An

Assignment
On
Multicollinearity
Course No.: Ag. Econ.- 507
Course Title: Econometrics

SUBMITTED TO
Shivraj Singh Rathore
Assistant Professor,
Department of Agril. Economics
C.P.C.A., S..D.A.U.
SUBMITTED BY
PATEL PARUL MUKESHBHAI
Degree - M. Sc. (Agril. Econ.)
Reg. No. – 04-AGRMA – 01804-2018
C.P.C.A., S.D.A.U.
 THE MEANING AND SOURCE

Perfect multicollinearity is the violation of Assumtion 6 (no explanatory variable

is a perfect linear function of any other explanatory variables (rxixj ҂ 1).

The multicollinearity in econometrics is defined as a situation of high correlation

between any one or more pairs of explantory variables. Due to co- movement of
variables in time series data or interdependence of economic magnitudes in cross
sctiona data, some degree of intercorrelation among the explanatory variables or
regresore is always present. Thus the problem of multicollinearity is always
present except for an extreme case where the explanatory variables are orthogonal (
two variables are called orthogonal, if the correlation between them is zero). But
such a case is never encountered where real world data are handelled.

Multicollinearity is not a condition that either exists or does not exists in economic
functions, rather a phenomenon inherent in most relationships due to the nature of
economic magnitudes. When any two explanatory variables are changing in nearly
the same way, it becomes extremely difficult to establish the influence of each one
regressor on Y seperately.

For example assume that the consumption expenditure of an individual depends on

his income and liquid assests change by the same proportion, the influence on
consumption of one of these variables may be erroneaously attributed to the other.
The effect of these variables on consumption cannot be sensibly investigated, due
to their high intercorrelation.

Multicollinearity may arise for various reasons:

1) There is a tendency of economic variables to move together over time.

2) The use of lagged values of some explanatory variables as separate
independent factors in the relationship.

There are two cases of multicolinearity:

1] Perfect Collinearity:

Perfect collinearity is a situation where any one pair of explanatory variables is

perfectly correlated i.e., simple correlation between them is unity. This situation
can be encountered in practice when observations on one of the explanatory
variables are exact multiples of observations on the other variable.

If the intercorrelation between explanatory variables is perfect (rxixj =1), then

a) The estimates of the coefficients are indeterminate ,

b) The standard error of these estimates become infinitely large.

Proof: a) Suppose the relation to be:

Y = b0 + b1X1+ b2X2 + u

X1 and X2 are in exact relation X2 = kx1 Where k is any arbitrary constant

number.

The estimates of coefficients :

Substituting kX1 for X2 :.

Therfore, the parameters are indeterminate (means no way of finding separate

values of each coefficient).

Proof b): If (rxixj =1) the standard errors of the estimates become infinitely
large.
Substituting kX1 for X2 :.

Thus , the variance of the estimates become infinite unless σu2 = 0 .

2] Less than perfect but higher collinearity:

Perfect collinearity is an extreme case rarely encountered in practice. But in the

cases of high collinearity are often encountered in agricultural economic analysis.

For example, in farm or crop production function studies the explanatory variables
viz., land, labour, bullock labour, operating capital etc. show the same broad
pattern of behaviour across various farms. In consumption or demand studies, the
explanatory variables like family incomes in cross section data and prices of
commodity in question and prices of other substitute/ complementary commodities
show the same broad pattern of behaviour across families or over time as the case
may be.

If the X’s are not perfectly collinear, but are to a certain degree correlated (0<
rxixj<1), the effect of collinearity are uncertain.

Two points should be stressed:

a) The estimates of the coefficients are statistically unbiased ( E(𝑏)= bi ) even

when multicollinearity is strong.
 b) When multicollinearity is present in the function the standard errors of the
estimates will, in general large.

THE CONSEQUENCES OF PRESENCE OF MULTICOLLINEARITY :

1) The estimates of regression coefficients are indeterminate if

multicollinearity is perfect,
2) The standard errors of regression coefficints are inflated which makes the
results less precise.i.e., interval estimates are very wide;
3) The OLS estimaors are unbiased;
4) The coefficients of multiple determination (R2 ) may be high but majority or
all of the regression coefficients may be insignificant. This is a situation
which will lead one to interprete that the explanatory variables taken
together explain very high variation in the dependent variable but
individually none is experting a significant effect. Thus presence of
multicollinearity may lead to wrong interpretation of the effect of some
explanatory variables.

KLEIN’S METHOD FOR TESTING MULTICOLLINEARITY :

The presence and extent of multicollinearity in the data, a simple zero order
correlation matrix of all explanatory variables should be worked. If none of the
correlation coefficients is high, the problem of multicollinearity is not considered
serious. But what magnitude of correlation coefficient is high and what is not is
subjective judgement. If correlation coefficient is 0.8 or more, it is usually to be
considerd high, but if it is less than 0.70 it can be considered as low.

Klein has suggested that multicollinearity should not be considered serious if the
simple correlation coefficient between a pair of variables is less than the multiple
correlation coefficient. Means collinearity is harmful if

r2 xixj ≥ R2y x1,x2,..xk

where, r2 xixj = simple correlation between any two explanatory variables and
R2 = overall (multiple) correlation of the relationship.
TESTS FOR DETECTING MULTICOLLINEARITY:

1] Frisch’s confluence analysis

The procedure is to regress the dependent variable on each one of the explanatory
variables seperately. Thus we obtain all the elementary regressions, and we
examine their results on the basis of a priori and statistical data.

We choose the elementary regression which appears to give the most plausible
results, on both a priori and statistical criteria. Then we gradually insert additional
variables and we examine their effects on the individual coefficients, on their
standard errors, and on the overall R2. A new variable is classified as :

a) Useful : if the new variable improve R2 without rendering the individual

coefficients unacceptable (wrong) on a priori considerations, the variable is
considered useful and is retained as an explanatory variable.
b) Superflous : if the new variable does not improve R2 and does not affect to
any considerable extent the values of the individual coefficients, it is
considered as superflous and is rejected (i.e., is not included among the
explanatory variables).
c) Detrimental : if the new variable affects considerably the signs or the
values of the coefficients, it is considered as detrimental. If the individual
coefficients are affected in such a way as to become unacceptable , then we
may say that this is a warning that multicollinearity is a serious problem.

2] The Farrar- Glauber test:

A statistical test for multicollinearity has been recently developed by Farrar and
Glauber. It is really a set of three tests, that is, the authors use three statistics for
testing for multicollinearity. The first test is a chi-square test for the detection of
the existence and the severity of multicollinearity in a function including several
explanatory variables. The second test is F-tests for locating which variables are
multi-collinear. The third test is a t-Test for finding out the pattern of
multicollinearity, which is for determining which variables are responsible for the
appearance of multi-collinear variables. Farrar Glaubar considers multicollinearity
in a sample departure of the observed X's from orthogonality. Their approach
emerged from the general ideals developed in the preceding paragraphs namely
that if multicollinearity is perfect, then the coefficients becomes indeterminate, and
that the inter-correlations among the various explanatory variables can be
measured by multiple correlation coefficients and partial correlation coefficients.
The Farrar Glauber test may be outlined as follows:

Steps in Carrying Out the Farrar-Glaubar Test

i. Conduct the Chi - Square test to detect the existence of severity of

multicollinearity in a function with several explanatory variables..
ii. Carry out F-test to locate the variables(s) inter-correlated, if Chi-Square
test is positive.
iii. Conduct T-test to detect the variables(s) that are responsible for
multicollinearity if the F-test is positive. (for the pattern of
multicollinearity).

SOLUTIONS FOR THE INCIDENCE OF MULTICOLLINEARITY:

The steps to be taken to take care of the presence of multicollinearity depends on:

1) The seriousness of multicollinearity

2) The purpose of the analysis
3) The importance of each factor which is collinear
4) Availability of other sources of data.

Some of the suggested methods are given below:

1) If the extent of multicollinearity is not considered serious, it may be

ignored.
2) If multicollinearity is present among unimportant variables (on the
context of one’s analysis), some of these variables may be dropped or
combined.
3) Increasing the sample size also reduces the problem of multicollinearity.
4) In time series studies specially when prices and monetary incomes
appears as explanatory variables, one can reduce the proble4m of
multicollinearity by taking relative prices and relative incomes in place of
absolute prices and incomes. Sometimes pooling of cross- section data
and time- series data is also done.
5) If multicollinearity is present and is expected to continue in future, as far
as prediction is comcerned, it does not create a serious problem.
6) When a series of past lagged values are supposed to be included as
explanatory variables, the Koyck’s distributed lag model can be used
which will be reduce the number of explanatory variables as well as solve
the problem of multicollinearity.