Chapter11 Regression Autocorrelation
Chapter11 Regression Autocorrelation
Chapter11 Regression Autocorrelation
Autocorrelation
One of the basic assumption in linear regression model is that the random error components or disturbances
y X + u , it is assumed that
are identically and independently distributed. So in the model=
u2 if s = 0
E (ut , ut s ) =
0 if s 0
i.e., the correlation between the successive disturbances is zero.
ut s ) =
In this assumption, when E (ut , = 2
u, s 0 is violated, i.e., the variance of disturbance term does not
When autocorrelation is present, some or all off diagonal elements in E (uu ') are nonzero.
Sometimes the study and explanatory variables have a natural sequence order over time, i.e., the data is
collected with respect to time. Such data is termed as time series data. The disturbance terms in time series
data are serially correlated.
Assume s and s are symmetrical in s , i.e., these coefficients are constant over time and depend only on
length of lag s. The autocorrelation between the successive terms (u2 and u1 ) ,
(u3 and u2 ),..., (un and un 1 ) gives the autocorrelation of order one, i.e., 1 . Similarly, the autocorrelation
Regression Analysis | Chapter 11 | Autocorrelation | Shalabh, IIT Kanpur
1
between the successive terms (u3 and u1 ), (u4 and u2 )...(un and un 2 ) gives the autocorrelation of order two,
i.e., 2 .
Source of autocorrelation
Some of the possible reasons for the introduction of autocorrelation in the data are as follows:
1. Carryover of effect, atleast in part, is an important source of autocorrelation. For example, the
monthly data on expenditure on household is influenced by the expenditure of preceding month. The
autocorrelation is present in cross-section data as well as time series data. In the cross-section data,
the neighboring units tend to be similar with respect to the characteristic under study. In time series
data, the time is the factor that produces autocorrelation. Whenever some ordering of sampling units
is present, the autocorrelation may arise.
2. Another source of autocorrelation is the effect of deletion of some variables. In regression modeling,
it is not possible to include all the variables in the model. There can be various reasons for this, e.g.,
some variable may be qualitative, sometimes direct observations may not be available on the
variable etc. The joint effect of such deleted variables gives rise to autocorrelation in the data.
3. The misspecification of the form of relationship can also introduce autocorrelation in the data. It is
assumed that the form of relationship between study and explanatory variables is linear. If there are
log or exponential terms present in the model so that the linearity of the model is questionable then
this also gives rise to autocorrelation in the data.
4. The difference between the observed and true values of variable is called measurement error or
errorsin-variable. The presence of measurement errors on the dependent variable may also
introduce the autocorrelation in the data.
Observe that now there are (n + k ) parameters- 1 , 2 ,..., k , u2 , 1 , 2 ,..., n 1. These (n + k ) parameters are
to be estimated on the basis of available n observations. Since the number of parameters are more than the
number of observations, so the situation is not good from the statistical point of view. In order to handle the
situation, some special form and the structure of the disturbance term is needed to be assumed so that the
number of parameters in the covariance matrix of disturbance term can be reduced.
ut ut 1 + t
=
where < 1, E ( t ) =
0,
ut ut 1 + t
=
= (ut 2 + t 1 ) + t
=
t t 1 + 2 t 2 + ...
=+
= r t r
r =0
E (ut ) = 0
E ( t2 ) + 2 E ( t21 ) + 4 E ( t2 2 ) + ...
E (ut2 ) =
=(1 + 2 + 4 + ....) 2 ( t' s are serially independent)
2
E (ut=
2
) =
2
for all i.
1 2
u
E (ut ut =
1) E ( t + t 1 + 2 t 2 + ...) ( t 1 + t 2 + 2 t 3 + ...)
= E { t + ( t 1 + t 2 + ...)}{ t 1 + t 2 + ...}
= E ( t 1 + t 2 + ...)
2
= u .
2
Similarly,
E (ut ut 2 ) = 2 u2 .
In general,
E (ut ut s ) = s u2
1 2 n 1
1 n2
.
E (uu ') = = u 2
2
1 n 3
n 1 n2
n 3 1
Note that the distribution terms are no more independent and E (uu ') 2 I . The disturbance are
nonspherical.
Regression Analysis | Chapter 11 | Autocorrelation | Shalabh, IIT Kanpur
4
Consequences of autocorrelated disturbances:
Consider the model with first order autoregressive disturbances
y X + u
=
n1 nk k 1 n1
ut = ut 1 + t , t = 1, 2,..., n
with assumptions
E (u ) = 0, E (uu ') =
2 if s = 0
= E ( t ) 0,=
E ( t t + s )
0 if s 0
where is a positive definite matrix.
b = ( X ' X ) 1 X ' y
= ( X ' X ) 1 X '( X + u )
b = ( X ' X ) 1 X ' u
E (b ) =0.
So OLSE remains unbiased under autocorrelated disturbances.
Since disturbances are nonspherical, so generalized least squares estimate of yields more efficient
estimates than OLSE.
The GLSE of is
=
( X ' 1 X ) 1 X ' 1 y
E ( ) =
=
V ( ) 2 ( X ' 1 X ) 1.
u
e =y Xb =Hy
(e e )
2
t t 1
d= t =2
n
e
t =1
2
t
n n n
et2 et 1 et et 1
=n
=t 2=t 2
+ n 2 n
=t 2
.
et et
2
=t 1 =t 1
2
et 2
=t 1
zero autocorrelation of et s d 2
As 1 < r <1, so
if 1 < r < 0, then 2 < d < 4 and
if 0 < r <1, then 0 < d < 2.
So d lies between 0 and 4.
Since e depends on X , so for different data sets, different values of d are obtained. So the sampling
distribution of d depends on X . Consequently exact critical values of d cannot be tabulated owing to
their dependence on X . Durbin and Watson therefore obtained two statistics d and d such that
d <d <d
and their sampling distributions do not depend upon X .
Considering the distribution of d and d , they tabulated the critical values as d L and dU respectively.
They prepared the tables of critical values for 15 < n < 100 and k 5. Now tables are available for
6 < n < 200 and k 10.
The test procedure is as follows:
H0 : = 0
Nature of H1 Reject H 0 when Retain H 0 when The test is inconclusive
when
H1 : > 0 d < dL d > dU d L < d < dU
H1 : < 0 d > (4 d L ) d < (4 dU ) ( 4 dU ) < d < (4 d L )
H1 : 0 d < d L or dU < d < (4 dU ) d L < d < dU
d > (4 d L ) or
(4 dU ) < d < (4 d L )
Values of d L and dU are obtained from tables.
Regression Analysis | Chapter 11 | Autocorrelation | Shalabh, IIT Kanpur
7
Limitations of D-W test
1. If d falls in the inconclusive zone, then no conclusive inference can be drawn. This zone becomes
fairly larger for low degrees of freedom. One solution is to reject H 0 if the test is inconclusive. A
better solutions is to modify the test as
Reject H 0 when d < dU .
Accept H 0 when d dU .
This test gives satisfactory solution when values of xi s change slowly, e.g., price, expenditure
etc.
2. The D-W test is not applicable when intercept term is absent in the model. In such a case, one can use
another critical values, say d M in place of d L . The tables for critical values d M are available.
3. The test is not valid when lagged dependent variables appear as explanatory variables. For example,
yt 1 yt 1 + 2 yt 2 + .... + r yt r + r +1 xt1 + ... + k xt ,k r + ut ,
=
ut ut 1 + t .
=
In such case, Durbins h test is used which is given as follows.
Durbins h-test
Apply OLS to
yt 1 yt 1 + 2 yt 2 + .... + r yt r + r +1 xt1 + ... + k xt ,k r + ut ,
=
ut ut 1 + t
=
(b ). Then the Dubins h -
and find OLSE b1 of 1. Let its variance be Var (b1 ) and its estimator is Var 1
statistic is
n
h=r
(b )
1 n Var 1
e e t t 1
r= t =2
n
.
e
t =2
2
t
et t 1 + yt .
=
Now apply OLS to this model and test H 0 A : = 0 versus H1 A 0 using t -test . It H 0 A is accepted then
accept H 0 : = 0.
4. If H 0 : = 0 is rejected by D-W test, it does not necessarily mean the presence of first order
autocorrelation in the disturbances. It could happen because of other reasons also, e.g.,
distribution may follows higher order AR process.
some important variables are omitted .
dynamics of model is misspecified.
functional term of model is incorrect.
The OLSE of is unbiased but not, in general, efficient and estimate of 2 is biased. So we use
generalized least squares estimation procedure and GLSE of is
= ( X ' 1 X ) 1 X ' 1 y
where
1 2 0 0 0 0
1 0 0 0
0 1 0 0
P= .
0 0 0 1 0
0 0 0 1
Note that the first observation is treated differently then other observations. For first observation,
( ) (
1 2 y1 = )
1 2 x1' + ( )
1 2 u1
where xt' is a row vector of X . Also, 1 2 u1 and (u1 u0 ) have same properties. So we expect
these two errors to be uncorrelated and homoscedastic.
1 2.
( y X ) ' 1 ( y X )
2 = .
nk
F =
( X ' 1 X ) 1 X '
1 y
e e t t 1
r= t =2
n
e
t =2
2
t
2. Durbin procedure:
In Durbin procedure, the model
yt yt 1 = 0 (1 ) + ( xt xt 1 ) + t , t = 2,3,..., n
is expressed as
yt = 0 (1 ) + yt 1 + x1 xt 1 + t
= 0* + yt 1 + xt + * xt 1 + t , t =2,3,..., n (*)
where 0* =
0 (1 ), * =
.
Now run regression using OLS to model (*) and estimate r * as the estimated coefficient of yt 1.
Another possibility is that since (1,1) , so search for a suitable which has smaller error sum of
squares.
e e t t 1
(ii) Estimate by r = t =2
n
.
e
t =2
2
t 1
e e t t 1
(III) Calculate by r = t =2
n
and substitute it is the model
e
t =2
2
t 1
yt yt 1 = 0 (1 ) + ( xt xt 1 ) + t
and again obtain the transformed model.
(IV) Apply OLS to this model and calculate the regression coefficients.
This procedure is repeated until convergence is achieved, i.e., iterate the process till the two successive
estimates are nearly same so that stability of estimator is achieved.
This is an iterative procedure and is numerically convergent procedure. Such estimates are asymptotically
efficient and there is a loss of one observation.
Suppose we get = 0.4. Now choose a finer grid. For example, choose such that 0.3 < < 0.5 and
consider = 0.31, 0.32,..., 0.49 and pick up that with smallest residual sum of squares. Such iteration
can be repeated until a suitable value of corresponding to minimum residual sum of squares is obtained.
The selected final value of can be used and for transforming the model as in the case of Cocharan-Orcutt
procedure. The estimators obtained with this procedure are as efficient as obtained by Cochran-Orcutt
procedure and there is a loss of one observation.
e e t t 1
(i) Estimate by = t =2
n
where et s are residuals based on OLSE.
e
t =3
2
t 1
( ) (
1 2 y1 = )
1 2 0 + ( ) (
1 2 xt + )
1 2 ut
yt yt 1 =(1 ) 0 + ( xt xt 1 ) + (ut ut 1 ), t =2,3,..., n.
1 1
=L exp 2 ( y X ) ' 1 ( y X ) .
2
n
( 2 )
n
2 2
2
1
Ignoring the constant and using = , the log-likelihood is
1 2
n 1 1
ln L( , 2 , ) =
ln L = ln 2 + ln(1 2 ) 2 ( y X ) ' 1 ( y X ) .
2 2 2
The maximum likelihood estimators of , and 2 can be obtained by solving the normal equations
ln L ln L ln L
= 0,= 0,= 0.
2
There normal equations turn out to be nonlinear in parameters and can not be easily solved.
One solution is to
- first derive the maximum likelihood estimator of 2 .
- Substitute it back into the likelihood function and obtain the likelihood function as the function of
and .
- Maximize this likelihood function with respect to and .
n 1 1 n
ln L *( , ) =
ln L* = ln ( y X ) ' 1 ( y X ) + ln(1 2 )
2 n 2 2
n
ln {( y X ) ' 1 ( y X )} ln(1 2 ) + k
1
=
2 n
n ( y X ) ' 1 ( y X )
= k ln
2 1
(1 2 ) n
n n
=
where k ln n .
2 2
( y X ) ' 1 ( y X )
1
.
(1 )
2 n
Using optimization techniques of non-linear regression, this function can be minimized and estimates of
and can be obtained.
If n is large and is not too close to one, then the term (1 2 ) 1/ n is negligible and the estimates of