skip to content

Prof. Dr. Dieter Hess (University of Cologne): 

"The performance of mechanical earnings forecasts"
(with Ashok Kaul and Martin Meuter)

Wednesday, June 10, 02:00pm - 04:00pm (WiSo Building, Room 410) 


Abstract: "We systematically evaluate the performance of cross-sectional earnings forecasting models for a large sample of firms between 1968 and 2014. Our performance evaluation is twofold: First, we analyze the difference in forecasting accuracy between the model proposed by Hou, van Dijk and Zhang (HVZ, 2012) and the allegedly superior earnings persistence (EP) and residual income (RI) models (Li and Mohanram, 2014) in a two dimensional framework. The first dimension attributes predictive ability to the choice of explanatory variables, whereas the second dimension captures the effect of forecasting different earnings specifications, i.e. firm-level earnings, earnings per share (EPS) and EPS excluding special items. Most importantly, we find that previously reported performance differences disappear, when we estimate the models at identical earnings definitions. However, we document that the predictive performance increases substantially (by about 10%), if we predict earnings per share, rather than firm-level earnings. Excluding special items leads to another 10% increase in forecasting accuracy. We provide evidence that the advantage of EPS vis-à-vis firm-level estimations is due to reduced heteroscedasticity, whereas the additional improvement stems from the exclusion of unpredictable items. While the better predictability of EPS before special items indicates a higher persistence in “street earnings”, this finding does not necessarily imply higher capital market relevance, too. Hence, in a second step, we measure two commonly used proxies for value relevance, i.e. earnings response coefficients (ERC) and realized return spreads, for each earnings definition. However, we do not find that lower forecast errors translate into higher ERCs and return spreads.