The Supraview of Return Predictive Signals

This study speaks to investment academics and practitioners by describing and analyzing the population of return predictive signals (RPS) publicly identified during the period 1970-2010. Our supraview brings to light a number of new facts about the population of RPS, including that more than 330 signals have been discovered and reported; the properties of newly discovered RPS remain stable over time; and RPS with higher mean returns have not only larger standard deviations of returns, but higher Sharpe ratios too. Using a sample of RPS, we estimate the average signed (absolute) cross-correlation of returns in the population of RPS to be just 0.05 (0.25). Abstracting from implementation costs, we show that this low of an average signed cross-correlation in RPS returns means that in theory an optimal portfolio of all RPS can have an equally-weighted (value-weighted) annualized Sharpe ratio as large as 3.0 (4.5). We also show that the probability that a given RPS has a positive alpha after being orthogonalized against five (25) other randomly chosen RPS is 62% (32%). Our study suggests that practitioners can expect to create value for their clients by hunting down new sources of alpha, and that academics testing for the existence of a new RPS do not need to orthogonalize the returns of that RPS against all pre-existing RPS. However, our findings also pose a challenge to academic theorists, since they imply that either U.S. stock markets are pervasively inefficient, or there exist a much larger number of rationally priced sources of risk in equity returns than ever previously thought