LAUSR

Abstract Forecasting the direction of stock returns is an important topic in the literature, and it is, to some extent, predictable. In this paper, we consider a new directional forecasting model that applies and extends the time-varying probability density function theory that was proposed by Harvey and Oryshchenko (2012). We capitalize on the relationship between the second order upper partial moment and the directional forecasts, and construct an adjustment mechanism for the forecasting model, which is an original work to the best of our knowledge. The empirical work using data from the Chinese stock market shows that both our forecasting benchmark model and the adjustment mechanism have statistically and economically significant out-of-sample predictive abilities for directional forecasts. Furthermore, the adjustment mechanism provides a great improvement and outperforms the benchmark model in general, and the results of the binary forecasting model are also provided for comparison purposes.