We address the problem of which subset of time series to select among a given set in order to forecast another series. The forecasts are evaluated in terms of Mean Squared Error. We propose a family of criteria for which weak and strong consistency results are proved. The criteria are compared to some well-known hypothesis tests by means of Monte Carlo experimentation and a real-data example.
Palabras clave / Key words: forecasting, model selection, VARMA modelsDetermining the MSE-optimal cross section to forecast (Pdf 316 KB)