Lag length in panel data

[Mario Ferri]

Dear Statalisters,
I have an unbalanced panel and was looking for a way to determine optimal lag length in panels.

I run Fisher-type test, Augmented DF, ADF

xtunitroot fisher varname, dfuller lags(0)

and it works for unit root test.

My question is how do I get the optimal lag length in panel?
I know how to do that in the VARs. What about panels, thought?
I will run a normal panel, if that is of interest

Thank you.
Mario Ferri

[Tasfaye Fikadu]

In the panel data set what comes before unit root test is cross-sectional dependence test; this is because the distinction between first-generation ( LLC and IPS) and second-generation is based upon cross-sectional dependence and you will use the latter in case the dataset has cross-sectional dependence. The optimal lag determination I found from different stata forum is :

forval i=1/11{

rdl Y X1 X2 X3 X4 if (country ==`i'), maxlags(2 2 2 2 2 )
matrix list e (lags)
di
}


where 11 is no. of panel unit; country: panel unit; 2, 2, 2, 2, 2, maximum lag as much as no. of variables

[Shashank Srivastav]

You may use pvarsoc command. It can be installed by using the command search pvarsoc. It provides optimal lag length using J-statistic and AIC criteria.