ARDL serial correlation

[tortoise]

hi,

could anyone tell me how we select the optimal lags for an ARDL specification? I'm talking about the initial stage when we run the F-test to determine the existence of a long-run relationship.. I chose the number of lags arbitraily till I got a good enough F-stat (i.e. long run relationship exists) but when I run the ARDL using the same number of maximum lags, I get serial correlation, and misspecification of the functional form..

thanks in advance !