How does loss rate evolve per cohort?
№ 109 · Loss rate cohort
Definition
MetricAnnualised net credit loss rate by origination cohort, tracked by months on book
Unit% of average outstanding balances, annualised, net of recoveries
Triangular grid showing cumulative default or loss rate per origination cohort. Reads upward over time. Build by tracking losses per cohort cumulatively. Newer cohorts beating older at same age = underwriting improving. Example: 2023 cohort 30% loss at M24; 2025b cohort 3% loss at M3.
Benchmarks
| Bottom 30% | Median | Top 30% |
|---|---|---|
| Above 6% annualised, or any book where successive cohorts peak later and higher, which signals adverse selection rather than a credit cycle | About 4.0% to 4.2% annualised net charge-offs on US bank credit card books, against a ten-year average of 3.40% | Below 3.0% annualised net charge-offs on a revolving consumer book, with each cohort's loss curve peaking earlier and lower than the one before it |
US commercial bank credit card loans, annualised net charge-off rate net of recoveries, Q4 2024 to Q4 2025, plus LatAm digital bank non-performing loan ratios for a fintech-native reference. Not comparable to secured or BNPL books. · Charge-off rate and NPL ratio are different metrics and are often confused on the same chart. The FRED series is annualised net charge-offs, so it is a flow. Nubank's 4.1% and 6.8% are 15-to-90-day and 90-plus non-performing stocks, which lead charge-offs by one to two quarters and cannot be read against the 4.11% US figure as if they were the same thing. Label the axis. The US series is also cyclical rather than structural: 10.54% at the 2009 peak, 1.63% at the 2021 trough, 4.11% now, so a cohort chart should always carry the vintage's macro context. Bank size barely separates the outcome, with all commercial banks at 4.11% and the 100 largest at 4.03% in Q4 2025.
| Category | Bottom 30% | Median | Top 30% |
|---|---|---|---|
| US bank credit card loans, annualised net charge-offs | 4.58% and above, the Q4 2024 level | 4.11% at Q4 2025 | 3.40% or below, the ten-year average |
| US bank credit card loans at the 100 largest banks | not separately published | 4.03% at Q4 2025 | not separately published |
| US commercial bank loans, all categories | 0.65% at Q4 2024 | 0.58% to 0.65% | 0.58% at Q4 2025 |
| LatAm digital bank, non-performing loan stock (not charge-offs) | 6.6% to 6.8% at 90 plus days | 4.2% to 4.4% at 15 to 90 days | 4.1% at 15 to 90 days |
When it looks bad
Successive cohort curves stack upward and to the right, so each new vintage peaks later and at a higher loss rate than the one before it, which means underwriting is loosening faster than the book is seasoning and the blended rate is being held down only by the size of the older cohorts.
The 2024 vintage peaked at 5.1% around month 14. The 2025 vintage is at 4.8% by month 9 and still climbing. Blended loss rate reads 4.3% and looks stable, but the vintage that is now 60% of new originations is tracking above 7%.
What to do about it
- Report blended loss rate and vintage loss rate side by side, and treat the blended figure as decoration. A blended rate that is flat while vintages deteriorate is the standard way a lending book gets into trouble without anyone seeing it in a monthly pack.
- Watch the early delinquency bucket as the leading indicator rather than waiting for charge-offs. Nubank's 15-to-90-day NPL moved 30 bps and then 20 bps down across successive quarters before the 90-plus stock followed, which is the lag structure a cohort chart should be built around (Nu Holdings Q2 and Q3 2025).
- Cut the vintage curves by acquisition channel and by approval band, since adverse selection almost always enters through a specific channel rather than across the whole funnel, and channel-level tightening is cheaper than a blanket score cutoff.
- Set the loss ceiling against the macro level, not against last year. US card charge-offs ran at 1.63% in 2021 and 4.11% in Q4 2025, so a model calibrated on 2021 vintages understates a 2026 book by more than two points of loss (FRED CORCCACBS).
Sources
- Board of Governors of the Federal Reserve System, via FRED Annualised net charge-off rate on credit card loans of 4.11% in Q4 2025, down from 4.58% in Q4 2024, against a 2009 peak of 10.54% and a 2021 trough of 1.63%. fred.stlouisfed.org ↗
- Board of Governors of the Federal Reserve System, via FRED Credit card charge-off rate of 4.03% at the 100 largest banks in Q4 2025. fred.stlouisfed.org ↗
- Board of Governors of the Federal Reserve System, via FRED Credit card delinquency rate of 2.94% at Q4 2025, easing from 3.08% at Q4 2024. fred.stlouisfed.org ↗
- Board of Governors of the Federal Reserve System, via FRED Annualised net charge-off rate across all loan categories of 0.58% in Q4 2025, down from 0.65% a year earlier. fred.stlouisfed.org ↗
- Zogby, tracking FRED series CORCCACBS The Q4 2025 credit card charge-off rate of 4.11% sits 0.71pp above the ten-year average of 3.40%, with decreases in three of the last four quarters. zogby.com ↗
- Nu Holdings (Nubank) ARPAC of USD 15.0 and cost to serve of USD 0.80 per active customer per month, with an efficiency ratio of 19.9% and ROE of 33%; 15 to 90 day NPL declined to 4.1%. international.nubank.com.br ↗
- Nu Holdings (Nubank) 15 to 90 day NPL of 4.2%, down 20 bps quarter on quarter, and 90+ NPL of 6.8%, up 20 bps in line with seasonality. international.nubank.com.br ↗
- Nu Holdings (Nubank) 15 to 90 day NPL of 4.4%, down 30 bps quarter on quarter, and 90+ NPL of 6.6%, reflecting earlier delinquency feeding through. international.nubank.com.br ↗