Autocorrelation PDF
Autocorrelation PDF
Autocorrelation
Chapter 4.6-4.11
24/9 2014
(AEM I)
24/9 2014
1 / 37
Overview
OLS properties.
What is autocorrelation?
Consequences of autocorrelation.
Solutions.
Testing for autocorrelation.
Illustration: the demand for ice cream (In EViews).
Illustration: incomplete dynamics (EViews program).
(AEM I)
24/9 2014
2 / 37
OLS properties
I
I
(AEM I)
24/9 2014
3 / 37
0
0
6 0 2
0 7
7
6
7
6
2
7
6
V fe j X g = I = 6
7
7
6
5
4
0
(AEM I)
24/9 2014
4 / 37
Autocorrelation
(AEM I)
24/9 2014
5 / 37
Autocorrelation
The error term picks up the inuence of those (many) variables and
factors not included in the model.
If there is some persistence in these factors, autocorrelation may arise.
Thus, autocorrelation may be an indication of a misspecied model
(omitted variables, incorrect functional forms, incorrect dynamics).
Accordingly, autocorrelation tests are often interpreted as
misspecication tests.
(AEM I)
24/9 2014
6 / 37
Autocorrelation
Demand for ice cream explained from income and price index
(AEM I)
24/9 2014
7 / 37
Autocorrelation
(AEM I)
24/9 2014
8 / 37
First-order autocorrelation
+ t
(AEM I)
24/9 2014
9 / 37
(AEM I)
2
1 2
24/9 2014
10 / 37
cov ft , t
1g
Thus, this form of autocorrelation implies that all error terms are
correlated. Their covariance decreases if the distance in time gets
large.
We see that if = 0, the non-diagonal elements of the covariance
matrix are equal to zero (no autocorrelation).
(AEM I)
24/9 2014
11 / 37
+ t
implies that all errors are correlated with each other, with correlations
becoming smaller if they are further apart.
Two alternatives:
I
I
(AEM I)
24/9 2014
12 / 37
With quarterly or monthly (macro) data, higher order patterns are possible
(due to a periodic eect). For example, with quarterly data:
t = t
+ t
+ 2 t
+ 3 t
+ 4 t
+ t
(AEM I)
24/9 2014
13 / 37
1
1,
or t
3,
(AEM I)
24/9 2014
14 / 37
Consequences
This also indicates three general ways to deal with the problem:
I
I
I
(AEM I)
24/9 2014
15 / 37
(AEM I)
24/9 2014
16 / 37
(AEM I)
24/9 2014
17 / 37
(AEM I)
24/9 2014
18 / 37
+ t
yt
= (1
) 1 + 2 (xt
xt
1 ) + t
With known , this produces (almost) the GLS estimator. Note: rst
observation is lost by this transformation.
(AEM I)
24/9 2014
19 / 37
= yt
= yt
yt
(b1 + b2 xt )
+ t
(AEM I)
24/9 2014
20 / 37
Change functional form (e.g. use log(x ) rather than x), see Figure
4.5.
Extend the model by including additional explanatory variables
(seasonals) or additional lags.
In many cases, including lagged values of y and/or x will help
eliminate the serial correlation problem. We will see an example of
this below.
(AEM I)
24/9 2014
21 / 37
1,
we obtain an
(AEM I)
24/9 2014
22 / 37
Some remarks:
If we also suspect heteroskedasticity, White standard errors may be
used in the auxiliary regression.
If the model of interest contains lagged values of yt (or other
explanatory variables that may be correlated with lagged error terms),
the auxiliary regression should also include all explanatory variables
(just to make sure the distribution of the test is correct). This is
typically refered to the Breusch-Godfrey autocorrelation test. In the
BG test the origional regressors are always included in the auxiliary
regression:
et = et 1 + xt0 + t
The BG test is valid also in the case where the model does not
include lagged values of yt !
(AEM I)
24/9 2014
23 / 37
(AEM I)
24/9 2014
24 / 37
Tt=2 (et et
Tt=1 et2
1)
(AEM I)
24/9 2014
25 / 37
Distribution is peculiar.
Moreover, it depends upon xt s.
In general, dw values close to 2 are ne, while dw values close to 0
imply positive autocorrelation.
The exact critical value is unknown, but upper and lower bounds can
be derived (see Table 4.8).
Thus (to test for positive autocorrelation):
I
I
I
(AEM I)
24/9 2014
26 / 37
(AEM I)
24/9 2014
27 / 37
+ 2 et
+ ... + q et
+ xt0 + t
Null hypothesis:
H0 : 1 = 2 = ... = q = 0
Test statistic: (T
with q DF.
(AEM I)
q)
24/9 2014
28 / 37
(AEM I)
24/9 2014
29 / 37
(AEM I)
24/9 2014
30 / 37
(AEM I)
24/9 2014
31 / 37
(AEM I)
24/9 2014
32 / 37
p
(T
1)
T = 2.19
R 2 = 4.32
24/9 2014
33 / 37
24/9 2014
34 / 37
= t
N (0, 0.01)
(AEM I)
24/9 2014
35 / 37
(AEM I)
24/9 2014
36 / 37
Summing up
Possible solutions:
I
I
I
(AEM I)
24/9 2014
37 / 37