Traditional Life Pricing

M-071 is the traditional life pricing model (TraditionalLifeModel, engine id insmodel.L4.traditional_life, internal P-006). A single engine prices both Term Life and Whole Life off one parameter, product_subtype, because the two share an intended use — new-business premium adequacy and profit testing for mortality-driven protection products — and differ only in cash-value mechanics. It composes five engines declared in get_required_engines(): MortalityEngine (death cost), LapseEngine (surrender/lapse decrement), InvestmentEngine (portfolio yield on reserves), ExpenseEngine (operating cost), and ReserveEngine (GAAP FPB reserve).

calculate() projects a cohort monthly. For each month it derives the attained age and policy year, then:

1. Mortality. Annual qx is built from a base rate, an age slope, and a multiplicative experience factor (_get_mortality_rate()), clipped to [0.0001, 1.0], then converted to a monthly decrement 1 − (1 − qx)^(1/12). Deaths drive death_benefit = deaths · face.
2. Lapse. Annual lapse comes from a duration-graded base curve (_get_lapse_rate()). Term carries a shock lapse at the end of the level term (the renewal cliff) and an elevated post-term lapse (3× base, capped at 25%); Whole Life lapse is halved (cash-value retention). Converted to a monthly decrement.
3. Premium. premium_inflow = in_force · annual_premium / 12 — level premium, the pricing lever the model solves around.
4. Investment income. reserve · monthly_yield, where monthly_yield = (1 + portfolio_yield)^(1/12) − 1.
5. Whole-Life-only cash mechanics. Cash surrender value (_get_cash_surrender_value(), linear growth from csv_start_year, capped at 95% of face) drives surrender outflow, policy-loan income (loan_balance · loan_rate / 12), and dividends (csv · dividend_rate / 12). Term sets all of these to zero.
6. Expense. Per-policy maintenance (annualised /12) plus a percent-of-premium load.
7. Net income and reserve roll-forward. net_income = premium + investment_income + loan_income − death_benefit − surrender − dividend − expense; the reserve rolls forward by the same flow and is floored at zero. DAC is set up at issue as in_force · annual_premium · dac_rate and amortised straight-line over dac_amortization_years.

In-force is decremented each month by deaths + lapses; projection stops at the term horizon (Term: term_years + 2) or when in-force falls below 1.

Profit testing (compute_profitability()) re-runs calculate() on a fresh pricing cohort and computes: PVFP (monthly net income discounted at the risk-adjusted rate), profit margin (PVFP / initial annual premium volume), an IRR approximation (total_profit / premium_volume / years — not a true cash-flow IRR solve), return on capital (PVFP over a capital charge of 3% of face for Term, 5% for Whole), breakeven year (first month cumulative net income turns positive), and loss ratio (total death benefits / total premiums). get_cash_flows() re-runs the projection across multiple interest-rate scenarios by averaging each scenario's yield.

The load-bearing assumptions for the canonical Term run. All are engine parameters with defaults in PRODUCT_CONTRACT["parameters"]; the snapshot uses the values asserted by tests/test_traditional_life_model.py::make_model.

Operational notes. Mortality and lapse are the two behavioural drivers that move pricing: mortality sets the benefit cost, lapse sets how long premium and reserve persist. The crediting/cash-value parameters are product-design inputs (contractual), not calibrated experience. The reserve is not a statutory or VM-20 reserve — it is a simplified roll-forward seeded at in_force · face · 0.01 · duration and is a pricing artefact, not a held reserve (M-001 owns VM-20 statutory reserves).

Deterministic single-cohort run of TraditionalLifeModel (P-006) in its TERM configuration — reproducible, requires no live firm data (python scripts/model_snapshots.py M-071 in InsModel; mechanics asserted by tests/test_traditional_life_model.py, 32 tests passing). The provider is a MagicMock.

Input: issue age 40 · Male · face $500,000 · level premium $1,200/yr · 20-yr term · base qx 0.002 · experience factor 0.80 · base lapse 8% Y1 → 2% · shock lapse 30% · portfolio yield 5% · 10,000 in-force · profit test at discount 6% over 240 months. Cash-flow rows shown are month 1; pricing metrics are from compute_profitability().

The two pricing drivers are visible directly: annual qx of 0.00172 (the benefit cost) and year-1 lapse of 8% (the persistency assumption). The PVFP and margin are pricing diagnostics from compute_profitability(), not statutory profit. Two numbers deserve honest reading. First, loss ratio 1.14 (>1): total death benefits exceed total premiums over the horizon, which on a standalone premium basis would be inadequate — yet net income is positive because the model credits investment income on a large seeded reserve ($50M at issue = 10,000 · 500,000 · 0.01 · 1). Second, profit margin 1.61 is implausibly high for term pricing and is a direct artefact of that reserve seeding inflating investment income relative to a deliberately low illustrative premium. These figures are internally consistent engine outputs on the documented canonical inputs; they are not calibrated to a real product's priced premium and should be read as mechanics demonstrations, not a profit claim. See Limitations 1, 3, and 5.

Captured 2026-06-04 · deterministic, no live data.

1. Reserve is a pricing artefact, not a statutory reserve. The reserve is seeded at in_force · face · 0.01 · duration and rolled forward by net cash flow, floored at zero. It is neither a VM-20 net-premium/deterministic reserve (M-001) nor a GAAP FPB reserve (M-020), despite ReserveEngine being a declared dependency. The seeded magnitude drives investment income and therefore inflates PVFP and profit margin — the snapshot's margin of 1.61 is a direct consequence and must not be read as a profit claim.

2. Mortality is base-plus-linear-slope, not a CSO table. _get_mortality_rate() applies (base_rate + age·slope)·experience with a single linear slope, clipped to [0.0001, 1.0]. It does not load the 2017 CSO select-and-ultimate structure, has no gender split beyond carrying a gender parameter that does not alter qx, and no smoker/non-smoker distinction. Acceptable as a pricing illustration; insufficient for a priced mortality basis.

3. Premium is an input, not solved. The model profit-tests a given level premium; it does not solve for the premium that hits a target margin or IRR. "Premium adequacy" is assessed only by inspecting the resulting loss ratio / PVFP. The canonical $1,200 premium on $500K face produces loss ratio > 1, illustrating exactly this gap — the harness reports inadequacy rather than correcting it.

4. Profitability metrics are approximations. IRR is total_profit / premium_volume / years, not a cash-flow IRR solve; return-on-capital uses a flat factor (3% face Term / 5% Whole) rather than a risk-based capital model; breakeven is read off cumulative undiscounted net income. Directional, not decision-grade.

5. Deterministic, single-path projection. No stochastic mortality, no dynamic (moneyness/rate-dependent) lapse, no stochastic interest. get_cash_flows() only averages each scenario's yield to a scalar — it does not run a genuine multi-path projection.

6. DAC is straight-line, not EGP-amortised. DAC amortises on a fixed straight-line schedule rather than in proportion to estimated gross profits/margins, diverging from LDTI/ ASC 944 acquisition-cost mechanics.

7. Whole-Life cash mechanics are linear approximations. CSV grows linearly from csv_start_year (capped at 95% of face); dividends and policy-loan income are flat-rate functions of CSV. No guaranteed nonforfeiture-law minimum test, no dividend-scale interest crediting, no policy-loan utilisation dynamics.

8. No live firm-data path (BV-032). All figures derive from documented canonical inputs and a MagicMock provider. Nothing here is a 10-K-reported or experience-calibrated value; the PRU calibration notes in the source docstring are illustrative targets, not validated outputs.

Tracked for ratification. Remediation status of the items noted in earlier passes: the tier-2 sensitivity suite on the top-3 load-bearing assumptions (mortality experience/base+slope, base lapse curve, shock lapse rate) is now DONE (tests/test_traditional_life_sensitivity.py, 13 tests; COND-071-02). The registry-vs-code engine-binding divergence is RESOLVED (COND-071-01; see Components). Still open: a Whole-Life-archetype deterministic snapshot (only TERM is snapshotted today); premium-adequacy / margin calibration (the engine profit-tests a supplied premium but has no premium-solver, and the canonical inputs produce loss ratio 1.14 / margin 1.61 inflated by a seeded reserve); the reserve-seeding pricing artefact (in_force·face·0.01·duration) that inflates investment income and PVFP; and the mortality basis itself (base+linear-slope clipped, not a CSO select-and-ultimate table; gender carried but inert; no smoker split; get_cash_flows() averages each scenario's yield to a scalar rather than running multi-path projection). None of these were changed in this pass.

Engine source: - ecosystem/InsModel/Models/firmmodel/products/traditional_life_model.py — TraditionalLifeModel (P-006), engine id insmodel.L4.traditional_life. - firmmodel/products/base_product.py — shared BaseProductModel contract (calculate, compute_profitability, get_termsheet, validate_input). - firmmodel/governance/legacy_metadata/TraditionalLifeModel.yaml — extracted component metadata.

Tests / snapshot: - tests/test_traditional_life_model.py (32 tests). - scripts/model_snapshots.py M-071.

Pricing / profit-testing literature: - ASOP No. 56 — Modeling (intended use, sensitivity, reliance, documentation). - ASOP No. 25 — Credibility Procedures (experience-factor application to base mortality). - ASOP No. 2 — Nonforfeiture and Policyholder Dividends (Whole-Life CSV / dividend context). - Atkinson & Dallas, Life Insurance Products and Finance (SOA) — asset-share / profit-testing methodology and PVFP construction for traditional life pricing. - Black & Skipper, Life Insurance — term vs whole-life premium structure, the renewal shock-lapse cliff, and persistency in pricing. - SOA 2017 Commissioners Standard Ordinary (CSO) Mortality Tables — the published mortality basis the model's base+slope approximates (not loaded; see Limitation 2).

Internal: - M-001 — VM-20 statutory reserves (engine-card depth exemplar; owns the held reserve this model only approximates). - M-070 — annuity pricing suite (sibling pricing card; shared BaseProductModel harness). - BV-032 — firm-data divergence (no live-data path).

Card change history. Code-side change history lives in git log of the component files.

• 2026-05-08 — stub created from registry data per Decision 023 Phase 5 / B-07.
• 2026-06-04 — hand-authored pass: Methodology, Key Assumptions, Output Snapshot, Limitations, Validation Packet, and References authored from traditional_life_model.py + TraditionalLifeModel.yaml legacy_metadata + bound A-NNN entries. Stub marker advanced to ``.
• 2026-06-06 — code-grounded documentation-accuracy pass: added Standards Coverage (CSO_2017 / ASOP_56 / ASOP_25 / ASC_944-40-LDTI, pointers only) and Dependencies (upstream models: none — registry declares no upstream_models) sections to match the M-001 section shape; reconciled the Components processing-engine roster to get_required_engines() (Mortality/Lapse/Investment/Expense/Reserve — benefit_engine is registry-bound but not returned by the code) with a registry-vs-code divergence note; added honest Validation Packet governance-status rows (2L challenge pending, sensitivity suite pending, back-test pending, next revalidation 2028-05-01) from registry lifecycle; added a tracked-for-ratification note under Limitations. No model outputs, back-test numbers, or validation results were fabricated or changed.
• 2026-06-06 — RAT-071-v1.0.0 remediation pass (1L, Decision 053 charter §2.2): COND-071-01 (binding) RESOLVED — model_registry.yaml components.processing.engines aligned to get_required_engines(): dropped benefit_engine (bound-but-unused), added investment_engine, expense_engine, reserve_engine_ldti (exercised-but-unbound); engine_registry.yaml model_membership lists made bidirectionally consistent; binding manifest regenerated; check_registry_integrity OK. COND-071-02 (sensitivity) DONE — authored tests/test_traditional_life_sensitivity.py (13 tests, shock-direction + bounded- magnitude on mortality / base lapse / shock lapse) and assembled the Tier-2 validation-evidence pack at modelling/validation_evidence/M-071/v1.0.0/README.md. Reconciled documentation_pack flags to disk reality (card/validation/change_log now present). Re-stamped doc-currency (fingerprint eb48b20cf61fbe10, current). COND-071-03 (independent sign-off) is left for 2L on re-review (1L may not record its own sign-off). The BV-032 back-test remains an honest pending item. No model outputs or validation results were fabricated.