Cochrane Orcutt - specify the lags

[chts0002]

Hello everyone,

I run a multipple autoregressive model including 40 observations and I want to fix the problem of autocorrelation through the Cochrane - Orcutt Procedure. I have detected serial correlation both the Correlogram - Q statistic and the Breusch - Godfrey LM test.
When I am doing the correlogram I put the number 15 in the lags specification and only in third order autocorrelation coefficient is insifnificant (p-value is really low - 0.012). If I do the Breusch Godfrey test I put 3 lags as lag specification test indicating corrrelation
So, Should I include AR(3) as an independent variable in the model in order to fix the problem or AR(1)???

Thanks a lot!!