- Based on our analysis of a sample of residential mortgage loans from our portfolio of rated RMBS transactions, the variation in score when using a bi-merge average rather than the median of a tri-merge is small, although loan-level observations could have greater differentials.
- Further, our analysis of this sample demonstrated that the variation among credit bureaus when using a bi-merge average compared to tri-merge median is also small.
- Timelines indicate that additional data will be released to the market in the fourth quarter regarding the transition to a bi-merge score. We will review this information, as well as changes in standard market practices, and incorporate into our analyses as appropriate.
Borrower credit score--which has typically been determined for mortgage underwriting using the classic FICO methodology--is an important factor to consider when modeling U.S. residential mortgage credit risk. Last fall, S&P Global Ratings wrote about the changes proposed by the Federal Finance Housing Agency (FHFA) as to how loan-level credit scoring will be assessed by government-sponsored enterprises (GSEs) Fannie Mae and Freddie Mac (see "New FHFA-Approved Credit Scores Will Require Mortgage Industry To Transition," published Oct. 27, 2022). Since the FHFA announcement last year, the market has received additional clarity about the timeline and the nature of the proposed changes (see "Enterprise Regulatory Capital Framework [ERCF] – Commingled Securities, Multifamily Government Subsidy, Derivatives, and Other Enhancements," Federal Housing Financing Agency, published during first-quarter 2023). In addition to the logistics related to integration within parts of the mortgage sector, the main points of discussion appear to be the use of two credit bureaus instead of three to determine a borrower's effective credit score, and an updated scoring methodology incorporating newer credit score models.
According to the ERCF, the comparative analysis carried out by the FHFA showed that there was negligible difference in effective credit scores under the different methods of calculation they considered. We too examined a population of loans and arrived at a similar conclusion at the aggregate level. In this article, we further examine the differences that arise when using two credit bureaus rather than three to determine an effective credit score. In our study, we used a population of roughly 23,000 observations (each observation represented a case for which three bureau scores were available) related to a subset of U.S. residential mortgage-backed securities (RMBS) transactions we rated, with loan origination dates ranging over the past 10 years.
The use of FICO 10T and Vantage Score 4.0 could be an improvement over the classic FICO because the new methods incorporate information such as rent, utilities, and telecom payments, which could be used to infer payment behavior patterns, as well as trended data and presumably the identification of revolving versus transacting credit in order to acquire greater insight into borrower behavior patterns. According to released timelines, the migration away from classic FICO is scheduled for 2025, while the introduction of an average bi-merge score versus the normal tri-merge bureau pull in certain mortgage loan underwriting is scheduled for 2024.
The Conventional Tri-Merge Credit Scoring Method
The conventional underwriting and pricing standard for the extension of residential mortgage credit through lending to consumers has typically relied upon the middle score (the median) of three bureaus, also known as the "tri-merge." While this has been the dominant approach for several decades, some lenders may have deviated from the practice in particular instances. For example, a lender might have relied upon a single bureau in second-lien mortgage lending within the non-agency lending space before the Global Financial Crisis.
Our understanding is that the three credit bureaus--Equifax, Experian, and TransUnion--originally provided credit reporting that could have depended on geography. That is, individual bureaus would have access to different types of information at different times depending on where a loan was originated. So even though all three credit bureaus used similar FICO scoring methodologies, this heterogeneity in trade lines likely led to variability in credit scores across bureaus. While reporting asymmetry hasn't altogether disappeared (especially in instances when a credit application is pending), the increase in regional overlap (attributable to improved technology, data warehousing, and reporting capabilities) may have reduced the regional variability.
Chart 1 presents the distribution of FICO score based on our sample set of loans and broken out by the three bureaus. We have masked the identities of individual bureaus by referring to them arbitrarily as Bureaus A, B, and C.
Based on our analysis of this loan sample, there is no marked difference between the score distributions of the three bureaus; moreover, no bureau consistently stands out as being the one with the highest or the lowest score. While the FICO score assumes a range of 300-850, most of the loans we examined are clustered at the higher end of the scale, as represented by the left skew shown in chart 1. This is to be expected because the population is conditioned on borrowers that qualified for a mortgage based in part on having acceptable credit scores. When stratified by vintage (see table 1), the difference between the highest and lowest bureau score for a given borrower was about 25 to 30 points, which is certainly a wide enough margin to affect loan pricing and mortgage eligibility. It also turns out that each bureau was roughly equally likely to represent the median, on average. In other words, a given bureau would rank in the middle some 30%-40% of the time in our sample, which is what one would expect statistically, assuming the differences are due to minor random variations.
The average FICO score across vintages in table 1 should be considered cautiously because of the potential credit variability between the portfolios we examined in our analysis. We provide more detail on FICO score across the credit spectrum below. For the overall data set, Bureaus A, B, and C represented the median score of the tri-merge 34.8%, 30.6%, and 40.5% of the time, respectively.
|Summary of bureau scores by vintage|
|Vintage||Number of tri-merge borrowers in sample||% of sample Bureau A is median(i)||% of sample Bureau B is median(i)||% of sample Bureau C is median(i)||Average difference between high and low scores||Average of median scores||Average of mean scores|
|(i)Percentages within a vintage can add up to > 100% due to instances when two or more bureaus agree exactly.|
Comparison Of Tri-Merge And Bi-Merge Credit Scores
Some score variation is expected among the bureaus because of information asymmetry (discussed above) and possibly other methodological factors. To examine the potential impact this could have under the FHFA proposal related to GSE financed loans, we compared the median score obtained under a tri-merge (the current practice) with the average (i.e., the mean or, in this case, the mid-point) using only two scores (the proposed bi-merge practice). Again, the use of only two scores isn't entirely novel, as circumstances may exist for which information from only two bureaus is available (for example, a consumer may have imposed a credit pull freeze on a particular bureau, or there may be insufficient reportable trade lines to obtain a score from one of the bureaus). However, when only two bureaus are used, the practice has been to use the lower of the two scores rather than the average proposed by the FHFA. Study results from the FHFA indicate that there is only a minor variation when using the average of (bi-merge or tri-merge) scores instead of the median of three bureaus.
Table 2 shows the variation that arises from specific combinations of bureaus (i.e., pairs of Equifax, Experian, and TransUnion, in no particular order). We show the average, high, and low scores, as well as the tri-merge score, that would be obtained from our sample using either a median or an average.
|Comparison of scores for various bureau combinations|
|Bureau||Higher of two scores||Mean of two scores||Lower of two scores|
|Bureau||Tri-merge median||Tri-merge mean|
|Bureau score averages|
The difference between an average bi-merge and a median tri-merge is small (within about a point), and there was no systematic bias in terms of which was larger. Note, however, that these are aggregate statistics that do not capture the variability that might come about at a loan level within the sample. At a more granular level, we have:
- Roughly half the time, the difference between the average bi-merge and the median tri-merge was within 5 points.
- Roughly a quarter of the time, the difference was between 5 and 10 points.
- Roughly a quarter of the time, the difference was greater than 10 points.
Chart 2 provides a histogram of the differences in average bi-merge and median tri-merge from our data set for each of the three credit bureaus. The distributions appear bell-shaped and roughly symmetric about the origin. Most of the mass is concentrated within a range of 10 points of the origin, again suggesting that an average of two scores is reasonably close to the median of three.
We also stratified by classic FICO band (increments of 25 points) to determine the localized variation between the average bi-merge and the median tri-merge scores. While data were more limited at the tails (below 600 and above 825), there was greater variance at lower FICO levels. This is evident from table 3, which shows that the average difference between high and low scores in a tri-merge is about 20 points for the 800-825 bucket, whereas for the 550-575 bucket the difference is about 45 points. One explanation could be that data for certain adverse events may be restricted to fewer than three bureaus, due perhaps to specific geographies. Another possibility is that higher scores will start to compress or cluster at the maximum level. For reasons discussed earlier, there weren't many FICO scores in our data set at or below 500, even though the minimum possible score is 300.
|Score distribution based on bureau combination|
|Tri-merge||Bureau A/B||Bureau B/C||Bureau A/C|
|Score range (bin)||% in bin||Median score||Average score||Average difference between high and low scores||% in bin||Average bi-merge||% in bin||Average bi-merge||% in bin||Average bi-merge|
While the average of two scores appears to be close to a median of tri-merge at the aggregate population level, increased variation at the loan level could arise with smaller loan pools. Furthermore, as we outlined in our publication last fall ("New FHFA-Approved Credit Scores Will Require Mortgage Industry To Transition," published Oct. 27, 2022) and as mentioned in the ERCF, the use of the lower of two scores within a portfolio of mortgages will generally result in a lower score than the median of a tri-merge. As with the broader mortgage industry, S&P Global Ratings considers the lower score when information from only two bureaus is available. We will continue to assess the proposed update and consider any related data to better understand the potential impact of a transition away from the current approach of interpreting multiple credit scores.
What Portion Of The Non-Agency Market Will Transition To The New Proposal Or Some Version Of It?
It's unclear at this early stage what portion of the non-agency market will transition to the new proposal (or some version of it). Because most originators of non-agency products typically focus on agency lending as well, the systems employed by these entities may include GSE-style conventional underwriting that may carry over to other mortgage products.
The transition to the new credit scoring under the FHFA proposal involves both conceptualizing the use of a bi-merge and understanding how the new FICO 10T and Vantage 4.0 scores compare to classic FICO, which the mortgage market has been using for decades. We appreciate that the change is accompanied by a timeline for integration so the industry can adapt. Advances in technology and reporting, as well as the wealth of residential mortgage performance data available today, should coincide with updates that better predict credit risk. Also, the use of a bi-merge could encourage some degree of innovation and competition among credit bureaus.
S&P Global Ratings research
- Enterprise Regulatory Capital Framework – Commingled Securities, Multifamily Government Subsidy, Derivatives, and Other Enhancements, Federal Housing Financing Agency, Federal Register, March 13, 2023
This report does not constitute a rating action.
|Primary Credit Analysts:||Jeremy Schneider, New York + 1 (212) 438 5230;|
|Sujoy Saha, New York + 1 (212) 438 3902;|
|Secondary Contacts:||Michael J Graffeo, New York + 1 (212) 438 2680;|
|Marco Kam, Toronto +1 4165072532;|
|Research Contact:||Tom Schopflocher, New York + 1 (212) 438 6722;|
No content (including ratings, credit-related analyses and data, valuations, model, software, or other application or output therefrom) or any part thereof (Content) may be modified, reverse engineered, reproduced, or distributed in any form by any means, or stored in a database or retrieval system, without the prior written permission of Standard & Poor’s Financial Services LLC or its affiliates (collectively, S&P). The Content shall not be used for any unlawful or unauthorized purposes. S&P and any third-party providers, as well as their directors, officers, shareholders, employees, or agents (collectively S&P Parties) do not guarantee the accuracy, completeness, timeliness, or availability of the Content. S&P Parties are not responsible for any errors or omissions (negligent or otherwise), regardless of the cause, for the results obtained from the use of the Content, or for the security or maintenance of any data input by the user. The Content is provided on an “as is” basis. S&P PARTIES DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE OR USE, FREEDOM FROM BUGS, SOFTWARE ERRORS OR DEFECTS, THAT THE CONTENT’S FUNCTIONING WILL BE UNINTERRUPTED, OR THAT THE CONTENT WILL OPERATE WITH ANY SOFTWARE OR HARDWARE CONFIGURATION. In no event shall S&P Parties be liable to any party for any direct, indirect, incidental, exemplary, compensatory, punitive, special or consequential damages, costs, expenses, legal fees, or losses (including, without limitation, lost income or lost profits and opportunity costs or losses caused by negligence) in connection with any use of the Content even if advised of the possibility of such damages.
Credit-related and other analyses, including ratings, and statements in the Content are statements of opinion as of the date they are expressed and not statements of fact. S&P’s opinions, analyses, and rating acknowledgment decisions (described below) are not recommendations to purchase, hold, or sell any securities or to make any investment decisions, and do not address the suitability of any security. S&P assumes no obligation to update the Content following publication in any form or format. The Content should not be relied on and is not a substitute for the skill, judgment, and experience of the user, its management, employees, advisors, and/or clients when making investment and other business decisions. S&P does not act as a fiduciary or an investment advisor except where registered as such. While S&P has obtained information from sources it believes to be reliable, S&P does not perform an audit and undertakes no duty of due diligence or independent verification of any information it receives. Rating-related publications may be published for a variety of reasons that are not necessarily dependent on action by rating committees, including, but not limited to, the publication of a periodic update on a credit rating and related analyses.
To the extent that regulatory authorities allow a rating agency to acknowledge in one jurisdiction a rating issued in another jurisdiction for certain regulatory purposes, S&P reserves the right to assign, withdraw, or suspend such acknowledgement at any time and in its sole discretion. S&P Parties disclaim any duty whatsoever arising out of the assignment, withdrawal, or suspension of an acknowledgment as well as any liability for any damage alleged to have been suffered on account thereof.
S&P keeps certain activities of its business units separate from each other in order to preserve the independence and objectivity of their respective activities. As a result, certain business units of S&P may have information that is not available to other S&P business units. S&P has established policies and procedures to maintain the confidentiality of certain nonpublic information received in connection with each analytical process.
S&P may receive compensation for its ratings and certain analyses, normally from issuers or underwriters of securities or from obligors. S&P reserves the right to disseminate its opinions and analyses. S&P's public ratings and analyses are made available on its Web sites, www.spglobal.com/ratings (free of charge), and www.ratingsdirect.com (subscription), and may be distributed through other means, including via S&P publications and third-party redistributors. Additional information about our ratings fees is available at www.spglobal.com/usratingsfees.