An auto loan delinquency rate measures the percentage of outstanding auto loan balances where borrowers have missed one or more scheduled payments. At 90+ days past due, the rate signals accounts approaching charge-off. In Q4 2025, the New York Fed reported 5.2% of outstanding auto debt at 90+ DPD — the highest level since the 2010 financial crisis recovery — up 7.7% from Q4 2024. For a broader view of how portfolio loss forecasting in credit decisioning software fits into the risk architecture these delinquency trends are stressing, the hub piece covers the full landscape of the entire auto industry.
In 2019, the delinquency rate for auto loans stood at 5.3%, but more revealing, the total outstanding debt balance stood at around $800 billion. Seven years later, a delinquency rate of 5.2% may seem similar to what we saw in 2019. Still, when viewed against a backdrop of total outstanding debt exceeding $1.6 trillion, even when adjusted for inflation, the total dollar amount of delinquent loans represents a significantly greater burden for lenders. In fact, in 2026, delinquent loans exceeded $94 billion, primarily driven by extended loan terms and increased originations.
The delinquency rate is a lagging indicator. It confirms what already happened. The question for a CRO managing a portfolio with extended-term cohorts is not what the current rate reads — it is which specific account combinations inside that rate are six to twelve months from charge-off, and whether the current model can surface them before the roll rate accelerates. The Signal Intelligence WorkBench surfaces those Precision Impact Segments directly from multi-table LMS, LOS, and bureau data — not from the aggregate delinquency figure every competitor’s scorecard also sees.
84-month loans now account for 22% of new originations — a record high — and loans of 72 months or more represent 27.5% of all originations per Dealertrack October data. The average new vehicle price now exceeds $50,000; the average loan term has climbed to 68.9 months for new vehicles and 67.7 months for used vehicles. Term extension is functioning as an affordability lever for more buyers, not a borrower preference, and every month of additional term adds to the negative equity window on a depreciating asset, increasing the total cost of financing over time.
Extended-term loans increase the risk of delinquency because more borrowers remain underwater for extended periods. The incentive for a car buyer to pay and make a missed monthly payment decreases when the vehicle’s value is below the outstanding balance. In addition, long terms slow the growth of equity, which means a lower Loss Given Default (LGD) if a repossession were to occur in the first 24 to 26 months of the loan. Finally, the nature of longer-term loans means that the probability of a disruption in income, such as unemployment or an unexpected large expense impacting the borrower’s ability to afford and repay the loan, increases.
Millman’s analysis from October of 2025 shows that from 2019 to 2024, the average maturity of a used-vehicle loan increased by 4 months, as a result of more 72- and 84-month term loans. The growth in longer-term loans has increased lenders’ vulnerability to depreciation and credit risk across the auto market.
The origination-point scorecard lacks a mechanism to identify this compounding effect. A borrower with a 720 FICO credit score taking out a 72-month used vehicle loan may appear acceptable at the time of closing. dotData’s Signal Intelligence Platform assesses how the loan’s LMS payment velocity, tradeline composition, and collateral depreciation trajectory interact during the first 18 months of the loan — pinpointing the specific Driver Signal combinations that forecast roll rate acceleration in extended-term cohorts before any delinquency migration is reflected in the portfolio dashboard.
The model assumption most loss forecasting teams are least prepared to abandon is that prime borrowers represent a stable anchor — low roll rate, high cure rate, predictable recovery. VantageScore’s November 2025 CreditGauge analysis has made that assumption untenable: early-stage delinquencies (30–59 DPD) reached 1.13% in September 2025 — the highest level in five years — with the prime borrower share falling from 34.0% in September 2023 to 32.7% in September 2025, while near-prime and subprime segments expanded as lenders encountered borrowers with a lower credit score. The composition of the portfolio is shifting underneath the model’s historical calibration. A reserve model that applies a standard roll rate assumption to prime borrowers is now systematically under-provisioning.
The mechanism is not a credit quality collapse. It is affordability compression applied to extended-term auto loan debt taken on at peak vehicle prices. Prime borrowers who financed new vehicles priced at $50,000 or more with 72- or 84-month terms in 2021 through 2023 are now 30 to 40 months into their loans but have not yet reached positive equity, especially as high interest rates compound the principal. During the same period, rising insurance costs, higher maintenance fees, and unabated inflationary pressures in the economy have made it harder for that same customers’ monthly budget to justify what was once a reasonably safe payment. The New York Fed’s Liberty Street Economics analysis of delinquency by lender type shows that non-captive auto finance companies, which tend to offer more used-vehicle and extended-term loans, are showing the highest levels of deterioration, exceeding pre-pandemic levels and accelerating.
The migration issue represents a challenge of signal discovery. A single delinquency indicator on a prime account is considered an anomaly. When that signal, however, is coupled with a distinct set of signals in tradelines, a partial payment flag in the ML model, and a higher LTV on a 72-month used loan, it indicates a precision-impact segment that might possibly have a two- to three-times higher anticipated prime roll rate. dotData’s Signal Intelligence Platform automatically highlights these compounding interactions—not after the tier migration is evident in aggregate data, but while the account-level signals remain within the intervention timeframe.
Not all delinquency is equal. A 5.2% portfolio-level 90 DPD rate indicates what has already happened. It does not tell the credit committee whether the 2022 origination cohort is performing worse than the 2019 cohort at the same months-on-book, whether the deterioration is concentrated in the 72+ month term bucket, or whether the CECL reserve model has captured the divergence. Vintage cohort analysis answers those questions — and under CECL, answering them is not optional.
The Federal Reserve’s CECL guidance requires institutions, including commercial banks and credit unions, to estimate lifetime expected credit losses from origination, incorporating reasonable and supportable forward-looking forecasts. That means tracking how specific origination cohorts perform as they season, and updating reserve models when cohort trajectories diverge from historical baselines. The 2022–2024 origination vintages are producing that divergence now — weaker cohorts replacing stronger pre-pandemic ones as they age into delinquency buckets — and institutions that are reverting to pre-pandemic historical loss rates beyond a 12-month forecast horizon are understating reserves on exactly the cohorts carrying the most extended-term exposure.
The Signal Discovery Console automates vintage cohort signal discovery across the full multi-table relational data space — joining LOS origination records, LMS payment history, and bureau tradeline data by cohort quarter — to isolate the Driver Signals that explain why specific vintages are deteriorating faster than the model predicts. What a manual analytics team would need weeks to build runs in hours. The Signal Intelligence WorkBench then translates cohort-level findings into Precision Impact Segments the credit committee can act on directly — as a PMA rule, a reserve flag, or a term-length policy adjustment — without a data science intermediary between discovery and deployment.
The highest-value delinquency prediction signals on extended-term portfolios are multi-table temporal interactions — not single-variable flags. LMS payment velocity changes in the first six months on book, combined with tradeline composition at origination and LTV-to-depreciation trajectory over the loan’s life, identify the specific account combinations that roll into 60+ DPD on extended-term cohorts while still appearing within tolerance in the aggregate portfolio dashboard. Standard scorecards evaluate these variables independently. They never account for compounding interactions — and that gap is where losses are forming.
One partial payment flag in month 4 of an 84-month used-vehicle loan means something completely different from the same flag in month 4 of a 48-month new-vehicle loan. The scorecard treats both as equivalent delinquency events. The Signal Intelligence Platform does not. In the dotData, a Driver Signal identifying accounts with an active secured credit card in tradelines within the past two years elevates 90 DPD risk from a 20% portfolio baseline to 39.3%. A second Driver Signal — an active education loan in the same window — independently raises it to 25.6%. Stack both simultaneously, and the resulting Precision Impact Segment represents only 0.443% of historical portfolio volume but carries a 50% default rate — a 30-percentage-point surge above baseline. That is not a hypothesis. It is the automated computation of a multi-table relational interaction that no manual analyst team has the bandwidth to test at scale.
Translate that to the P&L on a $400 million extended-term used vehicle portfolio: the 0.443% segment represents $1.8 million in outstanding balance running at 2.5x the baseline default rate — incremental expected losses the aggregate delinquency rate does not reveal, and the scorecard does not price. Identify five such segments across the book, and the unpriced loss exposure runs into the tens of millions, before CECL provisioning adjustments catch up.
Lengthening loan terms increases three risk dimensions at once: they extend the negative equity duration (pushing back the point at which collateral value surpasses the outstanding balance), slow down amortization (exacerbating loss given default if a default occurs in the first 24 to 36 months), and stretch payment obligations over more economic cycles — elevating the risk that an income disruption will occur before the loan reaches a recoverable equity level. According to Dealertrack data, 84-month loans now account for 22% of new originations, indicating this is a portfolio-level risk management concern, not merely an edge-case issue.
Extended loan terms do not necessarily result in delinquency, but they tend to amplify the negative effects when delinquency does arise. In addition, they extend the period during which borrowers are inclined to resolve their accounts, since a borrower with an outstanding balance significantly higher than the value of their vehicle has a reduced incentive to bring their account current. The pattern is illustrated in Milliman’s analysis: total subprime auto loan delinquencies and broader delinquent auto loan balances exceeded $60 billion in 2025, driven by heightened origination volumes and prolonged terms on depreciating assets.
Three structural forces are concurrently at play: the decline of post-pandemic vintages, where weaker cohorts from 2022 to 2024 are replacing stronger pre-pandemic originations as they transition into months-on-book delinquency; affordability compression, with average new car payments reaching $767 per month in Q4 2025, which is taking up an increasing portion of borrower income; and the accumulation of negative equity due to term extensions. The 90+ DPD rate of 5.2% in Q4 2025, against a $1.66 trillion balance sheet, indicates a significantly larger absolute dollar exposure than the same rate recorded in 2010 on an $800 billion balance sheet.
Vintage cohort analysis tracks the delinquency and loss performance of loans originated in the same period — grouped by quarter or year — as they age through months-on-book. It isolates whether a specific origination cohort is deteriorating faster or slower than historical equivalents at the same seasoning point, allowing risk teams to identify reserve model deficiencies before aggregate portfolio metrics reflect the cohort-level stress. Under CECL, vintage cohort analysis is a core methodology for estimating lifetime expected credit losses — a compliance requirement, not an analytical preference.
CECL requires institutions to estimate lifetime expected credit losses from origination, incorporating forward-looking forecasts that reflect current conditions. For extended-term portfolios, this means the reserve model must account for the full 72- to 84-month loan life — including depreciation curves, negative equity trajectories, and roll rate patterns specific to extended-term cohorts — not just the first 12 to 24 months of historical loss data. Institutions reverting to pre-pandemic historical rates beyond a 12-month forecast horizon are likely understating reserves on 2022–2024 origination vintages.
To predict risk, analyze how data from multiple sources and multiple tables interact over time. Instead of looking at individual risk factors, track how early LMS payment velocity, the composition of tradelines at origination, and LTV-to-depreciation trends affect one another. By using a multivariate approach, lenders can pinpoint the specific account-type profiles that drive 60+ DPD within extended-term cohorts.
dotData Signal Intelligence automates this cross-table analysis by processing millions of relational combinations in hours and uncovering specific driver signals and high-risk segments that static scorecards would miss. Moving to this type of analysis shifts the focus from individual risk variables to the complex interactions among variables.
The highest-value signals are temporal interactions across multiple data tables — specifically, how early LMS payment behavior (partial payments, payment method shifts) combines with specific tradeline types and vehicle collateral characteristics on accounts with 72+ month terms. A single partial payment flag at month 4 on an 84-month used vehicle loan with a specific tradeline composition identifies a Precision Impact Segment with default rates two to three times above the extended-term portfolio baseline — an interaction that no single-table scorecard variable captures independently.
Auto loan delinquency rate reveals the state of the portfolio. Vintage cohorts data reveal which originations are driving the portfolio, and multi-source, multi-table driver signals provide insight into which account types within those cohorts are next, before the roll rate confirms it in the dashboard.
dotData Signal Intelligence surfaces that third layer without displacing the risk team’s judgment or rebuilding the core stack. The SQL goes directly into the existing PMA workflow. Your credit committee gets a defensible, Glass Box rule with documented predictive lift to safeguard the portfolio’s future — and the examiner gets the audit trail they will ask for.
Talk to the dotData lending team →
Key Takeaways Adoption is ahead of delivery: Nearly two-thirds of financial institutions have not yet…
Key Takeaways EPD Vulnerability: Conventional credit decisioning software looks at applicant data in isolated silos…
The Macro Reality: Total U.S. auto loan debt has reached a historic $1.685 trillion, prompting…
Key Takeaways The Velocity Problem: Traditional 30-60 DPD (Days Past Due) reports are lagging indicators…
Key Takeaways The Equity Problem: Depreciation models overlook that 29.3% of current vehicle trade-ins have…
Lenders often focus on the strength of the loan origination scorecard when evaluating lending analytics.…