Will Bankers & the Financial Industry Ever Learn?
The short answer is probably “no.” The longer answer is probably also “no” and a bit more frustrating.
By Willie Costa & Vinh Vuong
Credit is one of the most notorious double-edged swords in finance: it’s a necessity for an overwhelming majority of Americans, but an unfortunate coincidence of unwise decisions and factors outside of one’s control can quickly turn debt into a burden, as we will see in this examination of a group of subprime obligors whose loans were securitized and sold off to investors. If that sounds like 2008, keep reading — when it comes to these loans, the devil truly is in the details. And while we are not in any way passing judgment on anyone, there are moments when one can be forgiven for thinking the system is rigged against them. It’s an old saw that the only people who consistently make money in finance are the bankers who facilitate and underwrite the deals, but only a select few could conjure the financial alchemy necessary to make money by selling loans that customers can’t repay, which then become part of a rock-solid investment.
A recent Bloomberg article outlines an example of how this scam strategy works: Banco Santander SA’s US unit bundles a host of subprime auto loans together — over 75,000 to be precise — and used this pool of debt to create an asset-backed security called Drive2019–3 (those who wish to follow along can pull the Drive Auto Receivables Trust 2019–3 data from EDGAR; the data analysis in this article is from a subset of approximately 34,000 subprime loans from the Drive Trust). Despite the likely comparisons to The Big Short, bundling and reselling loans as asset-backed bonds, called securitization, is a well-accepted practice on Wall Street. If done properly, securitization presents a massive opportunity for returns with minimal risk: Capital Group and BNY Mellon, for instance, were likely quite happy inhaling Drive2019–3 because the returns beat Treasuries. And car loans to those with credit scores that can be politely referred to as “not ideal” is a booming industry: more than $37 billion of bonds were backed last year by these spotty loans, up more than twice the amount from a decade prior according to Bloomberg data.
Any securitization based on subprime credit scores will naturally beg the question of risk: after all, most would assume that customers with low credit scores likely have those low credit scores due to a history of chargeoffs, delinquencies, debt consolidation or settlement, or some combination thereof. If borrowers can’t repay, it hardly matters to investors or banks — this is the institutional version of the late-night “No credit? No problem!” infomercials: complex tranching, overcollateralization, and levels of protection made it a virtual guarantee that investors would get back their principal plus interest, and Santander stood to earn tens of millions (maybe hundreds of millions) from arrangements like Drive2019. Meanwhile, the subprime borrowers would have their lives upended by a repossession, likely kicking off a deadly spiral of debt, savings drain, and restricted career mobility (and thus constrained wealth growth) due to restricted transportation mobility.
Pondering the fallout from this is hardly an academic exercise: car owners have fallen behind on payments at the highest rate on record: 6.11% of borrowers were at least 60 days delinquent in September 2023. Part of the problem with the surge is higher vehicle prices (even for used cars) and higher financing costs — unsurprising since we’re in a rising-rate environment, and despite the Fed’s fanciful lullabies to the markets, there is absolutely zero reason to assume 2024 should bring rate cuts, regardless of whether or not it will. Subprime borrowers can act as bellwethers for macroeconomic headwinds, often getting squeezed before the broader economy starts to suffer. In short, the risks posed by programs such as Drive2019 are not a footnote of history, but a rather chilling portent of what could happen in the near future.
From a risk management perspective, Santander and other offerers can hardly be blamed: the fees from such securitization can be staggering, and the bonds are designed to withstand a high rate of default (Santander assumed borrowers would be unable to repay 42% of the money they owed). And while the data presented herein come from Drive2019–3, the reader should keep in mind that this is just one of these bonds: it’s unlikely that other subprime loan securitization efforts will be much different. If letting customers incur more debt than they could ever possibly afford seems like a terrible idea, you are to be congratulated for possessing the ability to think logically. It’s also the sentiment shared by many industry insiders: Daniel Chu, CEO of subprime lender Tricolor Holdings, said it best — “It doesn’t really work for the customer. But it works for everyone else.”
Just how big is the problem? In August 2023, when consumer debt topped $1 trillion for the first time ever, auto loans were outpacing student debt and topped credit card debt by a factor of over 1.5x ($1.58 trillion for auto loans vs. $1.03 trillion for consumer credit). Bloomberg warned of a potential hit to subprime bondholders in July.
Echoes of 2008
Whenever the word “subprime” is mentioned, thoughts will inevitably turn toward the tragedy of errors that was the 2008 financial crisis. This is not entirely unwarranted: both the subprime mortgage implosion and the subprime auto lending hopefully-not-another-implosion were rooted in the same basic fact: no matter the layers of swaps, derivatives, technical jargon, and financial chicanery, at the end of the day institutions sold loans to customers who couldn’t repay. No amount of investor protections, media spin, bailouts, or target school ego is capable of changing this fundamental fact. Denying this would be as much a sin of intellect and dearth of logic as denying the sunrise. Amidst the flurry of golden parachutes, Congress passed the Dodd-Frank Act, which not only required enhanced scrutinizing of a borrower’s finances to ensure a mortgage was affordable, but created the Consumer Financial Protection Bureau (CFPB) to enforce compliance.
But automotive dealerships, where most of the subprime automotive loans are made, are exempt from CFPB oversight thanks to a successful lobbying campaign to convince lawmakers that regulation would hinder a critical sector of the economy — “regulation for Wall Street, not Main Street.” Some in Congress have publicly stated that this lack of oversight has enabled “risky and harmful practices to become commonplace.” To their credit, the CFPB and some state regulators have cracked down on subprime lenders like Santander, who reached a $550 million settlement in 2020 on the matter following allegations from attorneys general in 33 states and the District of Columbia. One of the allegations of this case was that the company knowingly made loans that were likely to default, echoing the “predatory lending” claims surrounding the 2008 meltdown.
But despite what appears to be progress, more than 30% of subprime auto loans default, which is near the highest default rate for subprime mortgages at the peak of foreclosures.
Anatomy of a subprime asset-backed security
As most are undoubtedly aware, the monthly payment for any loan is determined by three things:
- The loan amount, which is a function of the asset price and down payment
- The loan tenor
- The interest rate
The affordability, of course, is a bit more complicated and depends on an individual borrower’s income, spending habits, and so forth. Ideally, the car payment should be no more than 10–15% of the borrower’s take-home pay (called payment to income, or PTI). While common sense would argue that a rational borrower would select the cheapest car that would still meet functional requirements (including safety and reliability), the annals of human history show that we are anything but a rational species. We are also a myopic one: all things equal, a borrower will likely opt for a longer tenor to get a lower monthly payment, allowing them to either shell out less for the car they need or to bring the car they want into the range of affordability, conveniently overlooking the fact that a longer tenor means paying considerably more for the car over the life of the loan.
Credit scores
Unsurprisingly, the mean and median credit scores of subprime auto obligors examined (480 and 561, respectively) is well below the national average of 706 in 2019. But what’s rather surprising is that this difference is not actually statistically significant: the Z-score for this difference from the national average is 1.01, which denotes that at the 95% confidence level this difference in credit score is no different from randomness. Yet this difference in credit scores seals the fate of many subprime obligors.
Note that the magnitude of the difference in mean credit scores is massive regardless. It must also be noted that 5,779 rows within the dataset — more than 17% of the entire dataset — had a credit score of zero and were disregarded as outliers (the lowest credit score that can be attained with currently-used models is 300). Similarly, 15 entries had a credit score greater than 850 (another mathematical impossibility) and were similarly discarded. Therefore, while the lack of statistical significance appears noteworthy, the actual significance is unknowable unless approximately 5,800 credit scores are accounted for.
So how much of an effect does credit score have on the monthly payment? While the credit risk model(s) used by Santander and other similar lenders do not appear to be publicly available, the short answer seems to be “plenty.” In and of itself, this is not a novel observation, but it must certainly raise the question of whether conventional credit risk modeling can lock a borrower into an endless debt cycle and mitigate their chances at improving their financial situation.
Geographic distribution of obligors
Texas ranks 47th in average credit score, so it’s unsurprising that Texas had the highest number of subprime loans by a wide margin; similarly, Georgia ranks 45th in average credit score, making its place in the top five states with the most subprime loans understandable. Texas also has six of the twelve cities with the highest auto debt, but Georgia and Florida only have three and one, respectively. The average credit score in California in 2019 was 708, which is slightly above the national average and miles higher than the mean and median credit scores for subprime obligors. Similarly, the average credit score for Pennsylvania in 2019 was 713, also above the national average. So why do these five states account for over 44% of the total subprime loans in Drive2019?
Monthly payment
The variables reportingPeriodScheduledPaymentAmount and totalActualAmountPaid tell very different stories regarding the monthly payments of these obligors. While these payment values are higher than the national average of $391, as might be expected (although not to a statistically significant degree, as the Z-scores for the scheduled and actual monthly payment amounts were 0.621 and 0.570, respectively), the distribution of actual payments is skewed almost three times greater than that of the scheduled distribution (skew = 1.703 for actual, 0.5705 for scheduled). The question is why: what fees are tacked onto these loans that increase the mean monthly payment by over ten percent? What is the root cause of these differences? The reality is very clearly not matching the advertised payment, and it’s unlikely that the reasons for these differences are in the customer’s benefit.
Repos
It should come as no surprise that vehicle loans made to obligors with subpar credit will come with a high risk of repossession (for the current analysis, there was no distinction made between repossessions due to default and voluntary repossession, also called voluntary surrender). For the Santander dataset, the number of vehicles repossessed was approximately 17%, compared to the 2019 national average of 2.30%. This means that the repo rate for these subprime loans is more than seven times the national average. There is a question of correlation versus causation to be addressed: lower credit scores obviously correlate with higher interest rates and higher rates of repossession, yet the financial equivalent of the legal “but-for test” merits consideration — but for the existence of higher-than-average interest rates and monthly payments, would a repossession have occurred?
The effects of a repossession are obvious: most immediately, repossession will show up on credit reports as a derogatory mark, and remain there for up to seven years (a voluntary surrender will be listed as such, and may do slightly less damage to your credit — I suppose every little bit helps…). This will also affect the obligor’s ability to obtain future loans. While this may seem to be a minor concern, the nature of debt should be kept in mind: when used responsibly, the purpose of debt is to obtain an asset that will (hopefully) provide a greater return than the cost (with interest) of acquiring it. For instance, not only can most Americans not afford to acquire a home without a mortgage, most homeowners tend to enjoy capital appreciation of their homes that outpaces their mortgage (home prices tend to increase, although as we’ve seen since 2001 these increases are extremely location-dependent and can be erased by recessions and other catastrophes). A vehicle, on the other hand, is a depreciating asset that begins to lose value before the ink on the sales contract is dry, and unless the vehicle being acquired is along the lines of a 1955 Mercedes-Benz 300 SLR Uhlenhaut Coupé (auctioned for a record price of €135M [$145M]) it’s unlikely that any financed vehicle will become more valuable than its MSRP over time.
In other words, a vehicle is a necessary debt for most, even though they’re virtually guaranteed to be on the losing end of it. Losing a vehicle via repossession or surrender actually means losing at least two vehicles: the one that just got repossessed, and the one that the customer won’t be able to finance to replace it. Worse, a repossession won’t necessarily rid the borrower of financial obligation: they may still owe money to the lender, since the outstanding balance is a function of the post-repo sale price, repossession costs, storage fees, etc. It’s completely possible for a borrower to find themselves owing money on a car they no longer have.
The real question is why: why is this particular category of borrowers experiencing repossessions at more than seven times the national average? While most would be quick to point out that loan delinquency goes hand in hand with poor credit scores and borrowers being overleveraged (financing more car than they can afford) — assumptions that will be disproven later — it’s highly likely that the root cause is actually a lack of financial cushion. Financial demerits, in the form of derogatory credit marks, can often set off an avalanche of consequences that only make matters worse for the borrower. One can argue that the system is rigged, but the playing field is severely tilted toward catastrophe: for many, it’s much easier to reach ruin than success.
Popularity of vehicles by make and model
It’s unsurprising to see Ford and Toyota among makes occurring most frequently in the dataset, but what is surprising is that Nissan was the most-frequently financed manufacturer since it actually lost market share over the past several years (including 2019, when the Santander dataset was created). Nissan’s US market share decreased from 8.4% in 2018 to 6.1% in 2022, which makes its outsized presence in the dataset curious.
Worse yet for obligors, Nissan models are often panned by Consumer Reports for their subpar reliability (especially for model years that would have placed these cars right in the cross hairs of subprime lenders during the 2019 time period), yet Nissans account for more than 6% of all loans in the dataset. The Sentra, Altima, and Rogue — often touted as the most unreliable Nissan models — present yet another pitfall for obligors: despite the frequency with which they are financed (which alone should beg the question of why they’re so prevalent on the lots of subprime auto retailers), these vehicles will likely surprise owners with unforeseen repair bills that will ruin an already-taxed budget.
This maintenance and repair question is hardly academic when one considers the age of the vehicles in the dataset. While factory warranties transfer to the new owner of a vehicle, different warranties are valid for different lengths of time: bumper-to-bumper warranties are typically good for three years or 36,000 miles (whichever comes first), whereas powertrain warranties can last ten years or 100,000 miles. Why does this matter? Let’s look at the age of the vehicles in the dataset:
Despite the odd shape of the density plot, what’s critical is that over 51% of the vehicles in the dataset were at least three years old at the time of the Drive2019 filing. While this would still put them within the coverage of most powertrain warranties, many of the car parts most likely to fail are wear items that are not covered by powertrain warranties: brakes, brake pads, CV joints, shock absorbers, and similar items can make a car unsafe (or impossible) to drive, but whose repair or replacement must be paid for by the owner. This may require one to reexamine whether what is often assumed to be a lack of financial discipline on the part of the obligor might instead simply be a case of poor timing and unfortunate circumstances.
Interest rates
Much as with the distribution of credit scores for subprime auto debtors, the distribution of interest rates sees the mean and median skewed far from the national average. When it comes to interest rates, however, the difference is staggering:
Note that while the national average interest rate for car loans varied throughout 2019, its maximum value was 4.77% for a sixty-month term (a loan tenor that is not applicable to the current dataset, as shown later). Not only did the subprime obligors pay, on average, an interest rate four times higher than the national average, this difference is very statistically significant, with a Z-score of 3.663. This means that subprime borrowers are paying a higher interest even at a confidence level of 99.99%.
Loan tenor
The average car loan in 2019 was for 67 months, whereas the average subprime loan during the same time period was for about 70 months. This difference is not statistically significant (Z = 0.643), but when combined with the aforementioned issues of significantly higher interest rate and vehicle age (and likelihood of parts failing that are not covered under the balance of the factory powertrain warranty) an uncomfortable truth begins to emerge: subprime borrowers are not only paying more for a car, but their loan tenors are also virtually identical to those of borrowers deemed less risky, for vehicles that are older (and likely higher-mileage), and must additionally shoulder the cost of non-warranty repairs.
Loan amounts
The average loan payment for subprime obligors did not materially differ from the national average for used cars in 2019 (Z = 0.0585). Thus, while subprime obligors were very clearly disadvantaged from an interest rate standpoint, they were at least not gouged on price.
PTI
The distribution of PTI among the subprime obligors was somewhat surprising. Conventional wisdom recommends a PTI between 10–15% of a borrower’s take-home pay, and conventional assumptions hold that subprime borrowers are somehow being punished for the irresponsibility that led them to financing vehicles they couldn’t afford. The data do not support this assumption. In fact, more than 76% of borrowers had PTIs below 15%, indicating that the overwhelming majority of subprime borrowers were actually following “the rules” when it came to budgeting for a vehicle payment.
PTI, of course, does not tell the entire story: debt to income (DTI, calculated as all monthly debt payments divided by gross monthly income) is equally important, but unfortunately borrower DTI data were not available for this dataset. While PTI plays a critical part in vehicle affordability, its DTI (which includes credit card debt, student loan debt, etc.) can influence the ability of borrowers to make all of their payments on time. Eventually, obligors may find themselves in a financial Sophie’s Choice, where they must decide which of their bills will be ignored in order to make ends meet. Of course, missed payments accrue fees, making skipped bills more expensive and staying current on everything more difficult. But to write off subprime obligors’ payment issues as irresponsible borrowing is myopic at best and willfully ignorant at worst: they don’t appear to be borrowing more car than they should be able to afford.
Loan delinquencies
Most of the loans in Drive2019 were current at the time of filing, but it should be noted that this is another instance where it’s important to consider that the dataset is a snapshot in time of delinquencies, and that simply because a large portion of loans were current at the time of filing should not be taken to imply that a large portion of loans remained current throughout their tenor. The repossession rate of these loans should be enough evidence to argue this point.
Most of the loans were current at the time of filing, but the majority of the loans were very new (mean age of 1.03 years; median age of 0.99 years). Given that the tenor of these loans was about six years at origination, the apparent disconnect between repossession rate and loan delinquency is easily explained: the loans simply haven’t had enough time to default.
Furthermore, per the data dictionary a value of “True” for the variable repossessedIndicator
[…] indicates that the related financed vehicle has been repossessed and has not been reinstated, as of the last day of the reporting period. The repossessed indicator field will be marked “False”, if (i) the related financed vehicle has been repossessed but the account has been reinstated prior to the last day of the reporting period or (ii) the receivable will be repurchased on the next payment date and is no longer considered an active loan as of the last day of the reporting period.
In other words, a repossessed vehicle might not be considered as such if the account is reinstated or some sort of promissory note for the receivable has been received; thus, the actual repossession rate is likely higher since vehicles that will be repurchased are not flagged as “repossessed.” They certainly have been repossessed from the viewpoint of the obligor, but not necessarily in the eyes of the bank. This is likely for the best (at least, for Santander) since the net repossession recovery rate (the amount collected upon disposition of the repossessed vehicle net of expenses, normalized by the original loan amount) is tightly clustered around zero. Essentially, repossessed vehicles are eaten as a loss once disposition is complete — a failure for both the customer and the lender.
Credit score and PTI versus days arrears
Conventional logic dictates that the lower the credit score, the higher the likelihood of days in arrears. If one were to create a contour plot comparing the number of days delinquent with obligor credit score and monthly payment, the expected result would be that the highest number of days in arrears would appear near the top-left of the plot, viz. that lower credit scores and/or higher monthly payments correlated with a higher concentration of days in arrears. Once again, conventional logic appears to be based on assumptions that do not correlate with reality:
Based on Santander’s own data, there doesn’t appear to be a direct correlation between credit score and delinquency. Instead, the distribution of delinquencies is approximately symmetric about a credit score of roughly 575. In fact, most Americans have a credit score between 600 and 750, suggesting that credit score and rate of delinquency for these subprime loans has little — if any — correlation. This begs the question of how reliable a credit score is in terms of likelihood of default or delinquency, and whether a more holistic approach to an obligor’s ability to pay is warranted. Certainly, credit score matters, but the peak of delinquency near a score of 575 and seemingly unrelated to monthly payment — and the fairly symmetrical distribution around this point — merits further analysis.
Similarly, conventional logic must be disproven with regards to days in arrears as a function of credit score and PTI: recall that not only is the distribution of PTI for the subprime loans within the recommended range, but there was virtually no difference between mean and median PTI. As with the previous graph, a symmetric distribution about a credit score of approximately 575 is seen:
Not only does there appear to be the same unexpected location of the highest concentration of delinquent loans, but the highest concentration of delinquencies are at a PTI that is generally considered perfectly adequate. Again, if credit score and PTI explained delinquency to a reliable degree, one would expect to see the epicenter of the contour plot far below the national median credit score and higher than the generally recommended PTI range. That is not the case here.
One must be forced to conclude, then, that these delinquencies are not the result of irresponsible, high-credit risk customers buying more car than they can afford, but of other factors that lie outside of the conventional risk metrics and which may or may not be unique to subprime borrowers. The only other factor that could contribute to delinquency would be the monthly payment of the loan, which is directly affected by the interest rate. As shown previously, the mean and median interest rates of these loans greatly exceeds the national average, which may lead one to conclude that these loans are possibly predatory, and adversely affecting consumers beyond the financial penalties they already face due to their credit scores and likelihood of out-of-pocket repair costs. There is also the effect of time to consider: recall that the dataset is a snapshot of the Drive2019 loan bundle, which contained very young loans and a rather unique way of counting repossessions that does not necessarily fit a layperson’s understanding of the term. It would be very interesting to analyze traces of individual loans over time, to see if there are specific factors that repeatedly stress obligors and drive the loans toward delinquency or default.
Closing thoughts
Asset-backed securities offer a good way to theoretically diversify risk, as each individual facet of the security (auto loans, in this case) represent a small fraction of the total value of the pooled assets. Auto loan ABS are a particularly interesting corner of the investment universe, as they’re the second-largest subsector in the ABS market and are quickly outpacing most other forms of American consumer debt. Tranching these securities into instruments possessing theoretically different risk profiles also facilitates marketing the bonds to investors with varying risk appetites and time horizons.
But to treat securitization as a panacea for poor credit decisions is a verse we all heard sung in 2008. The difference with auto ABS is that there are quite a few more auto loans than mortgages, even though the risk of correlated default remains. This is especially damning when one considers that most American debtors, regardless of their credit history, are far more reliant on a vehicle than a mortgage; the fallout following a default on an auto loan can very quickly place the most financially vulnerable in a position of high interest rates and potentially reduced earnings from which they may never recover. This is exacerbated by the fact that the ability to earn high fees from origination and securitization couples quite nicely with the pronounced absence of any meaningful liability, driving originators and ABS packagers toward loan volume rather than loan quality. Not only is this flaw intrinsic to securitization, but when dealing with subprime borrowers the temptations of moral hazard are obvious: they’re going to default anyway (so the conventional thinking goes), so why bother with loan quality?
The data from over 30,000 subprime auto loans tell a different story: the link between credit score, PTI, and delinquency does not appear to be as strong as some might think — in fact, the highest number of delinquencies were centered around a credit score that barely reaches into subprime territory. The distribution of payment to income for the obligors was well within the recommended limits for the majority of the obligors, yet they pay interest rates five times the national average and face a rate of repossession more than seven times the national average. So while the amounts and tenors of these subprime loans match up very nicely with nationwide norms, the interest rates are practically usurious and put the obligors at a heightened risk of financial ruin for factors outside of their control — such as the fact that most of the cars available to them no longer maintain a warranty that covers the parts most likely to break. The difference is that when a Tier I obligor experiences such a problem, they are more likely to have credit cards, savings, or other means of resolving the issue; for a subprime obligor, an unexpected trip to the repair shop can spell doom.
As previously stated, these types of securitized products rarely, if ever, work in favor of the obligors. We would be so bold as to propose the suggestion that they’re perhaps intentionally designed in such a fashion: myopic short-termism gone mad with little regard for the root cause of the affordability problem. The knock-on effects of a force majeure-driven misstep for these obligors can quite literally last a lifetime, affecting not only their ability to earn, but eradicating what meager savings they may have and ensuring that career milestones such as promotions and retirement remain nothing but a pipe dream.
So are the delinquencies and repossessions — and attendant financial consequences for the obligors — correlated with or caused by the subprime lending structure? Arguments can certainly be made in either direction, but this is potentially another example of why, at its heart, the conventional lending system is broken. The concept of a conventional lender should be replaced with that of a lending partner that helps obligors create wealth to improve their position over time — an operating model that is very near and dear to us, philosophically as well as operationally. If nothing else, the data suggest that the conventional method of assessing credit risk for auto loans is incomplete and merits further refinement. If the conventional credit models were correct, there would be a much stronger correlation between delinquencies versus PTI, credit score, and monthly payment. This was not found to be the case, as the contour plots demonstrate, meaning that current credit risk models are at best incomplete and at worst (willfully?) ignoring reality. As a de minimis requirement, subprime lenders should be forced to examine an obligor’s ability to repay holistically, which includes not only DTI (as opposed to PTI) but also factors such as probabilistic non-warranty repair costs based on car model, obligor driving habits, etc. In other words, PTI should be reassessed from the standpoint of total cost of ownership. In reality, that is your actual PTI. Would this approach raise eyebrows? Certainly — but whereas a TCO-based approach to financing would raise eyebrows on Wall Street, the decision to do nothing about a problem when evidence suggests there might be a solution (in software-speak, this is often referred to as “executing the ostrich algorithm”) would undoubtedly raise eyebrows on Capitol Hill.
The reality is simple: investors must receive a return. Anything that will increase that return — which must include increasing the obligor’s ability to repay and decreasing their likelihood of default — must be viewed as worthy of investigation, a canonical example of how a financial institution can do well while also doing good. The situation of subprime borrowers is unquestionably dire, but refusing to deviate from conventional approaches to lending will never resolve the problem.
