Choosing the Best Forecasting Model

Refitting and Reevaluating Models

If you fit the ARIMA(0,1,0)(0,1,0)s and Airline models again in the same way as before, they will be added to the model list, with the same names but with different values of the model selection criterion. Parameter estimates will be different, due to the new fit range, and statistics of fit will be different, due to the new evaluation range.

For this exercise, instead of specifying the models again, refit the existing models by selecting Edit from the menu bar, then Refit Models and All Models. After the models have been refit, you should see the same two models listed in the table but with slightly different values for the selection criterion. The ARIMA (0,1,0)(0,1,0)s and Airline models have now been fit to the MASONRY series using data from June, 1978 to July, 1982, since this is the period of fit you specified. The statistics of fit have been computed for the period of evaluation, which was the same as the period of fit. If you had specified a period of evaluation different from the period of fit, the statistics would have been computed accordingly.

In practice, another common reason for refitting models is the availability of new data. For example, when data for a new month become available for a monthly series, you might add them to the input data set, then invoke the forecasting system, open the project containing models fit previously, and refit the models prior to generating new forecasts. Unless you specify the period of fit and period of evaluation in the Time Ranges Specification window, they default to the full data range of the series found in the input data set at the time of refitting.

If you prefer to apply previously fit models to revised data without refitting, use Reevaluate Models instead of Refit Models. This recomputes the statistics of fit using the current evaluation range, but does not reestimate the model parameters.

Chapter Contents
Previous
Next
Top