Latent factor models with functional single-index loadings
Abstract
We extend static linear factor models in a semiparametric framework, by assuming the loadings are unknown functions that depend on some exogeneous covariates. Since the latter covariates may be possibly numerous, their effect is summarized through individual univariate indices. We state the consistency and the asymptotic normality of our estimated factors and loading functions, when the number of individuals and their history length go to the infinity (large N, large T asymptotics). By simulation, the relevance of this approach is empirically assessed. On a large panel of U.S. stock returns, long–short portfolios constructed using our model achieve substantially higher Sharpe ratios than those based on standard PCA or even IPCA approaches.
Fields of Study & Subjects
Computational and Text Analysis Methods
Publication details
- Journal / source
- Econometric Reviews
- Publisher
- Informa UK Limited
- Volume
- 45
- Issue
- 9
- Pages
- 1338–1363
- Publication date
- 5 July 2026
- Language
- EN