Serial Correlation in VAR
1. Is there any serial correlation if the lag that we have chosen is 5?
DaifAllah Al-Houd posted the figure below.
Saeed Aas Khan Meo commented> Till lags 4 you have to face problem of serial
correction. Hence u can use 5 lags.
Saud Ahmad commented> Residuals are correlated with first 4 lagged terms which
violate the assumption of serial independence of residuals. So u have to re-specify ur
model and if u increase number of lags then this problem can be resolved if there are
not other misspecification in the model.
Sayed Hossain Thanks. What do you mean by 4 lags?
Saud Ahmad commented> I mean to say up till 4th lag there is a serial correlation.
Sayed Hossain That means you are asking us to use 4 lags for the VAR model to
eliminate serial correlation?
Saud Ahmad> No. I mean residuals are correlated with first 4 lagged terms which violate
the assumption of serial independence of residuals. So we have to re-specified our
model but if he increases number of lags then this problem can be resolved if there are
no other misspecification in the model.
Sayed Hossain commented> So this serial correlation problem may be solved on two
ways. 1. We can respecify our original VAR model and run the Bruesch Godfrey serial
correlation LM test again (the way we have done in the above) 2. If serial correlation still
persists, we can use 5 or 6 lags (not in VAR model) while running Bruesch Godfrey
serial correlation LM test so that we can overcome serial correlation. In the above, using
1 to 4 lags in LM test got serial correlation but as soon as we use 5 or 6 lags serial
correlation is removed. It means that LM test is highly sensative with the number of lags.
So you report lag 5 so that your model gets freed from serial correlation.
Sami Ozturk commented> your optimal lag should be greater than 4 since your model
has a serial correlation problem.
2. Is there any serial correlation in the VAR model?
Olasehinde Timilehin There is first order autocorrelation but no second order
Sayed Hossain commented> You better use 2 lags in LM test that you have used. And
report 2 lags so that your model is free from serial correlation.
3. Hello members, after reading I don't get something yet about "Correlation LM
test" for vecm in Eviews, everyone write to increase the numbers of "lags to
include" in this test if it still persists the autocorrelation. So in this vecm
sample, when I write 1 lag or 2 lag show autocorrelation but when I write 3 or 4
shows No Correlation. So what now ? how this influence my VECM with 1 lag?
J.r. Paladines posted it below.
Mine PD commented> After running unrestricted VAR on your variables, select the lag
that shows least autocorrelation by repeating VEC residual autocorrelation test, then
proceed to choose optimal lag.. Include the lag with least autocorrelation in "lags to
include" command then choose the optimal lag where majority of stars* appear.. That is
your optimal lag in unrestricted VAR. For vecm and johansen, you need to choose
optimal lag minus 1 (of unrestricted VAR),
J.r. Paladines commented> very nice, I suspected that, thanks a lot for your answer.
Sayed Hossain commeted> While you use Bruesch Godfrey LM test that you have done
take 3 lags there (Not in VECM model). But first 2 lags has serial correlation. So when
you report show 3 lags that has freed your model from serial correlation.
Meo School of Research
Serial Correlation in the VAR
Hossain Academy Note