Matlab arima estimate error

I am working on fitting a time series of data using arima model. I applied the AIC BIC selection process by looping the P and Q order from 0 to 4 for each one. However I found that in some iterations the model fitted is not corresponding to the model that I set. For example when I created a mod of ARIMA(3,0,1) by "mod=arima(3,0,1)" and then I estimate it using "estimate(mod,Y)", it gave me a fitting of ARIMA(0,0,1), which is not the one I expected. So turns out it gives the same AIC and BIC as the one fitted with ARIMA(0,0,1) and hence failed to select the right model.

Your help is appreciated.

Answer 1

[Hang Qian]

I guess that the lower bound constraints become effective under some model specifications. When the estimated model has a reduced order, the higher order coefficients might be zeros. When reduction occurs, I think there will be some warning messages shown on the screen.