Comparing forecast accuracy: A Monte Carlo investigation

The size and power properties of several tests of equal Mean Square Prediction Errors (MSPE) and of Forecast Encompassing (FE) are evaluated, using Monte Carlo simulations, in the context of nested dynamic regression models. The highest size‐adjusted power is achieved by the F‐type test of forecast encompassing proposed by Clark and McCracken (2001); however, the test tends to be slightly oversized when the number of out‐of sample observations is ‘small’ and in cases of (partial) misspecification. The relative performances of the various tests remain broadly unaltered for one‐ and multi‐step‐ahead predictions and when the predictive models are partially misspecified. Interestingly, the presence of highly persistent regressors leads to a loss of power of the tests, but their size properties remain nearly unaffected. An empirical example compares the performances of models for short term predictions of Italian GDP.