We show that returns to value strategies in individual equities, industries, commodities, currencies, global government bonds, and global stock indexes are predictable in the time series by their respective value spreads. In all these asset classes, expected value returns vary by at least as much as their unconditional level. A single common component of the value spreads captures about two-thirds of value return predictability and the remainder is asset-class-specifc. We argue that common variation in value premia is consistent with rationally time-varying expected returns, because (i) common value is closely associated with standard proxies for risk premia, such as the dividend yield, intermediary leverage and illiquidity, and (ii) value premia are globally high in bad times.

The flow of investment capital into the commodity futures market dramatically increased around 2004, and this event is referred to as the financialization of commodity markets. We study how this phenomenon has affected risk premia in this asset class by examining how returns to popular commodity futures strategies have evolved. We find that about 80% of commodity futures strategies that earned a statistically significant risk premia pre-financialization have earned an average of a zero return since. Using a six latent factor asset pricing model, we show that this deterioration in average returns can be wholly explained by an adverse change in the risk premia to systematically priced variation in the cross-section of commodity futures. In robustness tests, we show that the publication of commodity strategies in the academic literature can only explain about 25% of the 58 bps per month reduction that commodity futures strategies have experienced post-financialization. .

Despite the weak theoretical assumptions needed to guarantee the existence of a factor model that prices the cross-section of stock returns we show in a meta-analysis of fifteen state-of-the-art models that none of these models can price the mean-variance efficient portfolio of the other fourteen. No one model effectively describes the cross-section of US equities. Our results highlight the need to benchmark new factor models against each other, rather than anomaly portfolios or classic double sorted portfolios. When markets are incomplete, an infinite number of stochastic discount factors price assets, but empirically, we have yet to find just one. We refer to this as the \textit{factor model failure puzzle}. We present a theoretical model with many weak anomaly trading signals and show that slight methodological deviations lead to different factor models that result in different and flawed characterizations of the risk-return trade-off as observed empirically.

Previous research finds that machine learning methods predict short-term return variation in the cross-section of stocks, even when these methods do not impose strict economic restrictions. However, without such restrictions, the predictions from the models fail to generalize in a number of important ways, such as predicting time-series variation in market and long-short characteristic sorted portfolio returns across multiple horizons. I show this shortfall can be remedied by imposing economic restrictions in the architectural design of a neural network model and provide recommendations for using machine learning methods in asset pricing. Additionally, I shed light on the intermediate and long-run dynamics of the return forecasts generated by this model.

For many characteristics, like book-to-market, the persistence of return predictability does not match the persistence of the characteristic. Consequently, large alphas exist between new and old sorts, where new (old) sorts capture the return of a characteristic-sorted portfolio immediately (longer) after formation. These alphas (i) translate into large improvements in Sharpe ratio, (ii) are not captured by benchmark asset pricing models, and (iii) are linked to the return differential between new and old stocks. Since portfolios of new and old stocks are characteristic-neutral, we conclude that explanations of the cross-section based on recent observations of characteristics (and factors derived therefrom) are incomplete..