ACTIONS LENDERS CAN TAKE TO CONVERT UNCERTAINTY INTO RISK
Uncertainty will paralyze a company. Doing nothing, or worse, cutting back due to economic uncertainty, is typically the response when there’s a lack of confidence. However, we all know you can’t shrink yourself to growth. In order for companies to grow, they must continue serving customers and generating new products. So what do you do?
We think it begins with converting uncertainty into risk. You can price risk. You can act on risk. You freeze with uncertainty.
HOW TO LEND WITH CONFIDENCE
The question becomes how to make good decisions and take effective actions. Banks and financial services companies have significant support structures, but in times of disruption, those data, tools and assessments related to consumers may not be accurate. What can you do if the instrumentation is broken?
Start by evaluating tools and strategies that can be made readily available to make better decisions. As soon as the COVID-19 crisis began, our reporting revealed that lenders attempted to reduce their exposure by tightening their credit policies and approving fewer consumers and small businesses; that was a prudent, but blunt response. In this instance, here is the framework we use to support the actions taken with models and business strategies [Figure 1].
Figure 1: Risk Strategy and Model Changes
Here’s an alternative path. We believe loan originators can add an attribute or flag overlay on top of an existing strategy criteria. Over time, as more customer data is accumulated, lenders can keep their models. But they should adjust score bin risk levels based on the most recent customer performance and rebaseline their strategies. Eventually, models will be refit and rebuilt with new features that maximize consumer and lender value.
Identify the best attributes and flags
If you go with an overlay approach on existing models and scores, it’s critical to identify the best attributes and flags. Continuing with our COVID-19 use case, below is a method to get started.
- Generate a business hypothesis regarding what behaviors best capture the economic impacts of COVID-19.
- Then, leverage machine learning tools to complement it and identify additional features that may be worth a second look.
- Next, empirically validate the data through the use of historical samples.
Natural Disasters Provide Benchmarks
In my last blog, I wrote how recent natural disasters such as Hurricane Harvey, Superstorm Sandy and the California wildfires in 2018 provide several benchmark samples that share similarities with the current U.S. economic environment. All three disasters resulted in significant employment changes, as well as a form of private-sector funding (insurance checks), government support and stimulus that aided citizens.
While COVID-19 is a nationwide, exogenous event and the level of government stimulus is unprecedented, directional impact can be discerned from analytic solutions with respect to lending strategies.
Here’s how we did it:
CHOOSING THE RIGHT DATA
The massive fluctuations in employment and income reductions were the first place we turned when performing the analysis.
Why? Three primary reasons:
- Credit scores typically do not have income and employment information.
- The fact that nearly 50 million U.S. consumers have filed for unemployment since the first week in March is the underlying risk factor that lenders need to understand when extending credit.
- This information is actionable, as income and employment attributes and Automated Program Interfaces (API) can easily be leveraged as an overlay.
While we know employment and income information are valuable, our team wanted to understand what the data could disclose about consumers under duress. Since credit risk managers are tasked with taking quick effective action, we will share two of the most straightforward tools.
THE 1st TOOL: SEGMENTATION OVERLAYS
A segmentation overlay tool is typically labeled “the dual matrix.” Figure 2 shows how we segmented the VantageScore 3.0 credit score into five bands for consumers not impacted by Hurricane Harvey in 2017. Herein lies the value of the dual matrix: it is capable of further segmentation of the population to find pockets of opportunity and action.
The Near Prime consumers have a 90-day delinquency rate of 14 percent. When matrixing five VantageScore 3.0 credit score bands with five consumer income bands from The Work Number® (Equifax Workforce Solutions income and employment data), the high income and low credit score consumers have a 90-day delinquency rate of 9.2 percent. That is 34 percent lower than the population average. Lenders using a credit score along with The Work Number can continue lending to near-prime customers even when credit scores are below risk assessment thresholds.
To evaluate and benchmark a dual matrix, we evaluate the delinquency rate of the two corners—the low income and low scores (Lows) and the high income and high scores (Highs). Then, we evaluated the delinquency difference of those cells. The greater the difference, the more differentiation and value the attribute provided as a variable overlay. The ratio for the Lows to Highs is 80 (51.40/0.64). Remember that multiple because we will come back to it later.
Figure 2: Dual Matrix for Consumers Not Impacted by Hurricane Harvey
The hypothesis is that the income and employment data would be more informative and valuable during a natural disaster. People who have higher income tend to be more financially resilient to the economic stress of a flood or fire. As a result, we conducted the same cross-matrix analysis of The Work Number income ranges with VantageScore 3.0 on consumers impacted by Hurricane Harvey. We identified impacted consumers by evaluating those receiving government or lending accommodations, such as payment relief and forbearance. Figure 3 displays the dual matrix for consumers impacted by Hurricane Harvey in Houston, Texas. The Highs had a 90+ delinquency rate of 0.76 percent, while the Lows delinquency rate was 66.52 percent, yielding a multiple of 87.5. The multiple increase from 80 to 87.5 signals that income information was better able to differentiate among those consumers who could adequately navigate paying their credit obligations. These results are promising, and we activated this insight as an overlay.
Figure 3: Dual Matrix for Consumers Impacted by Hurricane Harvey
THE 2nd TOOL: DECISION TREES
We built a decision tree [Figure 4] using The Work Number income, VantageScore 3.0 and Debt-to-Income score. Decision trees are designed to evaluate the significance of each criteria that lenders are evaluating. They are easy to understand for business and compliance, and operationally simple to implement. This makes them a highly effective tool for creating an attribute overlay solution. Figure 4: Decision Tree Looking at the leaves of the tree, you see prospective actions the lender can take for existing consumers based on score and income ranges. The decision tree enables lenders to more confidently lend to consumers, while providing relief to those whose credit profile is severely challenged. A good example is consumers with credit scores between 690 and 738. If layoffs or furloughs don't reduce their income by greater than 25 percent, they would be eligible for a credit line increase. Alternatively, lenders could cross-sell a standard credit card if they don’t have a card. For those consumers with larger negative income changes, proactive accommodation would be warranted to help them through the crisis.
ONE SIZE DOES NOT FIT ALL
Far too many lenders have been using blunt instruments, as evidenced by the reduction in loan originations since COVID-19 began. However, if we embrace new tools, data and analytics, we can carve out lending opportunities for consumers to get the products they need and deserve. One size never fits all, and as an industry we can do better. At Equifax, we’re passionately looking for solutions so consumers can live their financial best in these difficult times.
In my next blog post, we’ll go deeper into debt-to-income ratios. I think you’ll find the insights quite interesting. In the meantime, take a look at our past blogs in this series: