“Go Wild for a While!”: A New Test for Forecast Evaluation in Nested Models

In this paper, we present a new asymptotically normal test for out-of-sample evaluation in nested models. Our approach is simple modification of traditional encompassing that commonly known as Clark and West (CW). The key point our strategy to introduce an independent random variable prevents the CW from becoming degenerate under null hypothesis equal predictive ability. Using developed by (1996), show test, impact parameter estimation uncertainty vanishes asymptotically. variety Monte Carlo simulations iterated multi-step-ahead forecasts, evaluated terms size power. These reveal reasonably well-sized, even at long horizons when may severe distortions. power, results were mixed but has edge over approach. Finally, illustrate use with empirical application context commodity currencies literature.