Function 9 · collections and recovery
How many loans will charge off in the next six months, whom should a collector call first, and how much comes back?
The roll rate chain expects about 1,053 charge offs from 2026-09 to 2027-02; over 19 backtest origins it missed by 6.7% on average, against 21.5% for repeating the last six months. 1,158 loans are past due today, and the first 300 calls in the cure model's queue cover $4,663,027 of the $6,687,255 likely to be lost without one. Two years after a charge off, installment loans return 20.4%.
Six month charge offs: forecast against what happened, by origin
Charge offs in the next six months
Source: generated, Harborline installment and card loans, September 2023 to August 2026, seed 20260831, as of 2026-08-31.
The next six months for the loans open today
Loans
Source: generated, Harborline installment and card loans, September 2023 to August 2026, seed 20260831, as of 2026-08-31.
Cumulative recoveries after charge off
Share of the charged off balance recovered
Source: generated, Harborline installment and card loans, September 2023 to August 2026, seed 20260831, as of 2026-08-31.
How the forecast methods did
| Method | Mean absolute error, share of actual | Mean error, share of actual |
|---|---|---|
| Roll rates with the seasonal adjustment | 7.1% | -6.9% |
| Roll rates without it | 6.7% | -5.5% |
| Naive: the last six months again | 21.5% | -18.6% |
| Seasonal naive: the same months a year earlier | 46.3% | -46.3% |
Monthly cure rates, stated and observed
| From | Stated monthly cure for an average loan | Observed monthly cure |
|---|---|---|
| 30 days | 42.0% | 38.4% |
| 60 days | 22.0% | 19.9% |
| 90 days and over | 10.0% | 5.9% |
The generator states each bucket's cure rate for a loan of average risk. Riskier loans are late more often and cure less, so the observed rates sit below the stated ones.
The collections queue: balance at risk times the chance of not curing
1,158 loans were past due at the end of 2026-08, holding $10,128,599. The first 300 calls in the queue cover $4,663,027 of the $6,687,255 the bank expects to lose without collection, because the order is by money at risk, not by bucket.
| Loan | Product | Bucket | Balance | Probability of curing | Balance likely lost without a call | Why |
|---|---|---|---|---|---|---|
| M0002403 | mortgage | 90+ | $388,967 | 7.6% | $359,423 | a mortgage; months in a row past due |
| M0002578 | mortgage | 60 | $394,693 | 25.5% | $294,141 | a mortgage; balance |
| M0003612 | mortgage | 60 | $369,285 | 24.1% | $280,398 | a mortgage; balance |
| M0003403 | mortgage | 90+ | $292,105 | 6.6% | $272,824 | a mortgage; months in a row past due |
| M0002426 | mortgage | 90+ | $231,470 | 5.0% | $219,891 | months in a row past due; a mortgage |
| M0000916 | mortgage | 30 | $363,804 | 59.3% | $147,965 | a mortgage; balance |
| M0000902 | mortgage | 90+ | $155,855 | 6.1% | $146,326 | months in a row past due; a mortgage |
| M0002595 | mortgage | 90+ | $126,786 | 3.4% | $122,419 | months in a row past due; a mortgage |
| M0003330 | mortgage | 90+ | $121,826 | 7.0% | $113,279 | a mortgage; months in a row past due |
| M0000807 | mortgage | 30 | $277,015 | 60.8% | $108,454 | a mortgage; balance |
| M0001061 | mortgage | 90+ | $79,938 | 11.6% | $70,666 | a mortgage; months in a row past due |
| M0001179 | mortgage | 30 | $86,148 | 63.9% | $31,076 | a mortgage; balance |
| N0015836 | installment | 60 | $47,583 | 37.7% | $29,622 | sixty days past due; balance |
| N0014428 | installment | 90+ | $30,660 | 19.4% | $24,725 | months in a row past due; ninety or more days past due |
| N0003139 | installment | 90+ | $21,111 | 5.0% | $20,065 | months in a row past due; ninety or more days past due |
Source: generated, Harborline loans past due at the end of 2026-08, seed 20260831, as of 2026-08-31. The model and its gates are in the cure validation report.
Method and limitations
- The book is generated. Its delinquency follows a stated process with no calendar season, which is why the seasonal adjustment could not help.
- The forecast holds the last six months' transitions and adds no new loans. A change in underwriting or the economy breaks the first assumption at once.
- Recoveries are observed for two years after charge off; anything collected later is not in the loss given default.