Modeling Risks of Earthquake and Hurricane in …...The New EQECAT Creating Value Through Increased...
Transcript of Modeling Risks of Earthquake and Hurricane in …...The New EQECAT Creating Value Through Increased...
The New EQECAT Creating Value Through Increased Transparency
Modeling Risks of Earthquake Modeling Risks of Earthquake and Hurricane in Mexico
Dennis E. KuzakSenior Vice PresidentSenior Vice President
25 Mayo 2010
Creating Value Through Increased TransparencyProprietary and Confidential
Agenda
Overview of Mexico hurricane and earthquake RisksModeling mechanicsThe future challenges in risk modeling
Creating Value Through Increased TransparencyProprietary and Confidential
Modeling MethodologyPropagation of Hazard to sitePropagation of Hazard to site
Estimation of Loss
Estimation of Loss
Hazard DefinitionHazard Definition Estimation of Damage
Estimation of Damage
• HurricaneStorm Tracks
• HurricaneSurface Wind Speed
• Vulnerability functions
Occupancy
• Loss dataValidated with historical loss dataStorm Tracks
• EarthquakeMagnitudes
Frequencies
Surface Wind Speed
Local Wind Gust
• EarthquakeAttenuation functions
Occupancy
Building age
Number of stories
Construction materialeque c es
Soil MapsConstruction material
Creating Value Through Increased TransparencyProprietary and Confidential 3
Exposures Follow Population Density
Creating Value Through Increased TransparencyProprietary and Confidential
Source: Wikipedia
Hazard: Historical Hurricanes in Mexico
<
.
.
.
. ..
....
.
<
...
.. .
..
<
.
..
. ...
.
<< .
.
.
<
..
.
.
.
.
..
.
.
.
.
.
.
....
..
..
. .
.
.. ..
.
..
.
<<
.
.
.
.
..
<.
..
.
......
... .
.....
... . .
.
<
<.
.
..
.
.
...
.<
.
.
.
.
..< .. .
.
.
. ...
.
..<. .
...
.
..
<.
... .
.
...
<...
.
.
...
..
..
..
.<.. ..
. .
. ....
.
.
..
. .. .
.< :::::::::
.
. .
.. .
...
.<
<<
.
.... . .....
. ..
.
.
.
..
. .
..
.
<
<
.
.
...
.
..
.
. .. ..
....
..<
.......
.
.
..
.
..
..
<
.
.
.
..
. .
.
....
.< .. .
.
.
..
...
.
..
.
.
.
.
...
<.
...<
.
. ..
<
<
.
...
< ..
..
.
.
..
. .
.
<.
...
. .
...
..... . .. .
.
... .
..
..
.....
Creating Value Through Increased TransparencyProprietary and Confidential
Black=tropical depression, Green=Tropical Storm, Yellow=Hurricane, Red= Intense Hurricane (Cat 4, 5)
(Data: (Caribbean: 1851-present;Pacific: 1949 – present)
Intense Hurricanes only
<
.
.
.
. ..
....
.
<
...
.. .
..
<
.
..
. ...
.
<< .
.
.
<
..
.
.
.
.
..
.
.
.
.
.
.
....
..
..
. .
.
.. ..
.
..
.
<<
.
.
.
.
..
<.
18 Cat 4/5 landfalls160 years (~0 11 / year)
..
.
......
... .
.....
... . .
.
<
<.
.
..
.
.
...
.<
.
.
.
.
..< .. .
.
.
. ...
.
..<. .
...
.
..
<.
... .
.
...
<...
.
.
...
..
..
..
.<.. ..
. .
. ....
.
.
..
. .. .
.< :::::::::
.
. .
.. .
...
.<
<<
.
.... . .....
. ..
.
.
.
..
. .
..
.
<
<
.
.
...
.
..
.
. .. ..
....
..<
.......
.
.
..
.
..
..
<
.
.
.
..
. .
.
....
.< .. .
.
.
..
...
.
..
.
.
.
.
...
<.
...<
.
. ..
<
<
.
...
< ..
..
.
.
..
. .
.
<.
...
. .
...
..... . .. .
.
. .. .
12 Cat 4/5 landfalls60 years (~0.2 / year)
160 years ( 0.11 / year)
..
..
.....
Creating Value Through Increased TransparencyProprietary and Confidential
Latin America Seismicity
IPGH ifi d th kIPGH unified earthquake catalog from 1800-2003Earthquakes defineEarthquakes define major tectonic boundariesEarthquake parameters:
– Magnitude
– Epicenter locationEpicenter location
– Hypocenter depth
– Special studies of significant earthquakes
Creating Value Through Increased TransparencyProprietary and Confidential
earthquakes
Types of Seismic Sources
SubductionInterface
Shallow InterfaceBackground
Shallow Crustal
Deep Crustal
Deep Interface Wadati-Benioff
Creating Value Through Increased TransparencyProprietary and ConfidentialCONFIDENTIAL
Background
Earthquake Recurrence Relations
Type of Relation Type of Distribution Type of Sources Type of Relation Type of Distribution Type of Sources
Exponential Gutenberg-Richter (abbreviated GR)
Area sources & background ( )
(log N = a – bM) g
seismicity Characteristic Faults: Gaussian (bell
f h l) Faults where
ti curve from school) Subduction interface: uniform (box)
recurrence times are available (Type A)
Exponential & Characteristic
33/67 weighted GR/characteristic
Faults where only slip-rate data are available (Type B)
Creating Value Through Increased TransparencyProprietary and Confidential
available (Type B)
Time-Dependent Recurrence
C diti l b bilit th t th k ill d iConditional probability that an earthquake will occur during a specified period of time in the future given that one has occurred at a known date in the past (i.e., “the earthquake cycle”)cycle )Accounts for the effects of secular strain accumulation and release (“the elastic rebound theory”)Updated from study by Stuart Nishenko (1991), who at the time was a seismologist with the US Geological Survey
– Updated with earthquakes to 2003
– Updated with IPGH unified earthquake catalog
Creating Value Through Increased TransparencyProprietary and Confidential
Faults Modeled as Time Dependent
Subduction ZonesMexicoCentral America
Crustal Fault ZonesCaribbean-N. American plate boundary (Guatemala)Central America
PeruChile*
Antarctic-S. American plate boundary (Tip of S. America))
*Time dependent frequencies near Chile being updated July 2010
Creating Value Through Increased TransparencyProprietary and Confidential
Time Dependence: Secular Strain AccumulationAccumulation
Lognormal probability Conditional Probability = A/(A + B)Conditional Probability = A/(A + B)
distribution of recurrence time– Conditional on elapsed time
since last eventLastLast
AA
BB
– Probability increases as strain accumulates & elapsed time increases
LastEarthquake
LastEarthquake
• Probability depends strongly on standard deviation (σ)
Decreaseswith σ
on standard deviation (σ)Larger standard deviation leads to smaller effectTypical value is 0.5
Increaseswith σ
Creating Value Through Increased TransparencyProprietary and ConfidentialCONFIDENTIAL
yp
Exceedence Probability Curve:Hurricane
1.00
Hurricaneab
ility
20% to 50% annual probability of experiencing losses as large as from hurricane Wilma
0.10
danc
e Pr
oba
R f 200
0.01
nnua
l Exc
eed Repeat of 2005
Hurricane Wilma($0.40-$0.50 bn loss)
0 00
An
Creating Value Through Increased TransparencyProprietary and Confidential
0.000 2 4 6 8 10
Nationwide Ground-Up Damage, USD billions
Exceedence Probability Curves:Earthquake and HurricaneEarthquake and Hurricane
11.00
100.10
rsbabi
lity
turn
Per
iod,
yea
Exce
edan
ce P
rob
Earthquake
Hurricane1000.01
Ret
Ann
ual E Hurricane
1% to 2% annual probability of experiencing losses as large as from the '85 quake
1,0000.00
0 20 40 60 80 100
Repeat of 1985 Quake:$15 to $25b Losses
Creating Value Through Increased TransparencyProprietary and Confidential
Nationwide Ground‐Up Damage, USD billions
Michoacan Earthquake, M8, 19 Sep 1985
We learn from collapses……and we learn…and we learn equally from buildings that remain intact
Creating Value Through Increased TransparencyProprietary and Confidential
Photos courtesy of EQE Incorporated EQ Reconnaissance report, 1985
Simulation of Shaking in Mexico City
Th f i t h ki iThe zone of intense shaking in this simulation of the 1985 quake reflects the shape of ancient Lake Texoco
::::::::::::::::::::::::::::::::::::
.
<
<
<
<
<
<
<:::::::::
<
<
<
.
.
..
.
.
.
.
..
.
.
.
.
.
. .
.
.<
.
.
.
.
.
.
.
<
.
.
..
.
.
.
.
.
<...
.
..
..
.
.<
.
.
.
.
. .
.
.
.
.
.
<
:::::::::
.
.
.
.
.
.
.
.
.
<
<
<
.
.
.
.
. .
.
..
..
.
.
.
.
.
.
.
.<.
..
..
.
.
..
.
.
.
.
.
.
.
<
.
.
.
..
. .
.
.
.
. .
Creating Value Through Increased TransparencyProprietary and Confidential
<
<
.
.
.
.
.
.
.
<
.
.
.
<
.
.
.
10 km 10 km300 km
Business Interruption
Function of % damage to structure & contents.Economic loss has to be defined by metrics such as revenue or gross profit margin.BI is calculated at damage level, so policy deductibles and limits can be modeled.Loss Amplification is based on estimates of supply and demand of labor and materials Calculation is done on an event by eventof labor and materials. Calculation is done on an event by event basis- i.e. no factor is applied.Confidence in modeled BI is lower than PD and Contents since detailed BI coverages are not usually modeled and claimsdetailed BI coverages are not usually modeled and claims settlement can be difficult to predict.Contingent BI ( damage to assets not owned) is more challenging to model but in some cases, critical to the insured- e.g. GOM oil
Creating Value Through Increased TransparencyProprietary and Confidential
and gas producers. In these cases, network models are used.
Using Portfolio Models for Specific Risks
Geocoding (i.e. CRESTA)Variability of hazard within CRESTALocalized site conditions- e.g. coastal site withLocalized site conditions e.g. coastal site with surge/flooding or soils liquefiable under earthquake.Variability in single site building characteristics.Limits on single site policy sub limits and policy limitsLimits on single site policy sub limits and policy limits.Notwithstanding limitations, can be used as first level risk estimate, especially when risk data is not detailed enough for detailed modelsfor detailed models.
Creating Value Through Increased TransparencyProprietary and Confidential
Non proportional modeling
Modeling starts with ground up damage estimates, followed by application of primary coverages.Mexico model supports– Primary policy conditions: deductibles, location, co-insurance,
sub limits
– Reinsurance policy conditions: Semi-automatic bothReinsurance policy conditions: Semi automatic both proportional and non-proportional and Cat XL, with reinstatements
– No facultative
Creating Value Through Increased TransparencyProprietary and Confidential
Return Period
For each peril, risks are calculated using event frequencies and underlying frequency (recurrence) models:– Poisson
N ti Bi i l– Negative Binomial
– Exponential
– Characteristic
Conditional probability (e g time dependent)– Conditional probability (e.g. time dependent)
150,000 years of losses calculated for each peril based on above event frequencies, resulting in loss probability per yearReturn period is inverse of loss probability per year but can beReturn period is inverse of loss probability per year, but can be misleading since it implies a steady state recurrence assumption.Most insurers and reinsurers use return periods of 100 to 250 years for hurricane and EQ. (CEA uses 500-1000 years)
Creating Value Through Increased TransparencyProprietary and Confidential
years for hurricane and EQ. (CEA uses 500 1000 years)
Modeling non-common structures
Structural and architectural considerations: height, design criteria, building materials, quality of constructionPerformance in other cat eventsIndentify major failure modesMap to building types in modelAdjust vulnerability using quality factors in modelAdjust vulnerability using quality factors in modelUsually assume constructive total loss with damage>50%
Creating Value Through Increased TransparencyProprietary and Confidential
Model vs Real losses
Limited insured loss information. Major Mexican events include 1985 Michoacan EQ and 2005 Hurricane WilmaEQECAT models based on damage info with limited claims data.Event characteristics created by EQECAT modelingEvent characteristics created by EQECAT modeling specialists- not by clientsBetter info on building valuation, design details, contents needed to improve model estimatesneeded to improve model estimates
Creating Value Through Increased TransparencyProprietary and Confidential
Conclusions
Model development keyed to:– Scientific research: historical data
supplemented with numerical modelssupplemented with numerical models
– More claims data
– Better GIS info on risks- locations, construction types, ages, values, contents
Fl d i k d li l d l d d tFlood risk modeling less developed due to greater information needs: GIS elevation data, topographic, hydrology, land use plus localized
Creating Value Through Increased TransparencyProprietary and Confidential
g y gyrainfall data