Board gender diversity and emissions performance: Insights from panel regressions, machine learning, and explainable AI

Battle Count: 73

The paper is primarily focused on corporate governance and environmental performance rather than financial markets or trading strategies. However, findings on board gender diversity thresholds could inform ESG-based stock screening, responsible investment strategies, and factor models incorporating governance quality. The ML methodology (XGBoost, SHAP) is transferable to quantitative finance applications, but the specific domain (emissions performance) has limited direct trading relevance.

The paper combines multiple methodological approaches: panel econometrics (fixed/random effects, Hausman test, Sargan-Hansen), three ML algorithms with hyperparameter tuning via grid search, XAI methods (SHAP, PD plots), path analysis for mediation, and interaction terms for moderation. The DataRobot platform simplifies ML implementation, but the overall pipeline requires expertise in both econometrics and machine learning. Missing data handling and cross-validation add complexity.

The paper provides detailed methodology, hyperparameter settings, and variable definitions. However, data is only available through LSEG Workspace subscription, limiting full reproducibility. No code or GitHub repository is provided. The use of DataRobot platform for ML optimization adds a proprietary layer. Descriptive statistics, correlation matrices, and regression tables are fully reported.