Abstract
This article examines the cross sectional predictability of stock returns within the framework of time varying risk premia and asymmetric risk. In bad times, small, value and cyclical stocks are riskier than big, growth and noncyclical stocks, thereby explaining the value, size and cyclical premia. In contrast, in good times, value, big and noncyclical stocks generate higher average returns than do growth, small and cyclical stocks, despite the fact that they are not riskier. Furthermore, empirical tests of macroeconomic models highlight that average returns are commensurate with risk only in bad times. These results are robust to different measures of risk and different sets of test assets.
Notes
1 See, e.g. Fama and French (Citation1989).
2 See, e.g. Braun et al. (Citation1995), Ang and Chen (Citation2002).
3 Negative dispersion of risk means a negative difference between the betas of value stocks minus the betas of growth stocks.
4 See, e.g. Lakonishok et al. (Citation1994), Jagannathan and Wang (Citation2007).
5 Lakonishok et al. (Citation1994), Ang and Chen (Citation2002) and Ang et al. (Citation2006).
6 Note that Petkova and Zhang (Citation2005) proceed further by estimating the beta premium sensitivity using the generalized method of moment, in order to show that asymmetric beta dispersions across economic cycle are statistically significant. This article does not challenge this result, but focuses mainly on the joint pattern of average returns and conditional betas.
7 The results are similar if we implement the Exponential General Autoregressive Conditional Heteroscedasticity (EGARCH) model of Nelson (Citation1991).
10 A growing number of macroeconomic models have recently been proposed by Vassalou (Citation2003), Jagannathan and Wang (Citation2007), etc. These models are not implemented here because they must be tested on quarterly frequency. In addition, data on many macroeconomic variables such as consumption are not available for the 1927 to 2006 sample.