There is no standard economic forecasting procedure that systematically outperforms the others at all horizons and with any dataset. A common way to proceed, in many contexts, is to choose the best model within a family based on a fitting criteria, and then forecast. I compare the out-of-sample performance of a large number of autoregressive integrated moving average (ARIMA) models with some variations, chosen by three commonly used information criteria for model building: Akaike, Schwarz, and Hannan-Quinn. I perform this exercise to identify how to achieve the smallest root mean squared forecast error with models based on information criteria. I use the Chilean GDP dataset, estimating with a rolling window sample to generate one- to four-step ahead forecasts. Also, I examine the role of seasonal adjustment and the Easter effect on out-of-sample performance. After the estimation of more than 20 million models, the results show that Akaike and Schwarz are better criteria for forecasting purposes where the traditional ARMA specification is preferred. Accounting for the Easter effect improves the forecast accuracy only with seasonally adjusted data, and second-order stationarity is best.