bg-arrow-down icon-arrow-up icon-back-to-top icon-linkedin icon-menu icon-search icon-twitter logo-white slider-arrow-left-gray slider-arrow-left slider-arrow-right-gray slider-arrow-right

Consumer Credit Conundrum: The Benefit of New Deposit Metrics in Assessing Credit Risk

Previous
Other Articles
Next

A slew of federal and state actions aimed at easing the financial burden on Americans amid the COVID-19 pandemic has upended traditional metrics that the financial-services industry relies upon to assess credit performance.

The upheaval is accelerating the need for banks to expand the data and develop improved methodologies to analyze creditworthiness, including evaluating deposit and cash-flow behavior. This will provide a significant advantage to lenders that hold the core deposit and spending accounts of their customers.

PANDEMIC ASSISTANCE CLOUDS CREDIT VIEW

As a result of these actions, U.S. consumer average credit scores now stand at record levels — a stark contradiction to the economic tumult blanketing the country.

Consumer and small business lenders overwhelmingly depend on credit scoring, supplemented with other indicative metrics, to statistically assess applicant and customer creditworthiness. Traditional methods incorporate several of the “5Cs” of lending (including character, capacity, capital) though empirical modeling of behavior — sometimes with overlays for collateral and conditions.

Foundational to this statistical approach is an assumption that data used when calculating a score mean the same thing they did when the score was initially developed. That assumption no longer holds. The unprecedented scale of mandated and voluntary forbearance and loan modifications from the CARES Act, as well as stimulus payments and enhanced unemployment benefits, have fundamentally changed the context of credit data throughout the pandemic.

The magnitude of data dislocation for consumer lenders is unprecedented. Four million home mortgages, more than 40 million student loans and tens of millions of credit card, personal and auto loans are in forbearance or other special payment arrangements. The CARES Act stimulus and unemployment provisions are providing hundreds of billions of dollars of direct financial support to individuals, some of which are aiding payment of loans and providing many individuals temporarily higher incomes than before.

The effects of these changes are opaque in customary credit bureau data, which are the source of most credit score attributes. The CARES Act amended FCRA reporting requirements to stipulate any account that has a payment accommodation be reported to credit bureaus as current — provided the account was current when the accommodation was made. Furthermore, all elements of the reported trade, like scheduled monthly payment and amount past due, must be updated to accurately reflect the reported delinquency level.

The disconnect among various traditional metrics of credit health also can be found elsewhere. (See Figure 1.) CoreLogic reports a sharp year-over-year increase in individuals who are 90+ days delinquent on their home mortgages (4.1% compared with 1.3% a year ago), yet Equifax reports mortgage delinquencies are declining. Unemployment reached stunningly-high levels in 2020, but personal bankruptcy fillings fell 22%. Beremberg Economics found Americans saved a significant portion of their first stimulus payments; meanwhile, Novantas’ Weekly Deposit Tracker has found that individuals who received their second stimulus payment quickly drew much of it down.

The effects of these changes are opaque in customary credit bureau data, which are the source of most credit score attributes. The CARES Act amended FCRA reporting requirements to stipulate any account that has a payment accommodation be reported to credit bureaus as current — provided the account was current when the accommodation was made. Furthermore, all elements of the reported trade, like scheduled monthly payment and amount past due, must be updated to accurately reflect the reported delinquency level.

As a result of these actions, U.S. consumer average credit scores now stand at record levels — a stark contradiction to the economic tumult blanketing the country.

Figure 1: While Unemployment Has Skyrocketed, Reported Delinquencies are Down and Average FICO Scores are Up

COVID-19 Impact on Unemployment and FICO Scores

Source 1: Equifax Market Pulse Webinar on October 8 (page 34), Equifax Credit Trends Monthly
Source 2: FICO Blog: “Average U.S. FICO Score at 711, But Uncertainty Abounds”
Source 3: Federal Reserve Bank of St. Louis – Federal Reserve Economic Data

As a result of these actions, U.S. consumer average credit scores now stand at record levels — a stark contradiction to the economic tumult blanketing the country.

While it is clear that these programs have helped many Americans meet their immediate financial obligations, the relief has obscured loan performance and payment history data, putting reliance on this data in doubt. While it is simply too early to tell how much or in what ways scores and other credit decisioning tools are being affected, it is naive to assume it won’t be material. Indeed, the developments are already complicating new loan approval and service analytics.

Some may argue credit scores should continue to work as expected because score distributions are stable. But the current environment embraces measures of non-payment that don’t reflect “true” non-payment. Furthermore, there is no assurance that loans that are now being paid with the assistance of stimulus and unemployment funds will continue be paid in the future when those funds dry up.

With this unprecedented confluence of data challenges, credit managers need to find alternative ways to assess credit risk. An enhanced focus on employment consistency and verification, greater attention to persistence of income, more conservative loan-to-value levels and constructing novel credit bureau attributes are all worth considering.

TACKLING THE PROBLEM

For organizations with deep customer relationships, views of behaviors apart from previous loan performance may be among the most promising. Savings and personal cash flow can provide an alternative window into a customer’s situation that is apart from and somewhat orthogonal to loan performance. While patterns of deposit, savings and cash flows may have been affected by the pandemic, recording of these metrics hasn’t been distorted and relationships to traditional delinquency and charge-off measures are more likely to remain unaffected.

Risk functions historically have focused on credit behavior attributes because they had direct information value. Traditional metrics are currently capturing credit performance in the light of recent public policy actions – the two are inseparable. Analyses of deposit behaviors, including deposit balance trends and transactions, may enable credit managers to separate these effects. Initial high-level evaluation from Novantas, a data and analytics company that tracks more than $2 trillion of consumer deposits, sug­gests deposit metrics are indeed likely to prove valuable for assessing credit risk. Specifically, the history of a customer’s deposit holdings, the amount, the stability and the distribution across products appear indicative of a customer’s ability to weather financial stresses and strains.

Using a combination of uniquely-constructed deposit and cash-flow metrics, Novantas identified several attributes that demonstrated statistically-significant discrimination among mortgage holders who were delinquent and those in good standing. (See Figure 2.) The research also reinforces the importance of studying and evaluating alternative ways to construct metrics that measure deposit levels and cash flows. For example, Novantas researchers identified a statistically-meaningful difference in the ratio of discretionary liquid balance (savings not needed for immediate expenses) and cash flow to predict mortgage performance six months out. Discretionary liquid balance alone, however, didn’t exhibit a statistically- meaningful difference. This exercise demonstrates how a set of deposit metrics can drive information value for credit applications.

The specific characteristics that may be strongly predictive of future credit performance may vary from product to product, portfolio to portfolio and lender to lender depending upon the unique circumstances of lenders’ markets, policies and prospect/customer demographics.

There is further evidence that this approach works. An empirical study by FinRegLab, a non-profit research organization, determined that “the cash-flow metrics were predictive of credit risk across the diverse set of providers, populations and products studied.” The research was based on loan-level data provided by six non-banks — primarily fintech lenders — and considered cash-flow metrics alone and in combination with traditional risk metrics to construct risk models. Using standard measures of model performance, the cash-flow variables and cash-flow-only models “suggest a relatively robust ability to predict likelihood of default independent of…traditional metrics,” the study found.

Figure 2: Customers Who Continue to Pay Their Mortgages Have, on Average, a Much Better Ratio of Cash Flow-to-Discretionary Liquid Balance

Customer Deposits (02/20) vs. Mortgage Status (08/20)

Source: Disguised Novantas dataset of client mortgages and deposits

When risk scores are working, the bar is high because risk scores are known to be highly predictive, stable over time and cost effective. But these aren’t ordinary times.

Using cash flow to make credit decisions isn’t new to the industry as a whole– it has been used extensively in corporate lending decisions. A growing number of fintech lenders are already using data aggregators to collect and use consumer deposit and spending data in underwriting. But traditional banks typically don’t mine deposit data in consumer credit decisions. And if they do, they tend to use simple ratios or raw data transformations that may limit its information value.

More broadly, efforts to incorporate cash-flow data into risk scores have accelerated as evidenced by UltraFICO, Fair Isaacs’s risk score that includes cash flow attributes. There have also been efforts by Experian and Equifax to pull in cash-flow data from data aggregators like Finicity, Envestnet/Yodlee and others.

These approaches have their own set of challenges. For example, a consumer must provide deposit and spending information to a data aggregator and also grant permission to the lender to access and use the consolidated data — a high hurdle for consumers who are concerned about privacy and cybersecurity threats. Banks that hold a consumer’s core banking relationship have a leg up in this area because they already have a broad view of a customer’s cash flow. This advantage has been underleveraged by many banks; the current challenges to traditional underwriting information make advancing this data opportunity more urgent.

Ordinarily, a risk manager will ask a key question when evaluating additional sources of predictive risk: how much does the new information add incrementally to what is already known and is the marginal cost of the new data less than the marginal benefit to predictive power? When risk scores are working, the bar is high because risk scores are known to be highly predictive, stable over time and cost effective.

But these aren’t ordinary times. Today, with concerns about the reliability of risk scores, the bar to considering alternative data is considerably lower. And the urgency for banks to incorporate this data is far greater.

Alan Schiffres is President of Avalon Group Advisors, LLC, which is focused on consumer and small business credit risk policy, analytics and management. Mr. Schiffres is a paid advisor to Novantas.

A New Strategy For Tackling The Credit Disconnect

By Hank Israel and Don Kumka

Novantas recently evaluated thousands of loans at origination to develop a customer credit scorecard that only uses deposit analytics to gauge the risk of default. The scorecard, which uses a database of roughly one-third of U.S. deposit accounts that goes back more than a decade, captures underlying consumer financial behavior to expand a lender’s understanding of current borrowers and prospects.

Because the scorecard captures “thin file” customers with little credit history, lenders who have a deposit relationship with these customers can expand into a credit relationship with a greater degree of comfort.

The score includes 12 months of deposit activity prior to a credit-account opening and predicts the 90-day delinquency rate in the total population of the evaluated unsecured line and loan portfolios. (See Figure 1.) In rank ordering the probability of 90-day delinquency, lenders have opportunities to approve applicants who would otherwise fail if a traditional score was used. They can also limit lines or decline consumers whose traditional credit scores may not adequately reflect their ability to repay the loan.

The Novantas scorecard provides a powerful rank ordering of risk in the approved loan pool independently of traditional score or attribute data. Even more valuable is that the deposit score can capture and explain broader financial behaviors and can discriminate within traditional consumer credit score bands. (See Figure 2.) Across prime bands (680-740), Novantas discriminates six-to-eightfold difference in default rate from the highest- to lowest-performing customers in each 20-point band.

Leveraging the deposit scorecard to approve unsecured credit, the model can drive 15% lift in annual value over current practices, across the following actions.

Figure 1: The Novantas Deposit Prototype Score for Underwriting has Similar Results as Traditional Mechanisms

Deposit Prototype Score for Underwriting

Traditional Credit Score (Origination)

Deposit Prototype Score for Underwriting

Traditional Credit Score (Origination)

Source: Disguised Novantas dataset of client mortgages and deposits

Figure 2: The Novantas Protoype Represents Incremental Discrimination to Traditional Credit Scores with Broader Financial Behaviors

20
40
60
80
100

Source: Novantas disguised client data (Personal Unsecured Loans and Lines)
Bad rate: if the account is ever 60 days past due over next 12 months
Good rate: never more than five days past due over next 12 months

These deposit analytic tools can have an immediate impact on credit decisions in the following ways:

  1. Underwriting: Enhance underwriting policy and/or scorecards to incorporate the most powerful features from deposit underwriting to advance the lender’s information advantage in approvals, the size of the credit lines and/or pricing.
  2. Pre-Screen: Leverage the data to more accurately pre-screen consumers for offers. This will be less expensive than traditional pre-screen services and will provide an advantage over competitors.
  3. Credit Management: Use the data to manage lines and pre-delinquency engagement more accurately.
  4. Portfolio Forecasting: Credit managers can assess more accurately the risk in their current portfolios independent of accommodation and delinquency, providing more accurate insight to inform forward-looking credit performance.
  5. Collections Management: Prioritize collections treatments based on insights to drive economic actions and improve visibility and outcome of accommodation programs.
don kumka

Don Kumka

Director, New York
dkumka@novantas.com

Previous
Other Articles
Next
Subscribe

Stay up to date on the latest banking news