This case study, and the accompanying Python code lab, focuses on mitigating biases in equity selection models and develops three models:
-
Traditional Quant Model. We test a traditional quant model based on linear factors, identifying problematic assumptions and violations.
-
Hybrid ML/Quant Model. We seek to improve this traditional factor approach by adding ML, which entails additional risks and biases.
-
Interpretable ML Model. We examine a purpose-designed ML model using the symbolic artificial intelligence (SAI) approach, developed to avoid or address many biases present in the preceding modeling approaches.
We have done our best to include the key themes of what we think are the critical stages of model development in this notebook, but clearly in practice many more checks and details would be added to each of the 5 model development stages to best ensure stakeholder's KPIs are met, and Governance standards would be as high as possible. Hopefully our example and key themes will provide insight to avoid many of the classic biases in model development.
References [1] Philps, D., Tilles, D., & Law, T. (2021). Interpretable, Transparent, and Auditable Machine Learning: An Alternative to Factor Investing. The Journal of Financial Data Science, 3(4), 84-100 https://jfds.pm-research.com/content/early/2021/09/22/jfds.2021.1.077
[2] Fama, E.F. and French, K.R., 2015. A five-factor asset pricing model. Journal of financial economics, 116(1), pp.1-22.
[3] Fabozzi, F. J., Focardi, S. M., & Kolm, P. N. (2010). Quantitative equity investing: Techniques and strategies. John Wiley & Sons.
[4] Israel, R., & Ross, A. (2017). Measuring factor exposures: Uses and abuses. The Journal of Alternative Investments, 20(1), 10-25.
[5] Levin A. (1995). Stock Selection via Nonlinear Multi-Factor Models. Advances in Neural Information Processing Systems, 8. https://proceedings.neurips.cc/paper/1995/file/d6ef5f7fa914c19931a55bb262ec879c-Paper.pdf