Credit: a mortgage pool modeled loan by loan
The same mortgage pool declared loan by loan, with the published pool schedule asserted against the aggregate the engine rolls up from its children.
Every number below is checked against an independent reference implementation on every commit — period by period, and on each metric, inside a declared tolerance. See benchmark methodology.
The case
A $100m agency mortgage pool — 8% weighted average coupon, 360-month term, 20% loss severity, twelve-month recovery lag, prepaying at a flat 1% single monthly mortality against a 1% monthly default rate.
It is the same pool as the mortgage pool conventions case, at a different grain. There it is one pool. Here it is four loans of $40m, $30m, $20m and $10m that belong to a pool, and the pool itself holds no contract. Every figure asserted against the pool is an aggregate.
The reference
The industry's own standard formulas for analyzing mortgage-backed securities — the document that defines CPR, SMM, PSA and SDA — and the complete 176-month cash flow schedule it publishes for this pool.
Not redistributable. The publisher forbids reproduction in any form, so the source is neither vendored nor quoted; its figures are carried as anchor values and cited as facts.
The reference publishes four columns: interest, scheduled amortization, voluntary prepayments and principal recoveries. The pool's cash in a period is their sum, so the anchors here are the published figures added together. Addition is the only step taken.
What it exercises
| Pack | credit |
| Declared | five typed assets, one of them a parent; four contract instances |
| Language features | part of hierarchy, typed entity fields, per-instance contract suffixes |
| Conventions | level-pay amortization, SMM on the gross balance, MDR, a lagged recovery |
Two aggregates are asserted, computed by unrelated code:
entity.asset.pool.net_cash_flow— the hierarchy rollup, aggregating the children apart ofrelation names rather than a matching name prefix.domain.credit.gross_collections— the category subtotal, the pack folding four contract instances into one domain line.
Both must reproduce the same published schedule. A defect in either shows as a divergence between them.
The result
25 anchor months on both columns, across a 372-period grid. Every one agrees with the published schedule within the tolerance the source's rounding allows.
The rollup is also exact against the single-pool model: over all 372 periods,
entity.asset.pool.net_cash_flow here and model.net_cash_flow there agree to
zero — not within a tolerance, exactly. Splitting $100m into four unequal
loans changes nothing about the pool's cash.
The delta
Largest residual anywhere: 1.76 dollars, against a tolerance of 2.01.
It is the source's rounding, not arithmetic. Each published figure is given to the whole dollar and up to four are added, so two dollars bounds the difference before any model is run.
Run configuration
{
"deterministic": {
"annual_discount_rate": 0.08
}
}Verified results
Checked period by period: 2 series across 25 periods — 50 values in all, each within ±2.01 of the reference.
entity.asset.pool.net_cash_flowdomain.credit.gross_collections