A Short Introduction to

Extreme Value Theory

Paddy Paddam
GIRO/CAS Convention 2001

Email: paddy.s.paddam@uk.pwcglobal.com Tel: +44 (0)20 7804 0830

Contents

• Introduction and Context

• Theory
• Short example and application issues
• Comments and discussion

Intro Context EVT Example Discussion

 Loss severity

• The distribution of the size of a loss

• Common problem:
– Pricing
– Reinsurance pricing and modelling
– Catastrophe models
– Securitisation
– Capital Management/DFA
– Operational Risk Management

Intro Context EVT Example Discuss

 Loss severity - Central Limit Theorem

• Use the normal distribution for modelling sample means

• Common problem:
– Pricing - Property damage claims on motor book
– Reinsurance pricing and modelling - low layers
– Securitisation - whole account portfolios
– Capital Management - attritional losses

Intro Context EVT Example Discuss

 Context - The problem

Empirical DF

100.00%
Cumulative probability

90.00%

80.00%

70.00%

60.00%

50.00%

40.00%

30.00%

20.00%

10.00%

0.00%
1 10 100 1000 10000 100000 1000000 10000000

Loss Severity (Log Scale)

Intro Context EVT Example Discuss

 The problem - The Tail

100.00%
Cumulative probability

99.50%

99.00%

98.50%

98.00%

97.50%
0 500000 1000000 1500000 2000000 2500000

Loss Severity

Intro Context EVT Example Discuss

Which distribution would you use to
model the extreme losses and why?

Lognormal?

Pareto?

Gamma?

Weibull?

7 PricewaterhouseCoopers
 What is Extreme Value Theory?

• Statistical Theory of Extreme Events

• Fisher-Tippet Theorem
– For many loss distributions, the distribution of the
maximum value of a sample is a generalised extreme value
distribution.
• Generalised extreme value distributions are
– Heavy tailed => Frechet
– Medium tailed => Gumbel
– Short tailed => Weibull
Intro Context EVT Example Discuss
Useful result in Hydrology
and Climatology

Not so useful in
Insurance?
 PBH Theorem

• Pickands-Balkema-de Haan theorem:

– For many loss distributions, the distribution of losses above
a high threshold is a Generalised Pareto Distribution.

GPD
Empirical DF

100.00%

Even more
90.00%

80.00%

70.00%

60.00%

50.00%
GPD
40.00%

30.00%

Normal 20.00%

10.00%

0.00%
1 10 100 1000 10000 100000 1000000 10000000

Intro Context EVT Example Discuss

 Peaks Over Threshold Method

For many loss distributions, the distribution of losses above a high

threshold is a Generalised Pareto Distribution

Pickands-Balkema-de Haan theorem

11
 Generalised Pareto Distribution

• P(X<x|X>u) = 1 - [1+g(x-m)/s]^(-1/g) for g <> 0

1 - exp[-(x-m)/s] for g = 0
• Parameters: 100%

Cumulative Distribution
– m = location 80%

– s = spread 60%

40%
– g = shape
20%
– u = threshold
0%
0.1 1 10 100

Shape = 0 Shape = 0.5 Shape = 1

Intro Context EVT Example Discuss

 Fitting EVT distributions

• Maximum likelihood methods

• Probability weighted moments
• Variety of methods and software

Intro Context EVT Example Discuss

 Applying EVT - Example
Reinsurance modelling

• Considered a portfolio of 11 classes including Liability and

Property accounts
• Performed a standard stochastic claims frequency and severity
analysis
• In addition attempted to fit a GPD to the claims severity
• In our exercise, for 9 out of the 11 classes, the GPD was about
as good or better than a standard loss distribution in modelling
the extreme tail values of the loss severity distributions.

Intro Context EVT Example Discuss

 Fit to Claims Severity of a Non-Marine
Liability Class
100%
Cumulative Probability

98%

96% Gamma
GPD
94% Data
92%

90%
0 100000 200000 300000 400000
Claim Size

Intro Context EVT Example Discuss

 Choosing the Threshold
A Balancing Act

Higher Threshold
Lower Threshold Smaller Model error
Larger Model error
Less Data
More Data Larger Parameter error
Smaller Parameter error

Intro Context EVT Example Discuss

 One approach
- Focus on shape parameter

• Shape parameter is the most important

• Higher shape parameter => Thicker tailed => More
losses
• Compare the fitted shape parameter with the change
in threshold (or equivalently the number of
exceedances).
• Identify a stable plateaux
• Threshold < Excess attachment point

Intro Context EVT Example Discuss

 Fitted GPD Shape Parameter Compared
to chosen threshold
1.6
1.4
1.2
Shape Parameter

1
0.8
0.6
0.4
0.2
0
0 2000 Decreasing
4000 6000 Threshold
8000 10000 12000 14000
Increasing Model Error
Exceedences
Decreasing Parameter Error
Intro Context EVT Example Discuss
 Simple Conclusion

• Fit a GPD to the tails of claims severity distributions

because it reduces Model error
• However Parameter error & Random fluctuations risks
remain

Intro Context EVT Example Discuss

 Discussion

• Applications?
• Concerns?
• Questions?

Intro Context EVT Example Discuss

 References and Further Reading

• McNeil AJ Estimating the Tails of Loss Severity Distributions using Extreme Value Theory ASTIN Bulletin,
Vol. 27 no 1, pp 117-137. http://www.casact.org/library/astin/vol27no1/117.pdf.

• J. Beirlant, G Matthys, G Dierckx, Heavy Tailed Distributions and Rating,ASTIN Bulletin, Vol 31, No 1, pp 37-
58, May 2001 (Comprehensive list of references)

• Patrik P and Guiahi F, An Extrememly Important Application of Extreme Value Theory to Reinsurance Pricing,
1998 CAS Spring Meeting Florida (A presentation of the analysis of ISO claims severity)

• McNeil AJ and Saladin T, The Peaks over Thresholds Method for Estimating High Quantiles of Loss
Distributions, 28 ASTIN Colloquium 1997 (POT method used in pricing reinsurance)

• Embrechts P, Extremes and Insurance, 28 ASTIN Colloquium 1997 (More examples)

• Embrechts P, Kluppelberg C, and Mikosh T, Modelling Extremal Events, Springer Verlag, Berlin 1997 (The
major text book on the subject)

• Reiss R, and Thomas M, Statistical Analysis of Extreme Values, Birkhauser, Basel, Boston, 1997