- Analyzing loan-level data for auto-loan asset-backed securities (ABS) provided granular insights into the drivers of collateral risk and overall performance.
- Auto-vehicle type, loan-to-value (LTV) ratio, gross coupon, term and issuer are examples of data that can shed light on an individual loan’s default probability.
- By incorporating this loan-level data into the latest MSCI models, we could sharpen our analysis of auto-loan ABS.
U.S. regulators required issuers by Nov. 23, 2016, to disclose data on individual loans bundled into publicly offered auto-loan ABS.1 With four years of this loan-level data now amassed, we looked at whether it can give investors new and more granular insight into the risks and performance of auto-loan ABS, versus the standard pool-level analysis. We found that, by incorporating loan-level data into the MSCI U.S. Auto Loan ABS Collateral Model, we could sharpen our analysis of this ABS segment.
The exhibit below shows the conditional default rates (CDR) of two auto-loan ABS deals, which on the surface bore similarity in their pool-level characteristics and collateral-loan descriptions.2 The two deals nevertheless diverged in default performance, due to loan-level differences that investors might have overlooked.
Auto-Loan ABS Deals Looked Alike at Pool Level, but Differed in Default Performance
Historical default rates for two auto-loan ABS deals.
Loan-Level Data Revealed Differences in These Pools
Investors equipped with loan-level data could have spotted key differences in these pools. In this case, we can model the default-rate differences and attribute them to Deal 1’s higher percentage of weaker loans and higher loan-to-value (LTV) ratio.
Differences in Loan-Level Credit Quality and LTV Ratio Led to Differences in Performance
New Disclosure Rules Led to a Wealth of Loan-Level Data
In the four years since the U.S. Securities and Exchange Commision’s regulation requiring the new disclosures took effect, issuers have provided loan-level data on borrower credit score, cosigner, income and employment verification; loan term, balance, coupon, age and subvention; vehicle make, model, model year and condition when bought (new or used); and the full performance history. Signals captured from the analysis into this loan-level data can help ABS investors potentially gain new insights.
What Does Vehicle Type Say About Loans’ Performance?
The exhibit below shows default rates by vehicle type. We focus on two performance periods, with March 2019 to Feburary 2020 representing the baseline period and March 2020 to Feburary 2021 representing the pandemic period.3 Which vehicle-type loans defaulted the most?
Car Loans Defaulted More than Truck or SUV Loans
Car loans had a consistently higher default rate, across tiers and during both analysis periods, than that for SUVs and trucks. For subprime tiers, SUV loans underperformed truck loans, and the opposite was true for prime tiers. Why? Trucks and SUVs could be used to run small businesses, which generate cash flows, and the demands are more inelastic than those for cars. Truck loans also have shorter terms than loans for the other two vehicle types. Stimulus checks, unemployment benefits and relief programs might also have helped reduce or postpone defaults during the pandemic, potentially explaining the drop in default rates observable in the subprime tiers.
Default Rates by Loan Size
The exhibit below presents a complex and nonlinear aspect of credit performance by auto-loan size.
Bigger Loans Performed Better in Lower Credit Tiers
A majority of auto loans have an original balance between USD 10,000 and USD 40,000. High-balance loan groups performed worse in prime tiers, but better in subprime tiers. A higher-balance loan statistically has higher LTV and lower gross coupon — loan characteristics that tend to have offsetting credit effects. For prime tiers, gross-coupon differences were minor for loans above USD 20,000, and default events increased monotonically with LTV as well as balances; for subprime tiers, LTV stabilized for loan sizes above USD 30,000, so the performance improved as balances increased and gross coupons dropped. We also noticed a small group of deep-subprime loans with low coupons, short terms and high balances that outperformed their cohort peers in all observable dimensions, suggesting that credit scores have not presented the full story.
Other Loan-Level Data Mattered Too
Besides these examples, our analysis4 also showed that incorporating data for such loan variables as LTV, loan rates, loan terms, origination years, loan subvention and deal issuers helped provide a more accurate view of auto-loan-ABS risk, during the period of our analysis, as shown in the exhibit below.
Below we use the statistical measure of a “gain chart” to show the overall gains in modeling accuracy from using loan-level data over pool-level data.
Loan-Level Data Improved Our Model’s Accuracy
In sum, loan-level data may give investors a more granular information set, enabling a better understanding of the drivers of collateral performance.
1“Asset-Backed Securities Disclosure and Registration.” Securities and Exchange Commission, Aug. 26, 2014.
2Deal 1 refers to INTEX deal identifier FAOT18A, and Deal 2 refers to INTEX deal identifier GMAR184.
3Analysis was based on 165 auto-loan ABS deals from 18 issuers. A total of 3,433 deal months were covered by the two periods. On average, there are 40,953 outstanding loans in a deal.
4Yang, Yini, and Zhang, Joy. 2021. “US Auto Loan ABS Performance Review with Asset Level Data: Insights from Granularity.” MSCI Research Insight, May 21. 2021.
Further Reading
US Auto Loan ABS Performance Review with Asset Level Data
COVID Stimulus Helped Resilience of US ABS
A new COVID-19 Regime for MBS?
Can MBS Duration Turn Negative?
Are Securitized Products Ready for the LIBOR-SOFR Transition?