Look-Ahead Benchmark Bias in Portfolio Performance Evaluation

Abstract

Performance of investment managers is predominantly evaluated against targeted benchmarks, such as stock, bond, or commodity indices. But most professional databases do not retain time series for companies that drop from the database and do not necessarily track changes in the benchmarks over time. Consequently, standard tests of portfolio performance suffer from look-ahead benchmark bias, meaning that a given strategy is naively backtested against the assets constituting the benchmark of reference at the end of the testing period (i.e., now), rather than at the beginning of the period. In this article, the authors report that look-ahead benchmark bias can be surprisingly large in portfolios of common stocks—up to 8% per annum when the S&P 500 Index is the benchmark. Using CRSP data for the running top 500 U.S. capitalizations over the period 1927–2006, the authors demonstrate that look-ahead benchmark bias can account for a gross overestimation in performance metrics, such as the Sharpe ratio, as well as an underestimation of risk measured, for example, by peak-to-valley drawdowns. A general methodology to test investment strategy properties is advanced by the authors in the context of several random strategies having similar investment constraints.