Nonlinearity in the Cross-Section of Stock Returns: Evidence from China

Abstract

We study which characteristics provide independent information for the cross-section of expected returns in the Chinese stock market based on nonlinear predictive functions. Using 100 commonly explored stock characteristics from January 2000 to December 2019, we identify 15 to 19 characteristics that provide incremental predictive information. We find significant alphas based on the most up-to-date four-factor model of Liu et al. (2019) when exploring these characteristics jointly using nonlinear predictive models. A long-short spread portfolio based on out-of-sample predicted returns by a nonlinear model delivers a higher Sharpe ratio than that by a linear model. We document more supportive evidence for the nonlinear model after exploring interaction effects with firm size, earnings-to-price ratio, turnover, state dependency of predictors, and various methods of predictive information aggregation, such as forecast combination, principle component regression, and partial least squares.

Keywords: Cross-sectional return predictability; firm characteristics; adaptive group LASSO; information aggregation