Explainable machine learning models to identify the key drivers of the implied cost of capital

Abstract

We propose an explainable machine learning model to assess the impact of financial and non-financial factors on a firm’s ex-ante cost of capital, a measure that reflects the perception of investors on a firm’s riskiness. To this aim, we combine the XGBoost machine learning model with two explainability tools: Shapley values and Lorenz Zonoids. The effectiveness of our proposal is tested on a sample of more than 1,400 listed companies worldwide. The empirical findings indicate
that the XGBoost model has high predictive accuracy, and that, on the basis of the Lorenz Zonoid method, the factors which affect the implied cost of capital the most are: firm size, ROE, firm portfolio risk, and the country’s institutional quality. They also indicate the relevance of non-financial factors in predicting the cost of capital at the company’s level, namely environmental performance and governance practices. This paper provides also evidence indicating that investors punish the most polluted firms, supporting the need to promote a transition to a more sustainable economic system. This conclusion will be in the interest of the general public and policymakers.

Keywords: Explainable AI, non-financial factors, Shapley values, XGBoost models, cost of capital

JEL Classification: G12, G32, C1