Board Activity and Bank Performance: An Australian Financial Sector Evidence Using Machine Learning-Based Imputation

This study examines the influence of board nationality diversity and the frequency of board meetings on the performance of Australian banks. The data set consists of 13 publicly listed Australian banks whose data was available from the time period 2008-2022. In order to address missing values, we use the latest machine-learning imputation method involving the mixgb framework that combines Extreme Gradient Boosting (XGBoost) with subsampling and Predictive Mean Matching (PMM) in multiple imputation modes. This is the state-of-the-art approach to imputations and therefore we adopt it to make our data most reliable. Employing advanced imputation methods and panel data econometric techniques through STATA including Fixed Effects (FE), Random Effects (RE) and two-step System Generalized Method of Moments (GMM) to address potential endogeneity, we assess the impact of governance provisions on Tobin’s Q, with return on assets (ROA) serving as a robust performance metric. The findings show that proportional variations in the nationality of directors increase market-based performance (Tobin’s Q) across all models, at the expense of accounting-based returns (ROA). This implies that while nationally diverse boards offer long-term strategic advantages through broader global insight and networks, they may incur short-term coordination costs that reduce operational efficiency. Meanwhile, the frequency of board meetings has a negligible impact on ROA and shows no strong relationship with market valuation (Tobin’s Q). The System GMM model further confirms the persistence in market valuation and strengthens the case that board diversity, not meeting intensity, drives long-term market performance, also confirmed by fixed effect and random effect.