Personal tools

Generators

Note: Return to reference manual view.

WORK IN PROGRESS: This manual contains descriptions needed by users and developers.

1. Claims Generators Overview

There are three different kinds of claims generators in RiskAnalytics.

Usage in Models

  • A line of business normally contains different kinds of claims generators, i.e. an attritional claims generator and a single claims generator or several event claims generator.
  • If an event affects several lines of business, the event generator will rather be part of a cat model than of a single line of business.
Attritional Claims Generator

generate one claim of type attritional per period. Attritional claims are covered by proportional reinsurance contracts such as quota share and surplus and stop loss contracts.

Single Claims Generator

generate a certain number of claims of type single per period. The number is normally received from a frequency generator. Single claims are covered by proportional reinsurance contracts and the working excess of loss (WXL) and stop loss contract.

Event Claims Generator

generate a certain number of claims of type event aggregate per period. The number is normally received from an event generator. Event aggregate claims are covered by proportional contracts, cat excess of loss (CXL) and stop loss contracts.

Common Parameterization Structure

Independent of the generated claim type the following parameters may be set: 

Base: Apart from an absolute parameterization it is possible to calibrate the claim size based on premium written, number of policies and sum insured. In a first step a random number is generated according to the selected distribution and parameters. This number is then multiplied with the ‘base’ value. If the underwriting information consists of several risk bands the selected base values will be added up.

Distribution: The following ‘distributions’ are included. Adding further distributions is easy.

  • chi square quadrat
  • constant
  • discrete empirical
  • discrete empirical cumulative
  • inverse Gaussian
  • lognormal (mean, standard deviation)
  • lognormal (mu, sigma)
  • negative binomial
  • normal
  • pareto
  • poisson
  • piece wise linear
  • piece wise linear empirical
  • student
  • triangular
  • uniform

Parameter labels will be are switched in the tree view, if a different distribution is selected.

Modification: There are several ways to modify the random value, namely to censor, shift or truncate.


Background Information on Random Numbers

We use the SSJ Library from the University of Montreal for random number generators. Currently all random number generators are using F2NL607.

 

2. Attritional and Single Claims Generator

This claim generator includes an attritional and single claims generator.

3. Attritional, Single and Events Claims Generator

This component allows to generate attritional, single, earthquake, flood and storm claims.

It is used currently used for the property line of business in the Capital Eagle model.

4. Attritional Claims Generator

This component generates one claim of type attritional per period. Attritional claims are covered by proportional reinsurance contracts such as quota share and surplus and stop loss contracts.

5. Single Claims Generator

This component generates single claims by using a frequency and a claims size generator.

In a first step the number of claims is generated using the frequency generator. This number is sent to the claims generator that will generate as many single claims as defined by the frequency generator.

Parameterization possibilities

Explanations on claim types

6. Frequency Generators

A frequency generator is typically used in the context of frequency severity models (single and event claims generation). Also frequency generators could be used for modeling causal dependencies.

Base: Apart from an absolute parameterization it is possible to calibrate the frequency based on number of policies. In a first step a random number is generated according to the selected distribution and parameters. This number is then multiplied with the ‘base’ value. If the underwriting information consists of several risk bands the selected base values will be added up.

Distribution: The following ‘distributions’ are included. In general you will select a discrete distribution.

  • chi square quadrat
  • constant
  • discrete empirical,
  • discrete empirical cumulative
  • inverse Gaussian
  • lognormal (mean, standard deviation)
  • lognormal (mu, sigma)
  • negative binomial
  • normal
  • pareto
  • poisson
  • piece wise linear
  • piece wise linear empirical
  • student
  • triangular
  • uniform

Parameter labels will be switched in the tree view, if a different distribution is selected.

Background Information on Random Numbers

We use the SSJ Library from the University of Montreal for random number generators. Currently all random number generators are using F2NL607.

7. Event Claims Generator

This component generates event claims by using a frequency, event and claims size generator.

In the first step the number of claims is generated by using the frequency generator. Afterwards the event generator fetches the according number of event severities. Finally the claims generator generates the event claims using an inverse function.

Claim size parameterization possibilities

For the severities there are the same distributions as for the claims size available.

Explanations on claim types

8. Event Generator

This component generates events according to the frequency it receives. Events have a date and severity and are normally attached to a claim.

Causal dependency modeling is possible using the same event to generate several claims in different lines of business.

Document Actions