Managing credit risk and the cost of equity with machine learning techniques

Chi, Yujing (2022) Managing credit risk and the cost of equity with machine learning techniques. PhD thesis, University of Glasgow.

Abstract

Credit risks and the cost of equity can influence market participants' activities in many ways. Providing in-depth analysis can help participants reduce potential costs and make profitable strategies. This kind of study is usually armed with conventional statistical models built with researchers' knowledge. However, with the advancement of technology, a massive amount of financial data increasing in volume, subjectivity, and heterogeneity becomes challenging to process conventionally. Machine learning (ML) techniques have been utilised to handle this difficulty in real-life applications. This PhD thesis consists of three major empirical essays. We employ state-of-art machine learning techniques to predict peer-to-peer (P2P) lending default risk, P2P lending decisions, and Environmental, Social, Corporate Governance (ESG) effects on firms' cost of equity.

In the era of financial technology, P2P lending has gained considerable attention among academics and market participants. In the first essay (Chapter 2), we investigate the determinants of P2P lending default prediction in relation to borrowers' characteristics and credit history. Applying machine learning techniques, we document substantial predictive ability compared with the benchmark logit model. Further, we find that the LightGBM has superior predictive power and outperforms all other models in all out-of-sample predictions. Finally, we offer insights into different levels of uncertainty in P2P loan groups and the value of machine learning in credit risk mitigation of P2P loan providers.

Macroeconomic impact on funding decisions or lending standards reflects the risk-taking behaviour of market participants. It has been widely discussed by academics. But in the era of financial technology, it leaves a gap in the evidence of lending standards change in a FinTech nonbank financial organisation. The second essay (Chapter 3) aims to fill the gap by introducing loan-level and macroeconomic variables into the predictive models to estimate the P2P loan funding decision. Over 12 million empirical instances are under study while big data techniques, including text mining and five state-of-the-art approaches, are utilised. We note that macroeconomic condition affects individual risk-taking and reaching-for-yield behaviour. Finally, we offer insight into macroeconomic impact in terms of different levels of uncertainty in different P2P loan application groups.

In the third essay (Chapter 4), we use up-to-date machine learning techniques to provide new evidence for the impact of ESG on the cost of equity. Using 15,229 firm-year observations from 51 different countries over the past 18 years, we document negative causal effects on the cost of equity. In addition, we uncover non-linear effects because the level of ESG effects on the equity cost decrease with the enhancements of ESG performance. Furthermore, we note the heterogeneity in ESG effects in different regions by breaking down our sample. Finally, we find that global crises change the sensitivity of the equity cost towards ESG, and the change varies in areas.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Colleges/Schools: College of Social Sciences > Adam Smith Business School > Accounting and Finance
• Supervisor's Name: Sermpinis, Professor Georgios and Tsoukas, Professor Serafeim
• Date of Award: 2022
• Depositing User: Theses Team
• Unique ID: glathesis:2022-83141
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 23 Sep 2022 15:07
• Last Modified: 23 Sep 2022 15:07
• Thesis DOI: 10.5525/gla.thesis.83141
• URI: https://theses.gla.ac.uk/id/eprint/83141