Augmenting Bank And Economic Data with Generative Adversarial Networks

This study addresses the challenges in projecting capital ratios during high-stress periods due to the scarcity of stress scenario data and the tail behaviour of these ratios. We employ Generative Adversarial Networks (GANs) for data augmentation to improve the accuracy of capital adequacy ratio projections. By augmenting datasets with synthetic data, we improve model performance across various methodologies. The approach is demonstrated using Bayesian model averaging on data from US banks, showing significant improvements in projection accuracy. This methodology offers valuable insights for both regulators and banks in evaluating capital adequacy during financial crises.