Are Characteristics Covariances? A Comment on Instrumented Principal Component Analysis
43 Pages Posted: 23 Jun 2020
Date Written: 2020
We present analytical and simulation-based evidence that instrumented principal component analysis (IPCA) cannot reliably distinguish between whether covariances or characteristics explain asset returns because the question has to be answered jointly with the question of how many factors have to be modeled. IPCA finds a covariance-based explanation when estimating too many factors (“alpha-eating”) and a characteristic-based explanation when estimating too few factors (“beta-eating”). Our results therefore call into question the empirical evidence recently obtained that stocks (Kelly et al., 2019), options (Büchner and Kelly, 2022), and bonds (Kelly et al., 2021) are explained by covariances.
Keywords: cross-section of stock returns, covariances, characteristics, IPCA
JEL Classification: C230, G110, G120
Suggested Citation: Suggested Citation