Econometrics II Reading Material for Distance Learners

CHAPTER TWO
REGRESSION ON DUMMY VARIABLES

Chapter Contents
Dear distance students, this chapter comprises the following major sub-topics:
 Introduction
 The nature of dummy variables
 Dummy Variables for Differences in Intercept Terms
 Regression on one quantitative variable and one qualitative variable with two classes,
or categories
 Regression on one quantitative variable and one qualitative variable with more than
two classes
 Regression on one quantitative variable and two qualitative variables
 Testing for structural stability of regression models: The Slope dummy variables
 Interaction effects
 Dummy Variables in Seasonal Analysis

Chapter Objectives
At the end of this chapter, students will be able to:
 define dummy variables
 formulate dummy variables models
 know how to use dummy variables for the difference in intercept term
 estimate various regression equations with dummy variables
 apply dummy variables to represent the difference in slope coefficients
 test structural stability
 use dummy variables to explain seasonal differences

107
 Econometrics II Reading Material for Distance Learners

2.1. Introduction
Dear learners, as you might remind from Introduction to Business Statistics or Research Methods
for Economists, in empirical analysis we may encounter four types of variables. These are ratio scale,
interval scale, ordinal scale and nominal scale variables. Up until now we have been assumed that all
variables used in regression analysis are quantitative in nature, which is to say that they have been
continuous and hence, in the preceding chapters, in one way or another, we have discussed the ratio
scale variables in detail. However, this does not mean that these variables are the only variables that
the regression models can deal with. Regression analysis can deal with many interesting variables
that are expressed in qualitative terms. These qualitative measures have to be transformed into some
proxy so that it could be represented and used in a regression. Dummy variables are discrete
transformations and used for this purpose. Thus, the purpose of this chapter is to describe different
techniques on how to use dummy variables in regression analysis and how to interpret them.

2.2. The nature of dummy variables

It frequently happens that some of the factors that you would like to introduce into a regression
model (that can influence the dependent variable) are qualitative in nature (e.g., sex, race, color,
religion, nationality, wars, earthquakes, strikes, political upheavals, and changes in government
economic policy) and therefore not measurable in numerical terms. Some illustrative examples are
the following:
1. You are investigating the relationship between schooling and earnings, and you have both
males and females in your sample. You would like to see if the sex of the respondent makes a
difference.
2. You have data on the growth rate of GDP per capita and foreign aid per capita for a sample of
developing countries, of which some are democracies and some are not. You would like to
investigate whether the impact of foreign aid on growth is affected by the type of government.

In each of these examples, one solution would be to run separate regressions for the two categories
and see if the coefficients are different. Alternatively, you could run a single regression using all the
observations together, measuring the effect of the qualitative factor with what is known as dummy
variables. Dummy variables are constructed by econometricians to be mainly used as proxies for
qualitative variables which cannot be measured in any particular case for various reasons They are
artificial variables that work as proxies for qualitative variables and since they are discrete we need to
be careful when working with and interpreting them. Dummy variables regression has the two
important advantages of providing a simple way of testing whether the effect of the qualitative factor
is significant and, provided that certain assumptions are valid, making the regression estimates more
efficient. Hence, in this chapter we will mainly deal with nominal scale variables, which are also
known categorical/qualitative/binary or dummy variables.
Dear Distance learners, how can you introduce these variables in your regression analysis?
------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------

108
 Econometrics II Reading Material for Distance Learners

To include qualitative variables into empirical analysis, one has to change them into quantitative
variables. One way to quantify nominal scale variables is by creating artificial variables that take on
values of 1 or 0, where 1 indicates the presence of a certain attribute/quality and 0 indicates the
absence of that attribute. This artificial representation is possible since nominal scale variables
usually indicate the presence or absence of a certain quality or an attribute.

Gender is a typical example of a qualitative variable that need to be transformed into a numerical
form so that it could be used in a regression. Since gender could be male or female it is a
categorical variable with two categories. We therefore need to decide what category the dummy
should represent and what category that should be used as a reference. If the dummy variable
should represent men, the discrete variable D would take the value 1 for men and the value 0 for all
other values in the data set. It is therefore important to be sure that all other observations really
represent what you want it to represent. A dummy variable for men could therefore be expressed in
this way:
1
= (2.1)
0 ℎ ( )

When running the regression you can treat the dummy variable D as any other variables included in
the model. This does not change anything for the theory of univariate/multivariate regressions.
However, the economic interpretation of the results can be a bit tricky.

The variable D could take other numerical values than 1 and 0, for instance 9 and 7, and it will
not have any effect on its coefficient. However, the interpretation is easiest when using 1 and 0,
which is the reason why you should follow the structure of (2.1) which is standard.
In passing not that dummy variables can also take other values than 0 and 1, where a linear
transformation function Y = α + βD (where, α and β are constants, β ≠ 0 and D = 1 or 0) can be used
to transform the pair (0, 1) to other pairs. This is the reason why qualitative/dummy variables are
called nominal scale variables; they have no natural scale of measurement.
By analogy dummy variables for other nominal variables can be created.
Variables that assume 0 and 1 artificially are called dummy variables. Hence, dummy variables
can be considered as a device to classify data on qualitative variables into mutually exclusive
categories such as presence or absence of an attribute.
Dummy variables can serve different purposes in regression analysis. They can be used to:
i. Allow for differences in intercept terms
ii. Allow for differences in slopes
iii. Allow for seasonal differences

109
 Econometrics II Reading Material for Distance Learners

2.3. Dummy Variables for Differences in Intercept Terms

Sometimes an investigator may deal with a regression analysis to assess the statistical significance of
the relationship between quantitative dependent variable and qualitative explanatory variables. In
such cases one often uses dummy variables models, with some assumptions about their use. For
instance, the implicit assumption in this case is that the regression lines for the different groups differ
only in the intercept term but have the same slope coefficients. The most basic form of application
using dummy variables is when only the intercept is affected.
Illustration: Y = α + βDi + Ui (2.2)
Where, Y is annual salary of accountant, D is 1 if male and 0 if female. So we will have one dummy
variable, the omitted category(female)become a base or benchmark to compare with male counterpart

 The mean salary of a female accountant by taking conditional expectation is:

E(Y/D=0) = α + β (0) = α (2.3)

 The mean salary of a male accountant is:

E(Y/D=1) = α + β (1) = α+ β (2.4)

The intercept term gives the mean salary of a female accountant while dummy coefficient indicates
how much the mean salary of a male differs from a female. The mean salary of males is the intercept
plus coefficient of dummy.

Note: We cannot put in 2 dummies (1 for male and 1 for female), else will have perfect collinearity.
If we construct D1 as 1 if male, 0 if female and D2 as 1 if female, 0 if male, we are creating a common
problem of dummy variables regression called the dummy variable trap.

Thus, for the dummy variable regression models we can note the following four features.
1. To distinguish the two categories, male and female, we have introduced only one dummy
variable Di . For if Di  1 always denotes a male, when Di  0 we know that it is a female

since there are only two possible outcomes. Hence, one dummy variable suffices to
distinguish two categories. The general rule is this: 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.
2. The assignment of 1 and 0 values to two categories, such as male and female, is arbitrary in
the sense that in our example we could have assigned D=1 for female and D=0 for male.
3. 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.
4. The coefficient  attached to the dummy variable D can be called the differential intercept
coefficient because 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.

110
 Econometrics II Reading Material for Distance Learners

Example 2.1: Suppose we want to compare gender differences in salary of an accountant. Assume
after regression, we found the following result:

Y = 35.20 + 10.25D
t (43.82) (3.45)
Where, Y is income in thousands and D is a dummy variable taking on the value of 1 if male.

Interpretation
The estimated mean salary of female accountants is $35,200 and the estimated mean salary of male
accountants is $45,450. Thus, t- statistics shows us gender differential is statistically significant since
β is significant (because the t-calculated value =β/se(β) =10.25/3.45=2.97 is greater than the
t-tabulated value at 5% significance level for the two tailed test).

2.4. Regression on one quantitative variable and one qualitative variable

with two classes, or categories

In similar fashion, using the categorical variable defined by (2.1) we can also form the following
model with two explanatory variables (the dummy for gender and for years of experience).

Yi=β0+β1Di+β2Xi+Ui (2.5)

Where: Yi  annual salary of ith accountant

X i  Years of experience

Di  1 if male and =0, otherwise

Model (2.5) contains one quantitative variable (years of experience) and one qualitative variable (sex)
that has two classes (or levels, classifications, or categories), namely, male and female. What is the
meaning of this equation? Assuming, as usual, that E (u i )  0, we see that

Mean salary of male accountant: E (Yi / X i , Di  1)  (  0  1 )   2 X i (2.6)

Mean salary of female accountant: E (Yi / X i , Di  0)   0   2 X i (2.7)

In words, model 2.2 postulates that the male and female accountants’ salary functions in relation to the
years of working experience have the same slope   but different intercepts. In other words, it is
assumed that the level of the male accountant’s mean salary is different from that of the female
accountant’s mean salary (by  1 ) but the rate of change in the mean annual salary by years of experience
is the same for both sexes. If the assumption of common slopes is valid, a test of the hypothesis that the
two regressions (2.6) and (2.7) have the same intercept (i.e., there is no sex discrimination) can be made
easily by running the regression (2.5) and noting the statistical significance of the estimated  1 on the
^
basis of the traditional t test. If the t test shows that  1 is statistically significant, we reject the null
hypothesis that the male and female accountants’ levels of mean annual salary are the same.

111
Econometrics II Reading Material for Distance Learners

The only thing that differs between the two expectations is the coefficient for the dummy variable.
When D=1 we see that the conditional expectation in (2.6) consist of two constants β0 and β1,
which sum represents the intercept in that case. However, when D=0, the conditional expectation
will be given by (2.7) which only contain one constant β0. Hence, the model as a whole contains
two intercepts βo and βo+β1. If we take the difference between the two conditional expectations we
receive: E[Y = 1, ] − [ =0, ]= (2.8)
which equals the coefficient for the dummy variable. Since our binary variable D is discrete, we
cannot take the derivative of Y with respect to D, since a derivative requires a continuous variable,
and therefore is undefined here. In order to find the corresponding marginal effect (derivative) in
this case we have to form the difference given by (2.8) and conclude that when D moves from 0 to
1, the conditional expectation of Y change by β1 units, which represent the marginal effect for the
linear model. When working with the linear model, it makes no difference if we treat the dummy
variable as it was continuous when calculating the marginal effect since they become the same.
But with other functional forms it makes a difference.

Example 2.2
Assume the following regression result from a model given by (2.2) with Y being the hourly wage
rate, D a dummy for men, and X a variable for years of schooling. The dependent variable is
expressed in USA dollar ($). Standard errors are given within parenthesis:
= 55.9 + 21.9 + 2.4 X
(8.16) (4.30) (0.63)
Use the regression results to calculate how much higher the average hourly wage rate is for men.
First we have to check if the coefficient for the male dummy is significant. With a t-value
(β/se(β)=21.9/4.3) equals to 5, the coefficient is significantly different from zero at any
conventional significance levels. The marginal effect measured with this regression says that men
earn 21.9 hours more than women do on average, maintaining years of schooling constant.

2.5. Regression on one quantitative variable and one qualitative variable

with more than two classes
The human capital model described above includes a continuous variable for the number of years of
schooling. When including a continuous variable for schooling it is under the belief that the hourly
wages are set and determined based on this measure. An alternative approach would be to argue that
it is the level of schooling, the received diploma that matters in the determination of the wage rate.
That calls for a qualitative variable with more than two categories.
0 ℎ
For instance: = 1 ℎ (2.9)
2 ℎ
In order to include D directly into a regression model we have to make sure that the effect of going
from primary holing to secondary schooling on the hourly wage rate is of the same size as going
from secondary schooling to a post secondary schooling. If that is not the case we have to allow for
differences in these two effects. There are at least two approaches to this problem.
112
Econometrics II Reading Material for Distance Learners

The first and most basic approach is to create three binary variables; one for each educational
level, in the following way:
1 ℎ 1 ℎ 1 ℎ
= = =
0 ℎ 0 ℎ 0 ℎ

We can now treat D1, D2 and D 3 as three explanatory variables, and include them in the regression
model. However, it is important to avoid the so called dummy variable trap.

Consequences of the Dummy Variable Trap

What would happen if you included a dummy variable for the reference category? There would be
two consequences. First, was it possible to compute regression coefficients, you would not be able to
give them an interpretation. The coefficient α is a basic estimate of the intercept, and the coefficients
of the dummies are the estimates of the increase in the intercept from this basic level, but now there
is no definition of what is basic, so the interpretation collapses. The other consequence is that the
numerical procedure for calculating the regression coefficients will break down and the computer
will simply send you an error message (or possibly, in sophisticated applications, drop one of the
dummies for you).
Suppose that there are m dummy categories and you define dummy variables D1 -------------- Dm.
Then, in observation i, ∑ = 1, because one of the dummy variables will be equal to 1 and all
the others will be equal to 0. As a consequence the model is subject to a special case of exact
multicollinearity, making it impossible to compute regression coefficients.
The dummy variable trap appears when the analyst-tries to specify and estimate the following
model: lnY= βo +β1D1 + β2 D2 + β3 D3 +β4 X +U (2.10)
It is a mathematical impossibility to estimate the parameters in (2.10) since there is no variation in
the sum of the three dummy variables, since D1+D2+D3=1 for all observations in the data set. Since
the model only can contain one constant, in this case the intercept, we cannot include all three
dummy variables. The easiest way to solve this is to exclude one of them and treat the excluded
category as a-reference category. We re-specify the model in following way:
lnY = Bo +β2D2 +β3D3 +β4X +U (2.11)
That is, if D1, is excluded, the other categories will have D1, as reference. Β2 will therefore be
interpreted as the wage effect of going from a primary schooling diploma to a secondary schooling
diploma, and β3 will represent the wage effect of going from a primary schooling diploma to a post
secondary schooling diploma.
An alternative way to get out of this trap is by introducing as many dummy variables as the number
of categories of that variable, provided we do not introduce the intercept in such a model. Thus, if
we drop the intercept term from (2.11), and consider the following model,
lnY=β1 D1 +β2 D2 +β3D3 +B4 X +U (2.12)
we do not fall into the dummy variable trap, as there is no longer perfect collinearity.
The three dummy variables will then work as three intercepts in this model; one for each educational
level. The coefficients can therefore not be interpreted as relative changes in this case.

113
Econometrics II Reading Material for Distance Learners

Which is a better method of introducing a dummy variable? (1) introduce a dummy for each
category and omit the intercept term? or (2) include the intercept term and introduce only (m − 1)
dummies, where m is the number of categories of the dummy variable?
As Kennedy notes (Cited by Gujarati, 2004): Most researchers find the equation with an intercept
more convenient because it allows them to address more easily the questions in which they usually
have the most interest, namely, whether or not the categorization makes a difference, and if so, by
how much. If the categorization does make a difference, by how much is measured directly by the
dummy variable coefficient estimates. Testing whether or not the categorization is relevant can be
done by running a t test of a dummy variable coefficient against zero (or, to be more general, an F
test on the appropriate set of dummy variable coefficient estimates).

Example 2.3
Estimate the parameters of (2.13) and (2.14) and compare and interpret the results.

Specification I: (2.13) ln Y = 3.929+0.154D2 +0.295D3 +0.009X

(0.043) (0.024) (0.022) (0.001)
Specification II: (2.14) ln Y = 3.929D1 + 4.083D2 + 4.224D3 + 0.009X
(0.043) (0.034) (0.036) (0.001)
The three dummy variables represent three educational levels, and X represents the age of the
individual. The first thing to notice is that βo=C1, βo+β2=C2 and βo+β2+β3=C3. Hence, the two
specifications are very much related. Furthermore C2-C1=β2 and C3-C2=β3. With help from
specification II, we can derive the effect of going from a high school diploma to a college diploma
by taking the difference between C3 and C2 which turns out to be equal to 0.14, i.e. a 14 percent
increase. However, that effect could also have been received by taking the difference between β3 and
β2. For the obvious reason there should be no change in the effect of the other variables included in
the model (β4 in this example) when alternating between specification I and II.

Example 2.4: If we want to look at the effect of location( Addis Ababa, Hawassa, Arba Minch etc)
on Person's salary in thousands of Birr (Y). We create variable for towns- three variables:
– Adiss Ababa- D1= 1 if from Adiss Ababa, 0 otherwise

– Hawassa –D2=1 if from Hawassa, 0 otherwise

– Arba Minch-D3= 1 if from Arba Minch, 0 otherwise

– Multiple Regression Model: Y= γ0 + γ1D1+ γ2D2+ e

– Result: Y= γ0 + 3.8D1+ 2.1D2

Activity 2.1: Why was Arba Minch dropped?

Interpretation: Adiss Ababa=γ1=3.8 means that people in Adiss Ababa earn birr 3800 more on
average relative to people in Arbaminch.

Activity 2.2: Interpret Hawassa= γ2=2.1

114
Econometrics II Reading Material for Distance Learners

2.6. Regression on one quantitative variable and two qualitative variables

As you might have noted, so far we have been dealing with regression analysis where there is only
one qualitative variable. In practice, however, we may have more than one qualitative variable
affecting the dependent variable. Then, how will you introduce more than one qualitative variable in
your regression analysis? The introduction of qualitative variables in regression analysis is as direct
as the introduction of quantitative explanatory variables. However, for each qualitative variable we
introduce one on more than one dummy following the rule mentioned above.

Illustration 1: Suppose that we want to analyze the determinates of household consumption (C),
and that we have data on the income of the household (Y), the sex of household head (male or
female), the age of the household head (given in three categories such as less than 30 years, 30 to 60
years and greater than 60 years) and the education of the household head (given in three categories
such as less than high school, greater than or equal to high school but less than collage degree and
greater than or equal to college degree). Following the rule of dummy variables (m-1), for each
variable the number of dummies included in the regression analysis is one less than the number of its
categories. The coefficients of the dummies are interpreted as the differences between the average
consumption of the omitted (base) category and the category represented by the dummy under
consideration, keeping other things constant.

Illustration 2: Let us revert to the accountants’ salary regression (2.5), but now assume that in
addition to years of experience and sex, the skin color of the accountant is also an important
determinant of salary. For simplicity, assume that color has two categories: black and white. We
can now write (2.5) as:
Yi   1   2 D2i   3 D3i   X i  u i (2.13)

Where Yi  annual salary, X i  years of working experience, D2  1 if male and =0 otherwise,

D3  1 if white and =0 otherwise.

Note that: Each of the two qualitative variables, sex and color, have two categories and hence need
one dummy variable for each. Note also that the omitted, or base, category now is “black female
accountant.”
Assuming E (u i )  0 , we can obtain the following regression from (2.13)
Mean salary for black female accountant:
E (Yi | D2  0, D3  0, X i )  1   X i
Mean salary for black male accountant:
E (Yi | D2  1, D3  0, X i )  ( 1   2 )   X i
Mean salary for white female accountant:
E (Yi | D2  0, D3  1, X i )  ( 1   3 )   X i
Mean salary for white male accountant:
E (Yi | D2  1, D3  1, X i )  ( 1   2   3 )   X i

Once again, it is assumed that the preceding regressions differ only in the intercept coefficient but
not in the slope coefficient  .

115
Econometrics II Reading Material for Distance Learners

An OLS estimation of (2.15) will enable us to test a variety of hypotheses. Thus, if  3 is statistically

significant, it will mean that color does affect a accountant’s salary. Similarly, if  2 is statistically
significant, it will mean that sex also affects a accountant’s salary. If both these differential
intercepts are statistically significant, it would mean sex as well as color is an important determinant
of accountants’ salaries.

Example 2.5: Assume the house with two attributes that can further be two qualitative variables.
Let us denote the first variable as D1 . This variable can take values 1 (if the house has a driveway)
or 0 (if it does not). And also denote the second variable as D2 . This variable can also take values 1
(if the house has a recreation room) or 0 (if it does not). X1 is the size of the garden and X2 is land
rent and Y is the price of the house.
Now, suppose we will run the regression of Y on the four explanatory variables and a constant. The
result will be:
Y  2736  12598 D1  10969 D2  5197
. X 1  10.562 X 2
That means that if the house has no driveway ( D1  0 ) and no recreation room ( D2  0 ), its value

will be Y  2736  5197

. X 1  10.562 X 2 . If the house has a driveway, its value will be, holding all
other variables constant (or ceteris paribus), $12598 more. If the house has a recreation room, its
value will be, ceteris paribus, $10969 more. If the house has a driveway and a recreation room, its
value will be, ceteris paribus, 12598+10969 = $23567 more.

What is peculiar in the above result is that the marginal effect of the size of the lot, or the number of
bedrooms, is the same for the four kinds of houses. Increasing the size of the garden by 1 square foot
will increase the price of the house by $5.197 whether the house has or not a driveway or a
recreation room. This assumption may be justified or not.

From the preceding discussion it follows that we can extend our model to include more than one
quantitative variable and more than two qualitative variables. The only precaution to be taken is that
the number of dummies for each qualitative variable should be one less than the number of
categories of that variable.
Thus, you can see that dummy explanatory variables do not change the econometrics. In this chapter
of module we have just introduced a few ideas on the best way to give good economic interpretation
for the result. However, dealing with an explained or dependent dummy variable is a much more
difficult problem. In this case the econometrics of OLS does not work anymore and we must use
another econometrics, called discrete choice or binary models. Actually, the basis of this new
econometrics is not very difficult. However, it is different and will not be investigated in this
undergraduate course. Rather it will be covered in the graduate studies.

116
Econometrics II Reading Material for Distance Learners

2.7. Testing for structural stability of regression models: The Slope

dummy variables
As could be seen in the previous section, the dummy variable could work us an intercept shifter.
Sometimes it is reasonable to believe that the shift should take place in the slope coefficient instead
of the intercept. If the slopes are in fact different, testing for differences in the intercepts may be of
little practical significance. Therefore, we need to develop a general methodology to find out
whether two (or more) regressions are different, where the difference may be in the intercepts or the
slopes or both.
If we go back to the human capital model it is possible to argue that the difference in wage rate
between men and women could be due to differences in their return to education. This would mean
that men and women have slope coefficients that are different in size.
A model that control for differences in the slope coefficient for different categories of the qualitative
variable could be expressed in the following way:
ln Y = βo +(β1 +β2D) X +Ui
= βo + β1 X + β2 (DX)+Ui (2.14)
In this case the slope coefficient for X equals β1 when D=O and β1+ β2 when D=1. Hence, a way to test if
the return to education differs between men and women would be to test if β2 is different from zero,
which should be tested before going on to test if β1 +β2s different from zero. Observe that the coefficient
for the cross product is interpreted differently if both variables had been continuous. Since D is binary,
DX is only active when D= 1, and the corresponding effect is therefore related to the category specified
by D=1.

Example 2.6
Use the same data set as in Example 2.5 and estimate the coefficients in (2.14). The results are
presented below with standard errors within parenthesis:
lnY = 4.11 + 0.024X + 0.014DX
(0.031) (0.003) (0.001)
Use the regression results to investigate if there is a difference in the return to education between
men and women. To answer that question we simply test the estimated coefficient for the cross
product. Doing that, we receive a t-value of 10.3 which is above any critical values of conventional
significance level. Hence, if this specification is correct, we can conclude that the returns to
education differ between men and women.

2.7.1. A model with Intercept and Slope dummy variable

Whenever working with cross products it is very important to always include the involved variables
separately to separate that kind of effect from the cross product. If it is the case that D in itself has a
positive effect on the dependent variable, that unique effect will be part of the cross effect otherwise.
Hence whenever including a cross product the model should be specified in the following way:
lnY = βo + β1X + β2D+ β3(DX)+U (2.15)
When we include the two variables, X and D separately and together with their product we allow for
changes in both the intercept and the slope. If it turns out that the coefficient of β2 is not significant
one can go on and reduce the specification to (2.14), but not otherwise.

117
Econometrics II Reading Material for Distance Learners

Example 2.7
Extend the specification of (2.14) by including D separately. That is, estimate the parameters of the
model given by (2.15) and interpret the results. Doing that, we received the following results, with
standard errors within parenthesis.
lnY = 4.006 + 0.033X + 0.210D-0.002DX
(0.045) (0.004) (0.062) (0.005)
By investigating the t-values we see that β1, and β2 are statistically significant from zero. But the
t-value from the cross product is not significant any more. Since D alone has a significant effect on
the dependent variable, there is little effect left from the cross product, and hence we conclude that
there is no difference in the return to education between men and women.

Example 2.6 and 2.7 should convince you that it is very important to include the variables that
appear in a cross product separately, since they might stand for the main effect. In Example 2.6 we
did not include D even though it was relevant. Omitting relevant variables has consequences and
biases the remaining coefficients. In this case it made us to draw the wrong conclusion about the
return to education for men and women.

2.7.2. Test for structural differences

An important application using dummy variables is to test if the coefficients of the model differ for
different sub groups of the population, or if they have changed over time. For instance, assume that
we have the following wage equation expressed with a semi logarithmic (log-linear) functional
form:
lnY= βo +β1X1 +β2X2 +U (2.16)
with Y being the wage rate, X1 the number of years of schooling, and X2 the number of years of
working experience. We would like to know if β1 and β2 differ between men (m) and women (w)
simultaneously. That is, we would like test the following hypothesis:
H0:β1m = β1w, β2m= β2w
H1: β1m ≠β1w and / or β2m ≠ β2w

In order to carry out this test using dummy variables we need to create an indicator variable D, for
let's say for men, and then form the following regression model:
lnY= βo +β1 D+β2 X1 + β3 (X1D)+β4X2 β5 (X2D)+U (2.17)
Equation (2.17) will be representing the unrestricted model, where men and women are allowed to
have different coefficients, and equation (2.16) will be representing the restricted case where men
and women have the same coefficients. We will now compare the residual sums of squares (RSS)
between the two models and use those in our test statistic given by:
F= (RSSR - RSSu)/ df1 ~ ( , ) (2.18)
RSSu / df2
If the RSSR is very different from RSSu we will reject the null hypothesis in favor of the alternative
hypothesis. If they are similar in size, the test value will be very small and we say that the coefficients are the
same for men and women.

118
Econometrics II Reading Material for Distance Learners

Example 2.8
Assume that would like to know if the coefficients of equation (2.16) differ between men and
women in the population. In order to test the joint hypothesis we need to run two regression
models; one restricted model given by (2.16) and one unrestricted model given by (2.17). Using a
sample of 1483 randomly selected individuals we obtained the following results:

Table 2.1 Regression results

Restricted model Unrestricted model
Residual sum of squares (RSS) 145.603 140.265
Degrees of freedom (n – k) 1483 – 3 = 1480 1483 – 6 = 1477

Using the results given in Table 2.1 we may calculate the test value using (2.18). The degree of
freedom for the numerator is calculated as the difference between degrees or freedom for the RSS
from the restricted and unrestricted model. That is 1480 - 1477 = 3. Another way to think about the
degrees of freedom for the numerator is to express it in terms of the number of restrictions imposed
by the restricted model compared to the unrestricted. The unrestricted model has 6 parameters,
while the restricted model has only 3, which means that three parameters have been set to zero in
the restricted model. Therefore we have 3 restrictions. The test value using the test statistic is
therefore equal to:
( )/ ( . . / .
= = = = 18.73
/ . / .

The test value has to be compared with a critical value. Using a significance level of 5 percent the F-
critical value equals 2.6. Hence, the test value is much larger than the critical value which means that
we can reject the null hypothesis. We can therefore conclude that the coefficient of the regression
model differ for men and women.

2.8. Interaction effects

Consider the following model:
Yi   1   2 D2i   3 D3i   X i  u i (2.19)

where Yi  annual expenditure on clothing

X i  income

D2  1 if female
= 0 if male
D3  1 if college graduate

= 0 otherwise
Implicit in this model is the assumption that the differential effect of the sex dummy D2 is constant

across the two levels of education and the differential effect of the education dummy D3 is also
constant across the two sexes. That is, if, say, the mean expenditure on clothing is higher for
females than males this is so whether they are college graduates or not. Likewise, if, say, college
graduates on the average spend more on clothing than non college graduates, this is so whether they
are female or males.

119
Econometrics II Reading Material for Distance Learners

In many applications such an assumption may be untenable. A female college graduate may spend
more on clothing than a male graduate. In other words, there may be interaction between the two
qualitative variables D2 and D3 and therefore their effect on mean Y may not be simply additive as
in (2.19) but multiplicative as well, as in the following model:
Yi   1   2 D2i   3 D3i   4 ( D2i D3i )   X i  u i (2.20)
From (2.20) we obtain
E (Yi | D2  1, D3  1, X i )  ( 1   2   3   4 )   X i (2.21)
which is the mean clothing expenditure of graduate females.
Notice that:
 2  differential effect of being a female
 3  differential effect of being a college graduate

 4  differential effect of being a female graduate which shows that the mean clothing
expenditure of graduate females is different (by  4 ) from the mean clothing expenditure of females
or college graduates. If  2 ,  3 , and  4 are all positive, 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. 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. Needless to say,
omitting a significant interaction term incorrectly will lead to a specification bias.

2.9. Dummy Variables in Seasonal Analysis

Many economic time series based on monthly or quarterly data exhibit seasonal patterns, i.e., they
show regular oscillatory movements. For example, household demand for money at holiday times
shows seasonal pattern. Hence, it is desirable to remove the seasonal component from a time series
so that one can concentrate on the other components, such as the trend. The process of removing the
seasonal component from a time series is known as deseasonalization, and the series thus obtained
is called deseasonalized or seasonally adjusted time series.
Usually the important economic series such as the unemployment rate, the consumer price index
(CPI), the producer price index (PPI), and the industrial production index that we see in different
reports are published in their seasonally adjusted form.
One of the methods used to deseasonalize a time series is a method of dummy variables.
To illustrate the dummy variables technique of deseasonalization, suppose that we have quarterly
data on consumption (C) and income (Y). The seasonal adjustment in this case can be estimated by
including among the explanatory variables three seasonal dummy variables, Q1, Q2 and Q3.
In order to deal with this, we fit the following the quarterly regression model:

Ct=β0+βlY1t+.....+βkYkt+α1Q1t+α2Q2t+α3Q3t+ut (2.22)

120
Econometrics II Reading Material for Distance Learners

where
1 ℎ
=
0 ℎ
1 ℎ
=
0 ℎ
1 ℎ ℎ
=
0 ℎ
Note that in equation 2.22, we have treated the fourth quarter as the base category. Consequently, the
coefficients attached to the various dummies are differential intercepts, showing by how much the
average value of C in the quarter that receives a dummy value of 1 differs from that of the base
category, holding the income level constant. In other words, the coefficients on the seasonal
dummies will give seasonal increase or decrease in the average value of C relative to the base
season, for a given value of Y. Furthermore, the coefficient of Y shows that, allowing for seasonal
effects, if income (Y) changes by one unit, on average, consumption (C) will change by β units.

Dear distance learners and readers, so far we have seen how to deseasonalize the dependent
variable (for example C in equation 2.22). Now, an interesting question is: Just as the dependent
variable, C, exhibits seasonal patterns would not the explanatory variable, Y also exhibit seasonal
patterns? How, then, do we take into account seasonality in the explanatory variable Y?
Don’t worry! Now you have got the key at your hand, that is, the dummy variables technique. The
interesting thing about equation 2.22 is that the dummy variables in the model not only remove the
seasonality in C but also the seasonality in Y, if any. It is just like killing two birds with a single
stone.

121
Econometrics II Reading Material for Distance Learners

Chapter Summary:
1. Dummy variables, taking values of 1 and zero (or their linear transforms), are a means of
introducing qualitative regressors in regression models.
2. Follow either of the two methods to introduce a dummy variable:
(a) Introduce a dummy for each category of a qualitative regressor and omit the intercept term or
(b) Include the intercept term and introduce only (m− 1) dummies, where m is the number of
categories of the qualitative variable.
3. The category for which no dummy variable is assigned is known as the base, benchmark,
control, comparison, reference, or omitted category. And all comparisons are made in
relation to the benchmark category.
4. Dummy variables are a data-classifying device in that they divide a sample into various
subgroups based on qualities or attributes (sex, marital status, race, religion, etc.) and
implicitly allow one to run individual regressions for each subgroup. If there are differences
in the response of the regress and to the variation in the quantitative variables in the various
subgroups, they will be reflected in the differences in the intercepts or slope coefficients, or
both, of the various subgroup regressions.
5. The value of the intercept represents the mean value of the benchmark category
6. The coefficients attached to the dummy variables are known as the differential intercept
coefficients because they tell by how much the value of the intercept that receives the value
of 1 differs from the intercept coefficient of the benchmark category.
7. If a model has several qualitative variables with several classes, introduction of dummy
variables can consume a large number of degrees of freedom. Therefore, one should always
weigh the number of dummy variables to be introduced against the total number of
observations available for analysis.
8. Since the dummy variables are non stochastic, they pose no special problems in the
application of OLS. However, care must be exercised in transforming data involving dummy
variables. In particular, the problems of autocorrelation and heteroscedasticity need to be
handled very carefully.

Check Your Progress for Activities

Activity 2.1:
Arba Minch is dropped because it serves as a benchmark or a base for comparison
Activity 2.2:
Hawassa γ2=2.1 implies that people in Hawassa earn birr 2100 more on average relative to people in
Arba Minch.

122
Econometrics II Reading Material for Distance Learners

Review Exercises
1. Explain the following concepts
a. Dummy variables
b. Seasonal dummy variables
c. Dummy variables trap
2. Consider the following model: Yi = α1 + α2Di + β Xi + ui, where Y = annual salary of a college
professor, X = years of teaching experience and D = dummy for gender
Consider three ways of defining the dummy variable.
a. D = 1 for male, 0 for female.
b. D = 1 for female, 2 for male.
c. D = 1 for female, −1 for male.
Interpret the preceding regression model for each dummy assignment. Is one method preferable to
another? Justify your answer.
3. Given the regression result: Y = 25.20 + 8.25D + 1.25X
t (32.85) (10.45) (15.11)
Where, Y is income in thousands, D is a dummy variable taking on the value of 1 if male 0 for
female and X = years of experience.

a. Calculate the mean salary of male and female assuming 0 years of experience
b. Test the significance of coefficient of years of experience and Calculate the salary addition
for male or female accountant after one year experience
c. Holding experience constant, check whether the mean salaries of male and female
accountants are different or not

4. If you have monthly data over 30 years, how many dummy variables will you introduce in your
regression analysis to test the following hypotheses?
a. All the 12 months of the year exhibit seasonal patterns.
b. Only April, June and August and December exhibit seasonal patterns.
5. Suppose that the regression program of a graduating class Economics student of Arba Minch
University refused to estimate four seasonal coefficients when he enters the quarterly data
Including zero to one dummy variables for each quarter. What is he supposed to do?

123
Econometrics II Reading Material for Distance Learners

CHAPTER THREE
DYNAMIC ECONOMETRIC MODELS

Chapter Contents
Dear distance students, this chapter consists of the following major sub-topics:
 Introduction
 Distributed lag and Autoregressive models
 The Role of Lag in Economics
 The Reasons for Lags
 Infinite Distributed Lag
 Estimation of Lag Models
 The Granger Causality Test

Chapter Objectives
At the end of this chapter, students will be able to:
 define dynamic models
 differentiate distributed lag models from autoregressive models
 explain why lags are commonly used in economic analysis
 recognize the use of infinite distributed lags in applied economics
 know the various estimation methods commonly used for lag models
 apply the Granger Causality test for different macroeconomic variables

124
Econometrics II Reading Material for Distance Learners

3.1. Introduction
So far in the standard regression model we ignored the effect of timing (lagged values) of the
explanatory variable(s) on the dependent variable. The standard linear regression implies that change
in one of the explanatory variables causes a change in the dependent variable during the same time
period and during that period alone. But in economic models, generally, a cause often produces its
effect only after a lapse of time; this lapse of time (between cause and its effect) is called a lag.
Many economic models have lagged values of the regressors in the regression equation.
For example, it takes time to build roads and highways. Therefore, the effect of this public
investment on growth in GNP will show up with a lag, and this effect will probably go on about for
several years. It takes time before investment in research and development pays off in new
inventions which in turn take time to develop into commercial products.
In studying consumption behavior, a change in income may affect consumption over several periods.
This is true in the permanent income theory of consumption, where it may take the consumer several
periods to determine whether the change in real disposable income was temporary or permanent.
For example, is the extra consulting money earned this year going to continue next year?
Also, lagged values of real disposable income appear in the regression equation because the
consumer takes into account his life time earnings in trying to smooth out his consumption behavior.
In turn, one’s life time income may be guessed by looking at past as well as current earnings.
Therefore, realistic formulations of economic relations often require the insertion of lagged values of
the explanatory or insertion of lagged dependent variables.
Autoregressive and distributed-lag models are used extensively in econometric analysis, and in this
chapter we take a close look at such models with a view to finding out the following:
 What is the role of lags in economics?
 What are the reasons for the lags?
 Is there any theoretical justification for the commonly used lagged models in empirical
econometrics?
 What is the relationship, if any, between autoregressive and distributed-lag models? Can one
be derived from the other?
 What are some of the statistical problems involved in estimating such models?
 Does a lead–lag relationship between variables imply causality? If so, how does one
measure it?

125
Econometrics II Reading Material for Distance Learners

3.2. Lagged Models: Distributed lag and Autoregressive models

Regression models that take in to account time lags are known as dynamic or lagged regression
models.
In regression analysis involving time series data, the regression model may include both the current
and the lagged (past) values of the explanatory variables (the X’s) as well as the lagged value of the
dependent variable.
There are two types of lagged models. These are: Distributed and Autoregressive lag models.
Dear students, in this particular module we will invest more of our time in discussing distributed
lag models since in one way or another, an autoregressive model was discussed in autocorrelation
section of chapter one.
Time series data are any set of data from a quantifiable or qualitative event that are recorded
overtime or
A time series is a sequence of data points/numerical data measured typically successive points in
time spaced at uniform time intervals.
Time series econometrics/analysis compares methods for analyzing time series data inorder to
extract meaningful statistics and other characteristics of the data.

3.2.1. Distributed lag models

Distributed-lag models are models which include lagged values of the exogenous variables and/or
lagged values of the dependent variables among the set of explanatory variables.
The general form of a distributed-lag model with only lagged exogenous variables is:
Yt = a + boXt + b1Xt-1 + b2Xt-2 + ... + bsXt-s + ... + Ut (3.1)

They are called distributed lag models because the influence of the explanatory variable on the
dependent variable is distributed over a number of past values of X.
The coefficient b0 is known as the short run, or impact, multiplier because it gives the change in

mean value of Y following a unit change in X in the same time period t. If the change in X is
maintained at the same level thereafter, then b0+b1 gives the change in the mean value of Y in the
next period, and b0+b1+ b2 in the following period, and so on. These partial sums are called interim,
or intermediate, multipliers.
Suppose that we observe Xt from 1955 to 1995. Xt−1 is the same variable but for the previous period,
i.e., 1954-1994. Since 1954 is not observed, we start from 1955 for Xt−1, and end at 1994.
This means that when we lag once, the current Xt series will have to start at 1956 and end at 1995.
For practical purposes, this means that when we lag once we loose one observation from the sample.
So if we lag ‘s’ periods, we loose ‘s’ observations. Furthermore, we are estimating one extra ‘b’ with
every lag.
Finally, after ‘s’ periods we obtain which is known as the long run, distributed-lag multiplier
provided the sum b exists.
s

bi 0
i  b0  b1  b2        bs  b (3.2)

The number of lags, s, may be either finite or infinite. However, in order to avoid explosive values
of Yt, we assume that the b's have a finite sum: ∑ < ∞

126
Econometrics II Reading Material for Distance Learners

The average lag is defined as the weighted average of all the lags involved, with weights being
the relative size of the respective b coefficients
∑
= ∑
= ∑ ∑
(3.3)

Some examples will illustrate the importance of lagged variables.

Example 1: The consumption function

The model: C t     0Yt  1Yt 1   2Yt  2  U t is a distributed lag model of consumption function.

This means that the value of the consumption expenditure (C t ) at any given time depends on the

current and past values of the disposable income (Yt ) .

Recent versions of the consumption function postulate that the current level of consumption
depends on past levels of consumption, due to 'habit persistence', on current income and past levels
of income and other factors
Ct = f(Ct-1, Ct-2, .... , Yt, Yt-1 ,Yt-2 ,… , Xlt, X2t ….) (3.4)

Example 2: The demand for durables

The demand for durables (Dd) depends, among others, on present income (Yt), past levels of income
(Yt-s ) which determine the amount saved for the acquisition of the durables, on the stocks of
durables or, equivalently, on past acquisitions of durables (Sd. t -1), prices (Pt), and so on
Dd = f(Yt , Yt-1 , …, Sd , t-1 , Pt … ) (3.5)

Example 3: The investment function

Investment projects depend on past outputs, on expectations about future profits, on capital stock
and other factors
It = f(Xt. Xt-1 , Xt-2 ,.... , , Kt-1 ,rt-4 ... ) (3.6)

where X = level of output

= profit
K = capital stock
r = interest rate

Example 4: Demand for nondurables

The habit-persistence is a characteristic of human behavior. The quantity demanded for food,
tobacco and other nondurables depends, among others, on the past levels of consumption of
these commodities
Qt = f (Qt-1 , Yt, Pt, ….. ) (3.7)

127
Econometrics II Reading Material for Distance Learners

3.2.2. Autoregressive models

It should be noted that distributed lag model is not to be confused with autoregressive model.
If the model includes one or more lagged values of the dependent variable among its explanatory
variables, it is called an autoregressive model.
Thus, Yt = α + β0Xt + β1Xt−1 + β2Xt−2 + ut is an example of distributed-lag model whereas
Yt     X t  Yt 1   t is an example of an autoregressive model.

In the former case the current and lagged values of repressors are explanatory variables while in the
latter case the regressors can be both explanatory (current and lagged) and lagged values of the
regresand; thus, the latter are also known as dynamic models since they portray the time path of the
dependent variable in relation to its past value(s).

3.3. The Role of lag in Economics

In economics the dependence of a variable Y (the dependent variable) on another variable(s) X (the
explanatory variable) is rarely instantaneous. Very often, Y responds to X with a lapse of time. Such
a lapse of time is called a lag. Indeed lags are involved in all economic behavior. We live in a
dynamic world of continuous adjustment. Clearly an adjustment process takes time, the length of the
time period depending on the nature of the particular phenomenon. Yet economic theory is mostly
static. In comparative statics where different equilibria are compared, adjustments are assumed
instantaneous.

The process of adjustment and the lags involved are scarcely discussed. Yet these lags are of
paramount importance for decision making. It is crucial for the government official to know how
fast, after how many time periods the economic units will react to changes of various policy
variables (instruments). How fast will consumers react to the imposition of a sales tax or a credit
squeeze? How fast will firms react to tax concessions and other incentives for investment and
innovation? How fast will be the effects of devaluation? How fast will investors react to changes in
the interest rates?
Similarly the lags involved in the demand for consumers goods following a change in the 'policy
instruments' (price, quality, style, advertising) of a firm are of crucial importance for managerial
decisions. Again the micro-economic theory of the firm is mostly static. The lag patterns in the
adjustment of buyers to changes in various variables are scarcely mentioned.
In view of the nature of economic behavior any realistic formulation of economic models should
involve some lagged variables among the set of explanatory variables. Lagged variables are one
way for taking into account the length of time in the adjustment processes of economic behavior and
perhaps the most efficient way for rendering them dynamic.
Lagged models have become increasingly popular in applied econometric research. Their
refinement has made possible the handling of expectations about future events, if only in a rigid and
not very satisfactory way. In short, lagged models offer much flexibility to the formulation of
models of economic behavior.
Distributed lag models involve a high degree of empiricism, which is due to the unsatisfactory
state of economic theory with regard to the length of the adjustment processes of economic
phenomena.

128
Econometrics II Reading Material for Distance Learners

Economic theory, even where it recognizes the importance of time lags, never suggests the
precise number of lags that should be included in a function. Rather the pattern of lags is
explored and determined from the available sample observations, by adopting an experimental
approach involving various lag patterns and then choosing among them the one that gives the
best statistical fit.

3.4. The Reasons for Lags

There are three major reasons why lags may occur.
1. Psychological reasons: As a result of the force of habit (inertia), people, for example, do not
change their consumption habits immediately following a price decrease or an income
increase perhaps because the process of change may involve some immediate disutility.
Thus, those who become instant milliners by winning lotteries may not change the lifestyle
to which they were accustomed for a long time because they may not know how to react to
such windfall gain immediately. Of course, given reasonable time, they may learn to live
with their newly acquired fortune. Also, people may not know whether a change is
“permanent” or ‘transitory”. Thus, my reaction to an increase in my income will depend in
whether or not the increase in permanent. If it is only a nonrecurring increase and in
succeeding periods my income returns to its previous level, I may save the entire increase,
whereas someone else in my position might decide to “live it up”.
2. Technological reason: suppose, for instance, the price of capital relative to labor declines
making substitution of capital for labor economically feasible. Of course, addition of capital
takes time (gestation period) moreover, if the drop in price is expected to be temporarily,
firms may not rush to substitute capital for labor, especially if they expect that after
temporarily drop the price of capital may increase beyond its previous level.
Sometimes, imperfect knowledge also accounts for lags. At present the market for personal
computers is glutted with all kinds of computers with varying features and prices. Moreover,
since their introduction in the late 1970s, the prices of most personal computers have
dropped dramatically. As a result, prospective consumers for the personal computer may
hesitate to buy until they have had time to look into the features and prices of all the
competing brands. Moreover, they may hesitate to buy in the expectation of further decline
in price or innovations.
3. Institutional reasons: These reasons also contribute to lags. For example, contractual
obligations may prevent firms from switching from one source of labor or raw material to
another. As another example, those who have placed funds in long term saving accounts for
fixed durations such as one year, three years or seven years are essentially “locked” in even
though money market conditions may be such that higher yields are available.
For the reasons just discussed, lag occupies a central role in economics. This is clearly reflected in
the short-run–long-run methodology of economics. It is for this reason we say that short-run price or
income elasticities are generally smaller (in absolute value) than the corresponding long-run
elasticities or that short-run marginal propensity to consume is generally smaller than long-run
marginal propensity to consume.
129
Econometrics II Reading Material for Distance Learners

Activity 3.1
All econometric models are essentially dynamic. Argue for or against this statement

3.5. Infinite Distributed Lag

Though so far we assumed a finite number of lags imposed on Xt, some lags may be infinite and it
may have its own practical importance. For example, the investment in building highways and roads
several decades ago may still have an effect on today’s growth in GNP. In this case, we write
equation (3.1) as:
Yt = α +∑ +ut t= 1,2, -------, T (3.8)
There are an infinite number of βi’s to estimate with only T observations. This can only be feasible if
more structure is imposed on the βi’s.
First, we normalize these βi’s by their sum, i.e., let wi = βi/β where β =∑ . If all the βi’s have the
same sign, then the βi’s take the sign of β and 0 ≤ wi ≤ 1 for all i, with ∑ =1. This means that
the wi’s can be interpreted as probabilities.
In fact, Koyck (1954) imposed the geometric lag on the wi’s, i.e., wi = (1 − λ)λi for i = 0, 1, . . . ,∞.
Substituting βi = βwi = β(1 − λ)λi in (3.8) we get:
Yt = α + β(1 − λ) ∑ λX + ut (3.9)
Equation (3.9) is known as the infinite distributed lag form of the Koyck lag. The short-run effect
of a unit change in Xt on Yt is given by β(1 − λ); whereas the long-run effect of a unit change in Xt
on Yt is∑ = ∑ = . Implicit in the Koyck lag structure is that the effect of a unit
change in Xt on Yt declines the further back we go in time.
For example, if λ = 1/2, then β0 = β/2, β1 = β/4, β2 = β/8, etc. Defining LXt = Xt−1, as the lag
operator, we have LiXt = Xt−i, and (3.9) reduces to:
Yt = α + β (1 − λ) ∑ (λL)iXt + ut = α + β (1−λ) Xt/(1−λL) + ut (3.10)
where we have used the fact that ∑ = 1/(1−c). Multiplying the last equation by (1−λL) one
gets: Yt − λYt−1 = α (1 − λ) + β(1 − λ)Xt + ut − λut−1
or
Yt = λYt−1 + α (1 − λ) +β (1 − λ)Xt + ut − λut−1 (3.11)
This is the autoregressive form of the infinite distributed lag. It is autoregressive because it includes
the lagged value of Yt as an explanatory variable.
NB: Thus, autoregressive models can be derived from distributed lag models.

Note that we have reduced the problem of estimating an infinite number of βi’s into estimating λ
and β from (3.11). However, OLS would lead to biased and inconsistent estimates, because (3.11)
contains a lagged dependent variable as well as serially correlated errors. In fact the error in (3.11) is
a Moving Average process of order one, i.e., MA(1).
Moving Average Process:
Suppose that Ut is a purely random process with mean zero and variance δ2, then the process Xt
defined by: Xt = β0Ut + β1Ut-1 +---, + βm Ut-m is called a moving average process of order m and
denoted by MA(m).

130
Econometrics II Reading Material for Distance Learners

Autoregressive process:
Suppose again that ut is a purely random process with mean zero and variance δ2, then the process Xt
given by: Xt = α1Xt-1 + α2Xt-2 +---, + αr Xt-r + Ut is called autoregressive process of order r and
denoted by AR(r).
Autoregressive Moving Average Process (ARMA) is a combination of the AR and MA models.

Note that: AR(r) models can be estimated using OLS, but MA(m) models are complicated, because
we do not observe the lags of the error terms for which we want to estimate the parameters; thus,
ARMA(r,m) models are estimated using Maximum Likelihoods estimation method.

We digress at this stage to give two econometric models which would lead to equations resembling
(3.11).

I. Adaptive Expectations Model (AEM)

The modeling of expectations is frequently an important and difficult task of the applied economist
using time series data. This is especially true in macroeconomics, in that investment, saving, and the
demands for assets are all sensitive to expectations about the future.
As a makeshift solution, some models use an indirect technique known as the adaptive expectations
process. This involves a simple learning process in which, in each time period, the actual value of
the variable is compared with the value that had been expected. If the actual value is greater, the
expected value is adjusted upwards for the next period. If it is lower, the expected value is adjusted
downwards. The size of the adjustment is hypothesized to be proportional to the discrepancy
between the actual and expected value.
Suppose that output Yt is a function of expected sales Xt* and that the latter is unobservable, i.e.,
Yt = α + βXt* + ut , where expected sales are updated according to the following method
Xt* − Xt−1* = δ (Xt − Xt−1*) (3.12)
that is, expected sales at time t is a weighted combination of expected sales at time t − 1 and actual
sales at time t.
In fact, Xt* = δXt + (1 − δ)Xt-1* (3.13)
Equation (3.12) is also an error learning model, where one learns from past experience and adjusts
expectations after observing current sales. Using the lag operator L, (3.14) can be rewritten as
Xt* = δXt/[1 − (1 − δ)L].
Substituting this last expression in the above relationship, we get:
Yt = α + βδXt / [1 − (1 − δ)L] + ut (3.14)
Multiplying both sides of (3.14) by [1 − (1 − δ)L], we get:
Yt − (1 − δ)Yt−1 = α[(1 − (1 − δ)] + βδXt + ut − (1 − δ)ut−1 (3.15)
NB: (3.15) looks exactly like (3.11) with λ = (1 − δ).

131
Econometrics II Reading Material for Distance Learners

II. Partial Adjustment Model (PAM)

In the partial adjustment model it is assumed that the behavioral equation determines the desired
(or "target") value, Y t* , of the dependent variable, rather than the actual value, Yt.
Under this model there is a cost of being out of equilibrium and a cost of adjusting to that
equilibrium, i.e., Cost = a (Yt – Yt* )2 + b(Yt − Yt−1)2 (3.16)
where Yt* is the target or equilibrium level for Y , whereas Yt is the current level of Y .
The first term of (3.16) gives a quadratic loss function proportional to the distance of Yt from the
equilibrium level Yt*. The second quadratic term represents the cost of adjustment. Minimizing this
quadratic cost function with respect to Y, we get Yt = γYt* + (1−γ)Yt−1, where γ = a/(a+b).
Note that if the cost of adjustment was zero, then b = 0, γ = 1, and the target is reached immediately.
However, there are costs of adjustment, especially in building the desired capital stock.
Hence, Yt = γYt* + (1 − γ)Yt−1 + ut (3.17)
where we made this relationship stochastic.
If the true relationship is Yt* = α + βXt, then from (3.17)
Yt = γα + γβXt + (1 − γ)Yt−1 + ut (3.18)
and this looks like (3.11) with λ = (1 − γ), except for the error term, which is not necessarily MA(1)
with the Moving Average parameter λ.

Draw backs of Lagged models

Lagged values of the variables are important explanatory variables in most economic
relationships, because economic behavior in anyone period is to a great extent determined by
past experience and past patterns of behavior.
However, there are some practical problems we may encounter using lag models. If we lag ‘s’
periods, we loose ‘s’ observations as we have stated earlier. Furthermore, we are estimating one
extra ‘b’ with every lag. Therefore, there is double jeopardy with respect to loss of degrees of
freedom. The number of observations fall (because we are lagging the same series), and the number
of parameters to be estimated increase with every lagging variable introduced. Besides the loss of
degrees of freedom, the regressors in (3.1) are likely to be highly correlated with each other. In fact
most economic time series are usually trended and very highly correlated with their lagged values.
This introduces the problem of multicollinearity among the regressors and as we saw in Chapter 1,
the higher the multicollinearity among these regressors, the lower is the reliability of the regression
estimates.
In this model, OLS is still BLUE because the classical assumptions are still satisfied. All we have
done in (3.1) is introduce the additional regressors (Xt−1, . . . ,Xt−s). These regressors are uncorrelated
with the disturbances since they are lagged values of Xt, which are by assumption not correlated
with ut for every t.

132
Econometrics II Reading Material for Distance Learners

3.6. Estimation of Lag Models

3.6.1. Estimation of Autoregressive Models

Both the AEM and the PAM give equations resembling the autoregressive form of the infinite
distributed lag. In all cases, we end up with a lagged dependent variable and an error term that is
either Moving Average of order one as in (3.11), or just classical or autoregressive as in (3.18).
In this section we study the testing and estimation of such autoregressive or dynamic models.
If there is a Yt−1 in the regression equation and the ut’s are classical disturbances, as may be the case
in equation (3.18), then Yt−1 is said to be contemporaneously uncorrelated with the disturbance term
ut. In fact, the disturbances satisfy assumptions 1-4 of classical linear regression and E(Yt−1 ut) = 0
even though E(Yt−1 ut−1) ≠0. In other words, Yt−1 is not correlated with the current disturbance ut but
it is correlated with the lagged disturbance ut−1. In this case, as long as the disturbances are not
serially correlated, OLS will be biased, but remains consistent and asymptotically efficient. This
case is unlikely with economic data given that most macro time-series variables are highly trended.
More likely, the ut’s are serially correlated. In this case, OLS is biased and inconsistent.
Intuitively, Yt is related to ut, so Yt−1 is related to ut−1. If ut and ut−1 are correlated, then Yt−1 and ut are
correlated. This means that one of the regressors, lagged Y, is correlated with ut and we have the
problem of endogeneity.
A model with a lagged dependent variable and an autoregressive error term is estimated using
Instrumental Variables (IV). This method will be studied extensively in you post graduate studies.
In short, the IV method corrects for the correlation between Yt−1 and the error term by replacing Yt−1
with its predicted value t−1. The latter is obtained by regressing Yt−1 on some exogenous variables
say a set of Z’s, which are called a set of instruments for Yt−1. Since these variables are exogenous
and uncorrelated with ut, t−1 will not be correlated with ut.

3.6.2. Estimation of Distributed Lag Models

Granted that distributed-lag models play a highly useful role in economics, how does one estimate
such models?
Specifically, suppose we have the following distributed lag model in one explanatory variable:
Yt     0 X t   1 X t  2   2 xt  2        s X t  s    U t (3.19)
In (3.19) the length of the lag, that is, how far back into the past we want to go hasn’t been defined.
Such a model is called an infinite (lag) model, whereas models with specified lags are called a finite
(lag) distributed-lag model.
How do we estimate  and  in (3.19)? We may adopt two approaches:
A. Ad Hoc estimation of distributed-lag models
B. A priori restriction on  ' s by assuming that the  ' s follow some systematic pattern.

133
Econometrics II Reading Material for Distance Learners

A. Ad Hoc estimation of distributed lag models

Since the explanatory variable X t is assumed to be non-stochastic (or at least uncorrelated with the

disturbance term Ut), X t 1 , X t  2 and so on, are non-stochastic too. Therefore in principle, OLS can

be applied to the above model (3.19).

This is the approach taken by Alt and Tinbergen. They suggest that to estimate (3.19) one may
proceed sequentially; that is:
First, regress/estimate Y on X t , then Yt on X t , X  1 , then Yt on X t 1 , and X t  2 , and so on.

This sequential procedure will continue until the regression coefficients of the lagged variables start
becoming statistically insignificant and/or the coefficient of at least one of the variables changes
signs from positive to negative vice versa.

Consider the following hypothetical example.

Yˆt  8.37  0.17 X t

Yˆt  8.27  0.111X t  0.06 X t 1

Yˆt  8.27  0.109 X t  0.071X t 1  0.055 X t 2

Yˆt  8.32  0.108 X t  0.063 X t 1  0.022 X t  2  0.020 X t 3

Alt choose the second regression as the “best” one because in the last two equation the sign of X t  z

was not stable and in the last equation the sign of X t 3 was negative, which may be difficult to
interpret economically.
Although seemingly straight forward, ad hoc estimation suffers from many drawbacks, such as the
following:
a. There is no guide as to what is the maximum lag length
b. As one estimates successive lags, there are fewer degrees of freedom left, making
statistical inference somewhat unstable
c. More importantly, in economic time series data, successive values (lags) tend to be
highly correlated; hence multicollinearity rears its hideous head.
As noted in Chapter 1, multicollinearity leads to imprecise estimation; that is, the
standard errors tend to be large in relation to the estimated coefficients. As a result, based
on the routinely computed t ratios, we may tend to declare (erroneously), that a lagged
coefficient(s) is statistically insignificant.
d. The sequential search for the lengths of lags opens the researcher to the charge of
data mining.
In view of the preceding problems, the ad hoc estimation procedure has very little to recommend it.
Some prior or theoretical considerations are brought to bear upon the various  ’s if we are to make
headway with the estimation problem.

134
Econometrics II Reading Material for Distance Learners

B. Methods based on A priori restriction on  ' s

Method I: The koyck approach to distributed lag models
Koyck has proposed an ingenious method of estimating distributed-lag models. Suppose we start
with the infinite lag distributed-lag model (3.19).
Assuming that the β’s are all of the same sign, Koyck assumes that they decline geometrically as
follows.
βk = β0λk k = 0, 1, . . . (3.20)
where λ, such that 0 < λ < 1, is known as the rate of decline, or decay, of the distributed lag and
where 1 − λ is known as the speed of adjustment.
What (3.20) postulates is that each successive β coefficient is numerically less than each preceding β
(this statement follows since λ < 1), implying that as one goes back into the distant past, the effect of
that lag on Yt becomes progressively smaller, a quite plausible assumption. After all, current and
recent past incomes are expected to affect current consumption expenditure more heavily than
income in the distant past.
In order to reduce the number of lags in the given distributed lag model, the present model makes an
assumption that the impact of explanatory variables (on the dependent variable) it the most distant
past is less than what is in most recent periods.
More specifically, the Koyck lag formulation assumes that the weights (impacts) are declining
continuously.
Assume the original model is:
Y   0   0 X t   1 X t 1   2 xt 2        s X t  s    U t (3.21)

U ~ N(0, 2 )
 (u i u j )  0

(u i x i )  0

According to Koyck:  i  i  0

  1   0 ,  2  2  0
As we saw in our previous section, the partial sums of the standardized βi tell us the proportion of
the long-run, or total, impact felt by a certain time period. In practice, though, the mean or median
lag is often used to characterize the nature of the lag structure of a distributed lag model.

The Median Lag

The median lag is the time required for the first half, or 50 percent, of the total change in Y following
a unit sustained change in X. For the Koyck model, the median lag is as follows:
Koyck model: Median lag= − log 2/log λ (3.22)
Thus, if λ = 0.2 the median lag is 0.4306, but if λ = 0.8 the median lag is 3.1067. Verbally, in the
former case 50 percent of the total change in Y is accomplished in less than half a period, whereas in
the latter case it takes more than 3 periods to accomplish the 50 percent change. But this contrast
should not be surprising, for as we know, the higher the value of λ the lower the speed of adjustment,
and the lower the value of λ the greater the speed of adjustment.

135
Econometrics II Reading Material for Distance Learners

The Mean Lag

Provided all βk are positive, the mean, or average, lag is defined as:
∑
∑ (3.23)

which is simply the weighted average of all the lags involved, with the respective β coefficients
serving as weights. In short, it is a lag-weighted average of time.
For the Koyck model the mean lag is Koyck model: Mean lag = λ/1 – λ (3.24)
Thus, if λ = 12 , the mean lag is 1.
From the preceding discussion it is clear that the median and mean lags serve as a summary measure
of the speed with which Y responds to X.

Note the following features of the Koyck scheme:

1. Assuming non-negative values for  , koyck rules out the  ’s from changing sign,
2. by assuming  <1, lesser weight is assigned to the decline  ’s than current one;
and
3. The sum of the β’s, which gives the long run multiplier, is a finite amount in
Koyck scheme and is.

 1 
 i  0 
1  
i 0 (3.25)
Substituting the values of  ’s in the original model (3.19) we obtain:

Yt   0   0 X t  ( 0 ) X t 1  ( 2  0 ) xt  2        U t
(3.26)
As it stands, the model is still not amenable to easy estimation since a large (literally infinite)
number of parameters remain to be estimated and the parameter λ enters in a highly nonlinear form:
Strictly speaking, the method of linear (in the parameters) regression analysis cannot be applied to
such a model.
Lagging by one period and multiply by  we get:
Yt 1   0  (  0 ) X t 1  ( 2  0 ) xt 2  (3  0 ) xt 3        U t 1 (3.27)
Subtracting (3.27) from (3.26) and rearranging we obtain:
Yt  Yt 1   (1   )   0 X t  (U t  U t 1 )

Let  *   (1   ), Vt  U t  U t 1

Yt   *   0 X t  Yt 1  Vt (3.28)
where vt = (ut − λut−1), a moving average of ut and ut−1.
The above procedure of transformation is known as koyck transformation. If the koyck hypothesis
concerning the lag scheme and assumptions concerning Vt are accepted, ordinary least square can be
applied to obtain estimators of  ,  0 and  . From these estimates, the estimates of the original

parameters  0 ,  1 ,  2 ,    k can be easily obtained through:

*
ˆ  ,  i  i  0
1 

136
Econometrics II Reading Material for Distance Learners

Comparing (3.25) with (3.19), we see the tremendous simplification accomplished by Koyck.
Whereas before we had to estimate α and an infinite number of β’s, we now have to estimate only
three unknowns: α, β0, and λ.
Now there is no reason to expect multicollinearity. In a sense multicollinearity is resolved by
replacing Xt−1, Xt−2, . . . , by a single variable, namely, Yt−1.
However, the following features of the koyck transformation may be taken in to account:
1. We started with a distributed-lag model but ended up with an autoregressive model because
Yt−1 appears as one of the explanatory variables. This transformation shows how one can
“convert’’ a distributed-lag model into an autoregressive model.
2. The appearance of Yt−1 is likely to create some statistical problems. Yt−1, like Yt , is stochastic,
which means that we have a stochastic explanatory variable in the model. Recall that the
classical least-squares theory is predicated on the assumption that the explanatory variables
either are non-stochastic or, if stochastic, are distributed independently of the stochastic
disturbance term. The lagged variable Yt 1 is not also independent of the error term Vt i.e.

(U t Yt 1 )  0 this is because Yt is directly dependent on Vt . Similarly Yt 1 on Vt 1 . But since

Vt and Vt 1 are not independent, Yt 1 will obviously be related to Vt .

Hence, we must find out if Yt−1 satisfies this assumption.
3. In the original model (3.19) the disturbance term was ut , whereas in the transformed model it
is vt = (ut − λut−1). The statistical properties of vt depend on what is assumed about the
statistical properties of ut , for, as shown later, if the original ut ’s are serially uncorrelated,
the vt ’s are serially correlated.
It can be seen as under
(VtVt 1 )  (U t  U t 1 )(U t 1  U t 2 )

 (U t U t 1  U t U t 2  U 2 t 1  2U t 1U t 2 )

  (U 2 t 1 )

  u2  0

Therefore, we may have to face up to the serial correlation problem in addition to the stochastic
explanatory variable Yt−1.
4. The presence of lagged Y violates one of the assumptions underlying the Durbin–Watson d
test. Therefore, we will have to develop an alternative to test for serial correlation in the
presence of lagged Y. One alternative is the Durbin h test, which we discussed in previously.
Due to these problems, the Koyck transformation of the distributed lag model will give rise to biased
and inconsistent estimates. In addition to these estimation problem, the Koyck hypothesis is quite
restrictive in the sense that it assumes that impact of past periods decline successively in a specific
way.

137
Econometrics II Reading Material for Distance Learners

Method II: Rationalization of the Koyck Model

A. Adaptive Expectations (AE) Model
The Koyck model is ad hoc since it was obtained by a purely algebraic process; it is devoid of any
theoretical foundation. But this gap can be filled if we start from a different perspective.
Illustration:
Suppose we postulate the following model:
Yt   0   1 X t*  u t (3.29)

Where: Y=demand for money (real cash balances)

X*= equilibrium, optimum, expected long-run or normal rate of interest
u= error term
Equation (3.29) postulates that the demand for money is a function of expected (in the sense of
anticipation) rate of interest.
Since the expectational variable X* is not directly observable, let’s propose the following hypothesis
(the hypothesis we made so far) about how expectations are formed as:
X *  X t*1   ( X t  X t*1 ) , where  , such that 0<   1, is known as the coefficient of expectation.

The hypothesis is known as the adaptive expectation, progressive expectation, or error learning
hypothesis, popularized by Cagan and Friedman.
The hypothesized equation states that expectations are revised each period by a fraction  of the gap
between the current value of the variable and its previous expected value. Thus, for our model this
would mean that expectations about interest rates are revised each period by a fraction  of the
discrepancy between the rate of interest observed in the current period and what its anticipated value
had been in the previous period. Another way of stating this would be to write the hypothesized
equation as:
X t*  X t  (1   ) X t*1 (3.30)

which shows that the expected value of the rate of interest at time t is a weighted average of the
actual value of the interest rate at time ‘t’ and its value expected in the previous period, with weights
of ‘  ’ and ‘1-  ’, respectively.

 If  =1, X t*  X t , meaning that expectations are realized immediately and fully, that is, in
the same time period.
 If, on the other hand,  =0, X t*  X t*1 , meaning that expectations are static, that is,

conditions prevailing today will be maintained in all subsequent periods. Expected future
values then become identified with current values.” Substituting (3.27) into (3.26), we obtain
 
Yt   0   1 X t  (1   ) X t*1  u t

  0   1X t   1 (1   ) X t*1  u t (3.31)

Now, lag equation (3.29) by one period, multiply it by 1-  , and subtract the product from (3.31).
After simple algebraic manipulations, we obtain:
Yt   0  1 X t  (1   )Yt 1  u t  (1   )u t 1

  0  1 X t  (1   )Yt 1  vt (3.32)

where vt  u t  (1   )u t 1 .

138
Econometrics II Reading Material for Distance Learners

Let us note the difference between (3.29) and (3.32). In the former,  1 measures the average
response of Y to a unit change in X*, the equilibrium or long-run value of X. In (3.29), on the other
hand, 1 measures the average response of Y to a unit change in the actual or observed value of X.
These responses will not be the same unless, of course,  =1, that is, the current and long-run values
of X are the same. In practice, we first estimate (3.32). Once an estimate of  is obtained from the

coefficient of lagged Y, we can easily compute  1 by simply dividing the coefficient of X t ( 1 )

by  .
The similarity between the adaptive expectation model and the Koyck model should be readily
apparent although the interpretations of the coefficients in the two models are different.
Note that like the Koyck model, the adaptive expectations model is autoregressive and its error term
is similar to the Koyck error term.

B. Stock Adjustment, or Partial Adjustment Model

The adaptive expectation model is one way of rationalizing the Koyck model. Another
rationalization is provided by Marc Nerlove in the so-called stock adjustment or partial adjustment
model. We developed this model in our previous section, but for the sake of elaboration we will try
to discuss it with the use of some economic theories.
To illustrate this model, consider the flexible accelerator model of economic theory, which
assumes that there is equilibrium, optimal, desired, or long-run amount of capital stock needed to
produce a given output under the given state of technology, rate of interest, etc. For simplicity
assume that this desired level of capital K t* is a linear function of output X as follows:

K t*   0   1 X t  u t (3.33)

Since the desired level of capital is not directly observable, Nerlove postulates the following
hypothesis, known as the partial adjustment, or stock adjustment, hypothesis:
K t  K t 1   ( K t*  K t 1 ) (3.34)

Where,  such that 0    1 , is known as the coefficient of adjustment and where K t  K t 1 is the

actual change in capital stock and ( K t*  K t 1 ) is the desired change. Since K t  K t 1 , the change in

capital stock between two periods, it is nothing but investment, (3.31) can alternatively be written as:
I t   (Yt *  Yt 1 ) (3.35)

Where, I t  Investment in time period t.

Equation (3.34) postulates that the actual change in capital stock (investment) in any given time
period t is some fraction  of the desired change for that period. If  =1, it means that the actual
stock of capital is equal to the desired stock; that is, actual stock adjusts to the desired stock
instantaneously (in the same period). However, if  =0, it means that nothing changes since actual
stock at time t is the same as that observed in the previous time period. Typically,  is expected to
lie between these extremes since adjustment to the desired stock of capital is likely to be incomplete
because of rigidity, inertia, contractual obligations, etc. – hence the name partial adjustment model.
Note that the adjustment mechanism (3.34) alternatively can be written as:
K t  K t*  (1   ) K t 1 (3.36)

139
Econometrics II Reading Material for Distance Learners

showing that the observed capital stock at time t is a weighted average of the desired capital stock at
that time and the capital stock existing in the previous time period,  and (1-  ) being the weights.
Now substitution of (3.33) into (3.36) gives:
K t   ( 0   1 X t  u t )  (1   ) K t 1

  0   1 X t  (1   ) K t 1  u t (3.37)

This model is called the Partial Adjustment Model.

Since (3.33) represents the long-run, or equilibrium, demand for capital stock, (3.37) can be called
the short-run demand function for capital stock since in the short run the existing capital stock may
not necessarily be equal to its long-run level.
Once we estimate the short-run function (3.37) and obtain the estimate of the adjustment coefficient
 (from the coefficient of K t 1 ), we can easily derive the long-run function by simply dividing  0

and  1 by  and omitting the lagged K term, which will then give (3.33).
The partial adjustment model resembles both the Koyck and adaptive expectation models in that it is
autoregressive. But it has a much simpler disturbance term: the original disturbance term u t

multiplied by a constant  . But bear in mind that although similar in appearance, the adaptive
expectation and partial adjustment models are conceptually very different.
The former is based on uncertainty (about the future course of prices, interest rates, etc.), whereas
the latter is due to technical or institutional rigidities, inertia, cost of change, etc. However, both of
these models are theoretically much sounder than the Koyck model.
Since in appearance the adaptive expectations and partial adjustment models are indistinguishable,
the γ coefficient of the adaptive expectations model can also be interpreted as the δ coefficient of
the stock adjustment model if we assume that the latter model is operative in the present case
The important point to keep in mind is that since Koyck, adaptive expectations, and stock adjustment
models – apart from the difference in the appearance of the error term – yield the same final
estimating model.

Method III. Combination of Adaptive Expectations and Partial Adjustment Models

Consider the following model:
Yt *   0   1 X t*  u t (3.38)

where Yt * = desired stock of capital and X t* = expected level of output.

Since both Yt * and X t* are not directly observable, one could use the partial adjustment mechanism

for Yt * and the adaptive expectations model for X t* to arrive at the following estimating equation.

Yt   0   1X t  (1   )  (1   )Yt 1  (1   )(1   )Yt  2  u t   (1   )u t 1 

  0   1 X t   2Yt 1   3Yt  2  v t (3.39)

where vt   u t  (1   )u t 1  .

This model too is autoregressive, the only difference from the purely adaptive expectations model
being that Yt 2 appears along with Yt 1 as an explanatory variable. Like Koyck and the AE models,

the error term in (3.39) follows a moving average process.

140
Econometrics II Reading Material for Distance Learners

Another feature of this model is that although the model is linear in the  ’s, it is nonlinear in the
original parameters.
A celebrated application of (3.38) has been Friedman’s permanent income hypothesis, which states
that “permanent” or long-run consumption is a function of “permanent” or long-run income.
The estimation of (3.39) presents the same estimation problems as the Koyck’s or the AE model in
that all these models are autoregressive with similar error structures.

Summary of Estimation of Autoregressive Models

From our discussion so far we have the following three models:
Koyck
Yt = α (1 − λ) + β0Xt + λYt−1 + (ut − λut−1)
Adaptive expectation
Yt = γβ0 + γβ1Xt + (1 − γ)Yt−1 + [ut − (1 − γ)ut−1]
Partial adjustment
Yt = δβ0 + δβ1Xt + (1 − δ)Yt−1 + δut
All these models have the following common form:
Yt = α0 + α1Xt + α2Yt−1 + vt
that is, they are all autoregressive in nature.
Therefore, we must now look at the estimation problem of such models, because the classical least-
squares may not be directly applicable to them. The reason is two-fold: the presence of stochastic
explanatory variables and the possibility of serial correlation.
Now, as noted previously, for the application of the classical least-squares theory, it must be shown
that the stochastic explanatory variable Yt−1 is distributed independently of the disturbance term vt.
To determine whether this is so, it is essential to know the properties of vt. If we assume that the
original disturbance term ut satisfies all the classical assumptions, such as E(ut) = 0, var (ut) = σ2 (the
assumption of homoscedasticity), and cov (ut, ut+s) = 0 for s ≠ 0 (the assumption of no
autocorrelation), vt may not inherit all these properties.
Consider, for example, the error term in the Koyck model, which is vt = (ut − λut−1).
Given the assumptions about ut, we can easily show that vt is serially correlated because
E(vtvt−1) = −λσ2 which is nonzero (unless λ happens to be zero). And since Yt−1 appears in the Koyck
model as an explanatory variable, it is bound to be correlated with vt (via the presence of ut−1 in it).
As a matter of fact, it can be shown that cov [Yt−1, (ut − λut−1)] = −λσ2 which is the same as previous
proof. The reader can verify that the same holds true of the adaptive expectations model.
What is the implication of the finding that in the Koyck model as well as the adaptive expectations
model the stochastic explanatory variable Yt−1 is correlated with the error term vt?
As noted previously, if an explanatory variable in a regression model is correlated with the
stochastic disturbance term, the OLS estimators are not only biased but also not even consistent;
that is, even if the sample size is increased indefinitely, the estimators do not approximate their true
population values. Therefore, estimation of the Koyck and adaptive expectation models by the usual
OLS procedure may yield seriously misleading results.

141
Econometrics II Reading Material for Distance Learners

However, the partial adjustment model is different. In this model vt = δut , where 0 < δ ≤ 1.
Therefore, if ut satisfies the assumptions of the classical linear regression model given previously, so
will δut . Thus, OLS estimation of the partial adjustment model will yield consistent estimates
although the estimates tend to be biased (in finite or small samples). Intuitively, the reason for
consistency is this: Although Yt−1 depends on ut−1 and all the previous disturbance terms, it is not
related to the current error term ut.
Therefore, as long as ut is serially independent, Yt−1 will also be independent or at least uncorrelated
with ut, thereby satisfying an important assumption of OLS, namely, non-correlation between the
explanatory variable(s) and the stochastic disturbance term.

Activity 3.2: Check whether the following question is correct or wrong and also substantiate your
answer with possible justifications.
The Koyck model will not make much sense if some of the distributed lag coefficients are positive
and some negative.

Method IV: The Almon approach to Distributed Lag Models

Although used extensively in practice, the Koyck distributed-lag model is based on the assumption
that the β coefficients decline geometrically as the lag lengthens. This assumption may be too
restrictive in some situations.
The Almon lag model possesses two advantages over the Koyck procedure:
First, it does not violate any of the ordinary least square basic assumptions concerning the
disturbance term.
Second, it is far more flexible than Koyck method in terms of the form of lag scheme. It is because;
this method does not hypothesize any form of lag in advance.
This model assumes that any pattern of lag scheme among  ' s can be described by polynomial.
This idea is based on a theorem in mathematics known as Weierstrass’s theorem, which states that
under general conditions a curve may be approximated by a polynomial whose degree is one more
than the number of turning points in the curve. Suppose that the  ' s in a given distributed lag model
are expected to decrease first, then increase and again decrease
Suppose our original model to be estimated is:
Y   0   0 X t   1 X t 1   2 xt  2        s X t  s    U t (3.37)

assuming the above polynomial:  i  a 0  a1i  a 2 i 2  a3 i 3 where a0 , a1 , a 2 , and a 3 are parameters

to be estimated.
We are, now in a position to obtain all  ' s by setting i equal to the value of the subscript of the
particular coefficient.
 0  a0

 1  a 0  a1  a 2  a 3
 2  a0  2a1  4a 2  8a3
 3  a 0  3a1  9a 2  27 a3

 k  a 0  ka1  k 2 a 2  k 3 a3

142
Econometrics II Reading Material for Distance Learners

Naturally therefore, what is needed to be estimated is only four parameters of the polynomial
function:  i  a 0  a1i  a 2 i 2  a3 i 3 .

Obtaining the values of a0 , a1 , a 2 , and a 3 , we are able to estimate all the parameters of the original

distributed lag model.

Substituting values of  ' s in the original model,

Yt   0   0 X t  (a 0  a1  a 2  a3 ) X t 1  (a0  2a1  4a 2  8a3 ) xt 2      U t

 k   k   k 2   k 3 
Yt   0  a 0   X t i   a1   iX t i   a 2   i X t i   a3   i X t i   U i
 i 1   i 1   i 1   i 1 
Yt   0  a 0 w0  a1 w1  a 2 w2  a 3 w3  U t (3.38)
k k
where: w0   X t i , w2   iX t i
i 1 i 1

This is the final form (or transformed form) of Almon Lag model. We can now apply OLS method
to estimate ˆ 0 , aˆ 0 , aˆ1 , aˆ 2 , and aˆ 3 to obtain  ' s in the original form.
Note that: vt remains in its original form.

3.7. The Granger Causality Test

Although regression analysis deals with the dependence of one variable on other variables, it does
not necessarily imply causation. In other words, the existence of a relationship between variables
does not prove causality or the direction of influence.
But in regressions involving time series data, the situation may be somewhat different because time
does not run backward. That is, if event A happens before event B, then it is possible that A is
causing B. However, it is not possible that B is causing A. In other words, events in the past can
cause events to happen today. Future events cannot.
This is roughly the idea behind the so-called Granger causality test. But it should be noted clearly
that the question of causality is deeply philosophical with all kinds of controversies. At one extreme
are people who believe that “everything causes everything,” and at the other extreme are people who
deny the existence of causation whatsoever.
To explain the Granger test, we will consider the often asked question in macroeconomics: Is it GDP
that “causes” the money supply M (GDP→ M) or is it the money supply M that causes GDP
(M→GDP), where the arrow points to the direction of causality.
The Granger causality test assumes that the information relevant to the prediction of the respective
variables, GDP and M, is contained solely in the time series data on these variables.
The test involves estimating the following pair of regressions:

GDPt = ∑ +∑ +u1t (3.39)

Mt =∑ +∑ +u2t (3.40)
Where: it is assumed that the disturbances u1t and u2t are uncorrelated. Since we have two variables,
we are dealing with bilateral causality.
143
Econometrics II Reading Material for Distance Learners

Equation (3.39) postulates that current GDP is related to past values of itself as well as that of M,
and (3.40) postulates a similar behavior for M.
Note that these regressions can be cast in growth forms, ̇ and ̇ , where a dot over a variable
indicates its growth rate. We now distinguish four cases by regressing/ estimating the above two
equations and check for the significance of respective coefficients:
1. Unidirectional causality from M to GDP is indicated if the estimated coefficients on the lagged
M in (3.39) are statistically different from zero as a group (i.e.,∑ ≠ 0) and the set of estimated
coefficients on the lagged GDP in (3.40) is not statistically different from zero (i.e., ∑ = 0)
2. Conversely, unidirectional causality from GDP to M exists if the set of lagged M coefficients in
(3.39) is not statistically different from zero (i.e., ∑ = 0) and the set of the lagged GDP
coefficients in (3.40) is statistically different from zero (i.e., ∑ ≠ 0).
3. Feedback, or bilateral causality, is suggested when the sets of M and GDP coefficients are
statistically significantly different from zero in both regressions.
4. Finally, independence is suggested when the sets of M and GDP coefficients are not statistically
significant in both the regressions.
More generally, since the future cannot predict the past, if variable X (Granger) causes variable Y,
then changes in X should precede changes in Y.
Therefore, in a regression of Y on other variables (including its own past values) if we include past
or lagged values of X and it significantly improves the prediction of Y, then we can say that X
(Granger) causes Y. A similar definition applies if Y (Granger) causes X.
The steps involved in implementing the Granger causality test are as follows. We illustrate these
steps with the GDP-money example given in equation (3.39):
Step 1: Regress/ Estimate current GDP on all lagged GDP terms and other variables, if any, but do
not include the lagged M variables in this regression. This is the restricted regression. From this
regression obtain the restricted residual sum of squares, RSSR.
Step 2: Now run the regression including the lagged M terms. In the language of Econometrics, this
is the unrestricted regression.
From this regression obtain the unrestricted residual sum of squares, RSSUR.
Step 3: The null hypothesis is H0: ∑ = 0, that is, lagged M terms do not belong in the regression.
Step 4: To test this hypothesis, we apply the F test given by (3.41), namely,
(RSS − RSSUR)/
F= (3.41)
RSSUR/( − )

which follows the F distribution with m and (n-k) degree of freedom for numerator and denominator,
respectively. In the present case m is equal to the number of lagged M terms and k is the number of
parameters estimated in the unrestricted regression.
Step 5: Decision: If the computed F value exceeds the critical F value at the chosen level of
significance, we reject the null hypothesis, in which case the lagged M terms belong in the
regression. This is another way of saying that M causes GDP.
Step 6: Steps 1 to 5 can be repeated to test the model (3.40), that is, whether GDP causes M or not.

Before we illustrate the Granger causality test, there are several things that need to be noted:
1. It is assumed that the two variables, GDP and M, are stationary.

144
Econometrics II Reading Material for Distance Learners

Stationarity is the most vital and common assumption in Time series analysis. A stationary process
defined as a series/sequence without trend and seasonality as well as constant mean, variance, and
autocorrelation structure over time. If either of these changes over time, the series is
non-stantionary.
Empirics and visual inspection tells us that most macroeconomic variables are non-stantionary
(either consistently increasing or decreasing on average). If we apply OLS for non-stantionary
series, the estimated parameters become significant and contain trend (a pattern that exhibits a
tendency either to grow or to decrease fairly steadily over time); which is called spurious
regression which is resulted in wrong conclusions and policy implications.
In standard linear regression modeling, stationarity is important to ensure consistent coefficient
estimates and avoid spurious results. Thus, before estimation, we have to assure stationarity of all
variables.
Sometimes taking the first differences of the variables makes them stationary, if they are not already
stationary in the level form.
2. The number of lagged terms to be introduced in the causality tests is an important practical
question. It should be added that the direction of causality may depend critically on the number of
lagged terms included.
3. We have assumed that the error terms entering the causality test are uncorrelated. If this is not the
case, appropriate transformation, may have to be taken.
4. Since our interest is in testing for causality, one need not present the estimated coefficients of
models (3.39) and (3.40) explicitly, rather we just stick to the results of the F test given in (3.41).

Example
R. W. Hafer used the Granger test to find out the nature of causality between GNP (rather than GDP)
and M for the United States for the period 1960-I to 1980-IV (as cited by Gujarati).
̇ and Ṁ and used
Instead of using the gross values of these variables, he used their growth rates, GNP
four lags of each variable in the two regressions given previously. The results were as follows:
The null hypothesis in each case is that the variable under consideration does not “Granger cause”
the other variable.
Direction of causality F value Decision
Ṁ →GNP
̇ 2.68 Reject
̇ → Ṁ
GNP 0.56 Do not reject
These results suggest that the direction of causality is from money growth to GNP growth since the
estimated F is significant at the 5 percent significance level; the critical F value is 2.50 (for 4 and 71
df). On the other hand, there is no “reverse causation” from GNP growth to money growth, since the
F value is statistically insignificant.

145
Econometrics II Reading Material for Distance Learners

Chapter Summary
1. For psychological, technological, and institutional reasons, a regressand may respond to a
regressor(s) with a time lag. Regression models that take into account time lags are known as
dynamic or lagged regression models.
2. There are two types of lagged models: distributed-lag and autoregressive.
In the former, the current and lagged values of regressors are explanatory variables. In the latter, the
lagged value(s) of the regressand appear as explanatory variables.
3. A purely distributed-lag model can be estimated by OLS, but in that case there is the problem of
multicollinearity since successive lagged values of a regressor tend to be correlated.
4. As a result, some shortcut methods have been devised. These include the Koyck, the adaptive
expectations, and partial adjustment mechanisms, the first being a purely algebraic approach and the
other two being based on economic principles.
5. But a unique feature of the Koyck, adaptive expectations, and partial adjustment models is
that they all are autoregressive in nature in that the lagged value(s) of the regressand appear as one
of the explanatory variables.
6. Autoregressiveness poses estimation challenges; if the lagged regressand is correlated with the
error term, OLS estimators of such models are not only biased but also are inconsistent. Bias and
inconsistency are the case with the Koyck and the adaptive expectations models; the partial
adjustment model is different in that it can be consistently estimated by OLS despite the presence of
the lagged regressand.
7. To estimate the Koyck and adaptive expectations models consistently, the most popular method is
the method of instrumental variable. The instrumental variable is a proxy variable for the lagged
regressand but with the property that it is uncorrelated with the error term.
8. An alternative to the lagged regression models just discussed is the Almon polynomial
distributed-lag model, which avoids the estimation problems associated with the autoregressive
models. The major problem with the Almon approach, however, is that one must pre-specify both
the lag length and the degree of the polynomial. There are both formal and informal methods of
resolving the choice of the lag length and the degree of the polynomial.

Check Your Progress for Activities

Activity 3.1
False
Econometric models are dynamic if they portray the time path of the dependent variable in relation
to its past values. Models using cross-sectional data are not dynamic, unless one uses time series or
panel regression models with lagged values of the regressand.

Activity 3.2
True
The Koyck model assumes that all the distributed lag coefficients have the same sign.

146
Econometrics II Reading Material for Distance Learners

Review Exercises
1. Differentiate the following concepts
a. Moving Average Process vs Autoregressive process
b. Autoregressive vs Distributed lag models
c. Adaptive Expectation (AE) vs Partial Adjustment(PA) Models
2. Explain with a brief reason whether the following statements are true, false, or uncertain:
a. If the Koyck and adaptive expectations models are estimated by OLS, the estimators will be
biased but consistent.
b. In the partial adjustment model, OLS estimators are biased in finite samples.
c. In the presence of a stochastic regressor(s) and an autocorrelated error term, the method of
instrumental variables will produce unbiased as well as consistent estimates.
d. In the presence of a lagged regressand as a regressor, the Durbin– Watson d statistic to detect
autocorrelation is practically useless.
e. The Granger test is a test of precedence rather than a test of causality.
3. Consider the model: Yt = α + β1X1t + β2X2t + β3Yt−1 + vt
Suppose Yt−1 and vt are correlated. To remove the correlation, suppose we use the following
instrumental variable approach:
First regress Yt on X1t and X2t and obtain the estimated t from this regression. Then regress
Yt = α + β1X1t + β2X2t + β3 t−1 + vt
where t−1 are estimated from the first-stage regression.
How does this procedure remove the correlation between Yt−1 and vt in the original model?
4. Assume that prices are formed according to the following adaptive expectations hypothesis:
Pt* = γ Pt−1 + (1 − γ) Pt−1*
where P* is the expected price and P the actual price.
Complete the following table, assuming γ = 0.5*:

Period P* P
t−3 100 110
t−2 - 125
t−1 - 155
t - 185
t +1 - —

5. What are the advantages of using Almon Polynomial Distributed-lag Model over that of
Koyck Model and what are its draw backs
6. Consider the Koyck (or for that matter the adaptive expectation) model given in (3.28), namely,
Yt = α(1 − λ) + β0Xt + λYt−1 + (ut − λut−1)
Suppose in the original model ut follows the first-order autoregressive scheme ut − ρu1−t = εt , where
ρ is the coefficient of autocorrelation and where εt satisfies all the classical OLS assumptions.
a. If ρ = λ, can the Koyck model be estimated by OLS?
b. Will the estimates thus obtained be unbiased? Consistent? Why or why not?
c. How reasonable is it to assume that ρ = λ?
7. Evaluate the median lag for λ = 0.2, 0.4, 0.6, 0.8.
147
Econometrics II Reading Material for Distance Learners

CHAPTER FOUR
INTRODUCTION TO SIMULTANEOUS EQUATION MODELS

Chapter Contents
Dear distance students, this chapter consists of the following major sub-topics:
 Introduction
 Simultaneous Dependence of Economic variables
 Definitions of Some Concepts
 Consequences of Simultaneity bias
 Problems of simultaneous equation models
 The Identification Problem

Chapter Objectives
At completion of this chapter, students will be able to:
 recognize the nature of simultaneous equations
 define the basic concepts commonly used in simultaneous equation models
 prove the consequences of simultaneity bias
 know problems of simultaneous equation models
 verify the identifiability of simultaneous equations using order and rank conditions

148
Econometrics II Reading Material for Distance Learners

4.1. Introduction
One important assumption of the basic linear regression model or the application of least squares to
a single equation assumes, among others, that the explanatory variables are truly exogenous (i.e., the
error term has to be uncorrelated with the explanatory variables); thus, there is one-way causation
between the dependent variable Y and the explanatory X's. If the explanatory variables are in fact
correlated with the error term, one of the classical linear regression (OLS) assumption: (Cov(Xu=0)
is violated and then become (Cov(Xu) ≠ 0), which is called endogenity problem. This is a very series
problem in model estimation even sever than multicollinarity, hetroscedasticity and autocorrelation problems.
If we have a two-way causation in a function this implies that the function cannot be treated in
isolation as a single equation model, but belongs to a wider system of equations which describes the
relationships among all the relevant variables. If Y = f(X), but also X = f(Y) we are not allowed to use a
single- equation model for the description of the relationship between Y and X. We must use a multi-
equation model, which would include separate equations in which Y and X would appear as
endogenous variables, although they might appear as explanatory in other equations of the model.
A system describing the joint dependence of variables is called a system of simultaneous equation
or simultaneous equations model. The number of equations in such models is equal to the number
of jointly dependent or endogenous variables involved in the phenomenon under analysis. 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.
Thus, the application of Ordinary Least Squares (OLS) estimation for the model with the endogenity
problem, caused by simultaneity yields biased and inconsistent estimates of the parameters.
The bias arising from application of such procedure of estimation which treats each equation of the
simultaneous equations model as though it were a single model is known as simultaneity bias or
simultaneous equation bias.
In this chapter we will relax this assumption by including additional equations to the model that
explains where the correlation is coming from, and discuss the conditions that need to be fulfilled to
receive consistent estimate.

4.2. Simultaneous Dependence of Economic variables

Given the nature of economic phenomena it is evident that in reality there might be bidirectional
causation among different economic variables since most economic variables are interrelated.
Economists formulate models for consumption, production, investment, money demand and money
supply, labor demand and labor supply to attempt to explain the workings of the economy.
These behavioral equations are estimated equations by equation or jointly as a system of equations.
Some examples will illustrate the meaning of simultaneous relationships and the violation of
assumption of ordinary least squares: (Cov(Xu)≠0), which creates what is known as
simultaneous equations bias.

149
Econometrics II Reading Material for Distance Learners

Example 1: Suppose we want to estimate the demand for food. We know from economic theory
that the demand for any particular commodity depends on its price, P, on other prices, Po, and on
income, Y, cetrius paribus. Thus we may write the demand function for food as
Q = β0 + β1P + β2Po + β3Y + u (4.1)
where Q = quantity demanded , P = price of food , Po = price of other
commodities , Y = income and u = random variable.
If we apply least squares to this equation we will obtain biased estimates of β0 and β1, because P
and u are' not independent. The demand for any commodity is a function of its price, but at the
same time the price in the market is influenced by the quantity demanded of that commodity.
Consequently the above single equation cannot be treated as a complete (single-equation) model.
There should be at least one more equation in the model giving the relationship between P and Q,
for example:
P = α0 + α1 Q + α2 W + ν (4.2)
Where, W= index of weather conditions.
Substituting Q in this equation with its equal, we obtain
P = α0 + α1 (β0 + β1 P + β2Po + β3 Y + u) + α2 W + ν (4.3)
Obviously P is dependent on u and hence we have violation of Assumption 6 of the method of
least squares. P is not an exogenous variable in the demand function.

Example 2: Suppose we want to estimate the supply of money. This of course is regulated by
the government in an attempt to avoid inflation. Thus we may say that the main determinant of
the decision of the government about the supply of money is the real level of income. Hence we
may write the supply function of money as
M=b0 +b1 Y+u (4.4)
Where, M = money supply and Y = level of real income.
However, the level of real income is in turn influenced by the supply of money as well as by other
real forces, like the investment decisions of businessmen, the welfare policies of the government,
and so on. Consequently the supply of money cannot be treated as a single-equation model. Y is
not truly exogenous. There is a joint dependence between M and Y and hence we must construct a
model with simultaneous equations, one of which would be
Y = a0 +a1M+a2lt+ ... +ν (4.5)
Substituting M, we obtain
Y = a0 + a1 (b0 + b1 Y + u) + a2lt + ------ + ν (3.6)
Obviously Y = f(u) and hence in the function of the supply of money the explanatory
variable Y is not independent of the random variable u:
As stated earlier, the bias arising from the application of classical least squares to an equation
belonging to a system of simultaneous relations is called simultaneous equations bias. It originates
from the violation of Assumption 6 of OLS, that is it arises from the dependence of the
explanatory variables and u, [Cov(UX) ≠ 0) . This creates several problems. Firstly, there arises
the problem of identification of the parameters of individual relationships. Secondly, there arise
problems of estimation. The application of OLS yields biased arid inconsistent estimates. One
should therefore choose other estimation methods stated under introductory part above.
150
 Econometrics II Reading Material for Distance Learners

4.3. Definitions of Some Concepts

Definition 1: Endogenous and exogenous variables
In simultaneous equation models variables are classified as endogenous and exogenous. The
traditional definition of these terms is that endogenous variables are variables that are determined by
the economic model (within the system) and exogenous variables are those determined from outside.
However, from Econometrics view point in the case of Simultaneous Equation Models (SEMs),
endogenous variable is a variable that is correlated with error term (i.e., the covariance between
explanatory variable and error term is different from zero, Cov(XU) ≠0) while Exogenous variable
is a variable that is independent from or uncorrelated with the error term (Cov(XU)=0). .
Exogenous variables are also called predetermined.
Consider the following simple Keynesian macroeconomic model of income determination:
Yt  C t  I t (4.7)

C t   0   1   1Yt  U t (4.8)

with Yt, being the national income, It investments, Ct, the consumption expenditure and Ut a
stochastic term. Equation (4.7) is an identity and an equilibrium condition. Hence this model is
formulated under the condition of being in equilibrium, and the equation show how national income
is related to consumption and investment in equilibrium. The second, equation given by (4.8), is a
behavioral equation since it defines the behavior of the consumption expenditure in this economy.
Equations with stochastic error terms are to be considered behavioral. Since Yt and Ct are left hand
variables in this system of equations, their values are determined by the model. We therefore say
that Yt and Ct are endogenous variables. We have an additional variable included in the model
which is the investment. Since it is a right hand variable in the consumption function we say that it
is an exogenous variable, which is to say that the value of investments are determined outside the
model, it is pre determined. Since investment is determined outside the model it is also uncorrelated
with the stochastic term Ut.
Predetermined groups can be divided into two categories which are considered in general as
exogenous variables. These are: current and lagged exogenous and lagged endogenous. For
instance; X t and X t 1 depict the current and lagged exogenous variables and Yt 1 depicts lagged

endogenous variable. This is on the assumption that X’s symbolize the exogenous variables and Y’s
symbolize the endogenous variables. Thus, X t , X t 1 and Yt 1 are regarded as predetermined

(exogenous) variables.
Since the exogenous variables are predetermined (determined outside the model), they are supposed
to be independent of the error terms/stochastic term, Ut in the model with the (uncorrelated).
.
Consider the demand and supply functions.
Q d   0   1 P   2Y  U 1 (4.9)

Q s   0  1P   2 R  U 2 (4.10)

Where, Q=quantity, Y=income, P=price, R=Rainfalls, U 1 &U 2 are error terms and we call them
unobserved shifters of demand and supply curves, respectively.

151
Econometrics II Reading Material for Distance Learners

Here P and Q are endogenous variables and Y and R are exogenous variables and are called
observed shifters of demand and supply curves, respectively.
Under the assumption that markets clear, what we actually observe is equilibrium values of price and
quantity demanded or supplied. The two equations are linked in econometric analysis only because
observed price and quantity in the “i” units are simultaneously determined by the interaction of
supply and demand by the condition: Qd=Qs.
Thus, the two equations constitute the simultaneous equation models (SEMs).

Definition 2: Structural models

A structural model describes the complete structure of the relationships among the economic
variables. Structural equations of the model may be expressed in terms of endogenous variables,
exogenous variables and disturbances (random variables).
The model is a structural equation in that it is derivable from economic theory and has a causal
interpretation. The parameters of structural model express the direct effect of each explanatory
variable on the dependent variable. Variables not appearing in any function explicitly may have an
indirect effect and is taken into account by the simultaneous solution of the system.
For instance, in the simple Keynesian model of (4.7 and 4.8) a change in consumption affects the
investment indirectly and is not considered in the consumption function. The effect of consumption
on investment cannot be measured directly by any structural parameter, but is measured indirectly
by considering the system as a whole.
As an illustration for structural models, we will use the following simple model for a closed
economy.
Ct = ao + al Yt + Ul (4.11)
It = bo + b1 Yt + b2 Yt-1 + U2 (4.12)
Yt = Ct + It + Gt (4.13)

The first equation is a consumption function, the second is an investment function, and the third
is a definitional equation. The system is complete in that it contains three equations in three
endogenous variables, Ct, It, Yt. The model contains two predetermined variables; government
expenditure, G, and lagged income, [Link] above equations are called structural equations.
The structural parameters (α’s and β’s) are, in general, propensities, elasticities or other parameters
of economic theory. A structural parameter expresses the direct effect of each explanatory variable
on the dependent variable. Indirect effects can be computed only by the solution 'of the structural
system, but not by the individual structural parameters.
For example a change in consumption will affect investment indirectly, through the increase that
the consumption, C, will produce on income, Y, which is a determinant of investment. The effect
of C on I cannot be measured directly by any of the structural parameters, but it will be taken into
account by the simultaneous solution of the system.
Traditionally the structural parameters are represented by β's when they refer to endogenous
variables and by γ's when they are attached to a pre-determined variable. Similarly endogenous
variables are denoted by y's while exogenous variables are represented by x's.

152
Econometrics II Reading Material for Distance Learners

Using the conventional notation (and ignoring the constant intercepts) the structural system above
becomes:
yl = + (4.14)
y2 = + + (4.15)
y3 = y I + y 2 + x 2 (4.16)

Where y1 = C y2 = I y3 = Y
x1 = Yt-1 x2 = G

Transferring all the observable variables to the left-hand side we may obtain the complete table of
structural parameters as follows:
+0 − +0 +0 =
0 + − − +0 =
− − + +0 − =0
Note: We introduced zeros for missed variables from the order of system of linear equations
Similarly transferring equations 3.11-3.13, we will have:
Ct +0It -al Yt +0Yt-1+0Gt = Ul
0Ct + It - b1Yt - b2Yt-1+0Gt = U2
-Ct - It + Yt + 0Yt-1 - Gt = 0
Table of structural coefficients
1 0 -a1 0 0
0 1 -b1 -b2 0
-1 -1 1 0 -1

Table of structural coefficients in standard notation

1 0 − 0 0
0 1 − − 0
-1 -1 1 0 -1

Values of the structural parameters may be obtained by using sample observations on the
variables of the model and applying an appropriate econometric method

Activity 4.1
Consider a simple Keynesian model with no government
Ct = α + βYt + ut, t = 1, 2, . . . ,T
Yt = Ct + It
Where Ct denotes consumption, Yt denotes disposable income, and It denotes autonomous
investment. α and β are constant parameters such that >0 & 0<<1.
a. Comment on system of equations
b. Identify the endogenous and exogenous variables
153
Econometrics II Reading Material for Distance Learners

Definition 3: Reduced form of the model:

The reduced form of a structural model is the model in which the endogenous variables are
expressed as a function of the predetermined variables and the error term only.
The reduced form equations can be obtained in two ways:
1. To express the endogenous variables directly as a function of the predetermined variables as:

Yi = + + ⋯.+ + ( = 1, 2, …. , ) (4.17)

and proceed with the estimation of the ′ by applying some appropriate technique to this

expression.
2. To solve the structural system of endogenous variables in terms of the predetermined
variables, the structural parameters, and the disturbance terms.
Let us see one by one:
The first is to express the endogenous variables directly as functions of the predetermined variables.
This is called Direct Method.
Consider in our example of the simple three- equation closed economy model of equations (4.11-
4.13). As we have said this model has three equations in three endogenous variables (Ct, It ,and Yt )
and two predetermined variables (Gt, andYt-1). Thus, express the endogenous variables (Ct, It, and Yt)
directly as functions of the predetermined variables (Gt, andYt-1), ignoring the constant terms of each
equations) and use ’s as the parameters of the reduced form model.
Thus, we can re-write the model as:
= + + (4.18)
= + + (4.19)
= + + (4.20)
Note: 11 , 12 , 21 , 22 , 31 , and 32 are reduced form parameters.
 The second method for obtaining the reduced form of a model is to solve the structural
system of endogenous variables in terms of the predetermined variables, the structural
parameters and the disturbances.
To derive of the reduced form model for each equation, we can follow the following procedures:
(a) Substitute Ct and It in the third structural equation
Yt = (a1Yt+ ul)+ (b1Yt + b2Yt-1 + u2) + Gt
By rearranging we obtain

= + + . (4.21)

This is the reduced form of the third structural equation.

(b) Substitute Yt into the consumption function

1 +
= + + +
1− − 1− − 1− −

(4.22)
This is the reduced form of the consumption function.

154
Econometrics II Reading Material for Distance Learners

Or
( )
= + + . (4.23)

This is the reduced form of the investment function.

Thus, the structural system of our example gives the following reduced form model:

= + +
( )
= + +

= + +

Clearly for the two reduced forms to be consistent- the following relationships between the ∏'S and
the structural parameters must hold

= =

( )
= =

= =

It should be clear that there is a definite relationship between the reduced- form parameters and the
structural parameters: the ∏’S are functions of the structural parameters.
The constants and coefficients of exogenous variables are called reduced form equation parameters
and are non-linear functions of the structural parameters if the constant term included in the model.
 The reduced-form parameters measure the total effect, direct and indirect, of a change in the
predetermined variable on the endogenous variables, after taking account of the
interdependences among the jointly dependent endogenous variables, while a structural
parameter indicates only the direct effect within a single sector of the economy.
For example measures the effect of a unit increase in Yt-1on the value of investment.
This effect consists of two parts: firstly, there is the direct effect on I through the coefficient
b2 as set out in the structural equation of investment; secondly there is the additional effect
due to the fact that an increase in Yt-1 affects It → and It influences Yt → which in turn affects
It; finally Yt affects Ct → which in turn affects Yt and hence It.
Thus the total effect (measured by ) of Yt-1on It may be split into the following
components
( )
= = 1+

Or

= + (4.24)

= +

155
Econometrics II Reading Material for Distance Learners

 The reduced form coefficients are used for forecasting and policy analysis, since it is the
total effect of a change in the exogenous variables on the dependent variable(s) that is of
interest to the policy maker.
In summary, the above two ways of defining the reduced form model suggests that estimates of
the reduced form coefficients may be obtained in two ways.
 Firstly. Direct estimation of the reduced-form coefficients.
We express all the endogenous variables as functions of all the pre- determined variables of
the system and we apply ordinary least squares to these reduced-form functions. This method
of obtaining the ∏’s is called least-squares- no-restrictions (LSNR), because it does not take
into account any information on the structural parameters, that is, it does not use any
restrictions imposed by the form of the structural system.
For example the structural equations define that some coefficients are zero if the respective
variables are not included in a function; this information is not taken into account by the
method of LSNR. This method does not require complete knowledge of the structural
system. What is required is knowledge of the predetermined variables appearing in the whole
system.
 Secondly. Indirect estimation of the reduced-form coefficients.
We saw that there is a definite relationship between the reduced-form coefficients and the
structural parameters. It is thus possible first to obtain estimates of the structural parameters by
any appropriate econometric technique and then substitute these estimates into the system of
parameters’ relationships to obtain (indirectly) values for the ∏’s. This indirect method involves
three steps:
(l) Solve the system of endogenous variables so that each equation contains only
predetermined explanatory variables. This, as we saw, may be done by continuous
substitution of variables, until we arrive at the reduced-form of all the equations. In this way
we obtain the system of parameters’ relations, that is to say the system which defines the
relations between the 's, the 's and γ’s.
(2) Obtain estimates of the structural parameters by any appropriate econometric method.
(3) Substitute the estimates 's and γ's into the system of parameters' relations to find the
estimates of the reduced-form coefficients.
• This method is more complicated, but it has several advantages over the direct estimation of 's
from LSNR. (a) The derivation of reduced-form 's from the structural 's and γ's is more
efficient because in this way we take into account all the information (that is all the a priori
restrictions imposed by the structure on the parameters) incorporated into the structural model.
(b) Structural changes occur continuously over time. If we know these changes in the 's and γ’s
we may easily re compute the 's. While if the 's are computed with the LSNR method it will
not, in general, be possible to take this information into account, because each is a function of
several structural parameters, and if the exact relationship between 's, 's and γ's has not been
established, we cannot incorporate into the former the changes that may have occurred to the
latter. (c) Extraneous information on some structural parameters may become available from
other studies; such information again will be useless if we have not estimated the 's from
previous estimates of ’s and γ's.
156
Econometrics II Reading Material for Distance Learners

Activity 4.2
1. Refer to activity 3.1, solve for Yt and Ct in terms of the constant and It. What are the resulting
two equations?
2. Find the reduced form of the following structural model:
C    Y  U

Y CZ, for  >0 and 0<<1

Where: C=consumption expenditure
Z=non-consumption expenditure
Y=national income
C and Y are endogenous variables while Z is exogenous variable.

Definition 4: 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. For example:
y1 = f (x1, x2- ... xk ; u1)
y2 = f (xl, x2 .... xk ;yl ;u2)
y3 = f (xl , x2 …..xk ;yl, y2;u3) and soon.
The special feature of recursive model is that its equations may be estimated, one at a time, by OLS
without simultaneous equations bias.
To understand this, let us write the above recursive model in its complete form. Assume that there
are G endogenous variables k exogenous variables in the model
y1 =γ11x1 + γ12x2 + ……+ γlk xk + u1
y2 = γ21x1 + γ22x2 + …….. + γ2k xk + 21y1 + u2
y3 = γ31 x1 + γ32x2 + …….. + γ3k xk + 31y1 + 32y2 + u3
. . . . . .
. . . . . .
. . . . . .
yG = γG1x1 + γG2x2 + ...... + γGkxk + G1y1 + G2y2 + ... + uG

Given values of the exogenous variables (Xi) we may apply OLS to each equation individually,
because by assumption the distribution variables ui and uj are independent, and hence the y's
appearing in the right-hand side of each equation are independent of this equation's error term.

For example, in the second equation y1 is independent of u2, given u1 and u2 are ex-hypothesi
independent (uncorrelated).
Recursive systems are also called triangular systems because the coefficients of the endogenous
variables (the 's) form a triangular array: the main diagonal of the array of β’s contains units, and
no coefficients appear above the main diagonal.

157
Econometrics II Reading Material for Distance Learners

For example assume that we have a model with four endogenous and five predetermined variables
yl = γ11x1 + γ12x2 + ul
y2 = 21y1 + γ21xl + γ22x2 + γ23x3 + u2
y3 = 31yl + 32y2 + γ31xl + γ34x4 + u3
y4 = 41yl + 42y2 + 43y3 + γ44x4 + γ45x 5 + u4
In the above illustration, as usual, the x’s and y’s are exogenous and endogenous variables
respectively. The disturbance terms follow the following assumptions:
(u1u 2 )  (u1u 3 )  (u1u 4 )  E (u 2 u 3 )  E (u 2 u 4 )  E (u 3u 4 )  0 (4.25)

The above assumption is the most crucial assumption that defines the recursive model. 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.
Consider the second equation. It contains the endogenous variable y1 as one of the explanatory
variables along with non-stochastic x’s. OLS can be applied to this equation only if it can be shown
that y1 and u 2 are independent of each other. This is true because u1, which affects y1 is by

assumption uncorrelated with u 2 , i.e. (u1u 2 )  0 . y1 acts as a predetermined variable in so far as

y 2 is concerned. Hence OLS can be applied to this equation. Similar argument can be extended to
the 3rd and 4th equations because right hand side y' s are independent of respective equations u' s .
In this way, in the recursive system OLS can be applied to each equation separately.
To see whether this model is recursive it suffices to examine the form of the array of the 's. If it is
triangular the system is recursive. The system may be rewritten in the form where we formulated
earlier as:

yl -γ11x1 - γ12x2 =ul

y2 - 21yl -γ21xl -γ22x2 -γ23x3 =u2
y3 - 31yl - 32y2 -γ31xl -γ34x4 =u3
y4 - 41yl - 42y2 - 43y 3 -γ44x4 -γ45x5 =u4

The array of structural parameters is shown below.

′ of endogenous variables γ’s of exogenous variables

y1 y2 y3 y4 x1 x2 x3 x4 x5

1 0 0 0 − − 0 0 0

− 1 0 0 − − − 0 0

− − 1 0 − 0 0 − 0

− − − 1 0 0 0 − −
158
Econometrics II Reading Material for Distance Learners

Obviously the array of 's is triangular: the main diagonal contains units and all the coefficients
above it are zero (the lower triangular matrix). Hence the system is recursive, and its equations
can be estimated by OLS, which will yield unbiased and consistent estimates.
Thus, in a very special type of simultaneous equations model called Recursive, Triangular or Causal
model, the use of OLS procedure of estimation is appropriate.

Activity 4.3
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: X 2 = climatic

conditions; and X 3 =last season’s price. The retail rice = Y2 may be assumed to be the function of

production level = Y1 and exogenous factor X 4 = disposable income. Finally the price obtained by

the producer = Y3 can be expressed in terms of the retail price Y2 and exogenous factor X j = the cost

of marketing the producer.

a. Describe the relevant equations of the model
b. Re-write the stated equation in a matrix form

4.4. Consequences of Simultaneity bias

Most of the time in the simultaneous equation models, the endogenous variables may depend on the
error terms of the model (if there is interdependence between the explanatory variables and the error
term); hence the OLS technique is not appropriate for estimation of an equation in a simulations
equations model. In such a case OLS suffers from simultaneity bias: results in biased and
inconsistent estimators of α’s and β’s.
To show simultaneity bias problem and its consequences, let’s consider the following simple
simultaneous equation model:
Y   0  1 X  U 
 (4.26)
X   0   1Y   2 Z  V 

Suppose that the following assumptions hold.

(U )  0 , (V )  0
2 2
(U )   , u (V 2 )   u2
(U iU j )  0 ,  (ViV j )  0, also  (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   0   1 ( 0  1 X  U )   2 Z  V

 0   0 1   2    U V 
X     Z   1  (4.27)
1   1 1  1   1 1   1   1 1 
Applying OLS to the first equation of the above structural model will result in biased estimator
because cov( X iU i )  ( X iU j )  0 .

159
Econometrics II Reading Material for Distance Learners

Now, let’s proof whether this expression holds or not.

cov( XU )  X  ( X )U  (U )
  X   ( X )U  , since E(U)=0 (4.28)

    0 1   2    U V   0   0  1   2  
   0    Z   1      Z U
  1  1  1  1   1 1   1  1 1  1   
1 1  1   
1 1   
Substituting the value of X in equation (3.27) into equation (3.28)
 U 
   (  0   0  1   2 Z   1U  V   0   1 0   2 Z )
1   1  1 

 U 
   ( 1U  V )
1   1  1 

 1 
  ( 1U 2  UV )
 1   
1 1 

 1  2
  (U 2 )  1 u  0 , since E(UV) = 0
 1   1 1  1   1 1

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 parameters.
If we apply OLS to the first equation of the above structural equation (3.26) Y   0   1 X  U , we
obtain:
xy x(Y  Y ) xY Y x
ˆ1    ; (since is zero)
x 2 x 2 x 2
x 2
x( 0   1 X  U )  0 x xU xU
   1 
x 2 x 2
x 2 x 2
xX
But, we know that x  0 and  1 , hence
x 2
xU
ˆ   1 
x 2 (4.29)
Taking the expected values on both sides;
 xU 
(ˆ )   1   2 
 x 
Since, we have already proved that (XU )  0 ; which is the same as ( XU )  0 . Consequently,
xu
when ( XU )  0 ; (ˆ )   , that is ̂ 1 will be biased by the amount equivalent to .
x 2
NB: With correlation between the error term and the explanatory variables in a single
equation model (endogenity problem due to simultaneity bias), using OLS would lead to:
Biased and inconsistent parameter estimates
Invalid tests of hypothesis
Biased and inconsistent forecasts

160
Econometrics II Reading Material for Distance Learners

4.5. Problems of simultaneous equation models

Simultaneous equation models create three distinct problems. These are:
1. 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.
2. Identification of each equation of the model: Many times it so happens that a given set of
values of disturbance terms and exogenous variables yield the same values of different
endogenous variables included in the model. It is because the equations are observationally
indistinguishable, what is needed is that the parameters of each equation in the system
should be uniquely determined. Hence, certain tests are required to examine the
identification of each equation before its estimation.
3. Statistical estimation of each equation of the model: Since application of OLS yield
biased and inconsistent estimates, different statistical techniques are to be developed to
estimate the structural parameters. Some of the most common simultaneous methods of
estimation/estimation methods to avoid simultaneity bias are:
i) The indirect least square method(ILS)
ii) The two-stage least square method(2SLS)
iii) The three-stage least square method(3SLS)
iv) Limited information maximum likelihood method (LIML)
v) The instrumental variable method (IV)
vi) The mixed estimation method; and
vii) The full information maximum likelihood method (FIML)

Of these three problems, we are going to discuss the second problem (the identification problem) in
the following section.
However, these estimation methods are beyond the scope of this module due to the fact that this
chapter will only scratch the surface of the issues involved in estimating simultaneously equations
and should therefore be seen as an introduction to the subject.

161
Econometrics II Reading Material for Distance Learners

4.6. The Identification Problem

In general, we can think of any structural equation, as having one left hand side endogenous variable
y1, g1 right hand side endogenous variables, and k1 right hand side exogenous variables. The right
hand side endogenous variables are correlated with the error term rendering OLS on this equation
biased and inconsistent. In order to be able to estimate the structural equation coefficients they need
to be identified. So, what do we mean by that?
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.
Normally, for each endogenous variable, there exists a corresponding structural equation explaining
its behavior in the model. We say that a system of simultaneous equations is complete if there are as
many endogenous variables as there are equations.

To give an intuitive feeling for meaning of identification, we will give an example before going into
any formal and mechanical tests.
Consider the following two equation system:
Q  A0  A1 P  A2 X 1  U1 (Supply) (4.30)

Q  B0  B1 P  U 2 (Demand) (4.31)
This system contains two endogenous variables (P and Q) and one exogenous (X1) variable. These
two equations represent a supply and demand system for a given market, respectively. The question
is if any of these two equations are identified? That is to ask if the parameters of the two equations
can be estimated consistently. It turns out that the demand function is identified while the supply
function is not. The supply function contains an exogenous variable X1, and the supply function takes
a new position for each value of Xl. In that process we identify the demand function. But in the
demand function we have nothing unique that does not appear in the supply function so it is
impossible to move the demand function while holding the supply function fixed. Hence it is the
presence of an exogenous variable in one equation that allows us to estimate the parameters of the
other equation. If Xl had been included in both equations there would have been no unique variation
in any of the equations and hence no equation had been identified. However, if another exogenous
variable, X2, had been introduced and placed in the demand function, we would receive some
exogenous variation that could help us to identify the supply function. In that case both equations
would have been identified.
In econometric theory two possible situations of identifiability are traditionally distinguished:
1. Equation under identified.
2. Equation identified
a. Exactly identified.
c. Over identified.

162
Econometrics II Reading Material for Distance Learners

An equation is under identified if its statistical form is not unique. A system is under identified
when one or more of its equations are under identified.
If an equation has a unique statistical form we say that it is identified. It may be exactly
identified or over identified. But in both cases it is identified. A system is identified if all its
equations are identified.
It should be noted that identification problems arise only for those equations which contain
coefficients which must be estimated statistically (from sample data). Identification difficulties do
not arise for definitional equations, identities, or statements of equilibrium conditions, because
such relationships do not require measurement.
Though identification doesn’t concern with estimation of the model, it is closely related to it.
(1) If an equation (or a model) is under identified it is impossible to estimate all its parameters
with any econometric technique.
(2) If an equation is identified, its coefficient can, in general, be statistically estimated.
In particular: (a) If the equation is exactly identified, the appropriate method to be used for its
estimation is the method of indirect least squares. (b) If the equation is over identified, indirect least
squares cannot be applied, because it will not yield unique estimates of the structural parameters.
There are various other methods which can be used in this case, for example two-stage least squares
(2SLS), or maximum likelihood methods.

4.6.1. Formal Rules (Conditions) for identification

Identification may be established either by the examination of the specification of the structural
model, or by the examination of the reduced form of the model.
Traditionally identification has been approached via the reduced form. 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.
In this section we will examine both approaches. 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 form first and then examination of the values of the
determinant formed from some of the reduced form coefficients. The structural form approach is
simpler and more useful.
In applying the identification rules we should either ignore the constant term, or, if we want to
retain it, we must include in the set of variables a dummy variable (say Xo) which would always
take on the value 1. Either convention leads to the same results as far as identification is
concerned.
The process of identifying equations can be formalized in a decision rule that specify the conditions
that have to be fulfilled in order to identify one or several equations in a system. In the literature,
two rules are described and one is slightly easier to use than the other.

163
Econometrics II Reading Material for Distance Learners

[Link]. The Order Condition of Identification

The first decision rule for identification is the so called order condition. 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, which means that it is possible that
the equation is undefined even though the order condition says it is identified. However, in a system
with only two equations, the order condition will work well and can be trusted.
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.
Given that in a complete model the number of endogenous variables is equal to the number of
equations of the model, the order condition for identification is sometimes stated in the following
equivalent form. .
An equation in any SEM satisfies the order condition for identification if the number of excluded
exogenous variables from the equation is at least as large as the number of right-hand side
endogenous variables. or
For an equation to be identified the total number of variables excluded from it but included in other
equations must be at least as great as the number of equations of the system 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.
Then the order condition for identification may be symbolically expressed as
(K-M) ≥ (G- 1)

≥ [ − 1]

K-M = the number of variables (endogenous and exogenous) in the model excluded from the
equation under consideration.
The order condition states that:
1) If K-M = G – 1 => the equation is exactly identified
2) If K-M > G-1 => the equation is over identified
3) If K-M < G – 1 => the equation is under identified
NB: When checking the order condition you have to do it for each equation in the system.

Example 1
Consider the following system:
Y1  A0  A1Y2  A2 X 1  U 1

Y2  B0  B1Y1  B2 X 2  U 2

Use the order condition to check if the equations are identified.

In order to do that, we need to determine the value of K, M and G. This system contains two
endogenous variables and the total number of variables, endogenous as well as exogenous, is 4.
The number of endogenous and exogenous variables in each equation is 3.
For the first equation we have G-1 = 2-1=1 and K-M =4-3=1 since X2 is excluded from the first
equation. Since G-1=K-M, we have that the first equation is exactly identified.

164
Econometrics II Reading Material for Distance Learners

For the second equation we have G-1=1 and K-M=1 since X1, is excluded from the second equation.
Since G-1=1 and K-M=1 we have that also the second equation is exactly identified. When all the
equations of' the model are identified we say that the model is identified; thus, we are able to
estimate all the structural parameters.

Example 2
Consider the following system:
Y1  A0  A1Y2  A2 X 1  A3 X 3  U 1

Y2  B0  B1Y1  B2 X 2  U 2

In this example we have two endogenous variables and three exogenous variables with a total of five
variables. G-1 will in this example equal 1 as before since we still have only two endogenous
variables. Will the equations be identified in this case?
The first equation contains four variables which means that one variable has been excluded from the
equation, that is, X2 does not appear in equation 1 and K-M=1. Since G-1 =K-M the equation is
exactly identified.

The second equation includes three variables which mean that two variables have been excluded.
That is, Xl and X3 are not included in equation 2. That is K-M = 2, which means that G-l < K-M which
leads to the conclusion that equation 2 is over identified.

The order condition for identification is necessary for a relation to be identified, but it is not
sufficient, that is, it may be fulfilled in any particular equation and yet the relation may not be
identified.

Activity 4.4
Given a system contains 10 equations with 15 variables, ten endogenous and five exogenous, an
equation containing 11 variables is not identified, while another equation containing 5 variables is
identified. Use the order condition to check whether the equations are identified or not.

[Link]. The Rank Condition of Identification

The rank condition is slightly more complicated when dealing with larger systems of equations, but
when using only two equations it is as easy as the order condition. The rank condition is a necessary
and sufficient condition, which means that if we can identify the equations using the rank condition
we can be sure that the equation really is identified.

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 non-zero 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. This condition is called rank condition because it refers to the rank of the
matrix of parameters of excluded variables. The rank of a matrix is the order of the largest non-zero
determinant which can be formed from the matrix. In our case the relevant matrix is the sub matrix
of coefficients of the absent variables.
165
Econometrics II Reading Material for Distance Learners

Hence the rank condition may be also stated as follows: a sufficient condition for identification of a
relationship is that the rank of the matrix of parameters of all the excluded variables (endogenous
and predetermined) from that equation be equal to (G - 1).
The rank condition investigates whether two or more equations are linearly dependent on each other,
which would be the case if the sum of two equations would equal a third equation in the model. If
that is the case it is impossible to identify all structural parameters.
The basic steps in this decision rule can be best described using illustrative examples. The practical
steps for tracing the identifiability 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 is equal to zero.

For example let a structural model be

y1 = 3y2 - 2x1 + x2 + u1

y2 = y3 + x3 + u2
y3 = y1 - y2 – 2x3 + u3

where the y's are the endogenous variables and the x's are the predetermined variables.

This model may be rewritten in the form

-y11 +3y2 +0y3 -2x1 +x2 +0x3 +u1 =0
0y1 - y2 + y3 +0x1 +0x2 + x3 +u2 =0
y1 - y2 - y3 +0x1 +0x2 – 2x3 +u3 =0

Ignoring the random disturbances the table of the parameters of the model is as follows.
Equations Variables
Y1 Y2 Y3 X1 X2 X3
st
1 equation -1 3 0 -2 1 0
2nd equation 0 -1 1 0 0 1
3rd equation 1 -1 -1 0 0 -2

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.
Thirdly, Strike out the columns in which a non-zero coefficient of the equation being examined
appears. 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.
Table of structural parameters Table of parameters of excluded variables
Y1 Y2 Y3 X1 X2 X3 Y1 X1 X2
  
st
1 -1 3 0 -2 1 0 -1 -2 1
nd
2 0 -1 1 0 0 1
3rd 1 -1 -1 0 0 -2 1 0 0

166
Econometrics II Reading Material for Distance Learners

Fourthly, Form the determinant(s) of order (G - 1) and examine their value. 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:
−1 − 2 −2 1 −1 1
∆ = ≠0 ∆ = =0 ∆ = ≠ 0
1 0 0 0 1 0
(the symbol "∆" stands for 'determinant').
We see that we can form two non-zero determinants of order G - 1 = 3 - 1 = 2; hence the second
equation of our system is identified.
Fifthly, to see whether the equation is exactly identified or over identified we use the order
condition (K - M) ≥ (G - 1). With this criterion, if the equality sign is satisfied, that is if (K - M) =
(G - 1), the equation is exactly identified. If the inequality sign holds, that is, if (K - M) > (G - 1), the
equation is over identified and if (K - M) < (G - 1), the equation is underidentified.
In the case of the second equation we have: G=3 K=6 M=3 and the counting rule
(K -M) ≥ (G -1) gives (6 – 3) > (3 -1)
Therefore the second equation of the model is over identified.
The identification of a function is achieved by assuming that some variables of the model have a
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.
NB: In many applications, an equation that satisfies order condition is identified.

Dear Learners, the above basic steps are relatively time consuming and a bit tedious; hence, we
can have another method that we can follow. Let us see it in the following section as usual by the
help of illustrations:

Example 2
Consider the following system of equations:
Y1  A0  A1Y3  A2 X 1  A3 X 3  U 1

Y2  B0  B1Y1  B2 X 2  B3 X 3  U 2

Y3  C0  C1Y2  A2 X 1  A3 X 2  U 3

This system contains three endogenous variables (Y1, Y2, Y3) and three exogenous variables (X1, X2,
X3), which means that we in total has six variables.
The first step in checking the rank condition is to put up a matrix that for each equation mark
which of the six variables that are included (marked with 1) and which that are excluded (marked
with 0) from the equation as usual. For our system we receive the following matrix:
Matrix for the rank condition
Y1 Y2 Y3 X1 X2 X3
Equation1 1 0 1 1 0 1
Equation 2 1 1 0 0 1 1
Equation 3 0 1 1 1 1 0

167
Econometrics II Reading Material for Distance Learners

In order to check the rank condition for the first equation we have to proceed as follows:
Delete the first row and collect the columns for those variables of the first equation that were
marked with zero.
For equation 1: Y2 and X2 was marked with zero, and if we collect those two columns we have:
1 1
1 1

If this matrix contains less than G-1 rows or columns where all elements are zero, equation 1 will
not be identified. G refers to the number of equation just as in the order condition, which means
that G-1 =2.

Since we have two rows and two columns and none of them contains only zeros we conclude that
equation 1 is identified.

For equation 2 we proceed in the same way. We delete the second row and collect those columns
where the elements of the second row were marked with a zero. For equation 2 that was the case for
Y3 and X1, which is to say that these two variables was not included in equation 2. The resulting
matrix for this case then becomes:

1 1

1 1

It looks in the same way as for equation 1, which means that we have two rows and two columns
that is not only zeros. The same procedure should be done for the third equation and if you do that
you will see that it is identified as well.
When using larger systems it is quite possible that the order condition says that a particular
equation is identified even though the rank condition says it is not. When that happens it might still
be possible to generate estimates, but those estimates will not have any economic meaning since
they will represent averages of those equations that are linear combinations of each other. Hence,
you should not be content that you have received identified results just because the order condition
says so and the econometric software generates results for you. When using systems of more than
two equations you should also confirm the identification using the rank condition.

Activity 4.5
Assume that we have a model describing the market of an agricultural product. From the theory of
partial equilibrium we know that the price in a market is determined by the forces of demand and
supply. The main determinants of the demand are the price of the commodity, the prices of other
commodities, incomes and tastes of consumers. Similarly, the most important determinants of the
supply are the price of the commodity, other prices, technology, the prices of factors of production,
and weather conditions. The equilibrium condition is that demand be equal to supply. The above
theoretical information may be expressed in the form of the following mathematical model.
Qt  a 0  a1 P1  a 2 P2  a3Y  a 4 t  u

Qt  b0  b1 P1  b2 P2  b3C  b4 t  w
168
Econometrics II Reading Material for Distance Learners

DS
Where: D= quantity demanded
S= quantity supplied
P1  Price of the given commodity
P2  Price of other commodities
Y= income
C= costs (index of prices of factors of production)
t= time trend. In the demand function it stands for ‘tastes’; in the supply function it stands
for ‘technology’.
The above model is mathematically complete in the sense that it contains three equations in three
endogenous variables, D, S and P1. The remaining variables, Y, P2, C, t are exogenous.
Identify the supply function using:
a. Order condition
b. Rank criteria for identification

[Link]. Establishing Identification from the reduced form

Like that of the identification conditions from structural equations, there are two conditions for
identification based on the reduced form of the model, an order condition and a rank condition. The
order condition is the same as in the structural model. The rank condition here refers to the value of
the determinant formed from some of the reduced form parameters, π‘s.

1. Order condition (necessary condition), as applied to the reduced form

An equation belonging to a system of simultaneous equations is identified if
(K  M )  (G  1)
Total number of   number of 
excluded var iables   equations less one 
   
where , K, M and G have the same meaning as before:
K= total number of variables, endogenous and exogenous, in the entire model
M= number of variables, endogenous and exogenous, in any particular equation
G= number of structural equations=number of all endogenous variables in the model
If (K-M) = (G-1), the equation is exactly identified, provided that the rank condition set out below is
also satisfied. If (K-M) > (G-1), the equation is over identified, while if (K-M)<(G-1), the equation
is under identified, under the same condition.

2. Rank condition as applied to the reduced form

Let G* stand for the number of endogenous variables contained in a particular equation. The rank
condition as applied to the reduced form may be stated as follows:
An equation containing G* endogenous variables is identified if and only if it is possible to construct
at least one non-zero determinant of order G*-1 from the reduced form coefficients of the exogenous
(predetermined) variables excluded from that particular equation.
The practical steps involved in this method of identification may be outlined as:

169
Econometrics II Reading Material for Distance Learners

Firstly, obtain the reduced form of structural model.

For example assume that the original model is:
y1  b12 y 2  y11 x1  y12 x2  u1

y 2  b23 y 3  y 23 x3  u 2

y 3  b31 y1  b32 y 2  y 33 x3  u 3

This model is complete in the sense that it contains three equations in three endogenous variables.
The model contains altogether six variables, three endogenous ( y1 , y 2 , y 3 ) and three exogenous

( x1 , x 2 , x 3 ).

The reduced form of the model is obtained by solving the original equations for the exogenous
variables. The reduced form in the above example is:
y1   11 x1   12 x 2   13 x3  v1

y 2   21 x1   22 x 2   23 x 3  v 2

y 3   31 x1   32 x 2   33 x3  v3

where the π’s are functions of the structural parameters.

Secondly, form the complete table of the reduced form coefficients.
Exogenous Variables
Equations x1 x2 x3
st
1 equation: y1  11  12  13
2nd equation: y 2  21  22  23
3rd equation: y 3  31  32  33

Strike out the rows corresponding to endogenous variables excluded from the particular equation
being examined for identifiability.
Also strike out all the columns referring to exogenous variables included in the structural form of the
particular equation.
After these deletions we are left with the reduced form coefficients of exogenous variables excluded
(absent) from the structural equation.
For example, assume that we are investigating the identification procedure are found by striking out
the first row (since y1 , does not appear in the second equation) and the third column (since x3 , is

included in this equation).

Complete table of reduced Table of reduced form

form coefficients coefficients of excluded
exogenous variables
 11  12  13  21  22

 21  22  23  31
 31  32  33  32

170
Econometrics II Reading Material for Distance Learners

Thirdly, examine the order of the determinants of the π’s of excluded exogenous variables and
evaluate them. If the order of the largest non-zero determinant is G*-1, the equation is identified.
Otherwise the equation is not identified.

Example:
Assume that a structural model is of the form
y1 = 3y2 - 2x1 + x2 + u1
y2=y3 + x3 + u2
y3 = y1 – y2 - 2x3 + u3

We will examine the identification of each structural equation using the reduced-form approach.
The reduced form of the structural model is

= + + =4 −2 +3

= + + =2 − +

= + + =2 −

First equation: y1 = 3y2- 2x1 +x2 +u1

1. Order condition: (K – M ) ≥ (G – 1)
K=6 M=4 G=3
So that (K- M) = (G - 1)
Or (6 - 4) = (3 - 1)
Thus the equation satisfies the necessary condition for identification.
2. Rank condition
Excluded endogenous variables: y3
Included exogenous variables: x1; x2
Table of all reduced form coefficients Table of 's of excluded exogenous variables

4 -2 3 3
2 -1 1 1
2 1 0

From the table of 's of excluded exogenous variables we can form two non-zero determinants
of order 1 x 1.
Now G* = 2, since the first equation contains two endogenous variables. Thus the order of the
determinant of 's relating to excluded X's is one:
(G* - 1) = l
The rank condition is satisfied and the equation is identified.
Furthermore, by virtue of the order condition we see that the first equation is exactly identified since
the equality sign holds for this condition (that is K - M = G - 1).
Second equation: y2 = y3 + x3 + u2
1. Order condition
K=6 M=3 G=3
(K - M) > (G - 1)
(6 - 3) > (3 - 1). Thus the order condition is satisfied.
171
Econometrics II Reading Material for Distance Learners

2. Rank condition
Excluded endogenous variables: y1
Included exogenous variables: x3

Table of all ′ Table of ′ of excluded exogenous variables

4 -2 3 2 -1
2 -1 1 2 -1
2 -1 0

The order of the determinant of ′ of the excluded exogenous variables is 2 x 2; however us

value is zero:
2 −1
∆= =0
2 −1
Hence the highest non-zero determinant which we may form from the ′ is 1x1.
The second equation contains two endogenous variables, that is G* = 2. Therefore the
order of highest non-zero determinant is G* - 1 = 1.
Consequently the second equation is identified. Furthermore, using the order condition
(6 - 3) > (3 - 1) we conclude that the second structural equation is over identified.

Third equation: y3 = y1 – y2 - 2x3 + u3

1. Order condition
K=6 M=4 G=3
so that (K – M) = (G – 1)
or (6 – 4) = (3 – 1)
The order condition is satisfied

2. Rank condition
Excluded endogenous variables: none
Included exogenous variables: x3

Table of all ′ Table of ′ of excluded exogenous variables

4 -2 3 4 -2
2 -1 1 2 -1
2 -1 0 2 -1

The highest non-zero determinant is of order 1 x l, since

4 −2 2 −1
∆= =0 ∆= =0
2 −1 2 −1

The equation contains three endogenous variables, that is G*=3. Hence

order of largest non-zero determinant) < (G* -1.
The rank condition is not satisfied and therefore the third equation is not identified,
despite the fulfillment of the first condition for identification.

172
Econometrics II Reading Material for Distance Learners

Chapter Summary:
1. In contrast to single-equation models, in simultaneous-equation models more than one dependent,
or endogenous, variable is involved, necessitating as many equations as the number of endogenous
variables.
2. A unique feature of simultaneous-equation models is that the endogenous variable (i.e.,
regressand) in one equation may appear as an explanatory variable (i.e., regressor) in another
equation of the system. As a consequence, such an endogenous explanatory variable becomes
stochastic and is usually correlated with the disturbance term of the equation in which it appears as
an explanatory variable.
3. In the above situation the classical OLS method may not be applied because the estimators thus
obtained are not consistent, that is, they do not converge to their true population values no matter
how large the sample size.
4. The problem of identification precedes the problem of estimation. The identification problem asks
whether one can obtain unique numerical estimates of the structural coefficients from the estimated
reduced form coefficients. If this can be done, an equation in a system of simultaneous equations is
identified. If this cannot be done, that equation is un- or under identified.
5. An identified equation can be just/exact/fully identified or over identified. In the former case,
unique values of structural coefficients can be obtained; in the latter, there may be more than one
value for one or more structural parameters.
6. The identification problem arises because the same set of data may be compatible with different
sets of structural coefficients, that is, different models. Thus, in the regression of price on quantity
only, it is difficult to tell whether one is estimating the supply function or the demand function,
because price and quantity enter both equations.
7. To assess the identifiability of a structural equation, one may apply the technique of reduced-
form equations, which expresses an endogenous variable solely as a function of predetermined
variables. However, this time-consuming procedure can be avoided by resorting to either the order
condition or the rank condition of identification.
8. Although the order condition is easy to apply, it provides only a necessary condition for
identification. On the other hand, the rank condition is both a necessary and sufficient condition for
identification. If the rank condition is satisfied, the order condition is satisfied, too, although the
converse is not true.
In practice, though, the order condition is generally adequate to ensure identifiability.

173
Econometrics II Reading Material for Distance Learners

Check Your Progress for Activities

Activity 4.1
a. This is a system of two simultaneous equations, also known as structural equations
with the second equation being an identity. The first equation can be estimated by OLS
b. Since It is autonomous, it is an exogenous variable determined outside the system, whereas
Ct and Yt are endogenous variables determined by the system.

Activity 4.2
1. The resulting two equations are known as the reduced form equations
Ct = α/(1 − β)+βIt(1 − β)+ut/(1 − β)
Yt = α/(1 − β)+It/(1 − β)+ut/(1 − β)
2. 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. To do this substitute Y=C+Z into consumption equation

C     (C  Z ) + U
C    C  Z  U
C  C    Z  U
C (1   )     Z  U
    U
C    Z  **
1  1   1 
Substituting again the above equation into income identity, we get;
  1  U
Y    Z 
1   1    1  ***

Equation (**) and (***) are called the reduced form of the structural model of the above. We can
write this more formally as:
Structural form equations Reduced form equations
C    Y  U     U
C    Z 
1  1   1 
Y CZ   1  U
Y    Z 
1   1    1 

Parameters of the reduced form measure the total effect (direct and indirect) of a change in
exogenous variables on the endogenous variable.
  
For instance, in the above reduced form equation (**),   measures the total effect of a unit
1  
change in the non-consumption expenditure on consumption. This total effect is, the direct effect (
 1 
 ) times the indirect effect   .
1  

174
Econometrics II Reading Material for Distance Learners

Activity 4.3
a. The relevant equations of the model may be described as under:

Y1   1   2 X 2   3 X 3  U 1 

Y2   4  1Y1   5 X 4  U 2 
Y3   6   2Y2   7 X 5  U 3 

In the first equation, there are only exogenous variables and are assumed to be independent of U 1 .

In the second equation, the causal relation between Y1 and Y2 is in one direction.

Also Y1 is independent of U 2 and can be treated just like exogenous variable. Similarly since Y2 is
independent of U 3 , OLS can be applied to the third equation.
b. we can rewrite the above equations as follows:

Y1   1   2 X 2   3 X 3  U 1 

 1Y1  Y2   4   5 X 4  U 2 
  2Y2  Y3   6   7 X 5  U 3 

We can again rewrite this in matrix form as follows:

 
 
 
 1 0 0 Y1    1   2   3 0 0   X 1  U 1 
  1 0 Y     0 0 5 0   X 2   U 2 
 1  2  4  
 0   1 Y3     6 0 0 0   7   X 3  U 3 
2   
Coefficient matrix of coefficient matrix of X4
endogenous var iables exogenous var iable  
X 5 
The coefficient matrix of endogenous variables is thus a triangular one; hence recursive models are
also called as triangular models.

Activity 4.4
a. For the first equation we have
G  10 K  15 M  11
Order condition:
( K  M )  (G  1)
; That is, the order condition is not satisfied.
(15  11)  (10  1)
b. For the second equation we have
G  10 K  15 M 5
Order condition:
( K  M )  (G  1)
; That is, the order condition is satisfied.
(15  5)  (10  1)
The order condition for identification is necessary for a relation to be identified, but it is not
sufficient, that is, it may be fulfilled in any particular equation and yet the relation may not be
identified.

175
Econometrics II Reading Material for Distance Learners

Activity 4.5
1. Order condition: ( K  M )  (G  1)
In our example we have: K=7 M=5 G=3
Therefore, (K-M) = (G-1) or (7-5) = (3-1) = 2
Consequently the second equation satisfies the first condition for identification.
2. Rank condition
The table of the coefficients of the structural model is as follows.
Variables
Equations D P1 P2 Y t S C
st
1 equation -1 a1 a2 a3 a4 0 0
2nd equation 0 b1 b2 0 b4 -1 b3
3rd equation 1 0 1 0
0 0 0

Following the procedure explained earlier we strike out the second row an the second, third, fifth,
sixth and seventh columns. Thus we are left with the table of the coefficients of excluded variables:

Complete table of Table of parameters

Structural parameters of variables
excluded from the
second equation
-1 a1 a2 a3 a4 0 0 -1 a3
0 b1 b2 0 b4 1 b3
1 0 1 1 -1 0
0 0 0

From this table we can form only one non-zero determinant of order
(G-1) = (3-1) =2
1 a3
  (0)(1)  (1)(a3 )  a 3
1 0

The value of the determinant is non-zero, provided that a3  0 .

We see that both the order and rank conditions are satisfied. Hence the second equation of the
model is identified. Furthermore, we see that in the order condition the equality holds:
(7-5) = (3-1) = 2. Consequently the second structural equation is exactly identified.

176
Econometrics II Reading Material for Distance Learners

Review Exercises
1. Identify the demand function of activity 4.5 using:
a. Order condition
b. Rank criteria for identification

2. Assume the following simple version of the Keynesian model of income determination.
Consumption function: C t  a 0  a1Yt  a 2Tt  u

Investment function: I t  b0  b1Yt 1  u

Taxation function: Tt  c0  c1Yt  w

Definition: Yt  C t  I t  Gt

This model is mathematically complete in the sense that it contains as many equations as
endogenous variables. There are four endogenous variables, C, I, T, Y, and two predetermined
variables, lagged income (Yt 1 ) and government expenditure (G).

a. Identify consumption function using order and rank conditions

b. Identify investment function using order and rank conditions

3. Detect the identificability of the tax equation of question 2 using both conditions.

4. The celebrated IS, or goods market equilibrium, model of macroeconomics in its non-
stochastic form can be expressed as
Consumption function: Ct = β0 + β1Ydt 0 < β1 < 1 (1)
Tax function: Tt = α0 + α1Yt 0 < α1 < 1 (2)
Investment function: It = γ0 + γ1rt (3)
Definition: Ydt = Yt − Tt (4)
Government expenditure: Gt = ̅ (5)
National income identity: Yt = Ct + It + Gt (6)
where Y = national income
C = consumption spending
I = planned or desired net investment
̅ = given level of government expenditure
T = taxes
Yd = disposable income
r = interest rate
a. Obtain the reduced form of IS equation
b. What would happen if we were to estimate, say, the consumption function (1) in isolation?
Could we obtain unbiased and/or consistent estimates of β0 and β1?

177
Econometrics II Reading Material for Distance Learners

5. The other half of the famous IS-LM paradigm is the LM, or money market equilibrium,
relation, which gives the combinations of the interest rate and level of income such that the
money market is cleared, that is, the demand for money is equal to its supply.
The IS and LM curves show, respectively, that a whole array of interest rates is consistent with
goods market equilibrium and a whole array of interest rates is compatible with equilibrium in the
money market. Of course, only one interest rate and one level of income will be consistent
simultaneously with the two equilibria.
Algebraically, the model, in the non-stochastic form, may be expressed as:
Money demand function: Mtd = a + bYt − crt (1)
Money supply function: Mts = (2)
Equilibrium condition: Mtd = Mts (3)

where Y = income, r = interest rate, and = assumed level of money supply, say, determined
by the Fed.
Derive the reduced form of the system of equations

6. G. Menges developed the following econometric model for the West German economy*:
Yt = β0 + β1Yt−1 + β2 It + u1t
It = β3 + β4Yt + β5Qt + u2t
Ct = β6 + β7Yt + β8Ct−1 + β9 Pt + u3t
Qt = β10 + β11Qt−1 + β12 Rt + u4t
Where, Y = national income
I = net capital formation
C = personal consumption
Q = profits
P = cost of living index
R = industrial productivity
t = time
u = stochastic disturbances
a. Which of the variables would you regard as endogenous and which as exogenous?
b. What is the reason behind including the variable P in the consumption function?
c. Determine whether the structural equations of the model are identified.

178