Finite-sample properties of tests for forecast equivalence

This paper uses Monte Carlo experiments to examine the small-sample properties of some commonly used tests of equal forecast accuracy. The study pays particular attention to test power, evaluated using both asymptotic and empirical critical values. In addition to evaluating different tests, this paper evaluates the performance of different methods of determining the bandwidth used in computing autocorrelation-consistent test statistics. The simulation results show that tests of equal forecast accuracy have somewhat inflated size and modest or even low power. Moreover, the performances of the different tests and the bandwidth selection criteria are broadly similar.