Comparing Mutual Fund Screening Methods Across Short and Long Samples
Summary
This article reviews seven methods for classifying active mutual funds as skilled, unskilled, or zero-alpha, with emphasis on how their accuracy changes when return histories are short. It describes a four-factor alpha benchmark, multiple-testing adjustments, bootstrap procedures, false discovery rate methods, and a mixture-model approach estimated with expectation maximization. The methods differ in their goals: some identify standout funds, while others estimate the proportions of funds in each performance group.
The study uses simulations calibrated to US equity fund data and tests both overlapping and nearly discrete ability distributions. For short samples, bootstrap methods associated with Kosowski and Fama–French, and one FDR method, generally show lower estimation errors and better coverage than the alternatives. The rankings differ for long samples and can also depend on significance thresholds. An empirical analysis of US active funds finds materially different classifications across methods. These findings are conditional on the models, sample design, and historical data; the article cautions against extrapolating long-sample comparisons to short rolling windows.
Key ideas
- Fund selection methods classify managers using estimated risk-adjusted alpha, but their objectives differ.
- Multiple testing can cause conventional significance tests to label zero-alpha funds as skilled or unskilled.
- Simulations find that some bootstrap and FDR approaches estimate performance groups more accurately in short samples.
- Method rankings change with sample length and may also respond to significance thresholds.
- Different screening methods produce notably different classifications in the historical fund data.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.