Home / Patterns / Catastrophe Correlation Blindness

Catastrophe Correlation Blindness

Goal Claims Processing Frequency Occasional Category Insurance Published View source on GitHub ↗

Issue: Catastrophe risk model assumes independent claims; hurricane hits coast, model hadn’t provisioned for 10k simultaneous claims; reserve exhausted within days

Frequency: Occasional

Symptoms

  • Model predicts $100M reserve needed based on average annual claims
  • Single catastrophic event (hurricane, earthquake, fire) generates $500M in claims
  • Reserve depleted; company forced to hold up payments or borrow
  • Historical disaster scenarios not captured in model

Root Cause Reserve models assume claims are independent random events. Catastrophes create correlated massive claims (hurricane hits region = 1000s of simultaneous claims). Models trained on normal claim distribution miss tail correlations. CAT models exist but rarely integrated into core reserve calculations.

Example

Insurer: Coastal homeowners, 1M active policies
Model: Average annual claims = $50M, reserve = $100M
Historical data: 10 years of data, no major hurricanes
Model assumption: Claims are independent Poisson process
2024 reality: Cat 5 hurricane hits coast
Actual claims in September 2024: $450M (900+ simultaneous claims)
Reserve remaining: $100M - $450M = DEFICIT
Company response: Emergency reinsurance, credit line drawdown
Impact: Regulatory capital requirements breached; company facing insolvency

Key Statistics

FindingSource
Catastrophe events vs frequency assumption: 10-100x tail riskInsurance industry reports
Reserves inadequate during disasters: 15-20% of carriersPost-disaster audits
Correlation during catastrophes: Near 1.0 (all claims spike together)Reinsurance data

Mitigation Strategies

  1. CAT modeling integration: Use separate CAT model for reserve calculations
  2. Stress testing: Model 100-year and 500-year disaster scenarios
  3. Reinsurance layers: Transfer tail risk above reserve capacity

References