Taming the Factor Zoo: A Test of New Factors
Guanhao Feng, Stefano Giglio, Dacheng Xiu
Published in The Journal of Finance, January 2020.
Investing in factors helps improve portfolio outcomes, reduce volatility and enhance diversification. According to the Invesco Global Factor Investing Study, over 70% of institutional investors surveyed in 2018 used factor strategies. However, hundreds of risk factors have been "discovered" over the last 30 years forming a “zoo of factors” to challenge existing asset pricing benchmarks.
"Investment practitioners have too many choices for possible anomalies and there will probably be more in the future,” says Feng.
“It is crucial for both practitioners and academics to understand the real common risk factors. Which of these new anomalies are genuine factors, and which are simply a repackaging of hundreds of existing ones?”
This paper shows how to use dimension-reduction techniques such as LASSO to discriminate between useful, useless, and redundant factors. It provides a statistical test that can be applied in practice to newly proposed factors or anomalies to evaluate their contribution to the investment opportunity set, benchmarking it against the hundreds of existing factors simultaneously. It is therefore of direct use to an investor that is evaluating trading on a newly proposed anomaly. When an anomaly is deemed redundant, their method also identifies a small number of factors that span it. Finally, the technique also works well out of the sample, which is a primary concern in the practice of investments.
Feng and co-authors apply their methodology to a large set of factors proposed in the literature, uncovering several interesting empirical findings. First, several newly proposed factors (especially different versions of profitability) explain asset prices, even after accounting for the large set of factors proposed before 2012. Second, they show that applying our test recursively over time would have deemed only a small number of factors proposed in the literature significant. Finally, they demonstrate how the results differ starkly from the conclusions one would obtain simply by using the risk premia of the factors or the standard Fama‐French three‐factor model as a control.
Guanhao (Gavin) Feng is an Assistant Professor of Business Statistics at the Department of Management Sciences. Gavin focuses on the interdisciplinary research between machine learning and asset pricing. This article introduces a statistical test framework for the enlarging factor zoo discovered in the asset pricing and investment world. This article is based on his co-authored paper "Taming the Factor Zoo: A Test of New Factors" with Stefano Giglio from Yale University and Dacheng Xiu from the University of Chicago. Their paper has won the 2018 AQR Insight Award and is published in the Journal of Finance in 2020.