Stablility and serial correlation test for vector autoregresive models.

One of the key features in vector autoregressive models is choosing lag lengths. A common way to do this is to use a criteria such as AIC or BIC. Often these disagree on which lag to choose. The author I am reading on VAR models suggest "stability test" and "serial correlation" test to chose among lags when this occurs. I am not sure what these are. Serial correlation might be the Box Ljung test although I am not sure.

Stability seems to be related to non-Stationarity, although this is not certain. I don't know how you could know a specific lag is right and another wrong based on a test of Stationarity.

Anyone who could comment on this I would appreciate it.