Lenders To Re-Calibrate Algo-Based Loan Underwriting After RBI Governor Flags Caution
A week since Das raised his conerns, lenders are attempting to re-calibrate their underwriting algorithms
![<div class="paragraphs"><p>RBI signage at its headquarters in Mumbai (Source: Vijay Sartape/ BQ Prime)</p></div>](https://media.assettype.com/bloombergquint%2F2023-04%2F0d07fa3b-162d-41a4-904b-42d9e514ba09%2Frbi.jpg?auto=format%2Ccompress&w=200)
The algo-based loan underwriting processes by lenders are set to undergo changes after the Reserve Bank of India Governor raised caution. Experts believe that machines have a clear disadvantage over human oversight when reviewing loan profiles.
“When we outsource our brains to math, it comes with a price and caution,” said Ari Lehavi, group business head of predictive analytics at Moody's Analytics, in a conversation with BQ Prime.
The model, aided by analytics, is a significant governance issue on the part of lenders, which is why it is being highlighted, according to Vivek Iyer, partner-financial services advisory at Grant Thornton Bharat.
What Drives The Problem?
Colloquially called algo-based lending, it is the practice of using mathematical models and algorithms to assess the creditworthiness of a borrower. Based on the borrower’s behaviour and payment patterns, the algorithm approves or disapproves an instant loan without any human intervention.
The idea behind this is to automate the loan approval process for a faster turnaround, according to Anurag Jain, an executive committee member at the Digital Lenders Association of India. It also makes the lending process much easier as there is not much information to be reviewed, he added.
Hence, small-ticket loans get approved faster, Jain said. And that’s where the problem builds.
"When lenders give a bigger-ticket loan, more due diligence is done as more money is at stake. But here, you’re giving sachet loans, as the idea is to get more credit customers under your fold," he said.
Virat Diwanji, group president and head of consumer banking at Kotak Mahindra Bank, agrees.
"The issue has more to do with loans below Rs 50,000, and the cost of recovering those loans is prohibitive as we have to deploy people to collect money from you. These are the loans that usually get into trouble," he said.
In FY23, the fintech industry disbursed loans worth Rs 92,267 crore, up 21% year-on-year, as compared with Rs 76,396 crore in FY22, according to a report published by the Fintech Association for Consumer Empowerment and Equifax.
Here, the share of small-duration loans—less than six months—stands out as it touched 88% of the disbursement volume, compared with 63% in FY22.
But an increased roll-out of small-ticket loans also poses the risk of more delinquencies as they are not backed by data on borrower's repayment capacity, according to Iyer.
One of the reasons behind this growth is an increase in banks' appetite for riskier loans.
"The balance sheets are clean and stress is lessened, and as a result, there was a desire to increase the pace of credit. When this happens, you usually have more relaxed underwriting requirements," Iyer said.
Moreover, since the interest rates are high, lenders are trying to jump volume by aggressively pushing products as market conditions are also good, Iyer explained.
A senior banker at a private lender agrees.
The asset quality of banks has historically been the best. With the gross non-performing asset ratio at around 3.9%, it is only looking to go down further, this banker said.
According to TransUnion CIBIL's Oct. 2023 report, consumption-led products except credit cards continued to show double-digit year-on-year growth in Q2 FY23, indicating a strong demand and supply for this cohort.
What Changes?
With more loans being judged in an automatic fashion, it is important for lenders to exercise discipline, do stress-testing and understand the pockets where there is risk, according to Lehavi of Moody’s Analytics.
“It is also important to be careful, apply human judgement at some point, and not blindly follow the algorithm,” he said.
Market participants, too, are attempting to follow suit.
"When someone says to be careful, we usually go back and evaluate our model again. This means that we put the data of deals that have gone through and those that haven’t and evaluate the model to see if it needs more tightening as the variables are far too many," Diwanji of Kotak Mahindra Bank said.
There will be a course correction, according to the senior banker quoted above. The bank will figure out what is good and what isn't and refine accordingly, the person said.
From an industry perspective, it is important to re-calibrate these models based on new risk information that comes in, according to Adhil Shetty, co-founder of Bank Bazaar. Hence, it is important to constantly evaluate the data as per the best practices of AI to make sure there is no bias created.
"Ideally, have a human audit element as well so that any signs of caution can be checked,” he said.
It is also important that companies tweak their models every six months, as the issue is a lack of monitoring and updating them, suggests Anil Pinapala, founder and chief executive of non-bank lender Vivifi India Finance.
Jinay Gala, associate director at India Ratings & Research, said that a medium-term slowdown in credit is expected. This is because many banks and NBFCs are using partnerships with fintechs as a tool to protect their margins in a competitive environment, he added.
"To manage margins in a rising interest rate scenario, unsecured lending provided that delta. So, they may go slow for some time, but they can't avoid it," he said.
Lehavi of Moody’s Analytics, too, explained that lenders may start working towards the identification of riskier profiles by taking a cautious approach.
A lot of lenders will also start looking at these models and periodically reviewing them, according to Iyer. This will lead to tighter underwriting criteria, he added.