It's mainly a discussion topic... the lag selection isn't a problem when i check specific variables, but when i want to check it within a repeated process with simulated model i find the lag selection criteria and the lag they propose unacceptable. That is by automatically using any of these criteria i don't make sure that the selected lag length in the ADF has "white" residuals.

Is there any other criterion i'm not aware of or should a new criterion be produced? One that mainly checks the whiteness of the residuals and at a second stage how well that lag selection fits the dataset??