Assumption Setting

Assumption Setting

When to Use

• Define or update assumption sets for pricing, reserving, capital, or ALM models
• Document assumption rationale, sources, and effective dates for governance or audit
• Classify assumptions by category (economic, demographic, behavioral, operational)
• Apply selection methodology (experience, industry tables, judgment, blending)
• Design sensitivity and scenario grids tied to material drivers
• Establish governance: owners, approval tiers, change control, version history
• Build assumption packs differentiated by use case (rate indication vs valuation vs ORSA)
• Reconcile emerging experience vs last-approved assumptions before model runs
• Benchmark assumptions against industry or regulatory reference points (overview)

When NOT to Use

• Execute full pricing indications, IBNR triangles, or reserve ultimates as primary deliverable → actuary
• Scope actuarial consulting engagements, SOW, due diligence, or opinion support governance → actuarial-consulting
• Teach P&C coverages, FNOL, or underwriting without assumption governance lens → property-casualty-insurance
• Teach life/health products, benefits, or distribution without assumption pack focus → life-health-insurance
• Pension funding, ERISA, or retirement plan design → pension-retirement-funds
• Corporate budgets, investor metrics, or non-insurance FP&A → financial-analyst (if installed)
• Executive strategy without actuarial/risk assumption workstream → business-consultant
• Legal interpretation of policy wording, filings, or statutory sign-off → commercial-counsel
• IFRS note legal wording or general ledger recognition → ifrs
• SOC 2 / ISO control evidence without actuarial models → compliance-engineer

Related skills

Core Workflows

1. Scope the assumption exercise

1. Purpose — Pricing, valuation, capital, ALM, planning, or stress testing 2. Model — Name, version, owner; which outputs consume each assumption 3. Materiality — Drivers ranked by impact on key metrics (LR, reserves, SCR, surplus) 4. Population — In-force, new business, closed block; segment dimensions 5. Effective date — Align with valuation date, rate filing, or budget cycle

See `references/assumption_setting_scope_and_principles.md`.

2. Inventory categories and sources

1. List assumptions by category (economic, demographic, behavioral, operational) 2. Map each to source (company experience, industry table, vendor, judgment) 3. Flag prescribed vs company-specific and jurisdictional constraints 4. Note dependencies (e.g., lapse tied to crediting rate; trend tied to inflation) 5. Record data period and quality caveats per assumption

See `references/assumption_categories_and_sources.md`.

3. Select and justify assumptions

1. Run or reference experience studies (A/E); coordinate with actuary for technical fitting 2. Apply credibility blending where data is thin 3. Document expert judgment when overriding data 4. Set level, trend, and volatility components separately when material 5. Compare to benchmarks and prior approved set; explain deltas

See `references/selection_methodology_and_judgment.md`.

4. Govern, document, and control changes

1. Assign owner and approver per assumption or pack tier 2. Version the assumption pack (ID, date, model compatibility) 3. Maintain change log: prior → proposed → approved, with rationale 4. Attach exhibits: A/E tables, sensitivity summary, peer review sign-off 5. Archive reproducible data cuts per model risk policy

See `references/governance_documentation_and_change_control.md`.

5. Sensitivities, scenarios, and stress

1. Identify top drivers from materiality or prior sensitivity work 2. Define base, adverse, favorable, and regulatory/stress scenarios 3. Keep grids parsimonious—avoid combinatorial explosion 4. Align scenarios with risk appetite and ORSA/capital narrative (overview) 5. Document correlations or explicit independence assumptions

See `references/sensitivity_scenarios_and_stress.md`.

6. Line-of-business assumption packs

1. Select pack template: P&C, life/health, pension, or multi-line 2. Pull LOB-specific drivers (e.g., development factors vs mortality improvement) 3. Cross-check pack against product and regulatory context skills 4. Separate packs for pricing vs reserving vs capital when conventions differ 5. Hand off model execution to actuary after pack is approved

See `references/line_of_business_assumption_packs.md`.

Assumption pack checklist

Before models run, confirm:

• [ ] Every material driver has owner, source, and effective date
• [ ] Pricing vs valuation vs capital basis differences documented
• [ ] Changes from prior pack bridged with quantified impact where possible
• [ ] Sensitivities on top 3–5 drivers completed or scheduled
• [ ] Judgment overrides explicitly approved and time-bounded if interim
• [ ] No legal, statutory, or accounting sign-off claimed by the agent

Deliverable standards

State uncertainty and data limitations. Do not present outputs as appointed-actuary, legal, or audit opinions.

When to load references

• Scope and principles → references/assumption_setting_scope_and_principles.md
• Categories and sources → references/assumption_categories_and_sources.md
• Selection and judgment → references/selection_methodology_and_judgment.md
• Governance and change control → references/governance_documentation_and_change_control.md
• Sensitivity and stress → references/sensitivity_scenarios_and_stress.md
• LOB packs → references/line_of_business_assumption_packs.md

Assumption categories and sources

Table of contents

1. Category taxonomy 2. Common drivers by category 3. Source hierarchy 4. Prescribed vs company-specific 5. Benchmarking and industry data 6. Data quality and metadata

Category taxonomy

Tag each assumption with one primary category; note secondary links (e.g., lapse behavioral + economic).

Common drivers by category

Economic

• Risk-free or portfolio discount rate curve
• Inflation (medical, wage, general)
• Interest rate paths for ALM and dynamic lapse
• FX (multi-currency blocks)

Demographic

• Mortality / longevity tables and improvement scales
• Morbidity incidence and continuance
• Retirement age and turnover (group benefits)

Behavioral

• Lapse / persistency by duration and product
• Withdrawal and partial surrender (annuities, UL)
• Renewal and re-underwriting take-up
• Claim reporting and settlement patterns (as behavior of emergence)

Operational

• Expense ratios (acquisition, maintenance)
• Commission and premium taxes
• Reinsurance terms embedded as assumptions (cession %, attachment)

Underwriting / claims (P&C-heavy)

• Frequency and severity by peril
• Loss development factors and tail
• Catastrophe models or load factors
• Large loss thresholds and pooling

Source hierarchy

Prefer the highest-quality source available for the decision:

Always record study period, exposure basis, and segmentation used to derive company experience.

Prescribed vs company-specific

Do not substitute legal advice on what is prescribed; outline questions for qualified reviewers.

Benchmarking and industry data

Use benchmarks to sense-check, not to copy blindly:

• Industry experience studies (mortality, LTC, auto, etc.)
• Ratemaking organizations and bureau filings (P&C)
• Reinsurance market commentary (trend, cat)
• Rating agency and public peer disclosures (high-level)

Document differences in mix, geography, and underwriting when deviating from benchmark.

Data quality and metadata

Minimum metadata per assumption in the register:

Flag stale assumptions when experience period ended before material market or portfolio shifts.

Assumption setting scope and principles

Table of contents

1. What assumption setting covers 2. Principles 3. Use-case boundaries 4. Relationship to models 5. Ethics and reliance

What assumption setting covers

Assumption setting is the discipline of choosing, documenting, and governing inputs that models treat as fixed for a given run—distinct from model structure (formulas, segmentation) and model execution (coding, runs).

Typical scope:

Also includes strategic and planning assumptions when they feed actuarial or risk outputs (e.g., new business volume, mix shifts)—not standalone corporate FP&A.

Principles

1. Purpose-aligned — Same driver may differ for pricing vs statutory valuation vs economic capital; document basis explicitly. 2. Traceable — Every assumption links to source (study, table, policy, judgment) and approver. 3. Consistent — Cross-model dependencies (e.g., trend in price and reserve) reconciled or divergence explained. 4. Materiality-driven — Effort scales with impact on decisions and metrics. 5. Conservative where required — Regulatory or risk frameworks may bias direction; state framework, do not invent rules. 6. Transparent judgment — Overrides require written rationale, not undocumented tweaks. 7. Versioned — Effective dates, pack IDs, and change logs for audit replay.

Use-case boundaries

Not in scope: legal interpretation of filing requirements, appointed-actuary sign-off, or external audit opinions.

Relationship to models

Data → Assumption pack → Model engine → Outputs → Decisions
         ↑ governance

• Assumption pack — Versioned bundle consumed by one or more models
• Model — Defines how assumptions combine (e.g., frequency × severity)
• Experience study — Often inputs to assumption updates; technical execution may sit in actuary

When the user only needs triangles, ultimates, or indications without governance documentation, route to actuary.

Ethics and reliance

• Disclose conflicts when the same party sets assumptions for both pricing and reserving without independent review
• Flag one-time events (pandemic, reform, catastrophe year) affecting experience
• Distinguish indicated assumption from implemented (commercial or regulatory constraint)
• Refer legal and filing questions to commercial-counsel and qualified humans
• Coordinate IFRS 17 presentation with ifrs; this skill owns assumption documentation, not note wording

Governance, documentation, and change control

Table of contents

1. Roles and ownership 2. Approval tiers 3. Assumption register 4. Change control 5. Documentation standards 6. Audit trail and retention

Roles and ownership

RACI should be explicit for material assumption changes affecting published metrics.

Approval tiers

Example tiering (adapt to company policy):

Define materiality thresholds (e.g., % impact on reserves, capital, or indicated rate).

Assumption register

Central register fields (minimum):

ID | Name | Category | Value | Unit | Basis | Source | Owner | Effective | Status | Model(s)

Statuses: draft → pending approval → approved → superseded

Link register rows to:

• Experience study ID
• Assumption paper section
• Model input file / worksheet cell range (if applicable)

Change control

For each change request:

1. Description — Driver, segment, old vs new value 2. Rationale — Study, benchmark, judgment memo 3. Impact — Quantified where possible (reserve, LR, SCR proxy) 4. Dependencies — Other assumptions or models affected 5. Effective date — Valuation date or rate filing date 6. Rollback — Prior version retained in pack history

Emergency changes (e.g., post-catastrophe) still require retrospective documentation within policy SLA.

Documentation standards

Assumption paper (outline)

1. Executive summary (what changed, why now) 2. Scope (models, segments, dates) 3. Summary table: prior vs proposed vs approved 4. Evidence (A/E, benchmarks, sensitivities) 5. Judgment sections (if any) 6. Governance sign-offs 7. Appendices (tables, curves, study excerpts)

Pack manifest

• Pack name and version
• Compatible model versions
• File list with checksums or repo tags
• Known limitations

Align narrative with model risk policy language; do not claim regulatory filing adequacy.

Audit trail and retention

Retain for reproducibility:

• Approved pack snapshot
• Input data cuts (with PII handling per policy)
• Change tickets and approver identity
• Model run logs referencing pack version

Retention period → company records policy; flag legal hold when litigation or exam is active.

Coordinate SOX or ITGC evidence with compliance-engineer only when controls are technical; financial control testing → finance skills.

Line-of-business assumption packs

Table of contents

1. Pack types by use case 2. Property and casualty pack 3. Life and health pack 4. Pension and retirement pack 5. Multi-line and reinsurance 6. Cross-skill routing

Pack types by use case

Same LOB often maintains separate packs:

Document differences when the same driver name differs across packs (e.g., pricing loss trend vs reserving trend).

Property and casualty pack

Core drivers (illustrative—not exhaustive):

Product context → property-casualty-insurance (coverages, FNOL, reinsurance structures).

Technical triangles → actuary (references/reserving_and_loss_development.md).

Life and health pack

Product context → life-health-insurance (benefits, networks, risk adjustment overview).

Experience studies → coordinate with actuary (references/assumptions_experience_studies.md).

Pension and retirement pack

Plan design and ERISA context → pension-retirement-funds—not duplicate plan administration detail here.

Assumption packs for pensions should state measurement basis (funding vs accounting vs economic).

Multi-line and reinsurance

Reinsurance assumed business: align ceding company experience with terms in treaty assumptions.

Cross-skill routing

When building a new LOB pack, start from the closest template above, then trim immaterial drivers via materiality workflow.

Selection methodology and judgment

Table of contents

1. Selection workflow 2. Experience-based selection 3. Credibility and blending 4. Level, trend, and volatility 5. Expert judgment 6. Coordination with actuary

Selection workflow

1. Hypothesis — What changed in portfolio, market, or regulation? 2. Evidence — A/E, external benchmarks, sensitivity of prior assumption 3. Proposal — Point estimate or surface (by age, duration, territory) 4. Challenge — Independent review, reasonability vs related assumptions 5. Approval — Per governance tier 6. Implementation — Pack version bump; model regression check

Experience-based selection

When company data supports the update:

Technical study design and triangle methods → actuary (references/assumptions_experience_studies.md).

Credibility and blending

Blend observed with prior (last approved, manual, industry):

Report Z, exposure, and full credibility threshold per segment in the assumption paper appendix.

Level, trend, and volatility

Decompose when decisions depend on it:

Examples:

• Health morbidity: separate utilization trend vs unit cost trend
• P&C: separate loss trend for severity vs exposure inflation
• Lapse: base rate vs dynamic sensitivity to rates (ALM coordination)

Avoid double-counting trend in both severity and frequency without documentation.

Expert judgment

Use judgment when data is insufficient, distorted, or forward-looking:

Judgment overrides should state:

• Who approved
• Why data was insufficient
• When to revisit (trigger or date)
• Impact vs data-only alternative

Coordination with actuary

Hand off when the user needs numerical model output without assumption documentation—or assumption work without model execution.

Sensitivity, scenarios, and stress

Table of contents

1. Purpose and materiality 2. Sensitivity testing 3. Scenario design 4. Stress and reverse stress 5. Correlation and ordering 6. Reporting standards

Purpose and materiality

Sensitivities and scenarios support:

• Governance — Show robustness of decisions before approval
• Risk — ORSA, risk appetite, board narrative (overview)
• Pricing — Margin adequacy under adverse drivers
• Reserving — Reasonability of central estimate

Start from materiality (from prior runs or one-way shocks) to limit grid size.

Sensitivity testing

One-way sensitivities

Tornado / ranking

Rank drivers by absolute impact on target metric; focus documentation on top 3–5.

Interaction (selective)

Test key pairs only when interaction is material (e.g., lapse × crediting rate).

Scenario design

Scenario narrative

Each scenario document should list:

• Assumption overrides (table vs base)
• Rationale (macro narrative, combined plausibility)
• Metrics impacted
• Limitations (not exhaustive of all risks)

Stress and reverse stress

Stress testing — Push drivers beyond plausible one-ways to find breaking points (capital, liquidity, reinsurance exhaustion).

Reverse stress — Identify scenarios that breach constraint (e.g., RBC ratio, internal limit); work backward to required assumption combination.

Do not invent regulatory stress parameters; reference internal or public framework and mark verify with compliance/actuary.

Correlation and ordering

State explicitly:

• Independent one-ways vs joint scenario
• Ordering if scenarios applied sequentially (e.g., shock then recalc dynamic lapse)
• Consistent economic paths (rates and inflation) when combined

Avoid double-counting the same economic shock in multiple assumption lines.

Reporting standards

Minimum sensitivity exhibit:

For scenarios, add scenario ID, description, and assumption override manifest linked to pack version.

Sensitivities support decisions; they are not substitutes for capital model validation or appointed-actuary work.