A1662
Title: Robust inference in predictive regressions for stock returns
Authors: Jenny Hau-Ruess - Technical University of Munich (Germany)
Aleksey Min - Technical University of Munich (Germany)
Rustam Ibragimov - Imperial College Business School, CEBA and New Economic School (United Kingdom) [presenting]
Abstract: The focus is on a detailed robust analysis of predictability of stock returns. While a prior study suggests that no regression model can forecast the equity premium more accurately than its historical average, another study concludes that there are, in fact, such regressions if one restricts the model appropriately. The contribution to this discussion is by applying the general heterogeneity and autocorrelation robust inference approaches developed in other studies. These approaches are based on t-statistics computed using group estimates of regression parameters dealt with. They have appealing finite sample properties in many heterogeneity and dependence settings observed in practice as compared to, e.g., widely used HAC inference methods. The approaches are simple to use and do not require the estimation of consistent standard errors of regression estimators. The robust t-statistic methods are applied to assess the significance of the regressions slope coefficients, and the conclusions are compared with the results obtained from widely used alternative methods. It is examined whether a regression forecast is significantly more accurate as compared to the simple historical average forecast. Results demonstrate that even a critical investor can find a regression model that predicts the equity premium more accurately than the historical average based on data up to the year 2005.
