Menu

CRO Outlook

Multiple Credit Scores in Mortgage Lending: Understanding the Risks

As policymakers consider allowing the use of multiple credit scores in mortgage underwriting, banks need to be alert to risks to lenders and consumers.

Friday, August 26, 2022

By Clifford Rossi

Advertisement

Expanding access to credit for millions of consumers with limited or no credit experience has been a long sought-after goal among policymakers and consumer advocates. Doing so in a manner that balances market access with effective credit risk management is critical for credit investors and consumers alike.

Over time, credit scores have become a key measure of a borrower’s creditworthiness – thanks in large part to their objectivity, consistency and statistical credibility. Consequently, they have found widespread use among credit providers, insurers, and other market segments.

The foundation of effective credit risk management is sound underwriting practices, which are driven by statistically-based automated underwriting scoring (AUS) systems – at least, for most consumer lending products. These scorecards rely heavily on borrower credit history, leveraging a credit score (from the likes of FICO or VantageScore) and/or detailed credit attributes. However, they are only as effective as the credit data on which they are built.

clifford_rossi-1Clifford Rossi

Without comprehensive and accurate data, it’s hard to gauge the soundness of a single or multi-AUS approach – particularly with respect to expansion of credit or the impact of a given strategy on risk management. But this need for comprehensive and accurate data is problematic, because of the relative inadequacy of information that is actually available for a certain segment of the borrower population.

Before we take a closer look at the potential impact of alternative credit-scoring models, it’s helpful for us to first understand the evolution of credit scoring.

Expanding Access to Credit, and Its Implications for Credit Scoring

Through the years, FICO – which requires borrowers to have at least six months of credit history and an account update within the last six months – has been the dominant provider of credit scores. But alternatives to FICO have emerged, largely because of credit-scoring data problems.

Roughly six years ago, in the aftermath of a detailed study of U.S. credit records, the Consumer Financial Protection Bureau (CFPB) deemed 26 million individuals so-called “credit invisibles” – or people without any credit record. What’s more, an additional 19 million were determined to be unscorable by credit-scoring models, because of their insufficient credit history.

VantageScore then jumped into the fray, via developing a model that could effectively score a large segment of the consumer market that had been previously viewed as unscorable.

The entrance of VantageScore expanded credit access and disrupted the way credit scores were developed and used across consumer lending markets. Importantly, it also rekindled the multiple-credit-score discussion within the mortgage industry, particularly regarding the two largest mortgage credit investors on the planet: Fannie Mae and Freddie Mac.

The multi-score discussion gained even more traction earlier this year, when the FHFA announced it was considering four different options for using credit scores in the AUS processes of both GSEs – including the delivery of a single score, the delivery of multiple scores (e.g., from both FICO and VantageScore), the delivery of one score via a lender’s choice, and a waterfall approach that would establish a primary credit score and a secondary credit score. Whichever option the FHFA ultimately chooses will have significant implications for consumers and credit investors alike.

Multi-Scores and Mortgage Underwiring: An Empirical Analysis

In an automated underwriting environment when credit decisions are made in seconds, credit investors and their lender partners must have confidence in the statistical accuracy of their scores.

To understand the effects of alternative credit scoring models, I recently conducted an empirical analysis of the potential impact of mortgage scores developed by two hypothetical mortgage score providers on credit risk, profitability and effectiveness at expanding access to credit.

This analysis leveraged a large loan-level sample of GSE loans originated between 1999-2015. To approximate how credit scores built on different consumer account segments might impact credit, financial and model performance, three mortgage scores were developed.

Two (Score 1 and Score 2, developed by different score providers) were built using mortgage borrowers with a typical credit profile, reflected by loans with debt-to-income (DTI) ratios less than or equal to 43%. Another score (Score 3) was developed on a segment of “stretch” borrowers with DTIs above 43%.

All scores were calibrated to the same 300-850 score range, and score performance aligned with the baseline 1999-2004 origination period.

Even though mortgage scores developed from typical and stretch samples can look and feel the same in terms of score range (and even initial score performance), significant differences between Score 3 and Scores 1 and 2 were apparent. The mortgage score built on loans for borrowers with >43% DTI exhibited model performance that was substantially inferior to the scores developed from loans with <=43% DTI.

Indeed, Score 3 loans (those based on borrowers with >43% DTI) resulted in more model errors, higher credit risk and lower profitability. What’s more, the same credit policy cutoff used in optimizing profitability for Score 1 and Score 2 would destroy shareholder value for Score 3 – which also had limited value in actually increasing access to credit for the stretch borrower segment.

Parting Thoughts

As our analysis demonstrates, contrary to the belief of some “experts,” credit-scoring systems are not interchangeable methods for expanding access to credit. While a credit score may be minimally viable for loan underwriting, it may also expose credit investors to risk from higher model errors.

Indeed, given the paucity of data available for borrowers with either no credit history or an extremely limited history, developing statistically-reliable scoring models for so-called unscorables is extremely challenging – and potentially exposes credit investors to greater risk from model miscalculations. Credit investors and policymakers should therefore exercise great caution in assuming the interchangeability of credit scores as a mechanism to expand access to credit.

The problem is that on the surface, multiple scores – even if presented on the same score continuum – are not the same from a risk perspective. That was clearly and unequivocally demonstrated in the aforementioned empirical analysis, which showed that, over time, credit scores tend to diverge (and yield different levels of risk) as economic conditions change.

The best way of ensuring credit risk is managed prudently, and that access to credit is expanded, is to choose scoring systems that are highly effective in distinguishing between good and bad loans. But we still need to be careful in our approach, particularly regarding the deployment of multiple credit scores in an AUS – because the use of multiple scores creates operational challenges in business and risk systems, potentially diluting the predictive performance of models.

Credit investors need to perform their own analysis of alternative scores to mitigate third-party risk and to more effectively manage credit risk. Before making any decisions about credit-scoring systems, they must carefully research and confirm the claims of credit-scoring providers.

Clifford Rossi (PhD) is a Professor-of-the-Practice and Executive-in-Residence at the Robert H. Smith School of Business, University of Maryland. Before joining academia, he spent 25-plus years in the financial sector, as both a C-level risk executive at several top financial institutions and a federal banking regulator. He is the former managing director and CRO of Citigroup’s Consumer Lending Group.




Advertisement

BylawsCode of ConductPrivacy NoticeTerms of Use © 2022 Global Association of Risk Professionals