Revisiting Hypothesis Testing With the Sharpe Ratio

Abstract

It is shown via numerical simulation, asymptotic approximations and an extension of the algorithm and codebase of the p-value estimator of Ledoit and Wolf (2008) that the hypothesis test that is based on the difference between the Sharpe ratios of a pair of portfolios is of such low power that type-II errors would, in most circumstances of practical interest, be all too frequent. The test is shown to be potentially feasible in practice only if the returns of the paired portfolios under study tend to be strongly positively correlated and if decades of data are available.

Keywords: Sharpe ratio, p-value, statistics, correlation coefficient, robust

JEL Classification: C120, C460, C150, C870