The Correlation Structure of Anomaly Strategies
55 Pages Posted: 19 Jul 2017 Last revised: 10 Aug 2018
Date Written: August 9, 2018
We construct 1,185 trading strategies based on previously documented anomalies, 19% of which yield mean returns with t-statistics above three. These significant anomalies can be grouped into 43 clusters based on their time-series correlations. The three largest clusters are momentum, profitability and issuance. Often a single cluster contains similar anomalies separately documented by many different prior studies. Anomaly clusters exhibit stability over time and across different market regimes. Benchmark asset pricing models leave the overall cluster structure largely intact but reduce correlations. Around two-thirds of cluster portfolios survive the stringent HXZ4 q-factor benchmark model with alpha t-statistics above three.
Keywords: return predictability; anomalies; correlation; asset pricing; benchmark models; cluster analysis; machine learning
JEL Classification: G12, G14, C38
Suggested Citation: Suggested Citation