A novel trading test of predictability

Abstract

Our paper introduces a novel out-of-sample (OOS) test designed to evaluate the predictive ability of financial return forecasts against the commonly used random walk model, a benchmark prevalent in the literature. Our proposed test, labeled WSEP, is based on modifications to the Anatolyev and Gerko (2005) Excess Profitability test (EP) by Pincheira et al. (2022) (SEP). WSEP employs a more conservative trading strategy that assigns weights to forecasts, thereby reducing risk exposure. This approach enhances statistical power due to variance reduction which compensate for the lower returns associated to our strategy. We construct weights for each forecast based on their magnitude, using either an exponential or folded normal cumulative distribution function. The WSEP test offers the advantage of providing an interpretation in terms of profitability, akin to both the EP and SEP tests. We evaluate WSEP size and power via Monte Carlo simulations, employing forecasts constructed from linear regressions estimated by ordinary least squares and random forests. Results demonstrate robust size properties and increased statistical power compared to natural benchmarks. Finally, we present an empirical application based on the commodity-currencies literature.