[Gretl-users] VAR in connection to AIC, BIC ...

[Allin Cottrell]

On Mon, 4 Aug 2008, Davor Horvatic wrote:

> OK. My guess was that var test goes like this get max lag 
> compute AIC etc. for each lag with the set you get minus lag 
> order. For example take set with 100 members. lag = 1 you have 
> 99 etc., so the method is always work with the sample number of 
> points minus max lag ?
> 
> This is why I asked because it did not soud right to me ... 
> Because I thought that AIC, BIC, HQC will have same value for 
> lag = 1 independet of maxlag. 

As Sven said, the comparison of AIC (and other criteria), for 
different lag lengths, will not be valid if a different sample 
range is used in each case.  So yes, the sample is of length 
(T - maxlag) for all calculations in any given comparison.

> As I said it is in the manual. My question was do I have any 
> freedom beside manual OLS and reading off AIC, BIC, HQC to see 
> correct lag for ADF. The method in manual for --test-down and it 
> is t-statiscis.

What you see is what you get ;-) Right now, there's no automatic 
option for test-down in the ADF test other than using the 
t-statistic on the last lag.  It wouldn't be too difficult to 
write a function that did it some other way, although use of AIC 
would not be very advisable since it tends to favour overly large 
models.

Allin Cottrell