Title: Unfolded skewness and kurtosis timings in out-of-sample density forecasts of financial returns
Authors: Xiaochun Liu - University of Alabama (United States) [presenting]
Abstract: The aim is to evaluate both statistical significance and economic relevance of distributional timings in out-of-sample density forecasts. We estimate a wide range of specifications with dynamic higher moments and conduct a variety of statistical tests. The test results consistently show that modeling time-varying skewness and kurtosis significantly improves both adequacy and accuracy of density forecasts. In the utility-based comparisons, we find that switching to time-varying skewness- and kurtosis-based portfolios from constant-higher-moments-based portfolios yields a gain of an extra 286 and 395 basis points on average per year, respectively. Moreover, an investor is willing to pay 90 and 140 basis points per year to acquire skewness and kurtosis information beyond volatility information. Among the competing models, the unfolded GARCH model, which decomposes returns into the product of their absolute values and signs, presents not only the strongest statistical significance for the left tail of financial returns, but yields the highest gains of about 8.7\% and 11.9\% on average for an investor unfolding skewness and kurtosis timings.