Banking the Unbanked Carries Risk (That AI Modeling Can Mitigate)

Banking regulators have set the financial inclusion as reform agenda for most of the countries across the globe for improving the standard of living, curbing poverty and reducing inequality by delivering access to formal savings and growth. Towards this end, financial institutions need to adhere to the relevant compliance demands and account for the risks involved with banking the unbanked. In absence of proper credit risk assessments, banks are facing challenges in terms of high non-performing assets due to lack of information about their customers. Let’s look at the impact of financial inclusion on credit risk and how the application of artificial intelligence (AI) and machine learning (ML) can be useful to solve the problem of information asymmetry.

Impact of Financial Inclusion on Credit Risk

In simple terms, financial inclusion (FI) means access and use of financial services such as banking, insurance and pension by households and individuals. In the United Nation’s Sustainable Development Goals (SDGs) of 2030, FI is applies to four of them: Goal 2: zero hunger, Goal 5: gender equality, Goal 8: decent work and economic growth, and Goal 9: industry, innovation, and infrastructure. In other words, the benefits of FI play out in multiple dimensions. Many countries across the world have also set financial inclusion as a reform agenda and financial regulatory agencies have adopted it in their mandates. 

But fast growth in financial inclusion can impair financial stability and increase the credit risk as demonstrated in the last decade by the subprime mortgage crisis in the United States and the Andhra Pradesh microfinance crisis in India.

The Indian government launched an initiative in 2014 called PMJDY (Pradhan Mantri Jan Dhan Yojana) to provide universal access to banking services. It gives every adult a basic banking account, access to need-based credit, remittances facility, and insurance and pension to the weaker sections and lower-income groups.

As of October 2021, over 0.44 billion accounts have been opened. PMJDY appears a very effective program in this regard. But on the other hand, the reality of private lending at high interest rates to low-income groups is still very prevalent. Also, the amount of non-performing assets (NPAs) at public banks remains high. The main reason is that this kind of initiative was forced by banking regulators without proper mechanisms to control and identify reliable borrowers who nevertheless have little banking history.

There is one more problem: some customers who know that the loans they receive to support their farms or small businesses will be waived off before the next election cycle, so they default on purpose. Hence, those who genuinely intend to repay do not receive credit incentives from the banks because they need to account for bad actors.

Financial inclusion, in other words, faces multiple challenges in terms of credit risk assessment for microlenders who have no collateral and for whom only asymmetric information is available. Identifying the challenges, the Basel Committee on Banking Supervision (BCBS) has proposed 29 Core Principles including credit risk for effective banking supervision. The organization states that banks must have an adequate credit risk management process that considers their risk appetite, risk profile and market condition.

Information Asymmetry and the Potential Application of Artificial Intelligence and Machine Learning

Information asymmetry is a phenomenon where agents do not have the same level of information. This is a very peculiar problem when banks are tasked with increasing financial inclusion but have very limited information to assess credit risk. The bank can minimize the credit risk only when the bank can collect and process the characteristics of the borrowers.

One solution is the use of big data and machine learning. Researchers Margarete Biallas and Felicity O’Neill describe how countries such as Kenya, Mexico, Nigeria, India and Tanzania have used a mobile application called Branch to address this problem to a certain extent. This application helps vulnerable groups such as small farmers, small business owners and young women access financial services and products at affordable prices. The application uses machine learning and an algorithm to factor text messages, call logs, contact details and GPS in combination with the credit history of the customer to arrive at a lending decision. Using the machine learning application, it became easier to sanction loan amounts as small as $50 USD for short durations (from a few weeks to a few months). This was not possible earlier using the traditional credit assessment methods only. 

In another success story, Ant Financial (a subsidiary of Alibaba Group) applied machine learning techniques to leverage online transaction data to assess the creditworthiness of loan applicants, even those without collateral. With traditional credit risk assessment methods, the collateral serves as a kind of mandate criteria for lending decisions. But the disadvantage of collateral criterion is that it excludes millions of high potential small businesses. On the contrary, Ant Financial uses AI and big data so that lending assessments are based on payment history. High-performing small businesses are captured in their customer base at a fast pace with minimal cost. This helped Ant Financial to increase its loan portfolio from $0.5 billion to $4.0 billion in just four years. 

Bias in AI Remains a Risk

Not every use is positive, and bias can creep into both applications of AI and risk assessments. The Apple Credit Card, launched in 2019, also used AI to assess credit. Within months, users noticed that it granted smaller lines of credit to women compared to men. Any use of predictive algorithms in the financial sector calls for rigorous bias testing. Otherwise, banks run the risk of increasing inequity instead of reducing it.

Given the above case studies, it can be recognized that the application of artificial intelligence and machine learning, though potentially imperfect, helps a lot in credit decisions using alternative data sources and non-traditional credit risk assessment methods. In the traditional approach, data related to customer identity, bank transaction, credit history, collateral and income were mostly used to generate the credit scores. But artificial intelligence and machine learning can tap data sources such as social media data, drone or satellite images, and publicly available data. The new approach in credit risk helps banks and financial institutions better perform credit risk assessments of customer behavior and evaluate their ability to repay the loans. Nowadays, the new social enterprises with customized financial solutions are more cost effective and better able to manage the credit risk by using artificial and machine learning. 

Conclusion

Understanding the impact on financial inclusion on the credit risk is vital for policymakers and the banking system. The advantage of financial inclusion using the application of artificial intelligence and machine learning relies on its adoption by banks in competitive markets. The application of these technologies in developing countries or under-developed countries allows banks to automate their business processes and leverage the new and big data sources to mitigate issues, including the high cost of serving rural and low-income customers and establishing customer identity and creditworthiness. These technologies have the potential to minimize the risk of accruing a high balance of non-performing assets and other exposures resulting from information asymmetry.