Parameter estimation and tests of equal forecast accuracy between non-nested models

This paper reviews recent work regarding the effects of estimated model parameters on tests of equal forecast accuracy between non-nested models. When accounting for these effects using standard methods, strong assumptions about the distribution of the observables may be needed. We provide a new method that enables us to avoid making those assumptions. Monte Carlo evidence indicates that the new method can lead to improvements in the size and power of the test across a range of sample sizes.