COVID-19 changed credit modeling forever — here's why

The full magnitude of COVID-19 on consumer credit is still incalculable, but one thing those focused on consumer credit markets know for sure — the pandemic will change credit modeling forever.

This week, strategic advisory firm, Oliver Wyman convened an excellent panel of credit risk executives to discuss COVID-19’s impact on consumer financial services. Many of their insights aligned with what we’ve seen at , where we’ve been using machine learning to monitor consumer credit behavior as the pandemic hit and then spread. Here’s a summary of the emerging consensus among credit risk professionals about COVID-19’s effect on consumer loan underwriting:

‍

1. To effectively originate and restructure loans in today’s world, lenders need to recalibrate their risk scores — even dramatic adjustments to business-as-usual models and policies may not be enough

Virtually no one had stress-tested their credit models for a downturn of this severity and speed, with unemployment levels spiking to 14 percent and GDP dropping five precent in a matter of weeks.

Many forecasts expect economic activity to remain dramatically subdued even after we bend the COVID-19 curve. Given how fast the deterioration came, much of the resulting consumer credit distress has yet to show up in credit bureau data and many loans now in forbearance are still being reported as current by lenders. Even once consumer credit distress is more accurately represented in the data, the model development and governance processes inside most financial institutions are not equipped with the speed needed to adapt to what’s happening. For this reason, lenders without sophisticated model monitors ought to treat their current model outputs with a high degree of skepticism for the foreseeable future.

‍

2. Lenders who develop the ability to better predict borrower risk in turbulent times will take more market share and — in the long run — enjoy higher growth, profitability, and customer loyalty

Most lenders are unable to finely parse risk in the event of global economic crises, particularly ones of this magnitude. As a result, we’ve seen many lenders stop originating entirely as they wait for recovery. This situation creates an opportunity for those lenders who use superior analytics and strong liquidity to gain market share by maintaining accuracy and staying agile no matter the market conditions.

There are some obvious actions lenders can take to get a better handle on credit risk. These include: (a) Get more rigorous about income and employment verification through increased reliance on checking account data from vendors like Yodlee and Plaid and products like Equifax’s The Work Number; (b) focus on customers with savings and not those who live paycheck to paycheck; and (c) for auto and other asset-backed lenders, adjust your LTV ratios.

3. The hardest task facing lenders today is determining whether and how to restructure loans in forbearance or default

Sorting the borrowers who’ve hit temporary economic stress from those who will never recover is very hard, in part because the data needed to validate most predictions isn’t available.

Additionally, most lenders are set up to manage portfolios with low levels of defaults and lack the automation and rigorous analytics to quickly enhance their loss mitigation strategies when defaults surge. A typical call-center operation staffed by humans doing manual account reviews is unlikely to efficiently and effectively determine which borrowers deserve workouts and which don’t — lenders will need machine learning and automation to efficiently restructure defaulted loans at scale.

‍

4. Consumers are likely to adjust their payment hierarchies

During the 2008 financial crisis, many borrowers were underwater on their homes, so for many it made financial sense to default on mortgages but continue to make auto and credit card payments.

In the current crisis, borrowers are sheltering-in-place at home and not driving as much, so mortgage payments might get prioritized over auto. Credit card payments are also likely to be prioritized over auto payments since most borrowers need to shop for food and supplies while they stay home.

‍

5. Going forward, credit models will need to incorporate variables they haven’t before

Together with the right math, more variables mean more accuracy and resilience in models. In a pandemic-prone world, important new data sources for credit models will include:  

• Checking account and other payment transaction data  
• Unemployment forecasts (including possibly by geography and industry, though these types of variables can trigger fair lending issues)
• COVID-19 transmission intensity rates (though once again, we’ll need to watch out for fair lending concerns — data shows that that minorities are more adversely affected by COVID, so variables like this could conceivably raise red flags)
• Variables related to other potential low-probability, high-impact events like climate risk

‍

6. Lenders who can get AI credit models live quickly will have an edge

The days of credit models that use 15-20 variables, rely on elementary math and take over a year to get into production are long gone.

The only way for lenders to thrive in this credit market is to use more data, better math, and tools that get models live quickly. Lenders who’ve automated key parts of the model development, validation, documentation and governance process will be able to react to changes in the marketplace faster than those who haven’t. Meanwhile, advanced fair-lending testing and multivariable model monitoring will become the norm. AI tools will be needed to monitor input and output distributions, feature drift, anomalies and outliers, as well as other fairness measures like approval rate ratios in real-time.

Those of us who work in credit risk don't know the long term effect of the pandemic on our workplaces, but the effects of COVID-19 on our jobs will be permanent. For lenders, a new era of credit modeling has already begun. As difficult as these times may be, COVID-19 presents lenders with an opportunity to accelerate their digital transformation and emerge well-positioned to win the future of consumer finance.