There are several criterion for choosing the optimal laglength in a time serie: AIC : Akaike information criterion ; BIC : Schwartcz information criterion ; HQ : Hannan-Quinn criterion ; RMSE : Root Mean Square Error ; MAE : Mean Absolute Error; BP : Bias proportion ; LIK : Log-Likelihood. The discrimination function differs from one to another criterion.
We can therefore obtain conflicting results. How to choose the optimal laglength? Are there tests that compare the criteria between them?