Dear members of the Gretl-Users list,

 

i'm currently working on my master thesis and have a question about the function of the AIC in the VAR lag selection.

 

In a part of my thesis i'm analysing two (quite short) time series of 12 observations each (yearly data) via a VAR model and impulse response functions. The VAR model is estimated with a constant, a time trend and the robust standard error HAC. The maximum lag to be tested down is determined by Schwert (1989, p. 151):

 

l4=int{4(T/100)^(1/4)}

 

which gives us a maximum lag of 2 for the lag selection. For this maximum lag Gretl selects via an AIC of 15,741186 a lag of 2 (for a lag of 1 the AIC is 17,62516). The problem i'm facing is, that i can't reproduce this AIC of 15,74... and therefore don't know the function in order to write it down in my thesis.

 

What is the function of the AIC that Gretl uses in its VAR lag selection? And how does Gretl compute the determinant of the variance-covariance matrix? My determinant is 3575.4722 and Gretl's is 2133.3137.

 

If tried two functions (so far) for the multivariate case of the AIC including the determinant of the variance covariance matrix but none of the seem to work (the Gretl User's Guide from Dec. 2017 gave me no clues). I've attached a file to this email (in different formats of which i hope one fits your requirements) including the residuals of my VAR model, the variance covariance matrix and my attempts of the AIC re-computation. I also added a Gretl-file with the data i'm using (for some reason my pc couldn't save a command record :/  ).

 

 

Best regards,

Marvin Berndt

 

 

P.S.: I'm using the Gretl version '2017d' for MS Windows x64.