How do I test for maxlags for AIC and BIC when I have negative autcorrelation? Would the negative autocorrelation have any effect on my test for maxlags?

I tested the maxlag and found that higher the lags, the lower the AIC, which assumes a better AIC for higher lags. Does this make sense?

  • $\begingroup$ What is your setting? What is maxlag in that setting? What kind of model do you have that allows you to test the maxlag? $\endgroup$ Commented Nov 15, 2016 at 11:43
  • $\begingroup$ I use the vector autoregressive model using data for inflation and gdp. When I increase lags in the model, AIC and BIC lowers. $\endgroup$ Commented Nov 15, 2016 at 11:53
  • $\begingroup$ Then I suppose you are not testing for maxlag, you are just selecting it upfront and then seeing which lag from zero (or one) up to maxlag delivers a model with the lowest AIC or BIC. But what you observe should not be happening. At least not for really high maxlag. Perhaps there is an error in your implementation. $\endgroup$ Commented Nov 15, 2016 at 12:49
  • $\begingroup$ @RichardHardy How would you test for maxlag? I tried testing for all of my variables (transformed and non-transformed) such as first difference, deltas and taking their logs, which all produce the same result, decreasing AIC or BIC for increasing lags. $\endgroup$ Commented Nov 15, 2016 at 22:20
  • $\begingroup$ If you insist on testing, then what is your null hypothesis? $\endgroup$ Commented Nov 20, 2016 at 9:53


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.