Spatial Uncertainty in Catastrophe

Modelling

Chris EwingChris Ewing

Aon Benfield

� What is Reinsurance?

� What is Catastrophe Modelling?

� Uncertainty in Catastrophe Modelling

� (Spatial) Uncertainty in Cat Modelling

– Hazard

Overview

– Exposure

– Vulnerability

– Loss

� ELEMENTS – Coping with Uncertainty

� Conclusions

� Reinsurance is simply insurance for insurers - it allows the insurer to remove risk

� Natural catastrophes can produce large losses for insurers and are

What is reinsurance?

large losses for insurers and are one type of risk where reinsuranceis needed

� Swiss Re (one of the world’s largest reinsurers) formed after a large fire in Glarus, Switzerland in 1863

� Recent global economic crisis =tighter regulation in banking and insurance on liquidity

� New regulations from EEC called Solvency II seek to reduce the risk of an insurer not meeting claims

Reinsurance in the ‘New Economy’

� Heavy losses from natural disasters need better catastrophe risk assessment

� Solvency II “standard formula‟ for calculating Nat Cat risk - but it is too simplistic

� Aon Benfield is an industry leader in treaty, facultative and capital markets, and act as reinsurance intermediaries and capital advisors

� Within Aon Benfield, Analytics offers clients catastrophe management, actuarial, rating

What do we do?

agency advisory and risk and capital strategy expertise

� Within Analytics, Impact Forecasting develop catastrophe models that help analyse the financial implications of catastrophic events

Risk

Credit Risk

Asset Risk

CatastropheNon

Business Risk Event RiskMarket Risk

Operating RiskLiability Risk

Types of risks faced by an Insurance corporation

•Corporate Bonds

•Reinsurance Ceded

•Derivative counterparties

•Other Receivables

•Equities

•Interest Rates

•Derivative Securities

•Foreign Exchange

•Losses due to Natural Disasters

•e.g. Flood, Windstorm, Earthquake

•Potential claims reported or paid at random times for random amounts

•Changes in volume

•Changes in Margin

•Fraud

•Unintentional Errors

•Systems Interruption

Credit Risk CatastropheNon

- CatastropheBusiness Risk Event RiskMarket Risk

� A Cat Model determines the potential loss to a client’s exposure from natural perils

– wind, earthquake, flooding, fire

� Cat Modelling is a tool/technique to estimate the potential loss to property and life following a catastrophic event

What is Catastrophe Modelling? - I

� 4 components of a Catastrophe Model

– Hazard

– Vulnerability

– Exposure

– Loss

� Spatial uncertainty is present in all components

Hazard

RISK

Peril Frequency and SeverityEQ shaking intensity

Wind strength

Flood depth inundation

Blast radius

Loss

Calculation

What is Catastrophe Modelling? – II

Vulnerability Exposure

RISK Risk Portfolio DataStructure values

Contents values

Time Element

Number of people

Deductibles / Limits

Reinsurance

Hazard SusceptibilityStructural classifications

Occupancy descriptions

Secondary Characteristics

� To help understand the risk faced by corporations to natural catastrophes

� To assist in determining insurance / reinsurance strategy

� Development of cat models following:

– European windstorms 1987 / 1990

Why use Catastrophe Modelling?

– European windstorms 1987 / 1990

– Hurricane Andrew 1992

– World Trade Centre terrorist attacks

– Hurricane Katrina 2005

� Catastrophe Models also cover Terrorism, Pandemic Influenza and Workers Compensation

� Most insurers and reinsurers will have some form of (basic) catastrophe model

� Catastrophe Models are required for Solvency II

� 3 commercial modelling companies

Who uses Catastrophe Models?

� 3 commercial modelling companies (AIR, EQECAT, RMS)

� Aon Benfield Impact Forecasting develops ELEMENTS - our in-house catastrophe model

Impact Forecasting Catastrophe ModelsCountry / Peril Crop

Cyclone /

HurricaneEarthquake

Offshore Oil

PlatformRiver Flood Storm Surge Terrorism

Tornado /

Hail / Severe

Winds

Wildfire /

Bushfire

Workers'

Compensation

Regional Models

Asia ●

Central and Eastern Europe ●

Greece and Cyprus ●

Western Balkans ●

Country models

Albania

Australia ● ● ●

Austria ● ●

Belgium ● ●

Bosnia and Herzegovina ●

Bulgaria ●

Canada ●

Croatia ●

Cyprus ●Cyprus ●

Czech Republic ●

France ● ●

Germany ● ●

Greece ●

Hungary ● ●

Japan ●

Kazakhstan ●

Montenegro ●

New Zealand ●

Norway ●

Poland ●

Romania ●

Serbia ●

Slovakia ● ●

Slovenia ●

South Africa ● ● ●

Switzerland ●

United Kingdom ●

United States ● ● ● ● ● ● ● ● ● ●

� All models are uncertain

� The key is to understand the uncertainty and handle it as best as we can

Vulnerability LossHazard

Event Generation

IntensityCalculation

Incomplete or incorrect data Default assumptions

Application

of limits and

Uncertainty in Catastrophe Modelling

Exposure

Generation Calculation

Risk Characterisation

Damage

Calculation

Policy

Conditions

Insured

LossToo much aggregated data,

insufficiently detailed geographical

position, missing data

of limits and

deductibles

� Due to complexity of nature all models are simplified mathematical representations of physical phenomena

� Primary vs. Secondary

– Primary = Hazard

Uncertainty in Catastrophe Modelling

– Primary = Hazard

– Secondary = Damage

� Aleatory vs. Epistemic

– Aleatory = inherent randomness which cannot be reduced

– Epistemic = uncertainty due to lack of information which can be reduced

� Spatial uncertainty is the lack of, or the error in, knowledge about an object’s geographic position or location

� Location is essential in Catastrophe Modelling to calculate potential losses

(Spatial) Uncertainty in Catastrophe Modelling

Modelling to calculate potential losses

– Where is the Earthquake epicentre?

– Where will the largest intensities be felt?

– Where are the insured locations?

– Where are the flood / tidal defences?

– Where will the tropical storm hit land?

– Where is the postcode which has the

most damage?

� Incomplete geographic data of the hazard

� Incorrect data - inherent errors in the data / data collection errors

� Flooding

– Digital elevation model uncertainty - valley bottoms and streamlines

DTM Comparison - Profile

350

Uncertainty – Hazard I

100

150

200

250

300

350

0 1000 2000 3000 4000 5000 6000

Distance (m)

Ele

vati

on

(m

)

10m FKB Data

25m Mestekart

� USGS - recent Japan EQ (Tohoku) epicentre uncertainty of 8.4 miles

� Hazard location uncertainty has impact on the calculation of intensities etc.

Uncertainty – Hazard II

� Tsunamis area even more problematic to model

– Wave inundation effects

– Effect of vegetation / debris on

wave run-up

– Back-wash

� Building stock - property characteristics or building modifiers

– roof structure

– dwelling type

– construction type

– number of storeys

Uncertainty – Vulnerability I

– number of storeys

– age

� Not enough detail - generally on a regional scale

� Could more accurate data be used?

� Use of Remote Sensing and GIS?……perhaps but costs are too high

Building modifier in

catastrophe model

vulnerability

component

Non-spatial data

(but typically

regional

aggregations)

Ordnance Survey

MasterMap (OS,

2011)

Cities Revealed

Building Class

Dataset (GIG, 2011)

Roof Type Yes No No

Window Type Yes No No

Type (occupancy) Yes Yes Yes

Construction Type Yes No No

Basement property No No No

Uncertainty – Vulnerability II

Basement property No No No

Housing Type

(detached/semi-

detached/terraced)

Yes No Yes

Age Yes No Yes

Building size (area) Yes Yes Yes

Number of stories Yes No No

Typical building modifiers used in a UK based model and

potential datasets

� Exposure = the properties which the insurer covers

� More accurate the address details = closer to actual location in the model

� Flood models especially need accurate addressing

Uncertainty – Exposure I

Province 2-digit postcode (CRESTA) 4-digit postcode

� Compare the 2 examples of a river flood

Uncertainty – Exposure II

Insured property in postcode area

=

affected by the river flood

Insured property building centroid

=

not affected by river flood

� (un)knowledge of the geographical location has the largest impact on accuracy of final losses

Uncertainty - Loss

The impact of improved data on Hurricane

Charley loss estimation

� ELEMENTS developed by Impact

Forecasting is “open-box” unlike other

commercial catastrophe models

� In ELEMENTS, exposure and hazard

uncertainty dealt with by the Master Table - probability of a particular event occurring

ELEMENTS - Coping with Uncertainty I

ELEMENTS

- probability of a particular event occurring

within a particular administrative area

� ELEMENTS deals with uncertainty in

vulnerability by using the Chance of

Loss concept

- the likelihood that a particular

property will be affected by an event

� ELEMENTS also allows the user to select

different vulnerability functions for a given

region to determine the effect of other building

stock information.

� Finally for the exposure component

ELEMENTS allows multiple geographic

ELEMENTS - Coping with Uncertainty II

ELEMENTS allows multiple geographic

resolutions to be used

� US-based models in ELEMENTS can be

matched at building centroid (longitude

and latitude), ZIP Code, County and State

administrative levels.

The ELEMENTS user interface

� There will always be uncertainty in

Cat Models – but we can deal with it

� Location uncertainty has the largest effect on calculated losses

� To communicate uncertainty it is sensible

Conclusions I

� To communicate uncertainty it is sensible

to give a range of potential losses rather

than a single monetary figure

� By understanding the catastrophe modelling

process (re)insurers can better understand

their capital requirements

� ELEMENTS allows uncertainty to be

modelled and understood by allowing each

component of the model to be seen and

modified by the user

� ELEMENTS copes with uncertainty by

allowing a range of probable maximum

Conclusions II

allowing a range of probable maximumloss figures to be generated

� With the advent of increasing regulatory

requirements (such as of Solvency II),

a fully-documented and understandable

catastrophe model is required –

ELEMENTS can provide this

Thank you!

Chris Ewing

Aon Benfield Analytics

Impact Forecasting

t: +44 (0)207 522 8305

e: chris.ewing@aonbenfield.come: chris.ewing@aonbenfield.com

twitter: @web_gis

References

AIR Worldwide (2010), Understanding uncertainty, http://www.air-worldwide.com/PublicationsItem.aspx?id=19060

AIR Worldwide (2011), http://www.air-worldwide.com/

Aon Benfield (2010), Solvency II for Reinsurance Managers, http://www.aon.com/attachments/reinsurance/201006_solvency_II_reinsurance_managers_full.pdf

Aon Benfield (2011), Information about Aon Benfield, http://www.aonbenfield.com

Aon Benfield (2011a), Impact Forecasting, Aon Benfield, http://www.impactforecasting.com

Aon Benfield (2011b), ELEMENTS: Natural Catastrophe Models documentation, Aon Benfield, 2011

Aon Benfield (2011c), Norway Flood Model Presentation, Aon Benfield, April 2011

Aon Benfield (2011d), Kazakhstan EQ Model Technical Documentation, Aon Benfield, July 2011

Dlugolecki, A. et al. (2009), Coping with Climate Change: Risks and opportunities for Insurers, Chartered Insurance Institute, London/CII_3112

EMSC-CSEM (2011), Mw 9.0 off the Pacific coast of Tohoku, Japan Earthquake, on March 11th, 2011 at 05:46 UTC, http://www.emsc-csem.org/Page/index.php?id=196

EQECAT (2011) http://www.eqecat.com/

FSA (2011), Solvency II, http://www.fsa.gov.uk/pages/About/What/International/solvency/index.shtml, Financial Services Authority, August 2011

References

GIG (2011), Geo-Information Group, Cities Revealed Building Class - http://www.geoinformationgroup.co.uk/products/building-class

Grossi, P. (2004), Sources, nature and impact of uncertainties in catastrophe modelling, 13th World Congress on Earthquake Engineering, http://www.iitk.ac.in/nicee/wcee/article/13_1635.pdf

Holmes, K.W., Chadwick, O.A., Kyriakidis, P.C. (2000), Error in a USGS 30-meter digital elevation model and its impact on terrain modeling, Journal of Hydrology 233 (2000) p154-173

Nurmagambetov A., Mikhailova N., and Iwan W., 1999: Seismic hazard of the Central Asia Region. In: King S. A. et al. (eds.): Seismic Hazard and Building Vulnerability in Post-Soviet Central Asian Republics, p1-43

ONS (2011), http://www.statistics.gov.uk/census2001/profiles/commentaries/housing.asp

OS (2011), http://www.ordnancesurvey.co.uk/oswebsite/products/os-mastermap/

Swiss Re (2011), Swiss Re – Established 1863 - http://www.swissre.com/about_us/established_1863/

RMS (2008), A guide to Catastrophe Modelling, Worldwide Reinsurance, http://www.rms.com/Publications/RMS%20Guide%202008.pdf

RMS (2011), http://www.rms.com/

Rudi, W. and Toksoz, M.N., 2001, Uncertainty Analysis in Seismic Event Location, In: 23rd Seismic Research Review: Worldwide Monitoring of Nuclear Explosions – October 2-5, 2001, p324-322

USGS (2011), Honshu EQ, http://earthquake.usgs.gov/earthquakes/eqinthenews/2011/usc0001xgp/

Zhunusov T., Taubaev A., Itskov I., Mikhailova N., and Nurmagambetov A., 1999: Seismic Hazard and Building Vulnerability in Kazakhstan. In: King S. A. et al. (eds.): Seismic Hazard and Building Vulnerability in Post-Soviet Central Asian Republics, p67–92.