When it comes to building financial models, factor selection is a crucial element to consider. With the integration of machine learning (ML) and data science in finance, the question arises: which factors should be considered for ML-driven investment models and how should we choose among them?
These questions are not only open but also critical. ML models can aid not just in factor processing but also in factor discovery and creation.
Factors in Traditional Statistical and ML Models: The (Very) Basics
Factor selection in machine learning is referred to as “feature selection.” Factors and features play a role in explaining the behavior of a target variable, while investment factor models outline the primary drivers of portfolio behavior.
One of the simplest methods for constructing factor models is ordinary least squares (OLS) regression, where the portfolio return acts as the dependent variable and the risk factors as independent variables. Ensuring low correlation among independent variables is crucial for different statistically valid models to explain portfolio behavior and determine the extent of the portfolio’s behavior each model accounts for, as well as the sensitivity of the portfolio’s return to each factor, as shown by the beta coefficient attached to each factor.
Unlike traditional statistical models, ML regression models can capture non-linear behavior and interaction effects more effectively. While they do not produce beta coefficients like OLS regression, they provide a more flexible approach to modeling.
Why Factors Should Be Economically Meaningful
While synthetic factors have gained popularity, economically intuitive and empirically validated factors have advantages over purely “statistical” factors. It’s essential for factors in traditional regressions and ML models to be economically distinct to avoid issues like multicollinearity in traditional regressions.
ML model construction differs from OLS regression with fewer strict assumptions, making it flexible in the selection of factors. Employing techniques like the least absolute shrinkage and selection operator (LASSO) in the pre-model stage can help distill a large set of factors into a smaller, more meaningful set for ML models.
Using decades of research and empirical validation, economically meaningful factors like Fama-French-Carhart factors offer a robust foundation for ML-driven models. These factors can help explain asset returns and lead to successful investment trading models.
Conclusion
Factor selection is pivotal in ML-based investment models. By focusing on economically meaningful factors, we can enhance the understandability and effectiveness of our ML-driven investment strategies.
If you enjoyed this post, remember to subscribe to Enterprising Investor.
All opinions expressed are the author’s own and should not be considered as investment advice. Views expressed do not necessarily reflect those of CFA Institute or the author’s employer.
Image credit: ©Getty Images / PashaIgnatov
Professional Learning for CFA Institute Members: CFA Institute members can self-report professional learning credits for reading this post on Enterprising Investor.