Cross-sectional Return Predictability Research Articles

Machine learning and the cross-section of cryptocurrency returns

We employ a repertoire of machine learning models to investigate the cross-sectional return predictability in cryptocurrency markets. While all methods generate substantial economic gains—unlike in other asset classes—the benefits from model complexity are limited. Return predictability derives mainly from a handful of simple characteristics, such as market price, past alpha, illiquidity, and momentum. Contrary to the stock market, abnormal returns in cryptocurrencies originate from the long leg of the trade and persist over time. Furthermore, despite high portfolio turnover, most machine learning strategies remain profitable after trading costs. However, alphas are concentrated in hard-to-trade assets and critically depend on harvesting extreme returns on small, illiquid, and volatile coins.

Leveraged trading and stock returns: Evidence from international stock markets

Are margin traders as well-informed as short sellers when it comes to leveraged investing? Our paper, utilizing a unique dataset on stock-level short selling and margin trading from three international stock markets, reveals that while short selling has cross-sectional return predictability, margin trading does not. In comparison to short selling, margin-trading activities demonstrate a stronger correlation across stocks and weakly predict firm fundamentals. This suggests that margin traders are less likely to possess a firm-specific information advantage. Our findings at the investor account level also indicate that margin traders are less sophisticated than short sellers.

Quants and market anomalies

Sell-side quantitative equity research analysts (Quants) conduct econometric analyses of stock returns to uncover market anomalies and assist equity analysts and institutional clients with stock selection. We present novel evidence that establishes their role in helping analysts and mutual fund clients discover market anomalies and capital markets evolve toward greater pricing efficiency. Specifically, we find that analysts and mutual fund clients with greater access to Quants make recommendations and trades that reveal greater knowledge of anomalous cross-sectional return predictability. More importantly, cross-sectional return predictability is weaker in stocks that have higher coverage (ownership) by analysts (mutual fund clients) with access to Quants, and strengthens when quasi-exogenous brokerage house closures reduce the availability of Quants.

Scale or Yield? A Present-Value Identity

Abstract We propose a loglinear present-value identity in which investment (“scale”), profitability (“yield”), and discount rates determine a firm’s market-to-book ratio. Our identity reconciles existing influential market-to-book decompositions and facilitates novel insights from three empirical applications: (1) Both investment and profitability are important contributors to the value spread and stock return news variance. (2) Any cross-sectional return predictability has a mirror image in cash-flow fundamentals, providing asset pricing theories with additional moments to match. (3) The investment spread significantly improves the predictability of time-series variation in the value premium and justifies the poor performance of value in recent years.

Machine learning goes global: Cross-sectional return predictability in international stock markets

We examine return predictability with machine learning in 46 stock markets around the world. We calculate 148 firm characteristics and use them to feed a repertoire of different models. The algorithms extract predictability mainly from simple yet popular factor types—such as momentum, reversal, value, and size. All individual models generate substantial economic gains; however, combining them proves particularly effective. Despite the overall robustness, the machine learning performance depends heavily on firm size and availability of recent information. Furthermore, it varies internationally along two critical dimensions: the number of listed firms in the market and the average idiosyncratic risk limiting arbitrage.

Man versus Machine Learning: The Term Structure of Earnings Expectations and Conditional Biases

Abstract We introduce a real-time measure of conditional biases to firms’ earnings forecasts. The measure is defined as the difference between analysts’ expectations and a statistically optimal unbiased machine-learning benchmark. Analysts’ conditional expectations are, on average, biased upward, a bias that increases in the forecast horizon. These biases are associated with negative cross-sectional return predictability, and the short legs of many anomalies contain firms with excessively optimistic earnings forecasts. Further, managers of companies with the greatest upward-biased earnings forecasts are more likely to issue stocks. Commonly used linear earnings models do not work out-of-sample and are inferior to those analysts provide. Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online.

Innovation links, information diffusion, and return predictability: Evidence from China

Based on the activities of patent citation in China, a novel type of cross-firm innovation links is generated to investigate the gradual diffusion of information along the innovation chain via tests of cross-sectional return predictability. Various signals are created to represent the value of the information contained in the innovation links; these signals are demonstrated to have robust cross-predictability for stock returns in both the cross-sectional regression model and portfolio strategies. The effect of predictability is found to be stronger for stocks with high institutional ownership and analyst coverage. Considering the minimum number of steps required to establish the cross-firm linkage, innovation links are further partitioned to represent different proximity of the linked firms. It is then found that information diffuses faster across closely-linked firms than across distantly-linked firms. Sophisticated investors are found to be able to properly process the relevant information and benefit from innovation links.

Financing anomaly, mispricing and cross-sectional return predictability

This study investigates the persistence of financing anomaly in the Chinese stock market. The results show that the long-short portfolio earns an average monthly excess return of 0.88% for the following 12 months based on this anomaly. We provide a behavioral mispricing-based explanation for this anomaly that financing anomaly is stronger in stocks or period with high investor overconfidence. Moreover, we find that financing anomaly is stronger in stocks with more investor attention, indicating that excessive investor attention increases mispricing and reduces market efficiency. However, idiosyncratic volatility cannot capture the arbitrage costs faced by financing anomaly, and investors' lottery-like preference cannot explain financing anomaly. Our findings are robust after controlling for other firm characteristic variables, considering shell value contamination and different sample periods.

Ripples into waves: Trade networks, economic activity, and asset prices

We exploit information in sovereign CDS spreads and the international trade network to provide causal evidence of the propagation of global economic shocks. We show that trade links are an important source of shock transmission using the natural experiments of the Japanese tsunami and the COVID-19 lockdown in China. We then confirm more general and gradual information flows along the trade network by showing extensive country-level credit/equity cross-sectional return predictability. News about country fundamentals flows primarily from importers to exporters, depends on both direct and indirect links in the trade network, and is magnified by the exporting country's financial vulnerability.

Comovement and return predictability in asset markets: An experiment with two Lucas trees

Abstract Using a laboratory experiment, we investigate whether comovement can emerge between two risky assets, despite their fundamentals not being correlated. The ‘Two trees’ asset pricing model developed by Cochrane et al. (2007) guides our experimental design and its predictions serve as our source of hypotheses. The model makes time-series and cross-section return predictions following a shock to one of the two assets’ dividend distributions. As the model predicts, we observe (1) positive contemporaneous correlation between the two assets, (2) positive autocorrelation in the shocked asset, and (3) time-series and cross-sectional return predictability from the dividend-price ratio. In line with the rational foundations of the model, the model's predictions have stronger support in markets with relatively sophisticated agents.

Navigating the factor zoo around the world: an institutional investor perspective

The literature on cross-sectional stock return predictability has documented over 450 factors. We take the perspective of an institutional investor and navigate this zoo of factors by focusing on the evidence relevant to the practicalities of factor-based investment strategies. Establishing a sound theoretical rationale is key to identifying “true” factors, and we emphasize the need to recognize data-mining concerns that may cast doubt on the relevance of many factors. From a practical investment perspective, much of the factor evidence documented by academics may be more apparent than real. The performance of many factors is dependent on the inclusion of small- and micro-cap stocks in academic studies, although such stocks would likely be excluded from the real investment universe due to illiquidity and transaction costs. Nevertheless, a parsimonious set of factors emerges in equities and other asset classes, including currencies, fixed income, and commodities. These factors can serve as meaningful ingredients to factor-based portfolio construction.

Contagion and Return Predictability in Asset Markets: An Experiment With Two Lucas Trees

Using a laboratory experiment, we investigate whether comovement can emerge between two risky assets, despite their fundamentals not being correlated. The ‘Two trees’ asset pricing model developed by Cochrane et al. (2007) guides our experimental design and its predictions serve as our source of hypotheses. The model makes time-series and cross-section return predictions following a shock to one of the two assets’ dividend distributions. As the model predicts, we observe (1) positive contemporaneous correlation between the two assets, (2) positive autocorrelation in the shocked asset, and (3) time-series and cross-sectional return predictability from the dividend-price ratio. In line with the rational foundations of the model, the model's predictions have stronger support in markets with relatively sophisticated agents.

Most claimed statistical findings in cross-sectional return predictability are likely true

Harvey, Liu, and Zhu (2016) “argue that most claimed research findings in financial economics are likely false.” Surprisingly, their false discovery rate (FDR) estimates suggest most are true. I revisit their results by developing non- and semi-parametric FDR estimators that account for publication bias and empirical correlations. These estimators provide simple closed-form expressions and reliably produce an upper bound on the FDR in simulations that cluster-bootstrap from empirical predictor returns. Applying these estimators to the Chen-Zimmermann dataset of 205 predictors, I find that most claimed statistical findings in the cross-sectional predictability literature are likely true.

Cross-sectional Return Predictability in Indian Stock Market: An Empirical Investigation

This paper provides a comprehensive analysis of stock return predictability in the Indian stock market by employing both the portfolio and cross-sectional regressions methods using the data from January 1994 and ending in December 2018. We find strong predictive power of size, cash-flow-to-price ratio, momentum and short-term-reversal, and in some cases of book-to-market-ratio, price-earnings-ratio. The total volatility, idiosyncratic volatility, and beta are not consistent stock return predictors in the Indian stock market. In cross-sectional regression analysis, size, short-term reversal, momentum, and cash-flow-to-price ratio predict the future stock returns. Overall, the two variables momentum and cash flow to price ratio demonstrate reliable forecasting power under all methods and both small and large size samples.

Technical trading index, return predictability and idiosyncratic volatility

This study examines the cross-sectional return predictability of technical trading index and tests whether the source and the persistence of technical trading effect is the result of idiosyncratic volatility limiting arbitrage in the Chinese stock market. By eliminating common noise components in technical indicators, we propose a new technical trading index, TECHIWC, which negatively predicts future returns from short to long terms. This predictive power is not subsumed by other well-known firm characteristics. Furthermore, we find that the relationship between the TECHIWC effect and idiosyncratic volatility is significantly positive, which is consistent with idiosyncratic volatility limiting arbitrage of the TECHIWC effect. Finally, this relationship is robust to consideration of other limits of arbitrage, market states, and alternative specifications of idiosyncratic volatility.

Overselling winners and losers: How mutual fund managers' trading behavior affects asset prices

We link a seemingly biased trading behavior to equilibrium asset prices. U.S. equity mutual fund managers tend to sell both their big winners and big losers. This selling pressure pushes down current prices and leads to higher future returns; aggregating across funds, we find that securities for which investors have large unrealized gains and losses outperform in the subsequent month. Funds with larger turnover, shorter holding period, and higher expense ratios, are significantly more likely to manifest this trading pattern, and unrealized profits from such funds have stronger return predictability. This cross-sectional return predictability is difficult to reconcile with alternative explanations.

Man vs. Machine Learning: The Term Structure of Earnings Expectations and Conditional Biases

We introduce a real-time measure of conditional biases in firms' earnings forecasts. The measure is defined as the difference between analysts' expectations and a statistically optimal unbiased machine-learning benchmark. Analysts' conditional expectations are, on average, biased upwards, and the bias increases in the forecast horizon. These biases are associated with negative cross-sectional return predictability, and the short legs of many anomalies contain firms with excessively optimistic earnings. Further, managers of companies with the greatest upward-biased earnings forecasts are more likely to issue stocks. Commonly-used linear earnings models do not work out-of-sample and are inferior to those provided by analysts.

The Cross-Section of Currency Appreciation Rates

We compute a cross-sectional variance decomposition for the average appreciation rate of foreign currencies. The cross-sectional dispersion in average nominal exchange rate growth is due to predictability of interest rate spreads among non-G10 currencies. However, among G10 currencies, we have an opposite pattern with cross-sectional return predictability playing the dominant role. The importance of the return channel among the G10 currencies is even stronger for real appreciation rates. By decomposing the cross-sectional return channel, pricing errors outweigh currency risk premia in terms of explaining nominal currency appreciation among G10 currencies. These results suggest important differences between G10 and other currencies.

Positive Feedback Trading and Stock Prices: Evidence from Mutual Funds

A mutual fund’s demand for a pricing factor, measured by the loading of the fund’s returns on the factor’s returns, is persistent over time. When stock characteristics are time-varying and change frequently, persistence in factor demand generates a need for rebalancing. This rebalancing motive, in turn, leads to predictable trading from mutual funds and contributes to cross-sectional return predictability. In particular, when there is a “mismatch” between a stock’s characteristic and the underlying funds’ demand for that characteristic, the “mismatched” stock will face selling pressure from the underlying funds and subsequently earn lower returns. Double-sorting on stocks’ characteristics and mutual funds’ factor demand refines value and momentum strategies, generating abnormal returns that cannot be explained by subsequent fundamentals or retail trading flows.

Resurrecting the Size Effect: Firm Size, Profitability Shocks, and Expected Stock Returns

Many studies report that the size effect in the cross-section of stock returns disappeared after the early 1980s. This paper shows that its disappearance can be attributed to negative shocks to the profitability of small firms and positive shocks to big firms. After adjusting for the price impact of profitability shocks, we find a robust size effect in the cross-section of expected returns after the early 1980s. Our results highlight the importance of in-sample cash-flow shocks in understanding cross-sectional return predictability. Received April 2, 2014; editorial decision August 6, 2018 by Editor Laura Starks.