Group items tagged

This paper discusses potential strengths and weaknesses of ML-based credit assessment

To develop the related technology for alternative credit scoring, the HKMA has commissioned the Hong Kong Applied Science and Technology Research Institute (ASTRI) to study the use of artificial intelligence in SME loan applications; and published a white paper to deliver the findings.

China's WeChat Pay has told its users that it will be giving them a credit rating, called "Payment Points." The system leverages artificial intelligence and uses users' data collected from the platform including personal and credit records, spending habits and other factors to determine a user's "credibility." Tencent hopes the system will be adopted by small commercial institutions including banks for personal credit evaluation.

This paper proposes a new framework for supervisors to measure the risk-adjusted performance of machine-learning (ML) credit assessment models, harnessing the process for validating internal ratings-based (IRB) systems for regulatory capital to detect ML's limitations in credit default prediction. From a supervisory standpoint, having a structured methodology for assessing ML models could increase transparency and remove an obstacle to innovation in the financial industry.

New agencies are trying to widen the range of financial data that's used to assess loan applicants. They hope to extend credit scoring to the unbanked, gig economy workers and young people with little credit history. These challengers use machine learning techniques to plough through financial transaction data, or to scrutinise online questionnaires. Machine learning is controversial in this area, as it could give rise to bias and discrimination in lending decisions. Nevertheless, large credit card companies such as Capital One and established agencies like Experian are showing interest in these new methods.

An article by a couple of Banco de España economists proposes a framework to assess how reliable machine learning (ML) technology is to credit assessments. It is based on generating datasets intended to resemble typical credit settings, in which the relationship between the variables is defined. It uses non-interpretable ML models on these generated datasets, and explain their results using two popular interpretability techniques. It then calculates to what extent the explanations given by the interpretability techniques match the underlying truth.