Alpha vs Alpha: Selection, Timing, and Factor Exposures from Different Factor Models

Abstract

Alpha and factor loadings differ in the assumed models for systematic risk and based on the estimation method. Using data on 1,312 US equity active mutual funds with $3.9 trillion in assets under management, the authors found that the estimated alphas are similar for commonly used regression-based factor models with combinations of value, size, quality, momentum, and low risk, but they differ significantly from results with holdings-based estimation methods. Holdings-based alpha measures are usually lower than measures with regression-based methods; can identify timing components that are largely positive, but modest, across managers; and are linked to systematic factor exposures. The authors found little evidence that certain areas of the market, such as small-capitalization stocks, offer more opportunities for alpha.

TOPICS: Mutual fund performance, performance measurement

Key Findings

• • Using data on 1,312 US equity active mutual funds with $3.9 trillion in assets under management, the authors analyzed the alpha and factor loadings from traditional approaches using return regressions and bottom-up approaches using individual security holdings.

• • Alphas produced by regression methods are similar, regardless of the factors chosen, but holdings-based alphas were generally lower than time-series estimates.

• • Most of the positive performance of the best managers comes from static factor tilts, and a small but modest component is added by managers from timing factors. Dynamic alpha is strongly negatively related to momentum.