eprintid: 10191236 rev_number: 12 eprint_status: archive userid: 699 dir: disk0/10/19/12/36 datestamp: 2024-05-17 12:53:01 lastmod: 2024-05-17 12:53:01 status_changed: 2024-05-17 12:53:01 type: thesis metadata_visibility: show sword_depositor: 699 creators_name: Sharma, Akash Sedai title: Stress Testing and Capital Adequacy Projections in Banking ispublished: unpub divisions: UCL divisions: B04 divisions: C05 divisions: F44 abstract: on banks’ loan supply, and specifically addressing the challenge of unreliable data used in key aspects of stress testing, including capital adequacy ratio projections. The thesis aims to improve the reliability and accuracy of these projections through the implementation of advanced methodologies, providing regulators and investors with better tools to assess financial stability and make informed decisions regarding lending practices. The first research paper (chapter 4) highlights the substantial influence of the stress testing on banks’ loan supply, emphasizing the crucial role of capital levels in supporting lending activities during stress test scenarios. The study recognizes the sensitivity of credit supply to stress tests and the significance of reliable outcomes. Underestimation can hinder credit supply, while overestimation increases systemic risks. Subsequent chapters focus on reducing estimation errors through various models and methodologies, with a particular focus on capital adequacy ratio projection—a key component of the CCAR stress test. This extension of the research expands the measurement of banks’ financial health beyond loan supply, providing reliable and more accurate outputs from projection models to evaluate financial stability. Informed lending decisions based on improved projections have substantial implications for the overall loan supply. The second research paper (chapter 5) introduces multi-output machine learning regression models for projecting capital adequacy ratios four periods ahead in time (quarter 5-8) per prediction. Comparative analysis of various models—neural networks, random forest, ridge regression, linear regression—reveals that Gaussian process yields the best outcomes. Unlike black box models, Gaussian process also provides point estimates along with uncertainty estimates in the form of a confidence band, enhancing transparency. Building on the second research paper, the third research paper (Chapter 6) leverages data augmentation through Generative Adversarial Networks (GANs) to improve the accuracy of capital adequacy ratio projections across different modelling approaches. This contribution tackles the challenge posed by the unbalanced nature of available data and the tail behaviour of crisis scenarios, which are crucial aspects of studying financial crises. The integration of synthetic data augmentation using GAN-generated samples demonstrates significant improvements in projection accuracy. This also highlights the significance of reliable capital adequacy ratio projections in relation to loan supply, reinforcing the findings from the first chapter. Overall, this thesis delves into the interplay between stress testing, loan supply, and capital adequacy ratios in banking regulation. The outcomes of this research offer valuable insights to regulators, investors, and banks, enabling them to assess financial stability, make informed lending decisions, and effectively manage risks within the banking sector. date: 2024-04-28 date_type: published oa_status: green full_text_type: other thesis_class: doctoral_open thesis_award: Ph.D language: eng primo: open primo_central: open_green verified: verified_manual elements_id: 2270267 lyricists_name: Sharma, Akash Sedai lyricists_id: ASHAC88 actors_name: Sharma, Akash actors_id: ASHAC88 actors_role: owner full_text_status: public pages: 200 institution: UCL (University College London) department: Civil, Environmental and Geomatic Engineering thesis_type: Doctoral citation: Sharma, Akash Sedai; (2024) Stress Testing and Capital Adequacy Projections in Banking. Doctoral thesis (Ph.D), UCL (University College London). Green open access document_url: https://discovery.ucl.ac.uk/id/eprint/10191236/1/AS_PhD_Thesis_Final.pdf