Number of observations used for ARIMA modeling

Question

Our professor keeps on writing in his slides that when you test different ARIMA models on your time series, one always has to keep T fixed.

I assume he is talking about fitting your model using the same number of observations. If you try a model with a higher lag, you loose one observation to fit the data on. Is this correct?

Is this why he writes restrict your sample? We have 214 observations, should I for example to be on the safe side only use 205 observations? And how does Eviews know to always use the same number of obersations?

Ref screenshot:

Thanks in advance.

Answer 1

When you test ARIMA models with different number of lags, you must make sure you always have the same data available to calculate the coefficients. If you don't specify the sample to be used in Eviews, you will use one observation less for estimating an AR(5) model than for estimating an AR(4) model, and hence the coefficients, and more important the significance of your coefficients, are not calculated on the same number of observations.

This should be avoided, so think first about the maximum number of lags you want to include in the model and set the sample size accordingly.