Multi-period forecasting using different models for different horizons: an application to US economic time series data

Most time series forecasters have used a single forecasting model at multiple forecast horizons. This practice is still largely in place, despite various suggestions that using different forecasting models for different horizons could improve forecast accuracy. Using monthly data on various US economic time series, we investigate empirically whether the procedure of using a multi-period-ahead forecasting autoregressive model selected separately for each horizon (instead of using a single autoregressive model for all horizons) may or may not improve forecast accuracy for economic time series. We find that the forecast performance of the procedure appears to depend on, among other things, optimal order selection criteria, forecast origins, forecast horizons, and the time series to be forecast.