Carbon Emissions and Firm Profitability: A SHAP and ALE-Based XAI Approach on the BIST Sustainability Index

Abstract

The primary objective of this study is to examine the relationship between carbon emissions and firm profitability using Explainable Artificial Intelligence (XAI) methods. The research analyzes various machine learning (ML) techniques to predict firm profitability. Additionally, XAI methods such as Shapley Additive Explanations (SHAP) and Accumulated Local Effects (ALE) are employed to enhance interpretability for decision-makers. The study focuses on firms listed in the Borsa Istanbul Sustainability Index (XUSRD), utilizing 189 firm-year data points collected from sustainability, operational, and integrated reports between 2021 and 2023. The findings indicate that low carbon emissions positively influence Return on Assets (ROA), while high emissions have mixed effects, positive in some firms and negative in others. Regarding Return on Equity (ROE), the analysis reveals a negative trend. Furthermore, Earnings Per Share (EPS) emerged as the variable with the highest contribution to both profitability models. The Random Forest algorithm was identified as the most effective method for predicting both ROA and ROE. This study contributes to the emerging literature by applying XAI techniques, specifically SHAP and ALE, to interpret machine learning-based profitability models within the context of carbon accounting, offering valuable insights for stakeholders and policymakers.