Key points are not available for this paper at this time.
ABSTRACT This study examines the joint determinants of audit quality by integrating auditor human and social capital through the application of ensemble machine learning algorithms, including Random Forest, GBRT, and XGBoost. Using a sample of 18574 firm-year observations from the Chinese A-share market between 2017 and 2023, we find that professional attributes, specifically industry expertise, workload, and tenure, combined with network centrality, are the primary predictors of audit quality. These factors substantially outperform natural and social characteristics in predictive power. Our analysis reveals robust non-monotonic relationships consistent with trade-offs between learning effects and cognitive overload. Specifically, professional attributes exhibit an inverted U-shaped association with audit quality, manifesting as a U-shaped pattern in absolute discretionary accruals and an inverted U-shaped pattern in audit reporting aggressiveness. Furthermore, we document significant heterogeneity in how network structure moderates individual capabilities. Central auditors leverage their structural advantages to enhance the benefits of expertise through knowledge sharing, whereas peripheral auditors face performance constraints regardless of their individual skills. By leveraging SHAP value analysis to interpret the black-box models, we provide empirical evidence supporting the Social Resource Management framework. These findings offer distinct implications for auditor selection and regulatory oversight in relation-based markets.
Niu et al. (Thu,) studied this question.