Lower AIK or lower Forecast Error?

[hpramos]

Hi everyone,

I'm ARIMA modelling some series and I've found the following trade-off: some models give me a lower information criterion (ex: AIK) but a high forecasting error (Mean Abs. Percent Error). In oppose to that, some models give me a higher information criterion but a low forecasting error.

Which one should I choose? I'll appreciate if you could explain why.

Thanks in advance,