Machine Learning and Credit Risk Calculation
Published on : Saturday 11-07-2020
Meghna Suryakumar elaborates upon why ML-driven early warning systems is the way forward for MSME and banks in India.
The coronavirus (Covid-19) pandemic has brought the Indian economy to an unprecedented halt. In the span of just a couple of months, cities have gone into complete lockdown, businesses have faced significant revenue losses, and numerous people have lost their jobs. The Covid-19-triggered economic downturn has also posed a major threat to India’s banking sector, which has already been grappling with US$ 123 billion worth of bad loans. Industry experts have warned that the ongoing pandemic has the potential to aggravate this bad loan crisis. Furthermore, it is reported that the non-performing assets (NPAs) of Indian banks can double to 18-20% by the end of the fiscal year. One of the main reasons for this is the inability of banks to detect the red flags and predict which accounts are likely to default.
Evidently, there is a dire need to reassess and redefine the current practices that are in place to detect loan defaulters. It is critical that banking organisations in India move from traditional mechanisms to advanced, digital systems that can identify credit risks early on. One of the easiest ways to achieve this is to deploy Early Warning Systems (EWS). In fact, the Reserve Bank of India (RBI) mandated the adoption of EWS in 2015 to mitigate the risks posed by Red Flagged Accounts (RFAs). Data, however, shows that archaic Early Warning Systems have failed to address the problem of India’s growing pile of bad loans. This necessitates incorporating advanced technologies such as machine learning into Early Warning Systems to predict default risks more accurately.
How does machine learning help in credit risk calculation?
When a business applies for a loan, the lender needs to evaluate whether the business can repay the loan amount within the given timeline. Two key parameters to consider are the capacity and willingness of the business. Lenders generally assess the capacity by using various measures of profitability and leverage. A profitable entity generates enough cash flow to cover the loan principal and the interest. A leveraged firm, on the other hand, has limited access to equity to weather economic shocks. Now, let us assume that there are two loan applicants – one with high profitability and high leverage, and another with low profitability and low leverage.
Which business has a lower credit risk? Answering this question becomes even more difficult when lenders have to weigh in multiple other factors – both external and internal. These additional dimensions include traditional financial data, non-financial datasets such as trade shipments, legal cases against the firm and/or promoter, as well as alternate datasets such as GST filings. Summarising all of these data points into a credit score can be challenging, and this is where machine learning comes into the scene.
Machine learning can yield better insights than a human analyst in a much more efficient and faster manner. It essentially analyses and compares large volumes of data sets to find out unusual patterns and risk elements. The biggest advantage of a machine learning-based mechanism is that it can learn continuously. So more data is fed into the system, the more accurate the results will be. There are different machine learning algorithms used to assess credit risk, the most common of which are artificial neural networks, random forest, and boosting.
Machine learning-driven Early Warning Systems
Machine learning-driven Early Warning Systems can retrieve important information from tens and thousands of data points to gauge and monitor risk, which is impossible for human analysts to replicate. It also has the ability to extract actionable insights from the accumulated information make predictions in near real-time.
Restructuring data from multiple data points
Banks often have access to a vast amount of data, but the data is unstructured and not useful to the credit decision making process. Machine learning can cleanse, restructure and validate this data to generate early-warning alerts. Based on these alerts, lenders or decision-making authorities can make an informed decision. Unlike human analysts, ML-based EWS can execute the tasks within minutes, which facilitates speedier credit-decision making.
Eliminating the scope of human bias
When it comes to the assessment of credit worthiness, human bias can play a detrimental role in the decision-making process. There have been several instances when this has proven costly for financial institutes, eventually adding to their share of bad debts. With EWS built on machine learning algorithms, human bias can be entirely eliminated, thereby leading to more unbiased and objective decisions.
Enabling access to alternate data
Up until recently, most banks would rely solely on traditional financial data to measure credit risk. While this particular method has worked well in developed countries, it is not effective in India, where the majority of the businesses have no credit score. ML-based EWS enables banking organisations to access the alternative data of the borrower. This data can include information as trivial as the applicant’s driving records or monthly subscription. Alternative data, when combined with traditional data, can help lenders look beyond the statistics and determine the true creditworthiness of a borrower.
Detecting fraudulent activities early on
It is a common practice for fraudsters to modify and present information in a way that matches the criteria set by the bank for credit disbursals. Such fraudulent activities can be easily detected by machine learning-powered EWS, making it easier for banks and financial institutions to tell genuine borrowers from the deceptive ones. Moreover, with little to no human intervention, the process can be completely error-free.
The Indian banking sector has long been reeling under the pressure of bad debts. With many financial institutes finding themselves in hot water in recent times, they are naturally wary of providing credits to MSMEs, which operate with thin revenue margins. With machine learning-driven Early Warning Systems, however, banks can not only protect their own interests but also extend their credit services beyond the metropolitans and move into India’s hinterlands. This will create a win-win situation for both the lenders and the hundreds of thousands of cash-starved MSMEs in the country.
Meghna Suryakumar is the Founder & CEO of Crediwatch a ‘Data Insights-as-a-service’ company that provides lenders, businesses with actionable credit intelligence on private entities they need to improve trust and increase their lending and trading activity. Crediwatch does this with no human intervention by deploying the latest practical AI and technology tools that provide the most reliable comprehensive real time inputs. Prior to Crediwatch, Meghna practised Law and had her own international law practice based in New York City and India. She is an alumnus of Columbia Law School and has a Certificate of Entrepreneurship from Stanford Business School. Her prior experience of finding alternate sources of data to find risk information on M&A targets led her to starting Crediwatch.
