Industry Risk in Asset Pricing: A Machine Learning Approach

We apply machine learning to the problem of building an industry risk factor for asset pricing models. Extending work by Fama and French (2020), we propose a novel approach for extracting valuable information from the cross section of industry portfolio returns. Using individual industries, we estimate a multifactor model and associated alpha parameters. Upon incorporating scaled alphas into a cross sectional regression of industry portfolio excess returns, we estimate a cross section industry factor as the market price of industry alpha risk.

In our empirical tests, we employ 30 industries in the sample period July 1963 to June 2024 from Kenneth French’s online data library. Standard Fama and MacBeth (1973) out-of-sample cross-sectional regressions are estimated to test the hypothesis that industry risk is a salient factor in pricing industry stock portfolios. Upon augmenting the CAPM as well as different multifactor models with a cross section industry factor, we find that their associated loadings are statistically significant and materially boost explanatory power in
terms of explaining industry portfolio excess returns. As robustness checks, we repeat the analyses using 47 industry portfolios as well as subperiods with similar results.

Our findings contribute to the large literature on multifactor models spanning traditional approaches based on observable firm and stock characteristics and machine learning approaches based on (for example) principal component analysis (PCA).3 However, our factors are not as opaque as those identified by PCA; instead, they are derived from samples of portfolios sorted on observable industry classifications and, therefore, are closely related to industry risk. Thus, our cross section factor construction methods take advantage of both traditional and machine learning approaches. Importantly, unlike PCA
factors, the economic interpretation of our cross section factor is readily identifiable to be attributable to industry risk contained in industry-sorted test asset portfolios.