Directional Predictability in Asset Returns

The (non-)predictability of asset returns is closely related to the validity of the efficient market hypothesis. While there is a large body of research confirming that the predictability of future asset returns is low, there is some evidence that direction-of-change forecasts are far more accurate. The objective of this research project is to gain further insights into the nature and the economic explanations of directional predictability and to analyze whether the evidence found is in violation of the efficient market hypothesis.

The research plan consists of several sub-projects. First, most studies on directional predictability focus on monthly stock returns and employ either lagged returns or economic variables as predictors in dynamic probit models. However, some more recent publications also find evidence for directional predictability in daily stock returns. The first part of our project, therefore, conducts a comparison of classification methods for daily data. We also test whether the inclusion of realized measures that are derived from high frequency data improves the prediction accuracy. Furthermore, we analyze whether combining sign forecasts and forecasts for the absolute variation of a process lead to more accurate predictions than the direct modelling of the returns.

In the second phase of the project we analyze whether directional predictability is a separate phenomenon or whether these findings can be explained by one of the other known sources of (cross-sectional) predictability such as momentum returns, value effects or firm size effects. If directional predictability turns out to be an independent source of excess returns, we analyze to which extend this is a violation of the efficient market hypothesis.

Project duration: 21 months