On choosing the optimal level of significance for the Durbin-Watson test and the Bayesian alternative