N° 17-37: Anomalies and False Rejections

We combine the information obtained from a large laboratory of over two million trading strategies with what we know about strategies that survive the publication process to measure two unobservable quantities: the proportion of false rejections and the $t$-statistic thresholds that researchers should apply to control for multiple hypothesis testing. We find the proportion of false rejections to be 45.3%, and thresholds to be 3.84 and 3.38 for time-series alphas and cross-sectional slope coefficients, respectively. The benefit of applying multiple hypothesis testing is, therefore, substantial.