Ensemble Models Using Symbolic Regression and Genetic Programming for Uncertainty Estimation in ESG and Alternative Investments

Trustable models exploit the diversity of model forms developed using symbolic regression via genetic programming to define ensemble models. These models have been shown empirically to have a strong predictive performance and the ability to extrapolate into regions of unknown parameter space or detect changes in the underlying system. This chapter demonstrates how the same technique for quantifying uncertainty is helpful in risk management workflows for alternative investing, especially when applying behavioral science principles. The use cases cover assets such as publicly traded private equities, specifically when the optimization objectives include financial and environmental, social, and governance (ESG) criteria, and ESG ETFs. This chapter provides an overview of these asset classes and a critical review of the issues with how current ESG ratings are formulated by rating agencies. Additionally, explicit uncertainty ranges are obtained, using an ensemble modeling approach, at a sufficiently high accuracy level to trust the uncertainty measurement. Future research is necessary to refine the approach presented as more data becomes available.

Venegas, P., Britez, I., Gobet, F. (2022). Ensemble Models Using Symbolic Regression and Genetic Programming for Uncertainty Estimation in ESG and Alternative Investments. In: Walker, T., Davis, F., Schwartz, T. (eds) Big Data in Finance. Palgrave Macmillan, Cham. https://doi.org/10.1007/978-3-031-12240-8_5

___________________________________________
Sources: https://doi.org/10.1007/978-3-031-12240-8_5