Modeling the Relation of Bank Governance and Risk Management

Abstract

The Great Recession (2008-2010) highlighted the complexity of risk management and the global impact of accumulated risk. The complexity of risk management in bank holding companies (BHC, or banks) often results in a lack of transparency for stakeholders, thus affecting their decisions. This research aims to simplify this complexity to help consumers, investors, regulators, management, policymakers, and taxpayers understand how well a firm structures its risk function and manages it. The research methodology uses Principal Components Analysis (PCA) and regression analysis to study the impact of various factors on bank risk management and performance. The analysis process starts with a replication of work done by other researchers, and moves to using a large, aggregated set of variables to determine a variable subset that creates a more effective risk management index (RMI), then to regression analysis, and concluding with sensitivity analysis. Three alternative RMI models were created and analyzed. However, none of them proved to be effective replacements for the existing RMI model. The correlation between these RMIs and tail risk was inconsistent, in terms of both strength and direction. Regression analysis was performed on the entire forty-six element factor set against tail risk, default risk, and return on assets (ROA). The former model resulted in fourteen statistically significant variables and explained just over 22% of the variation in tail risk. The second model resulted in seventeen statistically significant factors and explained just under 60% of the variation in default risk. The third model resulted in seven statistically significant variables and explained over 74% of the variation in ROA. Finally, a sensitivity analysis of the selected models was conducted. The removal of certain variables significantly reduced the strength of the models, indicating the importance of these factors in explaining the variation in the respective models. In particular, the removal of Tobin’s Q in the ROA model reduced the adjusted R2 from 73.72% to 26.07%. Similarly, the removal of the real estate loans variable significantly reduced the strength of the default risk model, with a nearly 50% reduction in adjusted R2.