Market participants have become increasingly concerned about subprime auto loan ABS given that the economy may be on the precipice of a recession wherein unemployment levels would likely rise and inflation could persist. Given the established relationship between unemployment and losses on collateral backing auto loan ABS, as evidenced in the 2007-2009 recession, there is market interest in understanding the likely impact of expected higher unemployment on consumer affordability, collateral performance, and auto loan ABS ratings. The subprime auto sector is of particular concern because these borrowers have more limited financial resources than prime-type borrowers, and some of the transactions include speculative-grade classes.

To test the resiliency of U.S. subprime auto loan ABS ratings to a potential economic downturn, the length and depth of which are unknown, we ran three hypothetical economic stress scenarios on 54 transactions issued from 2019 through April 2022.

Hypothetical base-case stress - 20% reduction in recovery rates 

• No rating downgrades across all ratings.
• Approximately 85% of the total number of classes rated in the 'BBB' to 'AA' categories upgraded into at least the next category.
• Approximately 76% of those in the 'BB' speculative-grade category upgraded to an investment grade rating.

2007-2009 hypothetical recession stress - 30% increase in remaining losses  

• Upgrades exceed downgrades for classes with investment-grade ratings.
• 3% of 'AAA' ratings downgraded one notch to 'AA+' due to pro-rata structures. The remaining 'AAA' ratings remain unchanged largely due to the sequential nature of the payment priority.
• 5% of those in the 'BBB' category downgraded to the 'BB' category.
• 40% of those in the 'BB' category downgraded into a lower category--36% to the 'B' category and 4% percent to 'CCC+'.

'BBB' hypothetical economic stress - remaining losses increased by the series 'BBB' rating multiple  

• 38% of those in the 'AA' category and 8% of those in the 'A' category upgraded into the next higher category.
• About 12% of 'AAA' ratings downgraded, but not below 'AA-'.
• About 17% and 36% of those in the 'AA' and 'A' category, respectively, downgraded into the next lower category.
• About 54% of classes rated in the 'BBB' category downgraded into a lower category--36% to the 'BB' category and 18% into the 'B' category.
• All rating movements within our credit stability criteria.

The Stress Scenario Sample

Our test sample included 54 subprime outstanding auto loan ABS issued between 2019 and April 2022 and rated by S&P Global Ratings, from all term subprime issuers. Given the relatively short life of auto loan ABS and deleveraging that occurs in these transactions, we opted to exclude any remaining outstanding 2018 deals, most of which have pool factors of 15% or less. Also, given the initial build in overcollateralization that occurs during the first six months of a transaction, we included only those deals that had at least six months of performance, and we excluded any transactions on CreditWatch negative. In total, these 54 transactions included 213 classes with ratings ranging from 'AAA'/'A-1+' to 'B'. These 54 deals were taken from a selection of 115 outstanding S&P rated subprime deals that closed during the time frame stated above.

Break-even cash flows were run on the selected transactions using collateral and bond balances as of September 2022. The break-even cash flow methodology and assumptions are discussed below, but one key element in the cash flows of the seasoned deals is that recoveries on future defaults were run with an approximate 20% haircut to the cumulative recovery rate realized to date.

The Three Stresses

Scenario 1 – hypothetical base case with 20% reduction in recovery rates

In this scenario, for transactions that closed prior to 2022, we projected out the deal's expected cumulative net loss (ECNL) percentage (as a percentage of the initial pool balance) based on the specific issuer's historical loss curves (both by month and by pool factor). For transactions that closed during the first four months of 2022, we used our original ECNL proxy. From there, we subtracted the losses taken through Sept 30, 2022, to arrive at remaining losses and then divided this by the pool factor as of Sept 30, 2022, to obtain the remaining net losses as a percentage of the current pool balance. This remaining net loss percentage became our hypothetical base case. We then compared the break-even cash flow results to this hypothetical base case.

Base Case Results  

Table 1

Table 2

Under the hypothetical base case scenario, we observed that there would likely be no downgrades on the sample we tested. Additionally, as is typical in auto loan ABS, there would likely be a high percentage of upgrades. For those classes rated in the 'AA', 'A', 'BBB', and 'BB' categories, approximately 89%, 86%, 80%, and 76%, respectively, would likely be upgraded at least to the next category. Our sample size included only two classes rated in the 'B' category, and under this scenario both of them would likely be upgraded, one to 'BB-' and the other to 'BBB-'. (While table 1 and 2 above are summarized by rating category, the appendix contains more granular ratings data).

Hypothetical Scenario 2 – 30% increase in losses

We chose to run this hypothetical scenario because, during the 2007-2009 recession, we found that subprime transactions performed 30% worse than we had originally expected, on average. This 30% increase in losses was applied to our expected remaining net loss level. When we compared the break-even CNL levels to these higher loss levels, we found that approximately 3% of the 'AAA' ratings (two A-2 classes) were susceptible to downgrade (one notch to 'AA+'). These particular 'AAA' classes employ a pro-rata structure. The remaining 'AAA' ratings remain unchanged primarily due to the utilization of a sequential-pay structure, which builds credit enhancement as the pool amortizes.

In addition, 5% in the 'BBB' category were vulnerable to a downgrade to the 'BB' category, and 40% in the 'BB' category could potentially be downgraded into a lower category (36% to single 'B' category and 4% to 'CCC+').

For the two classes rated 'B', one would be upgraded to 'B+', and one would not change. For purposes of this exercise, we used a 1.0x multiple for 'B-'. We used multiples of 0.75x-0.99x for 'CCC+', which makes 'CCC'-category ratings highly susceptible to default. These two transactions are from 2021 and are performing considerably better than our original ECNLs. If we were to give full credit to that improvement, these classes would likely be rated higher than they currently are today.

Scenario 2 Results  

Table 3

Table 4

Even under this mild stress, we found that upgrades would exceed downgrades for classes currently carrying investment-grade ratings. At the 'AA', 'A' and 'BBB' categories, upgrades to at least the next category would represent approximately 64%, 49%, and 29%, respectively, while downgrades to the next lower category equated to 0%, 0%, and 5%, respectively. At the speculative-grade category of 'BB', however, downgrades would exceed upgrades (40% downgrades compared to 24% upgrades).

For rating transition by notch, please see the appendix.

Hypothetical scenario 3 - 'BBB' Scenario

Under the third hypothetical scenario, we applied the appropriate 'BBB' multiple to the base case remaining losses.

Under this scenario, approximately 12% of the 'AAA' ratings could be downgraded, but none lower than 'AA-'. Of the eight downgrades, seven are on classes issued between August 2021 and March 2022. In addition, approximately 17%, and 36%, respectively, of those carrying ratings in the 'AA' and 'A' category could be downgraded to the next category. Consistent with our rating stability tests we ran at time of new issuance, none of the 'AAA' or 'AA' rated classes would be downgraded by more than one category. Quite interestingly none of the those rated in the 'A' category would be downgraded below 'BBB-'.

Approximately 54% of the outstanding BBB-category classes would be vulnerable to downgrade into a lower category with potentially 36% falling into the 'BB' category and 18% declining into the 'B' category. Based on our assumptions, none of the classes currently rated in the 'BBB' category would default under our 'BBB' scenario.

Not surprisingly 88% of the classes rated in the 'BB' category would be downgraded into a lower category (52% to the 'B' category and 36% to 'CCC'). Those downgraded to 'CCC' would have a very high likelihood of default.

Scenario 3 Results  

Table 5

Table 6

Under the above scenario, 38% and 8%, respectively, of classes currently rated 'AA' and 'A' actually qualify for upgrades into the next category. For rating transition by notch, please see the appendix.

Unbreakable Cash Flows

Interestingly there were many deals wherein classes currently rated 'AAA' were unbreakable, meaning that even with 100% defaults the class would still pay off on a timely basis. There were also a few classes rated in the 'AA' category that were unbreakable. Under all three scenarios, we show these 'AA' rated, unbreakable classes upgrading to 'AAA'. Some of these classes will likely be upgraded upon their next surveillance review while others may only be affirmed to the extent there are 'AAA' rated senior classes above them that are expected to be outstanding for another 18 months or more.

Break-Even Cash Flow Methodology

To test the resiliency of the structures, we ran break-even cash flows on the transactions based on their collateral, credit enhancement balances, and ratings as of Sept 30, 2022.

Assumptions For Break-Even Cash Flows - Loss Curve Timing

Loss timings were generally issuer-specific. We took into account the issuer's loss timing curves from their 2016 paid-off vintages. We recalibrated the 2016 loss timing curves to account for the current pool factor (percent of initial collateral pool remaining) of the specific transaction. For example, if an issuer's 2016 loss timing was 25% by end of year one, 55% by end of year two, 80% by end of year three, 95% by end of year four, and 100% by the end of year five, and we were running cash flows on a deal that had a similar pool factor after the first year, the recalibrated curve would assume that all the remaining losses would be taken at the following speed: 40% in year one of our cash flows, 33% in year two, 20% in year three, and 7% in year 5 (see table below). We used 2016's loss curves since it was not impacted by COVID-19-related stimulus, which led to low back-end losses on the 2017-2018 deals and in turn caused very front-loaded loss curves. If an issuer didn't have a paid-off loss curve from this vintage, we used loss curves from issuers with a similar collateral and loss profile.

Table 7

Recovery rate on future defaults

The recovery rates utilized on these break-even cash flows were generally 80% of the cumulative recovery rate (CRR) that the transaction experienced through Sept. 30, 2022. We did this to reflect that most transactions' cumulative recovery rates had benefited immensely from the run up in used vehicle values stemming primarily from a shortage of new vehicles and COVID-related stimulus in 2020 and 2021. Given that used vehicle values have already started to retreat and supply imbalances are likely to ease next year, we believed it would be prudent to haircut future recovery rates by 20% when solving for the break-even loss levels. This resulted in deal-specific recovery rates that typically ranged from a high of about 50% (for those low loss 2020 and 2021 subprime pools that have cumulative recovery rates [CRRs] of approximately 60%-62%) to a low of approximately 25% (for those high loss pools from second-half 2021 that are realizing CRRs of only 30%-32%). For the 2022 vintages, with limited performance, we generally assumed a recovery rate of 25%-35%. However, for a couple of the lower loss deals that were garnering elevated recovery rates, we ran higher rates (38% for one and 45% for the other one).

The 2007-2009 recession had a significant immediate impact on recovery rates but the subsequent bounce back in used vehicle values in the following years (due to low vehicle production in 2008-2011) lessened the overall deterioration on a vintage basis. For example, 2009's CRR at month 10 was only about 34%, a 17% reduction from 2005's CRR of approximately 41% at the same month. However, by month 40 the difference had narrowed to only about 9% (39% compared to 43% for 2005; see table 8 below). While we believe recovery rates will continue to normalize from where they are today and possibly decline slightly below pre-pandemic levels, depressed new vehicle production in 2020-2022, due to COVID-19 lock downs and semiconductor shortages, could be a supportive factor.

Table 8

Excess spread

In all scenarios, we haircut excess spread that wasn't used to build overcollateralization to its target by approximately 10%, which is consistent with our new issuance rating approach. We haircut the excess spread because higher APR loans may default or prepay to a greater extent than the lower APR loans, thereby causing a downward drift in the weighted average yield on the collateral, which we don't explicitly model. Haircutting the excess spread also provides a layer of conservatism because in reality the timing of defaults and prepayments may be different than we modeled.

Prepayments and other assumptions

For prepayments, we generally used the same prepayment rates that we used either at the time we rated the transaction, or that of a subsequent surveillance, if any. These prepayment speeds averaged 1.0% ABS and ranged between 0.60% ABS for the high-loss deep subprime issuers to 1.6% ABS for lower-loss subprime pools that have experienced higher levels of prepayments, some of which has been credited to trade-in activity due to strong used vehicle values (relative to historical levels). We believe that the prepayment rates that we ran through the cash flow model are conservative because generally when losses rise, prepayments decline.

For the few outstanding floating-rate classes, we assumed a starting secured overnight financing rate (SOFR) of 3.5% and applied the commensurate SOFR stress vector.

We did not assume a charge-off or recovery lag because the transactions are what we call "mid-stream" in that they are already collecting recoveries unlike at the time of issuance.

Comparison of Break-even Results to Remaining Losses

After running the cash flows, we compared the break-even CNL levels to the losses inherent in each of our three hypothetical scenarios. For example, if a transaction had a break-even CNL for the class A tranche of 55%, and our remaining CNL proxy for the deal under the base case was 12%, that would yield a multiple of 4.6x and would be sufficient for the class to retain its 'AAA' rating. Pools with an ECNL of 10%-13% generally should cover at least 3.50x losses at the 'AAA' rating (see table 9 below).

Table 9

As the loss level rises, the multiple required at each rating generally declines. For example, under hypothetical scenario 2, wherein losses increase to 15.6% (12% x 1.30), we'd generally require a multiple of approximately 3.30x at the 'AAA' level. As the class A coverage multiple of remaining losses exceeds 3.30x, the class would retain its rating. Under a hypothetical 'BBB' stress, wherein we'd use a 1.75x multiple of base case losses (given the base case of 12%), our 'BBB' level of losses would be 21%. Since the break-even can cover only 2.6x that level, which is lower than the approximate multiples of 3.0x and 2.8x needed for 'AAA' and 'AA+', respectively, for a 21% loss level, we'd likely assign a 'AA' rating to this class.

Limitations And Caveats

• This report does not constitute a rating action. The stress scenarios are hypothetical and are not meant to be predictive or part of any outlook statement.
• These stresses were with respect to losses only. Operational risk, counterparty exposure, and legal considerations were not part of this analysis.
• Our analysis excluded revolving transactions and two transactions with speculative-grade-rated classes on CreditWatch negative.
• The results are based primarily on the cash flow modeling and multiples we use to rate auto ABS. A small degree of analytical judgment was used in classifying the outcomes when multiples of remaining losses fell marginally short of the required amount. We also took into account structural features, particularly the sequential nature of the principal payments and assumed that the top-most A class would continue to benefit from the sequential pay structure within the A class to a greater degree than the longer dated class A notes, if any.
• A rating committee applying the full breadth of S&P Global Ratings' criteria, and including qualitative factors might, in certain instances, assign a different rating than the largely quantitative analysis undertaken herein.
• If loss timings come in faster or slower than the assumptions used in any of these scenarios, the results could have a positive or negative ratings impact depending on the transaction. A faster, more front-loaded loss timing could cause a slower build or a quicker erosion in credit enhancement but may be offset by lower expected remaining losses. Similarly, a slower, more back-loaded loss timing could allow further deleveraging and credit enhancement build, but this benefit could be offset by higher expected remaining losses. The ratings impact of loss timing, positive or negative, may depend on the degree of change in these two offsetting factors.
• Prepayment rates could also impact the results. To the extent prepayments slow, that could lead to a slower build in credit enhancement levels and keep the bonds exposed to economic conditions for a longer period of time.
• The implied rating transitions to the 'CCC' rating category are based solely on our model results. Our ratings analysis makes additional considerations before assigning ratings in the 'CCC' and 'CC' rating categories (see "General Criteria: Criteria For Assigning 'CCC+', 'CCC', 'CCC-', And 'CC' Ratings," published on Jan. 18, 2018).

Appendix

The author would like to thank Veer Umbargi and our cash flow team for their support on this scenario analysis.