Enhancing Forecast Accuracy by Using Long Estimation Periods

A tradeoff between forecast accuracy and the length of an estimation period always exists in forecasting. Longer estimation periods are argued to be less efficient, however, using the forecast encompassing and accuracy test, this study discusses the importance of considering the overall usefulness of information in the in-sample period. The empirical results demonstrate that forecasts using the correct model have reduced measurement loss and the mean of forecast errors decrease with an increase in in-sample period. Moreover, for the forecast accuracy and encompassing tests, reducing the use of observations in making estimates leads to the wrong model being easily accepted. Additionally, these analytical results are also consistent with the application in hedge performance, that is, the hedge effectiveness is optimized when the estimation period is longest, particularly under the recursive scheme