Function 2 · consumer lending
Is the installment book performing as priced, and what does it expect to lose?
Not quite. The two vintages written under the loosened cutoff reached 10.4% cumulative default by twelve months on book, against 4.0% for the vintages before them. Expected loss on the open installment book is $8,409,265, 5.53% of $152.1M exposure, after an observed to expected overlay that adds $3,484,210 because the scorecard under predicts those vintages. Every portfolio limit is within its threshold.
Cumulative default rate by months on book, one line per origination quarter
- Every other vintage
- 2024Q4 (loosened)
- 2025Q1 (loosened)
Cumulative default rate
Source: generated, Harborline installment loans, seed 20260831, as of 2026-08-31.
Monthly transitions between delinquency buckets
Rows: bucket this month. The further a cell sits from the background, the larger its share, on a log scale.
Source: generated, Harborline installment loans, seed 20260831, as of 2026-08-31.
Share of loans not yet defaulted, by score band at application
Share not yet defaulted
Source: generated, booked installment loans, seed 20260831, as of 2026-08-31.
Expected loss by score band
Probability of default from the scorecard with the vintage overlay, loss given default from recoveries two years after charge off, exposure as the balance at the as of date.
| Score band | Loans | Exposure | PD (with overlay) | LGD | Expected loss | EL rate |
|---|---|---|---|---|---|---|
| under 560 | 371 | $4.6M | 23.69% | 79.6% | $889,783 | 19.40% |
| 560 to 599 | 7,032 | $71.2M | 9.82% | 79.6% | $5,787,865 | 8.13% |
| 600 to 639 | 6,701 | $62.4M | 3.16% | 79.6% | $1,625,988 | 2.61% |
| 640 and over | 1,724 | $14.0M | 0.91% | 79.6% | $105,629 | 0.75% |
Source: generated, Harborline installment loans, seed 20260831, as of 2026-08-31.
Champion and challenger
The challenger ranks 0.005 AUROC better and agrees with the scorecard's ranking at 0.962. It was not promoted. A gradient boosted model cannot state Regulation B reasons directly, so a decline would be explained by a different model from the one that made it, and the gain is smaller than the promotion gate its report states.
| Measure, test months | Scorecard (champion) | Monotone GBM (challenger) |
|---|---|---|
| AUROC | 0.662 | 0.667 |
| Gini | 0.324 | 0.334 |
| KS | 0.238 | 0.250 |
| Brier score | 0.0626 | 0.0624 |
| Expected calibration error | 0.0286 | 0.0279 |
| Stability, development to test (PSI) | 0.0032 | 0.0035 |
| Validation conclusion | approved with conditions | not approved |
Both reports: pd_scorecard and pd_challenger.
A CECL style illustration, simplified
Twelve month default probability with the overlay, stretched to a lifetime by the ratio of cumulative default at thirty months to twelve on the vintages old enough to show both (2.71 times), then scaled by a macro factor of 0.898 for the supervisory baseline's average unemployment of 4.5%. It is labelled simplified because a real allowance adds qualitative factors, reversion and prepayment this does not.
| Vintage | Exposure | 12 month PD | Lifetime PD | Macro factor | Lifetime loss |
|---|---|---|---|---|---|
| 2023Q3 | $0.8M | 4.24% | 11.46% | 0.90 | $64,287 |
| 2023Q4 | $2.9M | 3.33% | 9.01% | 0.90 | $186,901 |
| 2024Q1 | $4.2M | 2.95% | 7.98% | 0.90 | $241,781 |
| 2024Q2 | $5.2M | 3.51% | 9.49% | 0.90 | $351,072 |
| 2024Q3 | $6.7M | 4.83% | 13.07% | 0.90 | $625,941 |
| 2024Q4 | $8.8M | 9.07% | 24.55% | 0.90 | $1,551,275 |
| 2025Q1 | $11.0M | 10.04% | 27.17% | 0.90 | $2,132,590 |
| 2025Q2 | $12.5M | 6.39% | 17.28% | 0.90 | $1,543,531 |
| 2025Q3 | $15.4M | 4.47% | 12.10% | 0.90 | $1,335,543 |
| 2025Q4 | $18.8M | 6.56% | 17.74% | 0.90 | $2,383,928 |
| 2026Q1 | $21.2M | 6.86% | 18.56% | 0.90 | $2,808,148 |
| 2026Q2 | $25.6M | 6.92% | 18.72% | 0.90 | $3,429,950 |
| 2026Q3 | $19.0M | 6.85% | 18.54% | 0.90 | $2,515,904 |
Source: generated, Harborline installment loans, seed 20260831, as of 2026-08-31.
What drives time to default
A Cox model on the same loans, concordance 0.704. A proportional hazards p value under the usual threshold means that covariate's effect changes with loan age and its ratio is an average.
| Covariate | Hazard ratio | 95% low | 95% high | PH test p value |
|---|---|---|---|---|
| Score, per 20 points | 0.595 | 0.570 | 0.621 | 0.990 |
| Loosened vintage | 1.960 | 1.761 | 2.182 | 0.003 |
| Unemployment at origination, per point | 1.120 | 1.043 | 1.201 | 0.016 |
Loans still at risk, by month
| Score band | At risk at 0 | 6 | 12 | 18 | 24 |
|---|---|---|---|---|---|
| under 560 | 587 | 428 | 294 | 189 | 77 |
| 560 to 599 | 9,893 | 7,214 | 4,805 | 2,945 | 1,538 |
| 600 to 639 | 8,782 | 6,464 | 4,368 | 2,793 | 1,616 |
| 640 and over | 2,216 | 1,656 | 1,174 | 756 | 446 |
Source: generated, booked installment loans, seed 20260831, as of 2026-08-31.
Portfolio limits
| Limit | Measured | Limit | Status |
|---|---|---|---|
| Loosened vintages' share of installment balances | 13.03% | 35.00% | ✓ within |
| Installment balances 30 or more days past due | 2.58% | 4.00% | ✓ within |
| Nonperforming loans to gross loans | 0.54% | 1.00% | ✓ within |
| Mortgages' share of gross loans | 72.02% | 75.00% | ✓ within |
Source: generated, Harborline installment loans, seed 20260831, as of 2026-08-31.
The overlay, vintage by vintage
Observed default over the scorecard's mean predicted default, by origination quarter. The most recent vintages carry 1.78.
| Vintage | Observed over expected | Loans |
|---|---|---|
| 2023Q3 | 1.15 | 399 |
| 2023Q4 | 0.95 | 1,321 |
| 2024Q1 | 0.81 | 1,455 |
| 2024Q2 | 0.99 | 1,494 |
| 2024Q3 | 1.34 | 1,580 |
| 2024Q4 | 2.20 | 1,897 |
| 2025Q1 | 2.42 | 1,998 |
| 2025Q2 | 1.72 | 1,822 |
| 2025Q3 | 1.19 | 1,222 |
Method and limitations
- The loan book is generated, with a stated risk structure the pipeline recovers before it is trusted. It is not any real lender's portfolio.
- Default is 90 days past due or charge off. A curve stops where its youngest loan's history ends; nothing is extrapolated past the as of date.
- The CECL figure is an illustration of the method on this book, not an allowance. It has no qualitative factors, no reasonable and supportable forecast period with reversion, and no prepayment model.