Unveiling the four-pillar framework: Machine learning evidence on personality, firm, governance, and financial origins of managerial overconfidence in China

This study investigates the key factors driving managerial overconfidence in Chinese A-share listed companies from 2011 to 2023. Utilizing advanced machine learning algorithms, including Random Forest and XGBoost, we analyze the effects of personal traits, firm characteristics, governance structures, and cost-effectiveness on managerial overconfidence. Our findings indicate that governance structure is the most significant determinant of managerial overconfidence across various models and datasets. Moreover, non-linear machine learning algorithms, particularly Random Forest, consistently outperform linear models in capturing the complex relationships between predictors and managerial overconfidence. The analysis identifies five critical secondary indicators: staff number, top shareholder ownership, enterprise size, operating income growth rate, and company listing age. Notably, managerial overconfidence is found to increase with company age, staff number, and enterprise size, while it decreases with operating income growth rate. The relationship with top shareholder ownership exhibits a more complex and non-linear pattern. These findings have important implications for corporate governance practices, investor decision-making, and regulatory policies.