ECONOMETRICS

Applications with EViews

Abdul Waheed
Econometrics: Applications with EViews by Abdul Waheed 1
Chapter 4: Multicollinearity
Learning Outcomes
1. Understand the meaning of multicollinearity.

2. Differentiate between perfect and imperfect multicollinearity.

3. Know the reasons for the multicollinearity problem in the
model.
4. Understand the consequences of multicollinearity.
5. Detect the multicollinearity in the regression problem model.
6. Learn the remedial measures of the multicollinearity problem.

Econometrics: Applications with EViews by Abdul Waheed 2

Chapter 4: Multicollinearity
Introduction
One of the assumptions of the linear regression model is that
there is no perfect relationship among the explanatory variable
included in the regression model.

If there is a strong relationship among explanatory variables of

the regression model, it means this assumption is violated and
there will be a multicollinearity issue in the regression model.

The multicollinearity creates a problem on the interpretation of

the regression results.

Econometrics: Applications with EViews by Abdul Waheed 3

Chapter 4: Multicollinearity
Because regression coefficients shows the mean effect on
dependent variable for one unit change in the respective
independent variable while all others independent variables are
held constant.

However, in the presence of multicollinearity the later part is

violated. So, this assumption is important in regression analysis
and interpretation of the results.

Econometrics: Applications with EViews by Abdul Waheed 4

Chapter 4: Multicollinearity
Perfect Multicollinearity

If one explanatory variable is some multiple of the other

explanatory variables, the multicollinearity will be perfect that is
r=1.

In the case of perfect multicollinearity, we cannot find estimates

of the regression model.

Perfect multicollinearity is rare especially on economic variables

therefore it is rare to believe a prior that the estimate of the model
cannot be obtained.
Econometrics: Applications with EViews by Abdul Waheed 5
Chapter 4: Multicollinearity
Imperfect Multicollinearity

In this case, there will be a strong correlation among explanatory

variables.

As a numerical example, consider the correlation of wheat supply

(WSUPPLY), support price of wheat (SPWHEAT), fertilizer
consumption (FERCON), water availability (WATAVA), and literacy
rate (LITR) for the period 1982 to 2015 using Time Series Data file.

This correlation matrix is given in Table 4.1. There is a strong positive

relationship among all variables, therefore, there may emerge imperfect
multicollinearity when we run a regression of WSUPPLY on other
explanatory variables.
Econometrics: Applications with EViews by Abdul Waheed 6
Chapter 4: Multicollinearity

Econometrics: Applications with EViews by Abdul Waheed 7

Chapter 4: Multicollinearity
Reasons of Multicollinearity
The multicollinearity may arise for various reasons. Some of the
reasons are discussed below.

1. The economic variables sometimes follow the same pattern

resulting in a strong relationship. Such as during the business
cycle, in a period of recession economic growth slow down, the
income, consumption, saving, invest, and employment all follow
the same trend resulting a strong correlation among these
variables.

Econometrics: Applications with EViews by Abdul Waheed 8

Chapter 4: Multicollinearity
2. The inclusion of lagged variables in the model may result in a
strong correlation in actual and lagged explanatory variables. For
example, in consumption function, sometimes lagged income
variable is included in the model resulting in a strong correlation
between actual and lagged income explanatory variables.

3. The use of time series data often results in multicollinearity

problems compared to cross-sectional data. Therefore,
multicollinearity is more common and more serious in the
regression analysis with time series data.

Econometrics: Applications with EViews by Abdul Waheed 9

Chapter 4: Multicollinearity
Consequences of Multicollinearity
If we have a multicollinearity issue in the regression results, then
we may face the following problems.
1. In the case of perfect multicollinearity (r=1) we cannot find
estimates of the regression model.
2. In the case of imperfect multicollinearity, we can still find
the estimates of the regression model, but
i. The OLS estimators will still be linear, unbiased, and
normally distributed.

Econometrics: Applications with EViews by Abdul Waheed 10

Chapter 4: Multicollinearity
ii. The OLS estimators will not be efficient. In the other
word, they do not have minimum variance relative to
other linear and unbiased estimators. There will be large
variances and standard errors of OLS estimators.

iii. The t statistic will be small, leading to the acceptance of

the null hypothesis that the true population parameters
are zero.

iv. The adjusted R2 will be high but few coefficients of the

regression model will be statistically significant.

Econometrics: Applications with EViews by Abdul Waheed 11

Chapter 4: Multicollinearity
v. There will be unstable OLS estimates and a wrong sign
of a few regression coefficients.

vi. The OLS estimator’s variance and standard errors

become sensitive to small changes in the data.

Econometrics: Applications with EViews by Abdul Waheed 12

Chapter 4: Multicollinearity
Detection of Multicollinearity

There is no unique method of detecting the multicollinearity

problem rather there are some rules of thumb, formal or informal,
that can be used to detect the multicollinearity issue in the model.

If the following happens, it signs the presence of multicollinearity

in the regression model.

Econometrics: Applications with EViews by Abdul Waheed 13

Chapter 4: Multicollinearity
1. High pair-wise correlations among explanatory variables.
If the correlation coefficient between two explanatory variables
is high such as more than +0.8 or -0.8, the multicollinearity
problem can be serious.

However, in the regression model involving more than two

explanatory variables, then simply high correlation will not
provide a guide to the presence of multicollinearity in this case,
we have to look at some other reason for multicollinearity.

Econometrics: Applications with EViews by Abdul Waheed 14

Chapter 4: Multicollinearity
2. A priori knowledge of the relationship among variables.
Sometimes we know a prior that there could be a strong correlation
between two variables.

For example, we expect a strong positive relationship between

income and wealth. If we include both variables in the
consumption model, there is a possibility of a multicollinearity
problem.

Econometrics: Applications with EViews by Abdul Waheed 15

Chapter 4: Multicollinearity
3. High R2 but few coefficients are significant.
If R2 is high such as more than +0.8 or -0.8, then the F test in
most cases will reject the null hypothesis ( i.e. the partial slope
coefficients are simultaneously equal to zero).

However, if the test of statistical significance of individual

coefficients shows that none or some coefficients are statistically
significant, then there can be a problem of multicollinearity in
the model.

Econometrics: Applications with EViews by Abdul Waheed 16

Chapter 4: Multicollinearity

Econometrics: Applications with EViews by Abdul Waheed 17

Chapter 4: Multicollinearity
4. Running the auxiliary regression in place of the main regression
of Yi on the Xi.
If we regress each independent variable on the other independent
variables such model is called auxiliary regression model.
(4.1) 𝑋1𝑖 = ∝0 + ∝1 𝑋2𝑖 + ∝2 𝑋3𝑖 + 𝜇𝑖
Then, we have to compute F statistics for corresponding R-square. If
the computed Fi exceeds the critical Fi at the chosen level of
significance, it will be considered that particular Xi is collinear with
other independent variables. It should be noted that multicollinearity
may be a troublesome problem only if the R2 obtained from the
regression of Y on X is less than the R2 obtained from an auxiliary
regression.
Econometrics: Applications with EViews by Abdul Waheed 18
Chapter 4: Multicollinearity

Quantitative Research Methods by Abdul Waheed 19

Chapter 4: Multicollinearity
Remedial Measures of Multicollinearity
1. Increase the Sample Size
Sometimes high correlation among variables can be reduced by
increasing the size of the sample. In this way, we can avoid the
multicollinearity problem.

However, if the variable in the population are highly correlated,

then obviously the variables of the sample will also be highly
correlated. In this case increasing the size of the sample will not
reduce the multicollinearity problem in the regression model.

Econometrics: Applications with EViews by Abdul Waheed 20

Chapter 4: Multicollinearity

Econometrics: Applications with EViews by Abdul Waheed 21

Chapter 4: Multicollinearity
2. Drop a Variable(s) from the Model
If the multicollinearity problem is very serious, then we can avoid
it by simply dropping a variable. However, it should be noted that
if we drop an important variable of the model, there will be an
issue of specification bias, that is, the model is not correctly
specified.

Thus, sometimes omitting a variable may reduce the

multicollinearity problem but result in another problem.
Therefore, this remedial measure should be used very carefully.

Econometrics: Applications with EViews by Abdul Waheed 22

Chapter 4: Multicollinearity

Econometrics: Applications with EViews by Abdul Waheed 23

Chapter 4: Multicollinearity
3. Transformation of the Variables
If we take the first difference transformation of the variables and
run the regression model, this often reduces the multicollinearity
problem as sometimes the level variables may be highly
correlated but the difference of the variables may not be highly
correlated.

Econometrics: Applications with EViews by Abdul Waheed 24

Chapter 4: Multicollinearity

Econometrics: Applications with EViews by Abdul Waheed 25

Chapter 4: Multicollinearity
The other method to get rid of the multicollinearity problem is to
use the ratio of the variables.

In many cases, we can use the investment-to-GDP ratio, tax-to-

GDP ratio, export-to-GDP ratio, debt-to-export ratio, and per
capita income (GDP to population ratio). This always help to
reduce the multicollinearity problem in the model.

Econometrics: Applications with EViews by Abdul Waheed 26

Chapter 4: Multicollinearity

Econometrics: Applications with EViews by Abdul Waheed 27

Chapter 4: Multicollinearity
In the difference transformation, there will be a loss of
observation, and therefore the degrees of freedom are reduced by
one.

Furthermore, the transformation may result in other problems

such as autocorrelation or heteroscedasticity, so the remedial
measures may result in another problem, therefore, the researcher
needs to be careful while using this transformation.

Econometrics: Applications with EViews by Abdul Waheed 28

Chapter 4: Multicollinearity
4. Combine the Cross-sectional and Time-series data
The multicollinearity problem can be reduced if we combine the
cross-sectional and time series data.

The combination of time series and cross-sectional data is known

as pooled or panel data. If we run a regression model with pooled
or panel data, then there are chances that the multicollinearity
problem will not be serious.

Econometrics: Applications with EViews by Abdul Waheed 29

Chapter 4: Multicollinearity
5. Express the Variable in the Deviation Form
If we express the explanatory variable (s) in deviation form (that
is, the deviation from the mean), we can reduce the
multicollinearity problem from the regression model.

But even after this the problem may persist, then we may add up
some other remedial measures to resolve the issue.

Econometrics: Applications with EViews by Abdul Waheed 30

Chapter 4: Multicollinearity
6. Do Nothing
Because the OLS estimates are still unbiased even in the presence
of high multicollinearity, therefore, sometimes we do nothing and
work with the available results.

Multicollinearity is sometimes essentially a data deficiency

problem, and we have no choice over the data, that’s why we do
nothing.

Furthermore, the multicollinearity is specific to the correlated

independent variable. Therefore, if a moderate multicollinearity is
present in some correlated independent variables that are not of
your major concern, then you may not need to resolve it.
Econometrics: Applications with EViews by Abdul Waheed 31
Chapter 4: Multicollinearity
For example, if multicollinearity is present in control variables
and not in the main experimental variable then you need not to
worry too much.

Similarly, if the objective of the model is to make prediction for

the dependent variable and you are not interested in the individual
effect of each independent variable then multicollinearity may be
ignored.

Econometrics: Applications with EViews by Abdul Waheed 32

End of the Chapter

Statistical Analysis in Business and Economics by Abdul Waheed 33