Natural Catastrophe Modeling Devashish Kumar, Feb 24, 2012 CEE, SDS Lab @ Northeastern University Education
. B.Tech. in Civil Engineering
. M.S. in Civil & Environmental Engineering – Computational Mechanics
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 2 Experience: Baker Engineering and Risk Consultants
. Quantitative Risk Assessment of Industrial Structures subjected to accidental explosions . BP, Shell, Conoco Philips, Exxon Mobil, Schlumberger, US Dept of State . Performed Finite Element Analysis of a storage tank for a law firm to support accident investigation . Designed a new Shock tube for the University of Ontario, Canada
© BP p.l.c (Texas City) Shock Tube@ Baker Engineering and Risk Consultants
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 3 Experience: Swiss Reinsurance Company (Swiss Re)
. Estimated unbiased risk premium of insurance portfolios exposed to natural perils such as hurricanes, earthquakes, and tornados
. Performed loss analysis of industrial risks to recalibrate state-of-the-art vulnerability curves
. Analyzed insurance portfolios of Ace Insurance Company for structuring of Catastrophe Bond (Cat Bond) for US HU and US EQ
. Liberty Mutual, AIG, USAA, Farmers, ZFS, Ace Insurance Company, FM Global
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 4
Motivation 1) Will climate change alter Nat Cat frequency and severity? 2) Eventually, how much thinner will my wallet be?
Climate Change
1)
Nat Cat: Frequency and Severity
2)
Source: Munich Re Loss Modeling www.climatechangebusinessforum.com/page/file/440/download
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 5 How risks are transferred? Living life and running businesses involve risks. Individuals & companies buy insurance. Insurers buy reinsurance.
Devashish Kumar| Mar 1, 2012| Nat Cat 6 Modeling
Transfer of risks: Policyholder, Insurance, & Reinsurance
Source: The essential guide to reinsurance by Swiss Re
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 7 Why Nat Cat Modeling?
Devashish Kumar| Mar 1, 2012| Nat Cat 8 Modeling Fire vs Natural Perils
. Fire losses occur relatively frequently (for an entire portfolio), are fairly consistently over time, and can be analyzed using statistical methods . Natural catastrophes occur rarely, and their losses fluctuate radically (no losses for decades and suddenly a year of enormous loss, Katrina 2005) . Natural perils loss data is not representative for statistical analysis
need for scientific risk assessment
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 9
Insured Catastrophe Losses 1970 - 2010
Increasing values Hurricane Katrina,2005
Concentration in
exposed areas Hurricane Ike, Gustav Ike, Hurricane
Insurance penetration 1992 Andrew, Hurricane
Attack onWTC,2001 Attack
Northridge Earthquake,1994 Northridge Chile Earthquake,2010 Chile Changing hazard . Climate variability . Climate change
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 10 Concentration in Exposed Areas
Ocean Drive, FL, 1926 Ocean Drive, FL, 2010 Population Growth Rate: 1930-2010 All US 150% Florida 1180%
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 11 Origins of Nat Cat Modeling
. In 1800s, residential insurers covering fire risk used pins on a wall-hung map to visualize concentrations of exposure; the practice ended in 1960s . Computer-based probabilistic catastrophe risk modeling started in the late 1980s
. Wide acceptance of Nat Cat models by re/insurance companies after unprecedented losses resulting from hurricane Andrew in 1992
. 3 major Nat Cat Modeling Companies:
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 12 Probabilistic Hazard Modeling: 4 Box Principle Wyndham Partners September 22, 2005
Natural Hazard Where, how often, and with intensity do events occur?
Photo 5. A near-coastal house collapsed by storm surge augmented with debris from buildings destroyed up stream.
Vulnerability What is the extent of damage at a given intensity?
Photo 6. A beachfront hotel just east of Gulfport, Mississippi. Note all that remains is the reinforced concrete frame and roof diaphragm. All cladding and interior improvements have been cleared from the structure by storm surge.
5 Value Where are insured properties located Distribution and how high is their value?
Insurance What portion is loss insured and Policy Limit Conditions what is deductible? Deductible
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 13 Where, how often, and how severe?
Natural Hazard Parameters that define Exposure to natural hazard risks: . Geographical distribution (FL HU, CA EQ) . Occurrence frequency (1 in 100 years) Vulnerability . Intensity (MMI for EQ, wind speed for HU)
Event Set: Value Distribution . Historical event catalogues and scientific research are used to quantify above parameters . Simulate all possible events that could unfold over Insurance thousands or tens of thousands of years Conditions . Model produces a “representative” list of event losses
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 14
Intensity Measure
Natural Hazard Earthquake: MMI Modified Mercalli Intensity
Vulnerability
Value Distribution
Insurance Flood: Atmospheric Perils: Conditions Water Depth (m) Wind Speed (m/s)
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 15 How is a Building’s Vulnerability Classified?
Through a vulnerability function that relates the expected amount of damage to the severity of the hazard, such as the peak wind speed or ground motion. Vulnerability Curve: How extensive will the damage be? Hurricane Vulnerability Function
Natural HazardHow is a Building’100 s Vulnerability Classified? )
% (
o i
t A building’s vulnerability at a given
a R
hazard intensity is measured by its
e
Through a vulng erability function that relates the expected
a 50 mean damage ratio. m
Vulnerabilityam ount of dama age to the severity of the hazard, such as the
D
average loss n
a MDR = peak wind speee d or ground motion. replacement value M Mean damage 10 0 Hurricane Vulnerability Function 120 Value 100 Peak wind speed (mph)
Distribution )
%
(
o
i
t Building Value = $200,000 A building’s vulnerability at a given
a © 2009 Risk Management Solutions, Inc. CONFIDENTIAL 4
R hazard intensity is measured by its
e
g
a 50 MDR = 10% @120mph mean damage ratio.
m a Insurance D average loss
n How is a Building’s Vulnerability Classified? a Loss = $20,000 MDR = Conditions e replacement value M Mean damage 10 0 A buildin1g20’s (or its contents) vulnerability is characterized by a vulnerability function (also Devashish Kumar| Mar c1,o 2012|mm NatoP nCatealky Modelingw icnda slpleee dd ( maph )damage curve) that relates the 16e xpected amount of damage to the severity of the hazard. For earthquakes, the hazard is quantified in terms of a ground
© 2009 Risk Manamgemoent tSoiloutionns, Inicn. tensity measure;CO NfFoIDErNT IAhL urricanes, the hazard is quantif4 ied by the peak wind speed. How is a Building’s Vulnerability Classified? As an example, let’s examine a hypothetical hurricane vulnerability function. Plotted along the horizontal axis is the peak wind speed and plotted along the vertical axis is the mean A building’s (ord iatsm caognet ernattsio) v(MulDneRr)a buislietyd itso c qhuaarancttifeyr iztheed abmy oa uvnutl noefr adbaimlitayg feu.n Actsi oind (iaclastoe d in this plot, the commonly callevdu lan edraambialigtye fcuunrvctei)o nth (asth roewlante isn tbhleu ee)x pineccrteeads aems ofruonmt ozfe rdoa mdaamgea gtoe tahte l ow wind speeds to severity of the nheaazarlryd 1. 0F0o%r e adratmhqaugaek east, htihgeh hwaiznadrd s pise qeudas.n tAifsie wd ein w tiellr msese osfh ao rgtrlyo,u tnhde shape of the motion intensitvyu mlneearsaubrielit;y f ofur nhcutriorinca ins edse, ftihnee dh abzya trhde i sc qouveanratigfiee dty bpye t(hbeu ipldeiankg w, cinodn tents, or time related speed. losses) and several key physical features of the building; that is, different classes of coverage and building types have different vulnerability functions. As an example, let’s examine a hypothetical hurricane vulnerability function. Plotted along the horizontal axis is the peak wind speed and plotted along the vertical axis is the mean damage ratio (TMoD fRu)r tuhseerd u tnod qeursatnatnifdy thhoew a tmhoeu vnut lnoef rdaabmiliatyg ef.u nAcst ionndi cisa tuesde din inth tish ep lroits,k t hmeo deling process, vulnerability fusnucptiopno s(es hthowe nh ainz abrldu em) oindcurleea (swesh ifcrhom fe zeedrso indtaom tahgee vautl nloewra wbiilnitdy smpoededusle t oa s indicated in the nearly 100% dapmreavgioeu ast s hliidgeh) wpirnedd iscptse ead ps.e Aaks wein wd isllp seeeed s ohfo r1t2ly0, mthpeh s hfoarp ea obfu tilhdein g of interest. With this vulnerability fuinncptiuotn, tish ed evfuinlneedr abbyi ltihtye fcuonvcetrioang ei st yupsee d(b tuoi ledsintigm, actoen ttehnatts t, hoer etixmpee crteeladt eadm ount of damage; losses) and seviner athl ikse cya pseh,y sthicea lv fuelanteurraebsi loitfy tfhuen bctuiioldni npgr;e dthicatts is1,0 d%if fedraemnat gcela.sses of coverage and building types have different vulnerability functions. As indicated in this hypothetical example, we quantify damage as a mean damage ratio To further unde(MrsDtaRn)d, hwohwic hth ies vsuimlnpelrya bthileit yr aftuinoc otifo tnh ies auvseerda igne t hane tricisipk amteodd elolisnsg tpor tohcee srse,p lacement value suppose the haozfa trhde m boudiludlien g(w. hich feeds into the vulnerability module as indicated in the previous slide) predicts a peak wind speed of 120 mph for a building of interest. With this input, the vu©ln2e0r0a9bRilisitky M faunnagcetmioennt iSso ulustieonds ,t Ion ce. stimate that tChonef ideexnptiealcted amount of damage; 4 in this case, the vulnerability function predicts 10% damage.
As indicated in this hypothetical example, we quantify damage as a mean damage ratio (MDR), which is simply the ratio of the average anticipated loss to the replacement value of the building.
© 2009 Risk Management Solutions, Inc. Confidential 4 Vulnerability curves = f (building characteristics, coverage)
Natural Hazard . Occupancy Type: residential, commercial, industrial
. Construction Class: Wood frame, masonry, reinforced concrete, steel, mobile home Vulnerability
. Building Height: 3 story RCC vs 15 story RCC
Value . Year of Construction: 1980 vs 2012 (building code) Distribution
. Insurance Coverage: Building, Contents, Business Insurance Interruption Conditions
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 17
Exposure Data and Geocoding
Natural Hazard Exposure Data: . Total Insured Value of assets that is at risk . Construction, occupancy, height, year built Vulnerability . Address . Perils to be modeled
Value Distribution Geocoding: . Process of finding Lat/Lon . Risk is at coast or inland (how far) Insurance . Higher the geocoding resolution, more precise the loss Conditions numbers
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 18
Deductibles & Policy Limit
Natural Hazard Deductible: . Share of loss among policyholder, insurer, and reinsurer . Cap the amount the re/insurer is liable to pay Vulnerability . Reduce the re/insurer’s administrative burden . Can be number of days for Business Interruption
Value
Distribution
Insurance Conditions
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 19 Loss Modeling Process
. 1000 insured buildings . Total sum insured = $1000 million . EQ prone region . 12 over a projected period of 200 years
Source: Natural Catastrophes and Reinsurance by Swiss Re
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 20 Vulnerability Output: Loss Frequency Curve Quantify
Value Insurance Hazard Vulnerability Natural Hazard Distribution Conditions
Combine
Loss Vulnerability
Combine
Value Distribution Frequency
Page 16 Insurance Conditions
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 21 Can we trust Nat Cat Models?
. A model is nothing more than a simplified representation of reality
. Yes if – models are calibrated (many times sufficient data is not available) – used within their limits – exposure data has sufficient detail and is of high quality – Un-modeled perils are properly considered – Aware of risks beyond the model parameter
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 22 Disproportionate changes in extremes in comparison to changes in mean climate conditions
Hazard Cause of Change in Hazard Resulting Change in Damage/Loss Doubling of wind speed Four-fold increase in damages Windstorm 2.2 °C mean temperature Increase of 5-10% hurricane wind increase speed Extreme temp 1 °C mean temperature increase 300-year temperature events occur every 10 years Flooding 25% increase in 30 minute Flooding return period reduced precipitation from 100 years to 17 years 1 °C mean summer temperature 17—28% increase wildfires Bushfire increase
Doubling of CO2 143%increase in catastrophic wildfires
Mills E, Lecomte E, Peara A., US Insurance Industry Perspectives on Climate Change, Feb 2001, US Dept of Energy
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 23 Source: Munich Re Summary www.climatechangebusinessforum.com/page/file/440/download
1) Will climate change alter Nat Cat frequency and severity 2) Eventually, how much thinner will my wallet be?
Climate Change Almost certain 1) for many perils Economic loss can significantly outpace
Nat Cat: Frequency and Severity climate change
2) Nonlinear Insurance loss can Loss Modeling increase even more
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 24
Devashish Kumar| Mar 1, 2012| Nat Cat Modeling 25 The [property/casualty] industry is at great risk if it does not understand global climate variability and the frequency of extreme events. -Franklin Nutter, President, Reinsurance Association of America (1999)
Devashish Kumar| Mar 1, 2012| Nat Cat 26 Modeling