COMPARING PREDICTIVE ACCURACY

We propose and evaluate explicit tests of the null hypothesis of no di fference in the accuracy of two competing forecasts. In contrast to pr eviously developed tests, a wide variety of accuracy measures can be u sed (in particular, the loss function need not be quadratic and need n ot even be symmetric), and forecast errors can be non-Gaussian, nonzer o mean, serially correlated, and contemporaneously correlated. Asympto tic and exact finite-sample tests are proposed, evaluated, and illustr ated.