|
Evaluating Model Performance
2004 Hurricanes: Characteristics Consistent With Stochastic Model Parameters
RMS Stochastic Event Set
 RMS hurricane
stochastic event set
contains several
similar events to the
actual hurricane tracks
and characteristics
Charley – Cat 4
Frances Frances – Cat 2
Ivan – Cat 3
Jeanne – Cat 3
Correlation of Risk Between Regions
RMS stochastic tracks represent the actual and potential
correlation of risk between different regions (Clash)
- Stochastic track selections below show Ivan’s path across
Jamaica and Grand Cayman, and Charley's potential
impact on western Florida
RMS Industry Survey of Modeled and Incurred Losses
Portfolio-level loss experiences have been gathered for
approximately 70% of the total industry loss (roughly even
percentages between personal and commercial lines)
 |
Books of Business |
|
Event-specific Losses |
|
| Total |
107 |
Total |
366 |
|
| Personal Lines |
43 |
Personal Lines |
154 |
| Commercial Lines |
53 |
Commercial Lines |
178 |
|
Summary by Event: All Lines of Business
Frances and Jeanne Claims
Summary by Event: Personal vs. Commercial Lines
Personal Lines Losses for the 2004 Season
Overall, personal lines consistent with modeled losses
- Personal lines losses account for 2/3 of the industry loss
However large variability around the mean
- Manufactured Housing
- Multi Multi-family vs. single family
- Year and type of construction
- Impact of building codes, practices and construction quality
Commercial Lines Variability
Significant variability in individual company losses
- 90% of modeled vs actual loss ratios between 0.14 and 1.42
Disproportionately high amount of loss results from a small number
of accounts in many commercial books
Commercial example of modeled / incurred losses
- Ratio for excess book: 0.5
- Removing just one account improves ratio by: 33%
Diagnosing model performance at portfolio level requires detailed
analysis of individual accounts
|
Business Owners
Policies (BOP) |
Standard COM |
Fortune 2000 |
Excess |
Surplus … |
|
Possible Drivers of Variability in Commercial Lines
Exposure data, and the modeling of
it, is more complex
Commercial lines more likely to
incur non-modeled losses
Slower settling of claims
Demand surge generally greater for
commercial classes
Reserves are traditionally set high
Excess layers are more sensitive to
uncertainties in the windfield
|
