Bridge (or waterfall): a chart that walks from one number to another in steps, each step one part of the change, so the parts add up exactly to the whole change.
The split in plain words
A rate over a pool is a mix of groups: each group has a share of the pool and its own rate. When the pool's rate moves, part of the move comes from the shares changing (mix) and part from the groups' own rates changing (performance). For each group:
- Mix = (share now − share then) × (rate then + rate now) ÷ 2
- Performance = (share then + share now) ÷ 2 × (rate now − rate then)
Add the parts over the groups and they equal the change exactly.
Worked example: a hundred extensions a year apart
The six-month Extension Default Rate: of extended loans, the share 30+ days late or charged off six months after the extension. Split by whether the loan was current or already late when it was extended.
| When extended | Last year: share | rate | This year: share | rate |
|---|---|---|---|---|
| Current | 60% | 10.0% | 50% | 12.0% |
| Late | 40% | 50.0% | 50% | 50.0% |
| All extensions | 26.0% | 31.0% |
The rate rose 5.0 points.
| Part | Working | Points |
|---|---|---|
| Mix: current | (0.5 − 0.6) × (10 + 12) ÷ 2 | −1.1 |
| Mix: late | (0.5 − 0.4) × (50 + 50) ÷ 2 | +5.0 |
| Performance: current | (0.6 + 0.5) ÷ 2 × (12 − 10) | +1.1 |
| Performance: late | (0.4 + 0.5) ÷ 2 × (50 − 50) | 0 |
| Total | +5.0 |
Mix adds 3.9 points and performance 1.1. Most of the rise came from more extensions going to loans that were already late, not from either group doing much worse.
The real numbers
The deck's bridge compares the latest six months of extensions with the same six months a year earlier, in five groups by lateness when extended. Each month is split with the midpoint rule and the six months are averaged. Counts of loans; an extension counts once its six-month status is reported.
Subprime (FICO under 660 at origination), extensions granted October 2024 to March 2025 against October 2025 to March 2026:
| Lateness when extended | Share then | Share now | Rate then | Rate now | Mix | Performance |
|---|---|---|---|---|---|---|
| Current or under 11 days | 31.4% | 30.5% | 18.5% | 19.1% | −0.18 | +0.17 |
| 11-29 days | 18.4% | 17.5% | 28.2% | 29.6% | −0.28 | +0.24 |
| 30-59 days | 26.9% | 28.6% | 52.7% | 53.3% | +0.89 | +0.17 |
| 60-89 days | 18.1% | 17.5% | 71.2% | 70.6% | −0.47 | −0.12 |
| 90+ days | 5.1% | 6.0% | 77.2% | 75.7% | +0.69 | −0.08 |
| Total | 42.0% | 43.1% | +0.64 | +0.39 |
The rate rose 1.0 point (42.0% to 43.1%): +0.64 from mix, mostly more extensions going to loans already 30-59 or 90+ days late, and +0.39 from performance. Prime rose from 19.7% to 20.4%: +0.15 from mix and +0.50 from performance. Shares and rates in the table are six-month averages, so multiplying them out reproduces the parts only approximately; the parts are computed month by month.
Common pitfalls
- Reading a rate change as better or worse lending. A pool's rate can rise with no group doing worse, just because the mix moved toward riskier groups. Split out the mix before drawing a conclusion.
- Splitting in one order only. Price the mix at last year's rates and it is +4.0 points in the example; at this year's rates, +3.8. Same data, two answers. The midpoint rule above takes the average, so the split does not depend on the order.
- Parts that do not add up. A bridge whose steps do not land exactly on the end number is leaving something out.
Where to see it
- Monthly State of Auto deck, "Extension Default Rate Bridge": prime and subprime, the latest six months against the same months a year earlier, split by lateness at extension.
Source: LoanTape, SEC Form ABS-EE loan-level data, through September 2026.