ECONOMETRICS

Applications with EViews

Abdul Waheed
Econometrics: Applications with EViews by Abdul Waheed 1
Chapter 5: Autocorrelation
Learning Outcomes
1. Understand the meaning of autocorrelation.

2. Differentiate between autocorrelation and serial correlation.

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

Econometrics: Applications with EViews by Abdul Waheed 2

Chapter 5: Autocorrelation
Introduction

One of the assumptions of the OLS estimation method is that the

error term is not correlated. This means the value, which the error
term takes in any one period is independent of the value of the
error term it took in any previous period.

This assumption means that the covariance of error terms of

𝜇𝑖 and 𝜇𝑗 is equal to zero, that is:

Econometrics: Applications with EViews by Abdul Waheed 3

Chapter 5: Autocorrelation
If this assumption is violated, that is,

Then we conclude that there is a problem of

autocorrelation or serial correlation in the regression
model.

Econometrics: Applications with EViews by Abdul Waheed 4

Chapter 5: Autocorrelation
Meaning of Autocorrelation
The word autocorrelation and serial correlation are used
synonymously.

However, the autocorrelation means lag correlation of a given

error term with itself.

The serial correlation is the lag correlation between two different

series of error terms.
Autocorrelation can be first-order or high-order. The first-order
autocorrelation is presented in the model if an error is
influenced by the error from the preceding period.
Econometrics: Applications with EViews by Abdul Waheed 5
Chapter 5: Autocorrelation
The autocorrelation problem is mainly observed when we used
the time series data. If the correlation is error term is observed in
the case of cross-sectional data, then it is called spatial
autocorrelation, which means the correlation of error terms is
related to space rather than time.

Positive or negative autocorrelation is possible. Positive

autocorrelation happens when the error tends to maintain the
same sign from period to period. If the error term's sign
frequently changes over time or have a negative relationship, then
it shows negative autocorrelation.

Econometrics: Applications with EViews by Abdul Waheed 6

Chapter 5: Autocorrelation

Econometrics: Applications with EViews by Abdul Waheed 7

Chapter 5: Autocorrelation

Econometrics: Applications with EViews by Abdul Waheed 8

Chapter 5: Autocorrelation
Reasons of Autocorrelation
1. Most economic time series follow the business cycle. In the
period of recovery, they show an uploaded trend and in the period
of recession, they start moving downward. Therefore, the
regression analysis of such time series data (GDP, CPI, WPI,
employment, unemployment, interest rate, money supply) is
likely to result in an autocorrelation problem.

2. If there is specification bias (such as excluding the relevant

variable or including the irrelevant variable or using an incorrect
functional term) in the model, then there may arise an
autocorrelation problem.
Econometrics: Applications with EViews by Abdul Waheed 9
Chapter 5: Autocorrelation
3. If there is a presence of lag-dependent variable in the
regression model it may result in an autocorrelation problem.

4. Sometimes the series has missing values and such missing

values are filled through interpolation or extrapolation.
5. The transformation of data (taking first different) may result in
autocorrelation problem.

6. Use of non-stationary time series (series whose, mean,

variance, and covariance are time variant) may also result in
autocorrelation problem.

Econometrics: Applications with EViews by Abdul Waheed 10

Chapter 5: Autocorrelation
Consequences of Autocorrelation
Following are the consequences of the autocorrelation problem.
1. Even when the residuals are serially correlated (presence of
autocorrelation in the model), the estimated OLS estimates are linear,
unbiased and consistent, but they are not efficient, that is, estimated
parameters do not have minimum variance. Hence, the estimators are
not BLUE.

2. As the presence of autocorrelation affects the variance of the

estimated parameters, the estimated parameters will have larger
variances. Consequently t-statistic will be under-estimated. Therefore,
the usual t test will no longer be valid and lead to misleading inference.
Econometrics: Applications with EViews by Abdul Waheed 11
Chapter 5: Autocorrelation
3. Furthermore, the insignificant parameters in the model will
decrease the value of 𝑅 2 . Thus, the R2 will be misleading for the
goodness of fit of the model.

4. As we cannot rely on R2, the F test will also be misleading for

the overall significance of the model.

Econometrics: Applications with EViews by Abdul Waheed 12

Chapter 5: Autocorrelation
Detection of Autocorrelation
Plot of the Residuals
If the plot of the residuals against time shows a systematic
pattern, then it means that autocorrelation is present in the model.

Figure 7.3 shows the plot of the error terms (from estimated
regression) for the supply-price model. The systematic pattern in
all the figures indicates the autocorrelation problem in the
estimated regression results.

Econometrics: Applications with EViews by Abdul Waheed 13

Chapter 5: Autocorrelation

Econometrics: Applications with EViews by Abdul Waheed 14

Chapter 5: Autocorrelation
Correlogram

The correlogram is the plot of the residual covariances

standardized by the residual variance. If the autocorrelation (AC)
that is correlation with lag values of a series, and partial
autocorrelation (PAC), that is correlation with two series, for all
lags are close to zero and Q-statistics are insignificant (that is, its
p-value is greater than 0.05), it confirms no autocorrelation in the
residuals. In this case, the correlogram plot will not show a
systematic pattern.

Econometrics: Applications with EViews by Abdul Waheed 15

Chapter 5: Autocorrelation
The correlogram and Q-statistic for the supply price model
(with the dependent variable as production of wheat WSUPPLY
and the independent variable as the support price of wheat
SPWHEAT) using the time series data of 1982 to 2015 are
shown below.

There is a systematic pattern in the correlogram (see

autocorrelation). Furthermore, the probability of Q-statistics is
all less than 0.05. both results sign the pressure of the
autocorrelation problem in the regression model.
Econometrics: Applications with EViews by Abdul Waheed 16
Chapter 5: Autocorrelation

Econometrics: Applications with EViews by Abdul Waheed 17

Chapter 5: Autocorrelation
Durbin-Watson Test

The Durbin-Watson (DW) test is the most common test

for the detection of autocorrelation in the regression
model. This is the test for first-order autocorrelation,
which is the most common problem in regression
analysis.

Econometrics: Applications with EViews by Abdul Waheed 18

Chapter 5: Autocorrelation
The conditions to apply the DW test are as follows.

1. The regression model should include an intercept term.

2. The lagged dependent variable should not be present as the
explanatory variable in the regression model.
3. The explanatory variable should be non-stochastic, i.e., fixed in
repeated sampling.
4. The error term Ui is assumed to be normally distributed.

Econometrics: Applications with EViews by Abdul Waheed 19

Chapter 5: Autocorrelation
Suppose the disturbance 𝜇𝑖 is generated by the first order to
regression scheme as follows.

(5.1) 𝜇𝑡 = 𝜌 𝜇𝑡−1 +et

The DW test statistic for first-order autocorrelation is given by

σ𝑇
𝑡=2(𝑒𝑡 − 𝑒𝑡−1 )
2
(5.2) DW = σ𝑇 2
𝑡=1(𝑒𝑡 )

Where et is the estimated residual from a sample regression model. The

larger is the time period (T) the better will be the approximation.

Econometrics: Applications with EViews by Abdul Waheed 20

Chapter 5: Autocorrelation
It can also be written as:

(5.3) DW = 2 (1 - 𝜌)
ො

The 𝜌ො on the right-hand side is the autocorrelation coefficient

from a first order autoregression scheme.

The sample first-order coefficient of autocorrelation ( 𝜌),

ො an
estimator of 𝜌 can take values between +1 and -1.

From (5.3) it is cleared that the DW test statistics only take a

value between 0 and 4.
0 ≤ DW ≤ 4
Econometrics: Applications with EViews by Abdul Waheed 21
Chapter 5: Autocorrelation
1. If the autocorrelation coefficient (𝜌)
ො is equal to 0, the DW test
statistic becomes 2. Then, this shows there is no autocorrelation
problem in the regression results.

2. If the anticorrelation coefficient (𝜌)

ො is equal to +1, the DW test
statistic becomes 0. Thus, if the DW test statistic is close to 0
there is clear evidence of positive anticorrelation.

3. If the autocorrelation coefficient (𝜌)

ො is equal to -1, the DW test
statistic becomes 4. Thus, if the DW statistic is close to 4, there is
clear evidence of negative autocorrelation

Econometrics: Applications with EViews by Abdul Waheed 22

Chapter 5: Autocorrelation

Econometrics: Applications with EViews by Abdul Waheed 23

Chapter 5: Autocorrelation
Serial Correlation LM Test

The Durbin-Watson test is no longer valid for the test of autocorrelation

if there is a lagged dependent variable in the regression model. In this
case, we can use the Breusch Godfrey Serial Correlation LM Test (in
short LM test). The advantage of the L-M test are:

1. Even if the regression includes the lagged dependent variable as an

explanatory variable in the regression model, the LM test is still
applicable.

2. The test is still applicable even if the error terms follow a simple or
higher order moving average process.
Econometrics: Applications with EViews by Abdul Waheed 24
Chapter 5: Autocorrelation
The EViews software can provides the result of such a test.

The null hypothesis of the LM test is that there is no

autocorrelation.

We accept that there is no problem of autocorrelation in the

regression model, if the value of the F-statistic and Chi-Square
is small, and its probability value is more than 0.05.

Econometrics: Applications with EViews by Abdul Waheed 25

Chapter 5: Autocorrelation
The LM test results of the supply-price model are shown in Table 5.3. Hence
there is the presence of the autocorrelation problem in the supply-price
Model.

Econometrics: Applications with EViews by Abdul Waheed 26

Chapter 5: Autocorrelation
Remedial Measures of Autocorrelation
Different approaches are used to treat autocorrelation. Some of
the approaches are discussed below.

Re-specification of the Model

To solve the autocorrelation issue, it is best to reformulate the
regression model and re-estimate the regression coefficients. By
introducing or eliminating new variables or altering the model's
functional form, we can reformulate it.

Econometrics: Applications with EViews by Abdul Waheed 27

Chapter 5: Autocorrelation

Econometrics: Applications with EViews by Abdul Waheed 28

Chapter 5: Autocorrelation
Cochrane-Orcutt Iterative Method

In some cases, the autocorrelation issue cannot be solved by

modifying the model's functional form or adding or removing the
variable(s). In these circumstances, the autocorrelation issue can be
resolved using the Cochrane-Orcutt approach.

The Cochrane-Orcutt is an iterative method to overcome the

autocorrelation problem.

Quantitative Research Methods by Abdul Waheed 29

Chapter 5: Autocorrelation
The step involved in this method is discussed below.
Step 1: Run the regression 𝑌𝑖 = 𝛽0 + 𝛽1 𝑋𝑖 + 𝜇𝑖 and obtain the
estimates of 𝛽0 and 𝛽1. Using the DW statistic, compute:

𝐷𝑊
(5.6) 𝜌ො = 1 −
2

Step 2: Now using 𝜌ො , transform the original data.

(5.8) 𝑌 ∗ = 𝑌𝑡 − 𝜌𝑌
ො 𝑡−1
(5.9) 𝑋 ∗ = 𝑋𝑡 − 𝜌𝑋
ො 𝑡−1

Quantitative Research Methods by Abdul Waheed 30

Chapter 5: Autocorrelation
Now, run the regression on transform variables Y* and X* and obtain the
estimates of 𝛽0 and 𝛽1 . The OLS estimators are now expected to be
efficient.

(5.10) 𝑌 ∗ = 𝛽0∗ + 𝛽1∗ 𝑋 ∗ + 𝜇𝑡

𝛽0∗
(5.11) 𝛽0 = 1−𝜌
(5.12) 𝛽1 = 𝛽1∗

If the residuals from the new estimated equation still show

autocorrelation, then the same procedure may be repeated. (i.e. estimate 𝜌
from the new DW, and define 𝑌 ∗∗ and 𝑋 ∗∗ and re-estimate until there is
no longer any autocorrelation.

Quantitative Research Methods by Abdul Waheed 31

Chapter 5: Autocorrelation
Example of Cochrane-Orcutt Iterative Method
Now consider the supply-price model in which we found that there
is a serious problem of autocorrelation. If substituting DW=0.504
in (5.6), then 𝜌ො is given below.

0.504
Then 𝜌ො = 1 − = 0.748
2

Now we will generate a new series WSUPPLY1 and SPWHEAT1 with

the above (𝜌ො =0.748) and using (5.10) we run the regression of
WSUPPLY1 on SPWHEAT1. The results are shown in Table 5.5.

Quantitative Research Methods by Abdul Waheed 32

Chapter 5: Autocorrelation

Quantitative Research Methods by Abdul Waheed 33

Chapter 5: Autocorrelation
It is clear that there is still a problem of autocorrelation. So, we
again obtain 𝜌ො using DW=2.692

2.692
𝜌ො = 1 −
2

2.692
Then 𝜌ො = 1 − = -0.346
2

Now we will generate a new series WSUPPLY2 and SPWHEAT2

with the above (𝜌ො = − 0.346) and run the regression on
WSUPPLY2 on SPWHEAT2. The results of Table 5.6 shows that
the value of DW is now 1.94. Hence, now there is no
autocorrelation problem.

Quantitative Research Methods by Abdul Waheed 34

Chapter 5: Autocorrelation

Quantitative Research Methods by Abdul Waheed 35

End of the Chapter

Statistical Analysis in Business and Economics by Abdul Waheed 36