Forecasting accuracy and the choice of first difference or percentage change regression models

Regression models used by forecasters are often formulated in terms of the first differences or percentage changes of the variables. A recently developed maximum likelihood procedure permits the researcher to determine whether the first difference or percentage change model is superior. In this paper the authors apply this new method to several forecasting models from the literature and then determine whether or not the correct functional form improves forecasting accuracy. Results indicate that when either the first difference or percentage change model can be rejected in favor of the other, then superior forecasts can be obtained by using the correct form.