Can heteroskedasticity impact forecast accuracy? This paper delves into the effects of heteroskedasticity on statistical tests used to assess the equality of forecast accuracy. The research addresses a common issue in empirical time series analysis. This study introduces two new Diebold–Mariano-type tests designed to enhance accuracy assessment. These tests utilize nonparametric estimation of the loss differential variance function. Through theoretical analysis and Monte Carlo simulations, the researchers demonstrate the potential for improved power compared to the original Diebold–Mariano test. The proposed tests are shown to be particularly effective for a general class of loss differential series. These new methods could improve the reliability of economic forecasts. By applying the new tests to forecasts of changes in the dollar/sterling exchange rate, the study highlights the practical value of the new procedures. The results provide greater evidence of differences in forecast accuracy compared to the original Diebold–Mariano test. These new tests can be valuable tools for practitioners looking for robust methodologies.
Appearing in the Journal of Applied Econometrics, this paper fits well within the journal's scope by offering advanced statistical methods relevant to econometric research. It contributes to the ongoing discussion of forecast accuracy and model evaluation, key areas of interest for applied economists using time series analysis.