Small-Sample Forecasting Regression or Arima Models?

Univariate models offer the most convenient options for forecasting and ARIMA models are still the most popular among them. The ARIMA modelling, however, requires long data series. This paper shows that a regression model may be estimated with a far greater efficiency in very small samples compared to the corresponding ARIMA model. As a result the larger information set used in a regression model may compensate for the small sample size and improve the forecast efficiency substantially. Three applications which utilize autoregressive forecasts on the exogenous variables highlight the gains in forecast efficiency in small samples from regressions over the ARIMA models.