Equity Style Payoffs And Stock Return Predictability: Evidence From The South African JSE Securities Exchange

Abstract

This paper undertakes to examine the structure of the payoffs to style attributes on the JSE Securities Exchange over the period from 1997 to 2007. Adopting the methodology of Haugen and Baker (1996), two expected return multifactor models are constructed to predict forward monthly returns of sample stocks using style attributes as model inputs. The Grinold model has an objective of selecting model inputs that will maximize the in-sample Grinold (1989) information ratio while the QH model has an objective of maximizing the in-sample Qian and Hua (2003) information ratio. The out-of-sample predictions of the models and their corresponding model inputs are evaluated over the period from 2002 to 2007. The permutations of the style attributes for the models are updated every 12 months based on training over the prior 60 months under the stepwise variable selection procedure proposed by Van Rensburg and Robertson (2003). The examination of the style attributes selected by the respective models reveal that most of the style attributes that exhibit significant univariate mean payoffs in our prior study are selected by the expected return multifactor models in this research. In addition, the value attributes with the most consistent univariate payoffs are found to be the most important contributors to equity return forecasting. It is also found that models that account for multicollinearity tend to outperform the models that fail to address this issue over the out-of-sample period. Overall, the QH model outperforms the Grinold model in terms of both information ratios over the out-of-sample period. The out-of-sample performance scores of the respective models are consistent with their corresponding in-sample performance scores, which suggest that successful expected return multifactor models, based on style attributes, can be developed to forecast ex-ante JSE equity returns.