Non-Performing Loan Prediction in the age of Artificial Intelligence

By Susan Hu

Introduction

A non-performing loan (NPL) is a loan in which the borrower is in default - in other words, the borrower has failed to make the scheduled payments for a specified period.

In 2019, 0.89% of the loans that banks in the United States held were non-performing.

For banks and financial institutions, higher percentage of NPLs directly compel banks to increase their levels of reserve funds, resulting in an increase in the bank's expenditures on loan and risk management and a decrease in the bank's profitability.

Being able to effectively predict the Non-Performing Loans can save banks millions of dollars every year.

In the age of Artificial Intelligence (AI), the forward-thinking bankers and leaders naturally want to explore how AI can help banks significantly increase the accuracy of Non-Performing Loan prediction.

Challenges the financial industry is facing

As mentioned above, higher numbers of NPLs force banks to increase reserve funds, expenditures on risk management, and as a result profits decrease.

Data from all around the world shows how NPL negatively affects profitability and increases constraints of maintaining higher reserve levels. In order to remain compliant with financial regulations, banks must increase their monitoring and human capital expenditures. Moreover, an increase in the number of NPLs elevates a bank's risk exposure, resulting in a negative impact on its stock as well.

The Basel III Accords focused on raising the level of required funds for banks, stress testing, and identifying inherent risks in market flow; in 2018-19, banks of member nations had to carry out self-risk assessment modeling, re-calculate necessary funds' levels, and use the results of these analyses to find ways to increase liquidity and decrease their leveraging. In this process, banks were also required to perform a strict risk analysis of their credit-related products.

From Europe to North America to Asia, banks and financial institutions are looking to adapt a holistic, well-developed solution that directly addresses risk exposure results from NPL. It calls to employ a precise risk-assessment and scoring methodology and exercise an increased awareness of unknown risks that exist in the bank's system.

Graphen's AI NPL

Graphen's NPL Solution is built upon AI Platform Ardi, which incorporates advanced graph computing and Machine Learning.

Graph storage and analysis tracks the relationships between creditors, which enables risk monitors to assess and track the risks of NPL. Via aggregation of creditors' core relationship data, the graph database stores and analyzes data in the form of a graph, analyzing mutual influences between accounts, and predicting the risk that each creditor may default or pay late. The solution evaluates customer's NPL risk via analysis of customer behavior and relationship changes, in addition to scans using domain-knowledge-based, best-of-practice rules.

Machine Learning allows the solution to automatically assign a risk score to customers based on the graph analysis. The Aggregate Risk score includes a customer's Entity Risk, Network Risk, Activity Anomalies, and Known High-Risk Patterns, based on machine learning that dynamically predicts potential NPLs.

Risk tracking and control, enabling banks to assess the risk status of an account, find potential NPLs, and take earlier preventative action (such as increasing the frequency of risk assessments, increasing guarantee collaterals, and requesting early repayments)

Based on implementation experience, Graphen's NPL risk model can make up for the limited data and business volume of small and medium-sized banks, thus reducing operating cost and time needed to collect and analyze bank data.

According to feedback of risk control personnel, because machine learning detects suspicious behavior and trading patterns, it can also play a role in bank risk preference decision-making, by providing references of risk patterns.

Simulated account default risk propagation can improve the forecasting of account repayment and provide a more accurate forecast of bank deposit reserves.

Built upon advanced AI technology, Graphen's AI NPL Prediction solution enables the bank to make earlier prediction and detections of NPL risks and take action as soon as possible.

Interface of Graphen AI NPL Prediction Solution

Conclusion

As banks and financial institutions recognize the need to increase the efficiency and effectiveness of NPL prediction, it is essential to utilize the advanced AI technology so they can sufficiently reduce risk, lower reserves and increase profitability. Combining Machine Learning and graph computing allows our next-gen NPL prediction to be much more cost effective.

If you want to learn more about AI, Graphen's AI NPL solution, and how it can help you increase your NPL accuracy, please reach out to me at susanhu@graphen.ai.