20.02.10
EC220
Tests for
Autocorrelation
2009/2010
This note extends the discussion of tests for autocorrelation in the text.
The material becomes part of
the EC220 syllabus.
However, it will not be covered in the 2010 examination.
Durbin’s ‘alternative’ test for autoregressive autocorrelation
The Durbin–Watson test is appropriate only when both parts of Assumption C.7 are satisfied.
When
the lagged dependent is used as one of the regressors, the second part of Assumption C.7 is
automatically violated.
Durbin proposed two tests for application in this case.
One is the Durbin
h
test described in the text.
This test has two limitations.
One is that the
h
statistic cannot be computed
if the denominator under the square root is negative.
The other is that it cannot be generalized to test
for higherorder autocorrelation.
It is strictly an AR(1) test.
The other test, which will be described
here as Durbin’s alternative test, since it does not seem to have a name, proceeds as follows.
Suppose
that the regression model is
t
s
t
s
t
r
r
t
u
Y
Y
X
X
Y
1
1
2
2
1
...
(1)
and it is hypothesized that the disturbance term may be determined by the AR(
p
) process
t
p
t
p
t
t
u
u
u
...
1
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
This is the end of the preview.
Sign up
to
access the rest of the document.
 Spring '10
 öcal
 Statistics, Econometrics, Normal Distribution, Regression Analysis, Autocorrelation, Durbin

Click to edit the document details