identification

245
 B. The rank condition for identification
 Fourthly, form the determinant(s) of order (G-1) and examine
their value.
 Guide line:
 If at least one of these determinants is non-zero, the equation
is identified.
 If all the determinants of order (G-1) are zero, the equation is
under identified.
 In the above example of exploration of the identifiability of the
second structural equation we have three determinants of order
(G-1)=3-1=2. They are:

246
 B. The rank condition for identification

247
 B. The rank condition for identification
 The identification of a function is achieved by
assuming that some variables of the model have
zero coefficient in this equation, that is, we
assume that some variables do not directly affect
the dependent variable in this equation.
 This, however, is an assumption which can be
tested with the sample data.
 We will examine some tests of identifying
restrictions in a subsequent section.
 Some examples will illustrate the application of
the two formal conditions for identification.
248
 B. The rank condition for identification
Example:

2
 Estimation of Simultaneous Equations Models

 To estimate the simultaneous equation models we adopt two

approaches.
 The first one is single equation method, also known as limited
information method.
 In this single equation method we estimate each question in the system
individually.
 The second one is system methods also known as full information
methods.
 In this case we estimate all equations in the model simultaneously.
 In practice system methods are not commonly used for variety of
reasons rather, single equation methods are often used.
 The major single equation methods applied in the estimation of
simultaneous equation methods are:
 1. Ordinary least squares (OLS)
 2. Indirect least squares (ILS)
 3. Two stage least squares (2SLS)

250
 1. Ordinary Least Squares

 We have seen that applying OLS on

simultaneous equation produce bias &
inconsistent parameters.
 But there is one situation OLS can be applied
appropriately even in the context of
simultaneous equation.

251
 1. Ordinary Least Squares

 In equation 10.51 the endogenous variables appear in the left & the
exogenous variables in the right hand side.
 Hence, OLS can apply straight forwardly to this question given all the
assumptions of OLS holds true.
 In equation 10.52 we can apply OLS provided that Y1 & U2 are
uncorrelated.
 Again we can apply OLS to the last equation if both Y1 & Y2 are
uncorrelated with U3.
 In this recursive system OLS can be applied to each equation
separately & we do not face a simultaneous equation problem.
 The reason for this is that clear, because there is no interdependence
among the endogenous variables.
 Thus, Y1 affect Y2 influence Y3 without being influenced by Y3.
 In other words each equation exhibits a unilateral causal dependence.

252
 2. Indirect least square (ILS method)

 ILS is applicable only for just/exact identified

equations [(K-M) = (G-1)].
 The method of obtaining the estimates of the
structural coefficients using OLS of the reduced
form coefficients is known as the method of
(ILS) indirect least squares & the estimates
obtained are known as the indirect least squares
estimates.
 Indirect Least Square method involves the
following Steps
2
 2. Indirect least square (ILS method)

1st:We first obtain the reduced form equation from

the structural equations. i.e. explaining the
endogenous variables as a function of explanatory
(exogenous variables) & a stochastic term.
2nd: Apply OLS to the reduced- form equations
individually. In this case the exogenous variables
are uncorrelated with the stochastic term.
3rd: we obtain estimates of the original structural
coefficients from the estimated reduced-form
coefficients obtained in step two. ILS derives from
the fact that structural coefficients are obtained
indirectly from the OLS estimates of the reduced
form coefficients.
254
 3. Two-Stage Least Squares (2SLS) Method

 The 2SLS procedure is generally applicable for

estimation of over-identified equations as it
provides unique estimators.
 Two-Stage Least Squares (2SLS) Method
involves the following steps.
st
Ŷ
1 : Estimate the reduced form equations by
OLS and obtain the predicted .
2nd: Replace theŶ right hand side endogenous
variables in the structural equations by the
corresponding and estimate them by OLS.
255
 Lab Session

Use “lecture_4.xls” data to

practice what we learnt in
previous sections
 END OF CHAPTER FOUR

Thank you very much for being with me

for a while!

Stay Safe!
257
 CHAPTER FIVE

INTRODUCTION TO PANAL DATA

REGRESSION MODELS
 5.1 Introduction

 The types of data that are generally available for empirical

analysis, namely, time series, cross section, and panel.
 In time series data we observe the values of one or more
variables over a period of time (e.g., GDP for several quarters or
years).
 In cross-section data, values of one or more variables are
collected for several sample units, or entities, at the same point in
time (e.g., crime rates for 9 regions in Ethiopia for a given year).
 In panel data, the same cross-sectional unit (say a family or a firm
or a state) is surveyed over time.
 In short, panel data have space as well as time dimensions.
 5.1 Introduction
A panel of data consists of a group of cross-sectional
units (people, households, firms, states, countries) who
are observed over time. We will often refer to such units
as individuals, with the term ‘‘individual’’ being used
generically, even when the unit of interest is not a
person.
 Let us denote the number of cross-sectional units
(individuals) by N, and number of time periods in
which we observe them as T.
 Panel data comes in several different ‘‘flavors,’’ each
of which introduces new challenges and opportunities.
 5.1 Introduction
Peter Kennedy1; describes the different types of
panel data sets as:
‘‘Long and narrow,’’ with ‘‘long’’ describing the time
dimension and ‘‘narrow’’ implying a relatively small
number of cross sectional units.
 ‘‘Short and wide,’’ indicating that there are many
individuals observed over a relatively short period of
time.
 ‘‘Long and wide,’’ indicating that both N and T are
relatively large.
A ‘‘long and narrow’’ panel may consist of data on
several firms over a period of time.
 5.1 Introduction

Data on 221 State of Oromia high schools in

2014 and again in 2015, for 442 observations
total; Data on 9 states of Ethiopia, each state is
observed in 3 years, for a total of 27
observations;
Data on 500 individuals, in five different
months, for 2500 observations total.
 5.2 Other names of Panel Data
Pooled data:- pooling of time series and cross-
sectional observations.
Cross-sectional time-series data:- Combination
of time series and cross-section data.
Micro-panel data, longitudinal data: A study
over time of a variable or group of subjects.
Panel data (also called longitudinal data) refers
to data for n different entities observed at T
different time periods.
 5.3 Balanced and unbalanced data

When describing the cross sectional data it was useful

to use a subscript to denote the entity; for instance, Yi
referred to be the variable Yi for the ith entity.
When describing panel data, we need some additional
notations to keep track of both the entity and the time
period.
This is done by using two subscripts rather than one:
The first, i refres to the entity., and the second, t, refers
to the time period of the observation.
Thus, Yit denotes the variable Y observed for the ith of
n entities in the i th of T periods.
 5.3 Balanced and unbalanced data

 Some additional terminology associated with

panel data describes weather some observations
are missing.
 A balanced has all its observations, that is, that
variables are observed for each entity and each
time period.
 A panel that has some missing data for at least
one time period for at least one entity is called
an unbalanced panel..
 5.4 Why we use panel data?
Panel data give:
 more informative data
more variability
 less collinearity among the variables
 more degrees of freedom and more efficiency. Time-
series studies are plagued with multi-collinearity.
Panel data are better able to:
 identify and measure effects that are simply not detectable in
pure cross-section or pure time-series data.
study the dynamics of adjustment.
 Cross-sectional distributions that look relatively stable hide a
multitude of changes.
 5.4 Why we use panel data?
Panel data allows you to:
 Controlling for individual heterogeneity.
Control for variables you cannot observe or
measure like cultural factors or difference in
business practices across companies; or variables
that change over time but not across entities (i.e.
national policies, federal regulations, international
agreements, etc.)
This is, it accounts for individual heterogeneity.
Time-series and cross-section studies not controlling
this heterogeneity run the risk of obtaining biased
results.
 5.4 Why we use panel data?
 Panel data models allow us to construct and test more
complicated behavioral models than purely cross-section or
time-series data.
 For example, technical efficiency is better studied and
modeled with panels.
Micro panel data gathered on individuals, firms and
households may be more accurately measured than
similar variables measured at the macro level.
Biases resulting from aggregation over firms or
individuals may be reduced or eliminated.
 5.6 Limitation of panel data
1. Design and data collection problems: These include:
 problems of coverage (incomplete account of the population
of interest)
nonresponse (due to lack of cooperation of the respondent or
because of interviewer error)
recall (respondent not remembering correctly)
 frequency of interviewing
interview spacing
reference period
the use of bounding and
time-in-sample bias.
 5.6 Limitation of panel data
2. Distortions of measurement errors:
Measurement errors may arise because of
faulty responses due to unclear questions
 Memory errors
Deliberate distortion of responses (e.g.
prestige bias)
Inappropriate informants
Misrecording of responses and interviewer
effects.
 5.6 Limitation of panel data
 3. Selectivity problems. These include:
(a) Self-selectivity: People choose not to work
because the reservation wage is higher than
the offered wage.
In this case we observe the characteristics of
these individuals but not their wage.
 Since only their wage is missing, the sample is
censored.
 However, if we do not observe all data on these
people this would be a truncated sample.
 5.7 Limitation of panel data

(b) Nonresponse: This can occur at the initial wave of the

panel due to refusal to participate, nobody at home,
untraced sample unit, and other reasons.
Item (or partial) nonresponse occurs when one or
more questions are left unanswered or are found not
to provide a useful response.
(c) Attrition: While nonresponse occurs also in cross-
section studies, it is a more serious problem in panels
because subsequent waves of the panel are still subject
to nonresponse.
Respondents may die, or move, or find that the cost
of responding is high.
 5.7 Limitation of panel data
4. Short time-series dimension:
 Typical micro panels involve annual data covering a
short time span for each individual. This means that
asymptotic arguments rely crucially on the number
of individuals tending to infinity.
Increasing the time span of the panel is not without cost
either. In fact, this increases the chances of attrition
and increases the computational difficulty for limited
dependent variable panel data models.
5. Cross-section dependence: Macro panels on
countries or regions with long time series that do not
account for cross-country dependence may lead to
misleading inference.
 5.8 Notation (Model specification ) for panel data
 Panel data consist of observations on the same n entities at two
or more time periods T.
 If the data set contains observations on the variables X and Y,
then the data are denoted

 Where the first subscript, i, refers to the entity being observed,

and the second subscript, t, refers to the date at which is
observed.
 A double subscript is used to distinguish entities (states, family,
country, individuals, etc.) and time periods.
 Consider the following simple panel data regression model:
 5.9 Estimation of Panel Data Regression

Where i= entity (state); n=number of entities, so

i=1,…,n; t= time period (year, month, quarter, and so
on); T= number of time periods, so that t=1,…, T
Panel data with k regressors:

We have three models to estimate panel data

1. Pooled model 2. Fixed Effect model 3. Random
effect model
A pooled model is one where the data on different
individuals are simply pooled together with no
provision for individual differences that might lead to
different coefficients.
 5.10 Pooled data
 For an equation with two explanatory variables X1
and X2, a pooled model can be written as
 5.10 Pooled data
for ( β0 ,β1 ,β2 ) has all its desirable properties.
It is consistent, and the usual t and F statistics are valid
in large samples for hypothesis testing and interval
estimation.
 If we also assume X1 and X2 are nonrandom, the least
squares estimator is the minimum variance linear
unbiased estimator in finite samples.
We will focus on large sample properties, however,
because it is typically unrealistic to assume X1 and X2
are nonrandom, and our sample sizes are usually large.
Hence, the main weakness of this model is it doesnot
capture the hetrogenity among the entity.
 5.11 The Fixed Effects (Entity/Time Fixed) Model

In the previous section we saw that one way to

recognize the existence of individual characteristics in a
panel data model is to allow individual errors in
different time periods to be correlated.
A second way is to relax the assumption that all
individuals have the same coefficients. Extending the
model in (11.4) along these lines, we can write

An i subscript has been added to each of the subscripts,

implying that ( ) can be different for each
individual. This model is a legitimate panel data model,
but it is not suitable for panels that are short and wide.
 5.11 The Fixed Effects (Entity/Time Fixed) Model

You may apply entity fixed effects regression when you

want to control for omitted variables that differ among
panels but are constant over time.
On the other hand, if there are unobserved effects that
vary across time rather than across panels, we apply
time fixed effects regression model.
Use fixed-effects (FE) whenever you are only interested
in analyzing the impact of variables that vary over
time.
FE explore the relationship between predictor and
outcome variables within an entity (country, person,
company, etc.).
 5.11 The Fixed Effects (Entity/Time Fixed) Model

 Each entity has its own individual characteristics that may or

may not influence the predictor variables (for example being a
male or female could influence the opinion toward certain issue
or the political system of a particular country could have some
effect on trade or GDP or the business practices of a company
may influence its stock price).
 When using FE we assume that something within the individual
may impact or bias the predictor or outcome variables and we
need to control for this.
This is the rationale behind the assumption of the correlation
between entity’s error term and predictor variables.
 FE removes the effect of those time-invariant characteristics
from the predictor variables so we can assess the predictors’ net
effect.
5.11 The Fixed Effects (Entity/Time Fixed) Models

 Another important assumption of the FE model is that those

time-invariant characteristics are unique to the individual and
should not be correlated with other individual characteristics.
 Each entity is different therefore the entity’s error term and the constant
(which captures individual characteristics) should not be correlated with
the others.
 If the error terms are correlated then FE is no suitable since
inferences may not be correct and you need to model that
relationship (probably using random-effects.)
 Think of the following two variables panel regression model in
fixed effect form:
5.11 The Fixed Effects (Entity/Time Fixed) Models

Average the data across time, by summing both

sides of the equation and dividing by T.
5.11 The Fixed Effects (Entity/Time Fixed) Models
In the last line of (10.16) note that the intercept
parameter has fallen out. These data are said to be
in ‘‘deviation from the individual’s mean’’ form, and if
we repeat this process for each individual, then we have
a transformed model.
 Sum….FEM
 FE:
intercept may differ across entity, but intercepts does not
vary overtime, that is it is time invariant.
Error terms are not correlated and each entity is
different
 the individual differences in the intercept values of
each entity are reflected in the error term.

 One way to take into account the “individuality” of

each entity or each cross sectional unit is to let the
intercept vary for each entity but still assume that the
slope coefficients are constant a cross firms.
 If T (is the number of time series data) is large and N
(the number of cross- sectional units) is small. The
choice is based on computational convince and FEM
 5.12 The Random Effects Model

If you believe that some omitted variables may be

constant over time but vary among panels, and others
may be fixed among panels but vary over time, then you
can apply random effects regression model.
 Random effects assume that the entity’s error term is
not correlated with the predictors which allows for
time-invariant variables to play a role as explanatory
variables.
In random-effects you need to specify those individual
characteristics that may or may not influence the
predictor variables.
 5.12 The Random Effects Model

What we are essentially saying is that the entities

included in our sample are a drawing from a much
larger universe of such population and that they have a
common mean value for the intercept (= α ) and
 the individual differences in the intercept values of
each entity are reflected in the error term.
In random effects model (REM) or error component
model (ECM) it is assumed that the intercept of an
individual unit is a random drawing from a much
larger population with a constant mean value.
 5.12 The Random Effects Model

 The individual intercept is then expressed as a deviation from

this constant mean value.
 One advantage of ECM over FEM is that it is economical in
degrees of freedom, as we do not have to estimate N cross-
sectional intercepts.
 We need only to estimate the mean value of the intercept and its
variance.
 ECM is appropriate in situations where the (random) intercept
of each cross-sectional unit is uncorrelated with the regressors.
 5.12 The Random Effects Model
 The basic idea of random effects model is to start
with
Y it = α i + β 1 X it + U it − − − − − (1 )
 Instead of treating as fixed, we assume that it is
a random variable with a mean value of α i (no
subscript i here).
 And the intercept value for individual entity can be
expressed as:
αi = α + ε i , i = 1,2,...,N - - - - - - - (2)
Where the random individual differences ε is a i

random error term which are called random

effects, are analogous to random error terms,
 5.12 The Random Effects Model

and we make the standard assumptions about them,

namely, that they have zero mean, are uncorrelated
σ 2
across individuals, and have a constant variance u , so
that;
Substituting equ. (2) into equ. (1), we get
Y it = α + β 1 X it + ε it + u it − − − − − (3 )
= α + β 1 X it + w it
Random
where wit = ε i + uit
Where effect

Regression
 RE…. Sum
 RE:
assume that the entity’s error term is not correlated with
the predictors which allows for time-invariant variables.
If it is assumed that the ε i and X’s are uncorrelated,
ECM may be appropriate where as if ε i and X’s are
correlated, FEM appropriate.
Each entity have a common mean value for the intercept.
If N is large and T is small, and if the assumptions
underlying ECM (REM) hold, ECM estimators are more
efficient than FEM estimators.
the individual differences in the intercept values of each
entity are reflected in the error term.
 5.13 Choosing between fixed and random effects

To check for any correlation between the error

component ui and the regressors in a random effects
model, we can use a Hausman test.
This test compares the coefficient estimates from the
random effects model to those from the fixed effects
model.
 The idea underlying the Hausman test is that both the
random effects and fixed effects estimators are
consistent if there is no correlation between ui and the
explanatory variables xkit.
 5.13 Hausman test

 If you are not exactly sure, which models, FE or

RE you should use, you can do a test called
Hausman test.
 To run Hausman test in Stata
 Run panel datalinear modelslinear
regression FE statistics post estimation
manage estimation resultstore in memory (save
as fixed) run RE statistics post estimation
manage estimation resultstore in memory (save
as random) statistics
postestimationtestsHausman specification test
 5.13 Hausman test

The hypothesis we use to test Hausman test:

H0: Random effect is appropriate model
H1: Fixed effects is appropriate model
Decision Rule: If P-value is less than 5%, we accept the
alternative hypothesis and we reject the null
hypothesis.
Example: Test: Ho: difference in coefficients not
systematic
chi2(2) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 18.35
Prob>chi2=0.000
 5.14 FE Vs RE
FE models RE models

Functional forms Yit = α i + β1 X it + uit Yit = α + β1 Xit + εit + uit

Intercepts Varying a cross Constant

groups/times
Error variances Constant Varying across
groups and/ times

Slopes Constant Constant

Estimation LSDV, within GLS, FGLS
effective model
Hyphothesis test Incremental F-test Breush-Pagan LM
tests
5.15 Additional Notes
Pooled Regression
Fixed effect or LSDV model
Hausman Test
Diagnostic Checking
 Summary
Balanced panel: If each cross sections unit has
the same number of time series observations.
Unbanced panel: If the number of observations
differ among panel data members. Friends
Initially, we assume that the X’s are non
stochastic and the error term follows the
classical assumptions, E (u ) ~ N (0, σ )
2
it

Estimation of panel data regression Models

1. FEM
Yit = β 1 + β 2 X 2 it + β 3 X 3 it + u it -------A
i = 1, 2 , 3,...., n
t = 1, 2 ,....., n
 FEM
Estimation of the above model depends on the
assumption we make about the:
 Intercept
 Slope of coefficients
 Error term
These possibilities are:
i. Assume that the intercept and slope coefficients
are constant a cross time and space and the
error term captures differences overtime and
individuals.
 FEM
ii. The slope coefficients are constant but the
intercept varies over individuals.
iii. The slope coefficients are constant but the
intercept varies over individuals, and time.
iv. All coefficients are vary over individuals
V. The intercept as well as slope coefficients vary
over individuals and time.
 Slope coefficients constant but the intercept
varies a cross individuals: the fixed effects or
Least Squares Dummy Variables (LSDV)
regression on model.
 FEM
One way to take into account the “individuality”
of each company or each cross sectional unit is
to let the intercept vary for each company but
still assume that the slope coefficients are
constant a cross firms.
 FEM
 To see this, we write model A as
Y it = β 1 i + β 2 X 2 it + β3X 3 it + u it − − − − − − − B

The subscript i on the intercept term to suggest that the
intercepts of say 4 firms may be different. This may be
different to special features of each company.
This may be dude to managerial style or managerial
philosophy.
In the literature model (B) is known as fixed effect due to the
fact that although the intercepts may differ across
individuals, each individuals’s intercept does not vary across
individuals; that is time invariant.
 FEM
If we write the intercept as β , it will suggest 1it

that the intercept of each company or individual

is time invariant.
How do we actually allow for the (FEM)
intercept to vary between companies?
 By using dummy variable tecniques (Differential
intercept dummies). We can write equation (B) as
follows:
Yit = α1 + α2 D2i + α3D3i + α4 D4i + β2 X 2it + β3 X3it + uit − − − −C
 FEM
Where D2i= 1 if the observation belogs to Say
Ethiopia, 0 otherwise; D3i= 1 if the observation
belongs to Kenya, 0 otherwise; D4i=1 if the
observation belongs to Uganda, 0 otherwise.
Since we have 4 dummies, we haves used only
three dummies to avoid falling into the dummy
variable trap.
FEM vs REM
The challenge facing researchers is that: Which
model is better, FEM or ECM?
 FEM vs REM
 The answer is that:
 The assumption that one makes about the likely correlation
between the individuals or cross section specific, error
component and the X regressors.
 If it is assumed that the ε i and X’s are uncorrelated, ECM may
be appropriate where as if ε i and X’s are correlated, FEM
appropriate.
 As Woodridge “ In many applications, the whole reasons for
using panel data is to all the unobserved effect ( ε i ) to be
correlated with the explanatory variables.
 ECM assumptions underlying ECM is that the ε i are a random
drawing from a large population.
 FEM vs REM
1. If T (is the number of time series data) is large
and N (the number of cross- sectional units) is
small. The choice is based on computational
convince and FEM may be preferable.
2. When N is large and T is small, the estimates
obtained by the two methods can differ
significant.
 Recall, that in ECM, β = β + ε , where ε i is
1i 1 i

the cross –sectional random component,

where as we treat β 1i
as fixed and not
random.
 FEM vs REM
3. ECM is appropriate if we strongly believed that the
individual or cross sectional units in our sample are not
random drawings from a larger sample.
4. If the individual error component, ε iand one or more
regressors are correlated, then the ECM estimators
are biased, where as those obtained from FEM are
unbiased.
5. If N is large and T is small, and if the assumptions
underlying ECM (REM) hold, ECM estimators are
more efficient than FEM estimators.
Hausman Tests
Hausman Tests
 Is there a formal test that will help us to choose between FEM and
ECM? Yes, a test was developed by Hausman in 1978.
 The hypothesis of Hausman test is that:
 H0: FEM and ECM estimateors donot differ substationally

 H1: FEM and ECM estimates differ substantially.

 Or H0: REM is an appropriate model

 H1: FEM is an appropriate model
 Decision Rule: If the null hyphothesis is rejected, the
conclusion is that ECM is not appropriate model and we may
be better off using FEM, in which case statistical inferences will
be conditions on the ε i in the sample.
1. Panel regression models are based on panel
data. Panel data consists of observations on the
same cross-sectional, or individual, units over
several time periods.
2. Advantages to using panel data
 They increases the sample size considerably
 By studying repeated cross sectional
observations, panel data are better suited to
study the dynamics of change.
 If enable us to study more complicated
3. Disadvantage of using panel data
-such data involve both cross sectional and time
series data, problems that plague cross sectional
data (hetroscedasticity) and time series data
(autocorrelation) needs to be addressed.
 Lab session

Use “lecture_5.xls” data to

practice what we learnt in
previous sections
 END OF CHAPTER FIVE

THANK YOU FOR BEING WITH ME

BEING @ COMMITTED
Stay Safe!
`~~~THE END~~~
GOOD LUCK!
Being @ committed!
All rights are reserved!