Lag selection in dynamic models with stepls

[m.j.greenwood]

Hi all,

I have a question about the stepls function. I'm using the unidirectional backward version to select the dynamic terms in an ARDL model with a 5% p-value stopping criterion. In some cases, I find that the procedure includes regressors that are not actually significant at the 5% level even though they are in the set of search variables. You can see this in the attached example for the variable DLEP_US(-1). Can anyone explain how this is possible and how I can ensure that the variables selected for inclusion out of the set of search variables are all significant at the 5% level?

Thanks for your help,
Matt

[EViews Gareth]

The User Guide has more details, but the final p-values that are shown in the step-wise output may not match the p-values that were used during the step-wise routine.