0 Introduction to Catastrophe Modeling Claims Club Asia Hong Kong, 23rd May 2013
0
Introduction to
Catastrophe Modeling
Claims Club Asia
Hong Kong, 23rd May 2013
1
Agenda
Basics of Catastrophe Modeling
Cat Modeling Outputs
Uses of Cat Models
Important Considerations
2
Basics of Catastrophe Modeling
3
What are Catastrophes?
Low frequency, extreme events, causes severe losses
Limited historical data
Normally related to natural disasters, such as Typhoon, Earthquake and
Flood
Can also describe concentrated or widespread damage from man-made
disasters, e.g. Terrorism, epidemics
4
11.03.2011 Japan EQ & Tsunami
Source: earthquakejapan2011
Tsunami
Source: Alertnet
Fire
Source: earthquakejapan2011
Landslide
Source: earthquakejapan2011
Flood
Source: earthquakejapan2011
Liquefaction
EQ
Nuclear
Source: Digitalglobe
5
What are Catastrophe Models?
Catastrophe Models (Cat Models) are simulation models based on
– Science of Peril
– Historical and Pre-Historical Data on Magnitude and Frequency of Peril
– Expert Knowledge
– Engineering Knowledge of Damageability
– Historical Loss Data and
– Insurance Policy Terms
For estimating
– Magnitude and Occurrence Rate of losses in events that are likely over a
long simulation period
6
Development of Cat Models
Prior to the advent of cat models, industry’s usual approach was to estimate
the Max % of Total Insured Value in an area that might suffer loss from a
realistic event, either based on past experience or expert’s judgment
Source: Nat Cat Risk Management from (Re) insurers perspectives
(2011) Dr He Hua, 9th Conference on Catastrophe Insurance in Asia.
7
Development of Cat Models (cont’d)
Other shortcomings of traditional ratemaking methods for Cat:
– Assumes Cat activities as “normal”
– Assumes population demographics were stable
– Assumes insured losses by peril were stable
– Does not take into account changes in building/construction codes
The introduction of fully probabilistic models represented a major step
forward by providing a scientific basis for assessing both the frequency and
severity of catastrophe risks.
When introduced, the use of catastrophe models was not widespread. Two
disasters in 1989 (Hurricane Hugo and the Loma Prieta Earthquake) sent a
warning signal to the insurance industry.
Catastrophe models gained rapid acceptance in the insurance and
reinsurance industries after Hurricane Andrew devastated parts of Miami in
1992, causing the largest insured loss experienced worldwide at that time.
Source: Issues in the Regulatory Acceptance of Computer Modeling in Property Insurance
Ratemaking (Rade T Musulin)
8
Catastrophe Model Vendors
AIR – (Applied Insurance Research); Founded in 1987;
Wholly-owned subsidiary of Insurance Services Office, Inc.
(ISO).
EQE – Founded in 1994; Affiliate of ABS Consulting
RMS – (Risk Management Solutions); Founded in 1988.
IF – (Impact Forecasting); Wholly-owned Aon subsidiary.
Non-black box model that can customize damage curves to
unique exposures.
9
Best Practice Modelling Suite
Class Peril Modelling Technique AB
Property
Per Risk Exposure
Quake Catastrophe
Cyclone Catastrophe
Storm Catastrophe
Bushfire Catastrophe
Hail Catastrophe
Flood Catastrophe
Tsunami Deterministic
Other Experience
Terrorism Catastrophe
Liability
Attritional Experience
Large Loss Exposure
Mega Loss Deterministic
Clash Exposure
Cat - Bushfire Catastrophe
PIDO Exposure
Motor
Own Damage Experience
TPPD Experience
CTP Exposure
Mega Loss Exposure
Marine
Cat – Cyclone Catastrophe
Cat – Quake Catastrophe
Attritional – Cargo Experience
Attritional - Hull Experience
Liability Experience
Agriculture
Attritional Experience
Cat - Cyclone Catastrophe
Cat - Fire Experience
Cat - Storm Experience
Global Model Builders
RMS
AIR
EQECAT
Bespoke Models
Aon Benfield Analytics
Impact
Forecasting
Consultants
10
Catastrophe Modelling Components
RISK = f ( hazard, exposure, vulnerability)
Exposure
data
Determine earthquake
motion and wind speed
Hazard Module
Calculate damage
Vulnerability Module
Quantify financial loss
Financial Module EP
Curve
11
Risk Data
Lo
ca
tio
n
Province
County
Postal Code
Street Address
Lat/Long
Att
rib
ute
Occupancy
Construction Type
Building Height
Year Build
Poly terms i.e. Limits, Deductible
Risk
12
Data Flow of Cat Model
Intensity of Hazard at Site
Building
Characteristics
Ground up
Loss
Policy Terms
Gross
Loss
Location of
Site
Parameters That Define Event
Including Magnitude, Physical
Location, Rate
Damage Curve
Items in light blue boxes are
user specified
Local
Conditions at
Site
Event Set
Industry Building
Inventory
Portfolio Vulnerability
Reinsurance
Structure
Net Loss
Ceded
Loss
13
Cat Modeling Outputs
14
Cat Modeling Outputs – Event Loss Tables
An Event Loss Table (ELT) is a table that contains for each event, the event
id, the annual rate of occurrence of the event, the expected loss caused by
the event, the affected exposure, and the uncertainty around the expected
loss as expressed by the standard deviation of the loss
15
Cat Modeling Outputs – The EP Curve
The OEP curve deals with individual
occurrences in a year. It shows the annual
probability that the losses for at least one
occurrence will exceed a certain amount. The
OEP curve is also known as the maximum
loss distribution.
The AEP curve deals with aggregate loss
dollars in a one-year time period. It shows the
probability that aggregate losses in a year
(i.e. the sum of all losses from all occurrences
in a year) will be greater than a certain
amount.
The AAL (Annual Average Loss) is the area
under the AEP curve. It is also known as Pure
Premium.
EP curves are cumulative distributions showing the probability that losses will exceed a certain
amount, from either single or multiple occurrences. These losses are expressed in the
Occurrence Exceedance Probability (OEP) and the Aggregate Exceedance Probability
(AEP) curves.
16
Uses of Catastrophe Models
17
Catastrophe Model Uses
Assess the risk in a portfolio of exposures:
– Total exposure and capital requirements
• Internal ERM, regulatory and rating agencies’
– Underwriting
– Primary pricing
– Aggregate management
– Reinsurance decisions
• Structure and pricing
• Alternative solutions
18
Total Exposure & Capital Requirements
0.00
5.00
10.00
15.00
20.00
25.00
30.00
0 100 200 300 400 500 600 700 800 900 1000
Mo
de
lled
Lo
sse
s (P
AK
Ru
pe
es B
illio
n)
Return Period (Years)
Net Retained - Without Event Limit
Earthquake
Typhoon
Combined Perils
Return Period Earthquake Typhoon Combined Perils
1000 23.12 15.48 24.64
500 17.40 12.85 19.28
250 12.12 10.01 14.66
100 6.06 6.57 9.67
50 2.75 4.10 6.38
20 0.79 1.60 2.82
10 0.22 0.44 1.15
Mean Loss 0.26 0.31 0.57
Std Dev 1.63 1.35 2.11
Modelled SI 406.25 410.83
250 Yr % of SI 3.0% 2.4%
100 Yr % of SI 1.5% 1.6%
Important note: A 1-in-250 Year event ≠ it only happens once
every 250 years. Instead, it means there is a 0.4% chance of the
event happening in any given year.
19
Regulatory/Rating Agency Requirements
Capital Model Return Period/Peril Basis
Australia 1:250 – All Perils Occurrence
Bermuda 1:100 TVaR – All Perils Aggregate
Canada 1:370 - Earthquake Occurrence
Japan
Greater of:
1:250 - Earthquake Occurrence
1:70 - Wind Occurrence
Lloyd’s RDS an 1-in-200 year all risk estimate within the ICA Aggregate
Solvency I None
Solvency II 1:200 – All Perils Aggregate
UK None for ECR; however ICA includes a 1-in-200 year all risk estimate
US None
AM Best BCar
Greater of:
1:250 - Earthquake Occurrence
1:200 - Wind Occurrence
S&P Enhanced 1:250 – All Perils Aggregate
20
Aggregate Management
21
Example Peak Zones 1 in 250 Year Event Loss
42
22
Underwriting
23
Pricing Cat Risk
Total Cat Cost = Net AAL + ( Reinsurance Premium ) + Net Capital Cost
= Net AAL + (Ceded AAL + Reinsurance Margin) + Net Capital Cost
= (Net AAL + Ceded AAL) + Reinsurance Margin + Net Capital Cost
= Gross AAL + Reinsurance Margin + Net Capital Cost
Model Output + Client
Model Miss
Allocation driven by
Ceded AAL
Volatility
Correlation to
industry
Allocation driven by
Volatility
Correlation to
portfolio
24
Reinsurance Design, Pricing & Cost Allocation
XYZ's Non Marine Excess of Loss Program
100m xs 200m
CNY 200m
80m xs 120m
CNY 120m
60m xs 60m
CNY 60m
30m xs 30m
CNY 30m
15m xs 15m
CNY 15m
0
50
100
150
200
250
300
0 200 400 600 800 1,000
Mo
de
lle
d L
os
s (
Millio
ns
)
Return Period (Years)
RMS EQ AIR TY
25
Reinsurance Design, Pricing & Cost Allocation
XYZ's Non Marine Excess of Loss Program
100m xs 200m
CNY 200m
80m xs 120m
CNY 120m
60m xs 60m
CNY 60m
30m xs 30m
CNY 30m
15m xs 15m
CNY 15m
Layer 1 Beijing Guangdong Guangxi Hebei Tianjin
Expected Loss 63 135 15 51 63
Standard Deviation 88.2 189 18.6 71.4 78.12
Layer 3 Beijing Guangdong Guangxi Hebei Tianjin
Expected Loss 110.5 97.75 61.2 75.65 97.75
Standard Deviation 154.7 136.85 75.888 105.91 121.21
Layer 5 Beijing Guangdong Guangxi Hebei Tianjin
Expected Loss 140 101 0 0 123
Standard Deviation 224 131.3 0 0 196.8
26
Important Considerations
27
Important Considerations when Pricing Cat
Cat Models DO NOT PREDICT future Cat losses
Important Considerations
– Model Misses
– Data Quality
– Demographic & Economic Changes
28
Model Misses
• Noticed during benchmarking of
individual events - Difference between
actual and modelled loss
• No cat model is able to fully reproduce a
historical loss
• Models differ from reality
• Distribution of possible damage not a point
loss
• Law of Large Numbers applies – run
enough events and on average the cat
model will converge towards reality
Modeled Loss
Actual Loss
Demand surge
Risk
Concentrations
Model
inaccuracies
-hazard
-vulnerability
Unmodelled Perils
Missing or
undervalued
exposures
Modelling
assumptions
29
Allow for Model Miss – Load Cat Model Results
• Unmodelled perils
• Fire following earthquake (FFE),
Tsunami
• FFE – one vendor advised 20% -
could be a lot higher
• Risk concentrations
• Cat models assume geographic
spread of risk with limited correlation
of loss between risks
• High correlation of loss in tight
clusters however
• Likely to under-estimate
• Demand surge
• Could be as high as 30%
30
Data Quality - Importance of Exposure Data
Enough emphasis cannot be put on the exposure
data; it in the end will determine the quality of the
catastrophe model.
"All discussions of catastrophic exposure
management begin with the accuracy and availability
of the exposure data. The most sophisticated,
complex catastrophe modeling systems cannot
estimate an insurer’s losses if the insurer cannot
identify what insurance coverages have been written
and where those risks are located.”
Source: Measuring and Managing Catastrophe Risk (1995)
Kozlowski &Mathewson, CAS.
31
Data Quality – The Need for Building Attributes
• Occupancy
• Construction
• Building Height
• Building Age • Varying building codes
• Policy Terms • Deductible
• Limit
• Number of Risk
• 。。。
32
Economic and Demographic Changes
Source: The World in 2050, PriceWaterhouseCoopers, 2006
0
5
10
15
20
25
30
US$
Trill
ion
2009
2050
Real GDP (2009 and 2050)
33
Rapid Development – Much of It in Risk Zones
Source: 快速城市化地区土地利用变化的水文响应模拟研究, Jing Zheng 2007
34
Summary: Economic and Demographic Changes
Images to the left are Shanghai Pudong, 1990
vs. today
35
Thank You!
Carole Ho FCAS | Executive Director
Aon Benfield China Limited | Aon Benfield Analytics
t +85228624183 | f +85222438924