Вы находитесь на странице: 1из 27

UsingRobustSimulationto CharacterizeUncertaintiesin CatastropheLossAssessments

Yajie J.Lee,CraigE.Taylor,Zhenghui Hu,WilliamP.Graf, CharlesK.Huyck RAACatModeling2014 February1113,2014,Orlando,FL ImageCat, Inc.


Supportingtheglobalriskanddisastermanagementneedsoftoday,usingthe technologiesoftomorrow

Timeline of EQ Catastrophe Modeling for Insurance


EQECat founded RMS founded AIR founded HAZUS Building Models
Dynamic Financial Analysis

Cloud Computing

R b Robust Si Simulation l i
(recognized uncertainty)

Usage

ATC-13
Government studies [Wiggins, ESSA, Friedman]

IBM PC

Risk Curves ('EP') for synthetic catalogs


Deterministic Cat Scenarios
Tracking Exposure Accumulations in Zones
Realistic Disaster Scenarios

1970
San Fernando Valley
USGS Probabilistic Hazard Maps
1976

1980
Loma Prieta
Algermissen & Perkins

1990

2000
Northridge

2010
Tohoku + Christchurch
2014

2020

1990

Frankel et al Petersen et al 1996 2002 2008

AFrameworkforMoreRobustUncertainty yAssessment
RobustSimulation

CurrentApproach Loss sAmount( ($M)

AverageReturnInterval(Years)

Agenda
Scientificchallenges WhatisRobustSimulation? Examples
USGSnationalseismicmaps Singleproperty Portfoliolosses

Conclusions l

Scientificchallenge:Whatassumptionsaremadeaboutmaximum events? AssumedMCEwaslessthan8.0 BUT!Japanesegeologistshad recognizedlargereventsinthe decadebeforetheearthquakefrom ancienttsunamirecords OtherstudieshadsuggestedM8 M8.5 5~ M9.6couldoccurinanysubduction zones Thoku wasM9.0,ruptured5segments

(Imagesource:Steinetal.,TectonicPhysics,2012)

Scientificchallenge:Howaccuratelyareseismicsourcesmapped?
(USGSNSHM,2014)

NewMadrid Magnitude M6.8toM8.0 Reoccurrencemodel 500 to50,000years Cluster&Noncluster


Pseudofaults
Modified M difi dfault f ltmodel d lfor f nuclearpowerplant

Scientificchallenge:Howstrongwillthegroundshake?
GMPEsforSubduction Earthquakes (Abrahamson2012)

GMPEsforCEUS(USGSNSHM,2014)

Scientificchallenge: g Howwelldoesabuilding gperform p duringgroundshaking?

(Forcebasedp ( paradigm) g )

(Displacementbased paradigm)

(A1970ConcreteFrame)

Increasingbuildingperiods

(CODA:CodeOrientedDamageAssessmentmethod, adaptedfromATC13,Graf&Lee,Spectra,2009)

WhatAssumptions p aremadeaboutmissing gdata?

Loss

Time

Loss s

Time

Los ss

Time

Scientificchallenge:Withalltheseoptions, options whatistheright right answer?


Like elihood

Asinglenumber?

Or
Likelihood

Severity Adistribution basedwhatwenowknow?

Severity

Scientificchallenge:Howdoweaccountforuncertainty? Twotypesof fmodeled d l duncertainties Innermodelrandomness(aleatory)


Randomvariabilityinpredictingthenaturalprocess Givenenoughtime,thevalueswillberealized afterasufficientnumberofeventcycles

Outermodelvariation(epistemic)
Lackofknowledge g (scientific/model ( / uncertainty) y) Onlytimeorfutureresearchwilltellthecorrectmodel

RobustSimulation:Representationof futureriskthroughsimulationofasuite ofpossibleanswersthatintegratesvalid scientificdisagreementandstochastic modelingofunknownvariables.

Severi ity

ReturnInterval

Application pp ofrobust estimation: S i i hazard Seismic h dassessment

The logic tree for the California earthquake models

2008USGSNationalSeismicHazardMapping(NSHM). Logic will be more complex in 2014 NSHM models!

C o u n t
14 10 12 0 0 .0 2 0 .0 6 0 .1 4 0 .1 0 .1 8 0 .2 2 0 .2 6 0 .3 3 0 .3 4 0 .3 8 0 .4 2 0 .4 6 0 .5 5 0 .5 4 0 .5 8 0 .6 2 0 .6 6 0 .7 7 0 .7 4 0 .7 8 0 .8 2 0 .8 6 0 .9 9 0 .9 4 0 .9 8 2 4 6 8

A site in San Francisco

Probabilistic seismic hazard results

PGA (g)

475y year(USGS) ( )

C o u n t
12 10 0 0 .0 0 2 0 .0 0 6 0 .1 0 .1 1 4 0 .1 1 8 0 .2 2 2 0 .2 2 6 0 .3 0 .3 3 4 0 .3 3 8 0 .4 4 2 0 .4 4 6 0 .5 0 .5 5 4 0 .5 5 8 0 .6 6 2 0 .6 6 6 0 .7 0 .7 7 4 0 .7 7 8 0 .8 8 2 0 .8 8 6 0 .9 0 .9 9 4 0 .9 9 8 2 4 6 8

PGA (g)

Groundmotiondistributionat475yearreturninterval
475year(USGS)

A site in Los Angeles

Applicationofrobustestimation: Earthquakelossassessmentfora singleproperty

Asample p midriseofficebuilding g
LosAngeles,1965
Concreteshearwall

Concreteshearwall: 35% Concreteframe: 17.5% Steelframe: 17.5% Woodframe: 10% Steelconcentricbracedframe:20%

Applicationofrobustestimation: ComparisonwithFederico Waismans 2010RAAearthquake results

(Waisman,2010)

ImageCat (ConventionalApproach)

(Waisman,2010)

ARobustEstimateofUncertainty
RobustSimulation LossAmou unt($M) 475year

ConventionalEPcurve

AverageReturnInterval(Years)

ARobustEstimateofUncertainty

unt($M) LossAmou

AverageReturnInterval(Years)

UsingmultipleanswersfromvariousCatmodelersdoesnotnecessarily disclosethefullrangeofuncertaintyinpotentialCatrisks.

R b tSimulation: Robust Si l ti
Accepts p modelimperfection p Acknowledgesmodel limitations Revealsuncertaintyassociated withimperfectknowledge Encouragesmodeldivergence Fewersurprises(Black Swans ) Swans) Creditordiscreditextreme modelsfromfutureresearch

RobustSimulation:
Providesacontrolled simulationprocess
Modeltraceability Preservescoherencyand integrity

Uses U nonparametric i statistics i i


Minimizesneedforcomplex classicalstatistics Propagatesuncertaintythrough layers,simplifyingcomplex accountingmethodtreatment

RobustSimulation:
Simplifiestimesequenceof losses
Clusteredevents,e.g.,NMSZ, Christchurch

Simplifiesmultiperilmodeling
Wind Flood Winterstorm Earthquake Tsunami Terror

Thedrunkardswalk

Conclusions
Asingleviewofrisksuppresses modeluncertainty yandimplies p illusoryprecisiontomodeling futureriskwithmanyunknowns. Regulatorsandmodelershave beguntostressunderstanding modellimitationsand uncertainty Uncertaintycanchangewhen moreinformationisacquired Robustsimulationprovidestools toassessuncertaintyadequately andallocateresources accordingly gy

Thailand

Thoku

Christchurch

Yajie (Jerry)Lee TechnicalDirector yjl@imagecatinc.com ImageCat,Inc.

Вам также может понравиться