Analyzing several developed and emerging international markets, I test the ability of global, regional, and local models to explain a large set of 134 cross-sectional anomalies. My main finding is that both global and regional factor models create substantially larger average absolute alphas than local factor models. Annual (absolute) anomaly portfolio alphas are on average 1.7 and 1.1 percentage points higher, respectively, with global and regional than with local factor models. Even for the most recent period, there is no evidence of a catch-up of global and regional factor models. There is substantial potential for international diversification of anomaly strategies.

This paper examines the estimation of global and local betas for a large set of Developed and Emerging international markets. Estimators based on daily data clearly outperform those based on monthly or quarterly data. For global and local market betas, the optimal window length is at roughly 24 and 12 months, respectively, for most Developed Markets. It tends to be somewhat longer for Emerging Markets. The best estimators include a double-shrinkage, a long memory (FI), and a simple combination approach. For hedging the market risk exposure in anomaly portfolios, the FI and combination estimators also perform overall best.

A stylized theoretical model with stochastic volatility suggests the existence of a trade-off between returns and volatility-of-volatility. Using the VVIX, a measure of the option-implied volatility of the volatility index, we confirm this prediction and detect that time-varying aggregate volatility-of-volatility commands an economically substantial and statistically significant negative risk premium. We find that a two-standard-deviation increase in aggregate volatility-of-volatility factor loadings is associated with a decrease in average annual returns of about 11%. These results are robust to controlling for aggregate volatility, jump risk, and several other characteristics and factor sensitivities, as well as various additional tests.

We conduct a comprehensive comparison of market beta estimation techniques. We study the performance of several historical, time-series model, and option-implied estimators for estimating realized market beta. Thereby, we find the hybrid methodology of Buss and Vilkov to consistently outperform all other approaches. In addition, all other approaches, including fully implied and dynamic conditional beta, based on generalized autoregressive conditional heteroskedasticity (GARCH) models, are dominated by a simple beta estimate based on historical (co-)variances and an approach based on the Kalman filter. Our conclusions remain unchanged after performing several robustness checks.

The standard full-sample time-series asset pricing test suffers from poor statistical properties, look-ahead bias, constant-beta assumptions, and rejects models when average factor returns deviate from risk premia. We therefore confront prominent equity pricing models with the classical Fama and MacBeth (1973) cross-sectional test. For all models, we uncover three main findings: (i) the intercept coefficients are economically large and highly statistically significant; (ii) cross-sectional factor risk premium estimates are generally far below the average factor excess returns; and (iii) they are usually not statistically significant. Overall, all new factor models are inconsistent with no-arbitrage pricing and cannot accurately explain the cross-section of stock returns.

Factors related to carry, duration, equity momentum, and the term structure are the most important risk factors in corporate bond markets. From a large set of factor candidates, we condense an optimal model with a two-step approach. First, we filter out factors that do not systematically move bond prices. Second, we use a Bayesian model selection approach to determine the optimal, parsimonious model. Many prominent factors do not move prices or are redundant. We document the new model’s good performance compared to that of existing models in time-series and cross-sectional tests and analyze the economic drivers of the factors.

A stock’s exposure to systematic risk factors is surrounded by substantial uncertainty. This beta uncertainty is both economically and statistically significantly priced in the cross-section of stock returns. Stocks with high beta uncertainty substantially underperform those with low beta uncertainty: a two-standard-deviation increase in the measure decreases average annual returns by 9.7%. These results cannot be explained by previously discovered determinants of cross-sectional stock returns. Aggregate beta uncertainty negatively predicts market excess returns in the short and medium term. We find supporting evidence for a mispricing explanation of the beta uncertainty premium.

In this paper, we study the asset pricing implications of persistence in the risk-neutral return distribution’s central moments. We detect a both economically and statistically significant premium of stocks with low over stocks with high such persistence. Annual value-weighted excess (risk-adjusted) returns are 4.38% (3.06%). These results cannot be explained by factors and characteristics documented in the previous literature. Furthermore, it is not the persistence of only one of the individual distributional moments but rather the joint persistence in all central moments of the risk-neutral distribution that is priced.

Machine learning-based stock market beta estimators outperform established benchmark models both statistically and economically. Analyzing the predictability of time-varying market betas of U.S. stocks, we document that machine learning-based estimators produce the lowest forecast and hedging errors. They also help to create better market-neutral anomaly strategies and minimum variance portfolios. Among the various techniques, random forests perform the best overall. Model complexity is highly time-varying. Historical stock market betas, turnover, and size are the most important predictors. Compared to linear regressions, allowing for nonlinearity and interactions significantly improves predictive performance.