Fair Isaac and two's not Company?

Loan-level auto ABS data surfaces a surprising finding: two borrowers may not be better than one

headshot of author
Steve McCarthy, CFA

Three's a crowd, but is two really company? Our analysis of loan-level auto ABS data challenges conventional wisdom that all else equal, loans with two borrowers should outperform those with only one. In fact, even when controlled for PTI and FICO, loans with two borrowers suffer from significantly higher delinquency rates than those with only one.

Ever 60+ rates and weighted average FICO

Loan-level data offers us new insight into the influence of a loan's credit characteristics by allowing us to tie them to the loan's subsequent performance. Here, we've plotted the weighted average FICO score of each deal in Elicient's universe with a cut-off date on or before August 1, 2017 against the first eight months of its performance.

We've represented performance as the percentage of loans that have ever been more than 60 days delinquent or suffered a loss (ever 60+). As we'd expect, deals with lower weighted average FICO scores have higher ever 60+ rates than deals with higher weighted average FICO scores.

Weighted average FICO seems to be a good predictor...

Looking at the population of 30 deals as a whole, the weighted average FICO score at origination appears to be highly predictive of the deal's performance over the first eight months.

In fact, a simple quadratic regression using only the weighted average FICO as an independent variable results in an R² of almost 87%.

...but significant performance differences remain unexplained

Still, it's clear that there are significant differences between the performance of the different shelves that cannot be explained by weighted average FICO scores alone.

For example, the AmeriCredit shelf (AMCAR) has a weighted average FICO that is more than 30 points below that of Santander's (SDART), but its default rate is less than 85% than that of Santander's.

Characteristics by shelf

Select shelves from dropdowns to compare

Among prime deals, WA FICO takes a backseat

Focusing only on deals with a weighted average FICO score greater than 720, the correlation between weighted average FICO and performance drops precipitously.

The R² of the quadratic regression limited to these deals is less than 18%.


So what's driving the performance of prime deals?

Since we are only looking at 8 months of performance data and default rates remain subdued, it's difficult to say with certainty which factors account for the variation that has been exhibited to date. Possibilities include the pool's ratio of new to used cars, or its PTI, LTV, loan term, or seasoning (among others).

One attribute that piqued our interest is the number of borrowers on a loan. From intuition and corollaries in the mortgage sector, we'd expect a loan with two borrowers to outperform one with a single borrower (all else equal). Access to loan-level data now lets us test that hypothesis.


Single borrower loans

To examine the relationship between performance and number of borrowers more closely, we grouped the relevant subset of single-borrower loans into 50 point FICO score and 1% payment-to-income buckets. Each bucket is labeled with its corresponding ever 60+ delinquency rate.

When examining only single borrower loans across all deals, it is apparent that PTI ratios and FICO scores discriminate future delinquency rates effectively. 5% PTI loans outperform 10% PTI loans within all FICO bands, and performance improves monotonically within each PTI band.

Focusing on the two extremes (the buckets in the upper-left and bottom-right corners), we can observe that 10% PTI loans with FICO scores between 450 and 499 are nearly 90 times more likely to default in the first eight months than 5% PTI loans with FICO scores between 750 and 799.

Multi-borrower loans

Relative to themselves, multi-borrower loans follow a similar pattern. Compared to their corresponding single-borrower buckets, however, multi-borrower loans appear to perform worse, with the effects being most pronounced in higher-FICO buckets.

The co-obligor effect

We can more easily compare the relative performance of multi-borrower loans to single-borrower loans by dividing the default rate of the former by that of the latter.

In buckets with FICO scores under 600, multi-borrower loan performance is generally in line with the respective single-borrower buckets: some buckets perform slightly better (up to 9% for the 500 FICO/10% PTI bucket), while others moderately worse (by as much as 25% for the 550 FICO/9% PTI bucket).

Multi-borrower performance begins to worsen in the 600-650 FICO bucket range and continues to deteriorate through the higher buckets.

In many of the buckets with FICO scores above 700, multi-borrower loans default more than twice as frequently as single borrower loans in the same bucket.


There's no obvious reason as to why loans with co-obligors seem to be less performant than those without, but one complicating factor that may help explain it is the way in which issuers compute a single FICO score for multi-borrower loans. By referencing each deal's exhibit 103 submissions, we identified three ways in which issuers seem to compute FICO scores for multi-borrower loans. A plurality of shelves (5 of 11*) use the primary borrower's FICO score only (though none elaborate on how the primary borrower is determined). Two shelves use the maximum FICO score of either borrower, while CARMX takes the average of the two. Finally, 3 shelves do not appear to disclose their methodology.

To explore the question just a little bit more, we categorized each shelf by its FICO calculation methodology, and then plotted the performance of both its single- and multi-borrower loans across FICO buckets. We also included the description each issuer provided of their methodology for calculating a single FICO score.

The pattern we observed of multi-borrowers' worsening performance relative to single borrowers' persists at the shelf-level for most issuers. We note that weighted average LTV is typically (though not always) higher for multi-borrower loans, which might explain some of the underperformance.

*We examined only loans with PTIs between 6% and 12% and FICO scores between 500 and 800. FICO buckets with fewer than 2,000 total or 750 single or multi-borrower loans that met this criteria were excluded. Three shelves (namely FITAT, USAOT, and CRART) do not appear at all because they had no buckets with enough loans that met our criteria.


Single- vs multi-borrower performance

Shelf
Reporting classification Issuer Reports Maximum Borrower Score
FICO reporting methodology narrative
Higher of applicant or co-applicant score. Business applications may be evaluated using sources other than bureau score. This field can be zero for a number of contracts, including but not limited to business loans, rebooked loans, and employee contracts

Though we can't say for sure with the data we have currently, we suspect that more times than not, the "primary borrower" is synonymous with the more creditworthy borrower on loans in pools that report the primary borrower's FICO. With that in mind, we'd expect multi-borrower loans across all three reporting methodologies (primary, maximum, or average) to have inflated FICO scores compared to single-borrower ones.


The introduction of loan-level reporting for auto ABS provides the opportunity for a deeper understanding of the underlying drivers impacting the performance of the securities to both investors and researchers alike. Investors can reduce their reliance on historical shelf performance and more confidently assess how migrations in the credit attributes of the collateral will ultimately influence their returns. The emergence of technological improvements and cultural shifts such as autonomous vehicles and the sharing economy is promising to change Americans' relationship with cars in the coming years, and a nuanced understanding of the complex credit attribute relationships at play will be critical for distinguishing newly persistent trends from mean-reverting noise.

Stay in the loop: enter your email address below and we'll let you know when we post new insights just like this one.