Measuring Bank Complexity using XAI

Abstract

Following the 2007-2009 global financial crisis, the complexity and opacity of banks have been under scrutiny due to their effects on financial stability. Measuring these attributes is challenging, especially complexity, as traditional methods focus on single aspects like organizational structure without considering the interplay between different factors. Utilizing Explainable Artificial Intelligence (XAI), we introduce a machine learning technique to assess both complexity and opacity in banking, revealing strong correlations between the two measures at both firm and industry levels. Our research identifies a counter-cyclical trend in bank complexity, peaking before financial crises and decreasing during distress. Given the proposed methodology to quantify bank complexity, we explore how it influences investor behavior, noting a significant decrease in trading activity in highly complex firms, a result consistent with the limited market participation argument. Overall, our findings indicate that greater complexity and opacity are associated with higher future stock returns, lower volatility, and better Sharpe ratios, and suggest a link between a bank's complexity and systemic risk.

Keywords: Information Processing, Machine Learning, FinTech, Marginal Expected Shortfall, Systemic Risk

JEL Classification: C18, G11, G14, G21