Testing the appropriate lag length in a VAR

[Szi]

Great, thanks to you both :)

[cychang]

I' m using EViews7 now. Before I fill the number of lags in the lag specification column, there is a default value in that column. And when the sample is large, the default value will be large, such as 8. When the sample is small, the default value will be small, such as 3. I'm wondering that how EViews decide the default value of the maximum lags to include?

thank you very much.
Andy

[EViews Gareth]

Obviously it is somewhat arbitrary, but I believe we use Min(8, (Obs - # of Endogenous)/10)

[gobbble]

For the Johansen Co-integration test, I determined the optimal lag length by creating a VAR in levels.

I found out that there is no co-integration between the variables and want to run a granger casuality test.

Now, when I want to rung a Granger Causality test I create VAR with first differenced logs. My question here is:

Do I have to determine the optimal lag length again now with first differenced variables or can I just use the optimal lag lengt dertermined before for the co-integration test with variables at levels?

Thanks!