The importance of large shocks to return predictability

Abstract

Based on the rare disasters literature of Barro and Ursúa (2008), Barro and Ursúa (2009), and Barro and Jin (2011), we show that the predictability of the S&P500 returns increases substantially when we control the regressions for major historical events, such as the Great Depression, World War I, World War II, the oil crisis of 1973-1974, and the subprime mortgage crisis. Controlling for these large shocks, the model with the dividend-earnings ratio as the regressor reaches an in-sample performance with an R2 of 27.6%, while all the other models increase their R2 after correcting for these large shocks. In addition, we show that controlling for major historical events improves the prediction performance, reducing the RSME in all of the 21 models we investigate. We check the robustness of our method by investigating the effects of controlling for the China trade shock of 2001 on the R2 and RMSE of the bias-corrected regressions. Our findings suggest that correcting for these shocks is critical to improve prediction performance.