June 2006 CARE

1
June 2006 CARE

• Robert Eramo - Risk Assessment Strategies,Inc.
  and Representative of Insureware

2
Method Background

• Measure Process Parameter Risk
• Use Insurewares ICRFS
• Apply to Individual Triangles First
• Compare Triangles to Find Parameter Correlations

3
Process Parameter Risk

• Coin Flip Example
• Variability of of Heads due to the basic
  process and Knowing Fairness of Coin
• Triangles likewise have similar contributors to
  Variability of Outcome

4
Loss Outcome Variability

• Size of Book Main Source Of Process Risk
• Relative Variance Higher For Book of Claims With
  100 Expected Claims vs. 1000 Expected Claims
• Trends in Development and Calendar Inflation are
  Key parameters
• Knowledge Of Parameters Uncertain

5
Triangle Parameter Risk

• Usually Not dependent of co.s size of book
• Therefore Increased Size Does Not Diversify
• Way to Improve Knowledge of Parameters
• ICRFS Example

6
Large Company

• Two Major Subsidiaries BOT POT
• Can Analyzing Both Simultaneously Improve
  Knowledge of Parameters
• First Note Initial Separate Models
• Look at Parameters For BOT Explicitly

7
Separate Models BOT POT

• Note t-statistics of development and calendar yr.
  parameters
• Specifics
• Specifics
• Specifics

8
Combined Model Benefits

• Note New Model Displays
• T-statistics specifics
• Comparison of Independent Models

9
Pot Development Parameter t-StatisticsModele
d Alone
Dev Period 12-24 Dev Period 24-72 Dev Period 72-120
Trend -.5753 -.4864 -.2867
T-Statistic -16.87 -21.00 -5.075



10
Pot Development Parameter t-StatisticsModele
d in Composite
Dev Period 12-24 Dev Period 24-72 Dev Period 72-120
Trend -.6255 -.5102 -.3809
T-Statistic -18.90 -49.54 -14.16



11
Pot Calendar Parameter
t-StatisticsModeled Alone
Cal Period 90-93 Cal Period 97-99
Trend .1827 .1827
T-Statistic 10.09 10.09



12
Pot Calendar Parameter
t-StatisticsModeled in Composite
Cal Period 90-93 Cal Period 97-99
Trend .1792 .1792
T-Statistic 10.18 10.18



13
Application to Company vs. Statewide Experience

• Model Company State Separately
• If there are reasonable correlations parameter
  uncertainty for company A model can be reduced

14
Application to Excess Layers

• Model Layers Separately
• Improve Knowledge of Parameters in XS Pricing