Actuarial Modeling Analysis Complete

Step 1.2 -- Model Inventory

Catalog every actuarial model found. For each model, record:

• Model type (loss reserving, pricing, life valuation, capital, catastrophe, reinsurance)
• Risk classification (materiality: high/medium/low, complexity, frequency of use)
• Owner and last review date (from comments, git history, or documentation)
• Input data sources and output consumers

Step 1.3 -- Data Infrastructure

Map actuarial data sources:

• Loss development triangles (paid, incurred, reported, closed)
• Exposure and premium data (earned, written, in-force)
• Mortality/morbidity tables (SOA tables, company-specific experience)
• Economic assumptions (interest rates, inflation, yield curves)
• Industry benchmarks (ISO, NCCI, AM Best aggregates)
• Experience studies (lapse, mortality, morbidity, disability)

============================================================ PHASE 2: LOSS RESERVING ANALYSIS

Step 2.1 -- Reserving Methodology

For each reserving model, determine the method and assess appropriateness:

• Chain Ladder (paid and incurred development) -- check for stability of development factors
• Bornhuetter-Ferguson (expected loss ratio method) -- check ELR source and reasonableness
• Cape Cod (Stanard-Buhlmann) -- verify weighting methodology
• Generalized linear models for development patterns -- check model fit
• Individual claim-level reserving (case reserves + IBNR) -- verify completeness
• Frequency-severity methods -- check independence assumption
• Berquist-Sherman adjustments -- verify adjustment rationale

Decision criteria: Flag any model using a single method without cross-validation against alternatives.

Step 2.2 -- Triangle Analysis

Assess loss development data quality:

• Triangle construction: verify accident year/quarter, development period, evaluation date alignment
• Data segmentation: confirm line of business, coverage, claim type, state splits are appropriate
• Development factor selection: compare volume-weighted, simple average, medial, optimal selections
• Tail factor selection: verify methodology is documented and reasonable
• Diagonal effects: check for calendar year trends that distort development
• Outlier identification: confirm treatment is documented and consistent

Step 2.3 -- Reserve Adequacy

Evaluate reserve quality against these benchmarks:

• Actual vs. expected analysis (reserve runoff testing) -- flag if AVE ratio deviates > 5% for 2+ years
• Reserve range estimation -- verify point estimate, low, high, and percentile ranges exist
• Discount rate application -- confirm methodology matches regulatory requirements
• Salvage and subrogation offsets -- verify they are not double-counted
• ULAE/ALAE reserve calculations -- check allocation methodology
• Actuarial opinion documentation -- verify NAIC Statement of Actuarial Opinion compliance
• ASOP compliance -- check ASOP 36, 43 (P&C) and ASOP 25 (health)

============================================================ PHASE 3: PREMIUM PRICING METHODOLOGY

Step 3.1 -- Ratemaking Process

Evaluate the pricing pipeline end to end:

• Pure premium vs. loss ratio approach -- confirm appropriate for the data volume
• Loss trend analysis -- verify frequency, severity, and mix shift trends are separated
• Loss development to ultimate -- confirm consistency with reserving ultimates
• Expense loading -- verify fixed, variable, profit, and contingency loads
• Credibility weighting -- check method (classical, Buhlmann, Buhlmann-Straub) and minimum thresholds
• Rate level history -- verify on-level adjustments are complete and accurate
• Indicated rate change -- confirm calculation ties to exhibits

Step 3.2 -- GLM Rating Models

If GLMs are used for pricing, assess each model for:

• Distribution selection appropriateness (Tweedie, Poisson-Gamma, Logistic)
• Link function selection with justification
• Variable selection -- check for multicollinearity and interaction terms
• Model fit statistics (deviance, AIC, BIC, residual analysis) -- flag poor fits
• Relativities stability -- compare across model iterations
• Cross-validation -- confirm out-of-sample testing is performed
• Comparison to one-way and two-way factor analysis for reasonableness

Step 3.3 -- Rate Filing Support

Evaluate regulatory compliance readiness:

• Rate indication documentation per state requirements
• Support for "not excessive, inadequate, or unfairly discriminatory" standard
• Filing exhibit preparation (loss data, trend, development, expense)
• Competitive analysis and market impact assessment
• Implementation planning (rate capping, grandfathering, transition rules)

============================================================ PHASE 4: LIFE AND HEALTH ACTUARIAL MODELS

Skip this phase if no life/health models are found. Otherwise:

Step 4.1 -- Mortality and Morbidity Tables

Evaluate table usage:

• Table sources: verify SOA tables (2017 CSO, VBT, ILEC) or company experience are current
• Experience study methodology: check exposure calculation, graduation, credibility
• Mortality improvement assumptions: verify Scale MP or custom improvement is applied
• Morbidity assumptions: check by condition and duration
• Lapse and persistency: verify assumptions match recent experience
• Selection vs. ultimate: confirm appropriate period is used

Step 4.2 -- Valuation Models

Assess reserve methodology against applicable standards:

• GAAP (ASC 944), Statutory (VM-20, AG43), IFRS 17 -- confirm correct standard is applied
• Cash flow projections -- verify both deterministic and stochastic runs exist
• Net premium reserve calculations -- check for accuracy
• DAC modeling -- verify amortization methodology
• PBR implementation -- confirm exclusion test and stochastic reserve calculations
• Asset adequacy analysis -- verify cash flow testing scenarios

Step 4.3 -- Product Pricing

Evaluate product pricing models:

• Profit testing methodology (profit margin, IRR, embedded value)
• Assumption sensitivity analysis -- confirm key assumptions are stress-tested
• Product design optimization (benefit structure, rider pricing)
• Reinsurance pricing and treaty optimization
• Competitive positioning analysis

============================================================ PHASE 5: STOCHASTIC MODELING AND CAPITAL ADEQUACY

Step 5.1 -- Stochastic Framework

Evaluate stochastic modeling infrastructure:

• ESG: identify interest rate model (CIR, Hull-White, Black-Karasinski) and calibration
• Monte Carlo engine: check scenario count (minimum 1,000 for screening, 10,000+ for production)
• Convergence testing: verify results stabilize with increasing scenario count
• Correlation structure: confirm risk factor correlations are justified
• Random number generation: check seed management and quasi-random sequence usage
• Runtime performance: assess parallelization and bottlenecks

Step 5.2 -- Capital Modeling

Assess capital adequacy models:

• Risk categories covered: insurance risk, market risk, credit risk, operational risk
• Capital metric: VaR, TVaR/CTE, economic capital, regulatory capital -- confirm appropriate metric
• Confidence level and time horizon: verify alignment with regulatory requirements
• Diversification benefit: check correlation assumptions and methodology
• Stress testing: confirm both prescribed and reverse stress tests exist
• DFA framework: verify Dynamic Financial Analysis integration if present

Step 5.3 -- Regulatory Capital Compliance

Evaluate compliance with applicable capital standards:

• Solvency II: SCR calculation, internal model approval status, ORSA documentation
• NAIC RBC: verify formula components and action level calculations
• IFRS 17: risk adjustment methodology and confidence level
• OSFI (Canadian): capital requirements if applicable
• ORSA: verify Own Risk and Solvency Assessment is current and comprehensive
• Capital allocation: confirm allocation methodology by business unit or product line

============================================================ PHASE 6: MODEL GOVERNANCE AND CONTROLS

Step 6.1 -- Model Risk Management

Assess governance against regulatory expectations (SR 11-7 / SS3/18):

• Model inventory with risk classification -- flag any models not in the inventory
• Development standards and documentation -- check for completeness
• Independent peer review or validation -- verify independence and qualifications
• Change control and version management -- check for audit trail
• Assumption setting governance and sign-off -- verify approval chain
• Model limitation documentation -- confirm limitations are disclosed to users

Step 6.2 -- Actuarial Controls

Evaluate the control framework:

• Data reconciliation: source-to-model tie-out procedures
• Reasonableness checks: automated bounds checking on outputs
• Back-testing: historical validation results and trending
• Audit trail: assumption change logging with justification
• SOX controls: financial reporting model controls documented and tested
• Certification process: actuarial opinion sign-off workflow and timeline

============================================================ PHASE 7: WRITE REPORT

Write analysis to docs/actuarial-modeling-analysis.md (create docs/ if needed).

Structure the report as:

1. Executive Summary -- 3-5 bullet points of critical findings
2. Model Inventory -- table of all models with risk classification
3. Loss Reserving Assessment -- methodology evaluation and adequacy findings
4. Pricing Methodology Review -- ratemaking and GLM assessment
5. Life/Health Model Evaluation (if applicable)
6. Stochastic Modeling Capabilities -- ESG and Monte Carlo assessment
7. Capital Adequacy Assessment -- regulatory compliance status
8. Model Governance Review -- control gaps and recommendations
9. Prioritized Recommendations -- with actuarial standards references (ASOP, SOA, Solvency II)

============================================================ SELF-HEALING VALIDATION (max 2 iterations)

After producing output, validate data quality and completeness:

1. Verify all output sections have substantive content (not just headers).
2. Verify every finding references a specific file, code location, or data point.
3. Verify recommendations are actionable and evidence-based.
4. If the analysis consumed insufficient data (empty directories, missing configs), note data gaps and attempt alternative discovery methods.

IF VALIDATION FAILS:

• Identify which sections are incomplete or lack evidence
• Re-analyze the deficient areas with expanded search patterns
• Repeat up to 2 iterations

IF STILL INCOMPLETE after 2 iterations:

• Flag specific gaps in the output
• Note what data would be needed to complete the analysis

============================================================ OUTPUT

Actuarial Modeling Analysis Complete

• Report: docs/actuarial-modeling-analysis.md
• Models inventoried: [count]
• Reserving methods reviewed: [count]
• Capital model components assessed: [count]
• Governance gaps identified: [count]

Summary Table

NEXT STEPS:

• "Run /underwriting-analysis to evaluate risk selection and pricing implementation."
• "Run /catastrophe-modeling to assess natural disaster exposure and reinsurance adequacy."
• "Run /claims-workflow to analyze loss development drivers and claims handling impact."

DO NOT:

• Do NOT modify any actuarial models, assumptions, or reserve estimates.
• Do NOT produce actuarial opinions or certifications -- flag findings for credentialed actuaries.
• Do NOT access or display individual claimant or policyholder data.
• Do NOT skip ASOP compliance assessment even for internal management models.
• Do NOT assume reserve adequacy from point estimates alone -- always check ranges and uncertainty.

============================================================ SELF-EVOLUTION TELEMETRY

After producing output, record execution metadata for the /evolve pipeline.

Check if a project memory directory exists:

• Look for the project path in ~/.claude/projects/
• If found, append to skill-telemetry.md in that memory directory

Entry format:

### /actuarial-modeling — {{YYYY-MM-DD}}
- Outcome: {{SUCCESS | PARTIAL | FAILED}}
- Self-healed: {{yes — what was healed | no}}
- Iterations used: {{N}} / {{N max}}
- Bottleneck: {{phase that struggled or "none"}}
- Suggestion: {{one-line improvement idea for /evolve, or "none"}}

Only log if the memory directory exists. Skip silently if not found. Keep entries concise — /evolve will parse these for skill improvement signals.