Parameter Uncertainty, Financial Turbulence and Aggregate Stock Returns
DGF 2017, FMA Europe 2017, SFA 2015, WFC 2015 // (SSRN)
Abstract: In this paper, we develop a novel, intuitive and objective measure of time-varying parameter uncertainty (PU) based on a simple statistical test. Investors who are averse to parameter uncertainty will react to elevated levels of PU by withdrawing from the market and causing prices to fall, a behavior that is well described by the model of portfolio selection with parameter uncertainty of Garlappi et al. (2007). We show that this model in combination with our measure, outperforms all other tested variables including the strongest known predictor to date. Additionally, it is the only predictor that fulfills all criteria generally expected from a stable predictor of the equity premium. All our results are statistically and economically significant and robust to a large variety of different specifications.
Higher Moments Matter! Cross-Sectional (higher) Moments and the Predictability of Stock Returns
(with Lars Kaiser)
SGF 2017 // (SSRN)
Abstract: In this paper we investigate the predictive power of cross-sectional volatility, skewness and kurtosis for future stock returns. Adding to the work of Maio (2016), who finds cross-sectional volatility to forecast a decline in the equity premium with high predictive power in-sample as well as out-of-sample, we highlight the additional role of cross-sectional skewness and cross-sectional kurtosis. Applying a principal component approach, we show that cross-sectional higher moments add to the predictive quality of cross-sectional volatility by stabilizing the predictive performance and yielding a positive trend in in-sample and out-of-sample predictive quality since the burst of the dot-com bubble. In particular, we observe cross-sectional skewness to span the predictive quality of cross-sectional volatility over short-forecasting horizons, whereas cross-sectional kurtosis significantly contributes to long-horizon forecasting of 12 months and above. Results are both statistically and economically significant.