Title: Performance of predictive regressions when models are uncertain
Authors: Benjamin Hillmann - Kiel University (Germany) [presenting]
Abstract: In spite of well-developed theory, predictability of stock returns using fundamental variables is difficult to establish in practice. One possible explanation for this fact is potential nonlinearity of the predictive relation; another possible explanation is of a rather statistical nature, namely the uncertainty associated with estimation, and in particular model selection. The effectivity of various forecasting procedures is explored taking these aspects into account. In an out-of-sample forecasting exercise, proper accounting for nonlinearity and model uncertainty is shown to somewhat improve evidence on predictability.