NAIC Catastrophe Computer Modeling Handbook

1
NAIC Catastrophe Computer Modeling Handbook
2
Purpose of the Handbook
What on Earth do we need this for?

• The purpose of the Catastrophe Modeling Handbook
  is to explore in some detail catastrophe computer
  models and to discuss issues that have arisen or
  can be expected to arise from their use.
• Used by insurance regulators to review models
  primarily used in underwriting, rate-making and
  solvency procedures.

3
User Perspectives of Models
Well heres what we think...

• Insurers Perspective
• Old rate making methods underestimate the cost of
  insuring catastrophe prone areas.
• Consumers Perspective
• Modelers must disclose input and output so
  independent tests can be run to assure that
  results are reasonable.

4
User Perspectives of Models (continued)

• Regulators Perspective
• Must learn to replace traditional models with a
  methodology that is in its relative infancy in
  terms of producing consistently reliable results.
• Must work with modelers to provide enough
  disclosure to make informed decisions while
  preserving the confidentiality of proprietary
  details.

5
Components of the Model
The whole is greater than the sum of its parts.

• Science Module
• Physical characteristics about the catastrophe
  (factors include wind speeds, landfall
  location, magnitude, location of fault,
  liquefaction potential).
• Engineering Module
• Estimates the effect of catastrophic events on
  different types of structures (factors include
  age of structure, construction type, attachment
  anchoring).
• Insurance Module
• Estimates the insured damage at a location
  (factors include guaranteed replacement cost
  multiplier, deductible, reinsurance limits).

6
Considerations of Model Inputs
Whatever goes in...

• Models use expected future exposures as the basis
  for analysis whereas more traditional techniques
  use historical data.
• Models can be modified to utilize exposure data
  maintained by companies and statistical agents.
• Currently, exposure data is not collected
  specifically to be used in catastrophe modeling
  which can significantly impact a models results.

7
Model Output
must come out.

• Average Annual Losses
• Provides an estimate of the amount a company
  needs to collect from policyholders each year to
  fund long-run catastrophe losses.
• Loss Costs
• Defined as Average Annual Losses divided by the
  number of exposure units giving rise to those
  losses.

8
Model Output (continued)

• Distribution of Losses
• The probability distribution that a company will
  experience losses less than or equal to a
  specified amount, used to
• estimate parameters when the form of the
  distribution has been assumed
• create an empirical loss distribution.
• Exceeding Probability Distribution
• Represents the probability that a client
  companys portfolio of risks will experience
  losses greater than a given amount.

9
Model Output (continued)

• Individual Event Losses
• Shows the event characteristics as well as
  detailed estimates on the losses for a specific
  simulated catastrophic event for a specified
  portfolio or risks.
• Historical Event Losses
• Characteristics of a particular historical event
  are used to simulate the loss experience for a
  specific exposure distribution, typically the
  clients current distribution.

10
Model Validation, Updating and Evaluation
Because you can never be too sure.

• Accuracy and Comparison to Historical Data
• Compare modeled results to actual results for a
  historic event.
• Convergence
• Confirm the model was calculated using sufficient
  iterations.

11
Model Validation, Updating and Evaluation
(continued)

• Expert Opinion/Peer Review
• Confirm that modelers disclose the extent to
  which the model has been verified or
  substantiated by independent expert opinion.
• Input Data Provided by Company
• Modelers should perform validation checks on
  provided data and disclose how invalid data is
  treated in their model.

12
Model Validation, Updating and Evaluation
(continued)

• Logical Relationship to Risk
• Verify that modeled loss costs increase as
  exposure to potential loss is assumed to increase
  and should decrease as exposures decrease.
• Model Updates
• Verify that modelers update zip code databases on
  a regular basis.

13
Model Validation, Updating and Evaluation
(continued)

• Probabilistic Range
• Verify that models produce results over ranges
  that can be reasonably expected.
• Sensitivity
• Test to see how a change in input or in a model
  parameter affects the outputs.
• Stability
• Verify that sampling and aggregation have a
  negligible contribution to error.

14
Model Validation, Updating and Evaluation
(continued)

• Questions to Ask About
• Company data input to models
• Computer simulation models
• Model output

15
Proprietary Information
Everything is Sacred.

• Two Conflicting Sides of the Coin
• Catastrophe modelers are dependent upon keeping
  their model assumptions and parameters out of the
  hands of competitors.
• Regulators are governed by Freedom of Information
  of Sunshine laws concerning what information must
  be disclosed to the general public.

16
Proprietary Information (continued)

• Three Regulator Options to the Dilemma
• These laws occasionally contain a waiver from
  disclosing data if it considered to be a trade
  secret.
• Modelers may allow regulators to perform an
  on-site visit and assist regulators in reviewing
  and understanding a specific model and its
  application to insurer data.
• Regulators may ask state seismologists and
  geologists to assist them in understanding
  scientific aspects of the modeling process.

17
Education and Outreach

• Explanation of how computer models are used in
  rate-making or portfolio management.
• An overview of the technical information that the
  models contain should be supplied.
• Educational materials should be prepared to
  increase public awareness of catastrophe risk.
• Educational materials should be developed and
  distributed through the mail and the Internet.
• Establish speakers bureaus for cities, towns,
  civic/business/religious organizations and
  educational institutions to identify available
  resources.

18
Other Suggestions
Let us not forget the kitchen sink!

• Related Activities
• Actuarial Standards for Model Use
• Using Models Outside the Actuarys Area of
  Expertise (Property and Casualty)
• Pre-Tax Loss Reserves for Companies
• NAIC has been working on a proposal for a
  statutory tax deductible pre-event catastrophe
  reserve.

19
Contact Information
As if this isnt more than you wanted to know!
For a copy of the NAICs Catastrophe Computer
Modeling Handbook, contact Davin Cermak at
dcermak_at_naic.org, or NAIC 2301
McGee, Suite 800 Kansas City, MO 64108