Re: [Gretl-users] VAR lag selection
Hi,
there are two issues here: First, in your example of course it is
understandable that if you don't allow the fourth lag, the choice will
be different if in the "unrestricted" setup the optimal lag order is four.
Second, the max lag order may change your effective sample (depending on
whether your dataset is subsampled). It can happen that the initial
observations make a difference.
cheers,
sven
Am 22.10.2010 08:14, schrieb Talha Yalta:
I am not sure if this question makes sense but is it normal that
AIC,
BIC and HQC values change when a different "maximum lag" selected in
VAR lag selection. I mean, if these statistics change based on the
initial choice of the maximum lags to be tested, then we also need a
test for the maximum number of lags to be tested, no?
Here is an example of the potential inconsistency:
VAR system, maximum lag order 4
The asterisks below indicate the best (that is, minimized) values
of the respective information criteria, AIC = Akaike criterion,
BIC = Schwarz Bayesian criterion and HQC = Hannan-Quinn criterion.
lags loglik p(LR) AIC BIC HQC
1 145.63131 -10.924461 -9.937075 -10.676137
2 159.38080 0.03626 -10.728765 -8.951470 -10.281780
3 182.94995 0.00006 -11.386952 -8.819748 -10.741307
4 224.97540 0.00000 -13.650034* -10.292921* -12.805730*
VAR system, maximum lag order 3
The asterisks below indicate the best (that is, minimized) values
of the respective information criteria, AIC = Akaike criterion,
BIC = Schwarz Bayesian criterion and HQC = Hannan-Quinn criterion.
lags loglik p(LR) AIC BIC HQC
1 149.21357 -10.767797 -9.786086* -10.507349*
2 163.67384 0.02448 -10.639487 -8.872406 -10.170680
3 184.17688 0.00055 -11.014740* -8.462290 -10.337574
Cheers
Talha