|
Using Cat Simulation Models After the Loss
1. Models Prior to Loss
Some Definitions
| Event Losses |
A table of all simulated events, with estimates of portfolio loss
amounts, descriptive code for the event, and the annual rate
of the event recurring. Usually sorted by loss amount. |
| Exceeding Probability |
Annual probability of a loss equal, to or exceeding, a
given amount. |
| Return Time |
Period (in years) = 1 / Annual E.P. This does not have an easy
interpretation as a probability. Don’t try! |
Event Loss Table
EP Results
Models Are Very Good at Considering:
- Coverages A and B
- Experimental data from wind tunnel/shake table tests
- Number of major historical events for large geographic regions
(e.g. a state or fault)
- Reconciliation to industry data, such as PCS or Sigma
- 25-year to 75-year return times
Structure Loss vs. Construction Data
Two Houses, Two Fates
Two houses on Belmont Street
in Pensacola had distinctly
different fates when Ivan hit.
The house on the right was
built in 1903 and refurbished.
The house on the left is only a
few years old.
2. Limitations to Any Model
Model Limitations
More Subtle Factors That the Models Can Currently
Only Implicitely Reflect:
- Coverage D and Risk Excess layers
- Secondary uncertainty/ Correlation issues
- The degree of enforcement of local building codes
- Foliage
- Weather patterns before and after loss events
- Physical alignment of structures along events’ force vectors
- Local variations of concentrations or hazard (street address detail)
- Changes in claims handling and other industry practices
Also:
- Data for medium-sized events, 0-10 year return time losses, are not collected as
consistently as for larger events. Modelers must look to larger events and back into
these events.
- Data for mega-events, 250+ year return time losses, are also missing due to limited
history. Modelers must extrapolate loss potentials from smaller events.
Demand Surge (for Blue Tarps)
3. Models After the Loss
- Pre-existing simulated “events” selected to most closely match the actual or
expected event
- Industry loss estimates based on after-the-event reconaissance
- Marketshare estimates overall
- “Back casting” physical event details
- Industry Loss in geographic detail – local damage factors
- Footprint files – localized market shares
4 Standards of Practice: ASOP 38
3.1 When Using a Model, the Actuary Should Do All of
the Following
- Determine appropriate reliance on experts;
- Have a basic understanding of the model;
- Evaluate whether the model is appropriate for the intended application;
- Determine that appropriate validation has occurred; and
- Determine the appropriate use of the model.
3.3 Understanding of the Model
Be reasonably familiar with the basic components of the model and understand both
the user input and the model output, as discussed below.
- Model Components — Identify which fields of expertise were used in developing
or updating the model, and should make a reasonable effort to determine if the
model is based on generally accepted practices within the applicable fields of
expertise. The actuary should also be reasonably familiar with how the model
was tested or validated and the level of independent expert review and testing.
- User Input — The actuary should understand the user input that is required to
produce the model output. This understanding includes the level of detail
required in the user input to produce results that are consistent with the
intended use of the model.
- Model Output — The actuary should determine that the model output is
consistent with the actuary’s intended use of the model.
3.5 Appropriate Validation
Refer to ASOP No. 23, Data Quality
Examine the model output for reasonableness:
- Results derived from alternate models
- How historical observations compare
- Consistency and reasonableness or relationships among various output
- The sensitivity of the model output to variations
Comments or Questions:
Paul J. Kneuer
(212) 797-2285 or paulk@holborn.com
|
