Abstract: We develop an approach to determine whether a particular predictor represents a proxy for fundamental risk. We build on the assumption that risk-based predictors should be linked to new information about economic conditions. We show that most predictors forecast returns on either days with macroeconomic announcements or the remaining days, indicating that sources of return predictability differ across predictors: few are driven by fundamental risk; most have other origins. We show that Shiller’s excess volatility is confined to non-announcement days, suggesting that the ability to forecast stock market’s noise component underlies much of the predictability documented in the literature.
Abstract: A linear factor model is closer to the mean-variance frontier the larger the Sharpe ratio of its mean-variance efficient portfolio. We show that a nonlinear factor model is closer to the mean-variance frontier the larger the Sharpe ratio of its mimicking portfolio. Based on this metric and economically meaningful nonlinearities, we consider a large number of factors to document that: (i) nonlinear factor models significantly outperform their linear counterparts, where the Sharpe ratio metric can even double; (ii) the preferred factor model can change under nonlinearities, depending on the test assets; and (iii) the stochastic discount factor is more likely to be sparse on observable factors when nonlinearities are contemplated.
Discussant: Alain-Philippe Fortin, University of Geneva
