The Risk Associated with Bank Loans and Mitigating Them through Quantitative Models

Banks in Ghana are going through tumultuous times. As anyone even tangentially involved in business knows, in August of 2017, the Bank of Ghana consolidated five banks facing liquidity issues; Sovereign Bank, Royal Bank, The Beige Bank, Construction Bank and Unibank into one bank, creatively named Consolidated Bank. The changes underfoot are not over. Only sixteen banks have met the new minimum paid-up capital requirement of GH¢400 million. This time of change presents many challenges, but also many opportunities for banks to modernize their management practices. The banking crisis may be just the push that Ghanaian banks need to adopt advanced risk modeling leveraging technologies such as machine learning.

Whenever a bank issues a loan a variable simultaneously appears with the loan issuance called “credit risk.” Credit risk basically asks if the loan can be repaid fully, partially or not at all. It uses the mathematical tools of probability theory to assign a probability of 0 (loan will not be repaid) to 1 (loan will be repaid) to assign the chances that a loan can be repaid back or not. The quantitative framework of calculating the likelihood of payback of a loan is called credit risk modeling.

As it has since the beginning of banking, the process of credit risk modeling begins by collecting historical data on a fairly homogeneous group of borrowers. This historical data captures their demographic data, their level of education, and their monthly income, the properties owned by the borrower and any outstanding debt or loans. Additionally, the historical records have an indicator field that depicts whether the borrower has defaulted on a loan in the past 3 to 5 years. In the absence of the statistical models, underwriters have to painstakingly measure the liquidity (current/assets), total assets to liabilities ratio, the amount of income of the customers, and other measures of business performance. The process of assessing the credit risk of a customer required enormous resources and time.  And this meant that loans could be issued at a slow rate which did not meet the needs of the market place or the bank. Under the status quo, banks make less money and take on more risk.

The introduction of statistical models began with simple but effective models such as logistic and probit models. Later, models known as generalized linear models dominated the field, but with the advent of machine learning and its highly sophisticated algorithms, credit models have become extremely accurate especially when high quality historical data is available. Despite the improvements in the risk quantification of loan paybacks, banks still wrestle with how to deal with uncontrollable factors such as job losses due to shifts in the macroeconomic climate and external shocks such as interest rate parity which under value the loans issued. Machine learning models make it possible to factor in these and other macroeconomic factors when predicting risk. With more advanced risk modeling, banks can expand their customer base to non-traditional loan seekers, such as small business owners and create more balanced portfolios. When applied in Ghana, machine learning models will help banks weather this period of uncertainty and rise from the tumult stronger than ever before.

 

Dr. Albert Essiam is the Data & Analytics Lead at OZÉ, a Ghana-based business that helps businesses and banks use data to make more profitable decisions. Dr. Essiam holds a PhD from MIT. If you want to learn how OZÉ can help your bank better model risk, email meghan@oze.guru.