Commercial property · historical backtest
Validation
What the guideline would have decided, and what happened afterwards.
- Bound property policies
- 27
- Premium
- $6.7M
- Incurred losses
- $12.6M
- Aggregate loss ratio
- 1.89×
Historical outcome cohort
Across 27 policies
Paid plus reserved, including expense
Incurred divided by premium
PR-2026-1081
The policy met appetite when it arrived. Its later incurred losses show why appetite fit alone cannot establish a profitable risk.
- Incurred losses
- $629,200
- Premium
- $58,800
- Loss ratio
- 10.70×
Outcomes by engine decision
Each policy was scored as of its received date. Loss history includes only earlier claims.
| Decision | Policies | Premium | Incurred | Loss ratio |
|---|---|---|---|---|
| accept | 1 | $59K | $629K | 10.70× |
| open | 0 | $0 | $0 | No policies |
| decline | 26 | $6.6M | $12M | 1.81× |
| All property | 27 | $6.7M | $12.6M | 1.89× |
Small historical sample. These results do not establish predictive accuracy or production readiness.
Where the book breaks the guideline
Bound policies that each written factor would decline. A policy can appear under several factors.
What the existing enrichment changed
The recorded comparison switches FEMA, USGS, weather and portfolio adjustments on and off across 38 scored submissions.
no decision tier changed: the property declines are structural hard fails (state, age, loss history), which external layers cannot move
| Open submission | Largest contribution | Without enrichment | With enrichment | Midpoint change |
|---|---|---|---|---|
| Case #143 | earthquake · ×1.06 | 30 | 24.1 | -5.9 |
| Case #126 | severe weather · ×1.06 | 30 | 26.1 | -3.9 |
| Case #141 | severe weather · ×1.06 | 30 | 26.1 | -3.9 |
The new county and climate map layers are context only. This report does not test them, and they do not change the appetite score.
Human decline reasons · 11 decisions
3 broker withdrawals are excluded. This cohort spans several business lines; routing a case outside the property guideline is not agreement with a human decline.
| Human reason | Cases | Business lines | Desk responsibility | Matching reason |
|---|---|---|---|---|
| loss history | 3 | cyber 1, health 1, property 1 | Appetite | 1 |
| cat exposure aggregation | 2 | auto 1, health 1 | Portfolio | 0 |
| outside appetite | 2 | health 2 | Appetite | 0 |
| insufficient controls | 4 | auto 1, cgl 2, health 1 | Intake | 0 |
| broker withdrew | 3 | cgl 1, cyber 1, lpl 1 | Excluded | Excluded |
Method, cohorts & limitations
Outcome cohort: 27 bound property policies. Status mix: active 15, expired 9, non renewed 3. Enrichment cohort: 38 scored property submissions. These are different denominators.
Loss ratio is incurred losses divided by premium, not a probability of a claim. The uncertainty range comes from written rules and incomplete facts; it is not a statistically calibrated confidence interval.
Pre-registration reference: 5351f2340c0009d42e5a68c8d8eb41738f6ffa18
Recorded backtest snapshot against the 2025 property guideline. Changing the current rulebook or adding map layers does not rerun this report. The report JSON contains the underlying policy and case rows.