Mediterranean Earthquake Industry Practice and Use of Multiple Models

1
Mediterranean Earthquake Industry Practice and
Use of Multiple Models
Lucian Chiroiu, PhD
2
INTRODUCTION Mediterranean Seismicity
3
INTRODUCTION Mediterranean Seismicity

• Greece
• Athens High seismic hazard, 45 of
  countrywide exposure
• Turkey
• Istanbul Moderate High seismic hazard, 40
  of countrywide exposure
• Italy
• Milano (and Lombardia) Moderate low seismic
  hazard, 35 of countrywide exposure
• Portugal
• Lisbon Moderate Low seismic hazard, 40
  of countrywide exposure

4
INTRODUCTION Mediterranean Seismicity
Earthquake History Athens September 7th, 1999
earthquake, M 5.9 Istanbul Izmit , August
17th, 1999 earthquake, M 7.4 Italy Friuli, May
6th 1976 earthquake, M 6.5 Lisbon Nov 1st,
1755 earthquake, M 8.5 - 9
Lisbon 1755
5
MARKET PRACTICE

• Cat XL protection is growing every year ( 10b
  )
• All 3 commercial models are available
• What is the tool used for accumulation control
  reinsurance purchase?

RI In general, a fixed PML expressed as of
key zone aggregates Accumulation control same
PML as of key zone aggregates

• No national regulators to fix minimum
  requirements (Solvency II ? )

6
DATA QUALITY
Geographical resolution - Postcodes are
available for all countries (but not in all the
models) - Exposure data available in average 70
- 80 at postcode level, but still 20 30 at
Cresta level only !
Building types and other secondary modifiers -
building types generally available at countrywide
level, based on assumptions (e.g. 70 of SI is
RC, 20 is URM and 10 is STL) - very rarely
more information available (e.g. number of
stories, year of construction)
7
CAT MODELS INPUT
Turkey Model A (detailed, postcode level), Model
B (aggregated, Cresta level), Model C (detailed ,
postcode level but incomplete data)
Greece Model A (detailed, postcode level), Model
B (detailed, county level incomplete), Model C
(detailed , county level incomplete)
Italy Model A (detailed, postcode level), Model
B (detailed, postcode level), Model C (detailed ,
postcode level)
Portugal Model A (detailed, postcode level),
Model B (detailed, county level), Model C
(detailed, postcode level)
8
CAT MODELS INPUT

• Secondary modifiers
• Not always available (year of construction, no of
  stories, etc)
• Standard building types available
• Standard Occupancies available

• Financial conditions
• Not always working properly
• Alternative solutions have to be made for
  modelling first loss policies (e.g. TCIP)

9
CAT MODELS INPUT
Case study TURKEY

• Geographic resolution postcode for Model A (but
  numerical postcodes do not work) , Cresta for
  model B (Istanbul is one Cresta), postcode for
  model C but missing 15 postcodes for Istanbul.

• Secondary modifiers Model A only accept Res, Com
  and Ind LOB Model B cannot accept year of
  construction or number of stories

• Financial conditions Model A and B cannot deal
  correctly with first loss type policies (TCIP).

10
CAT MODELS RESULTS
PORTUGAL
11
CAT MODELS RESULTS
TURKEY
12
CAT MODELS RESULTS
GREECE
13
CAT MODELS RESULTS
ITALY
14
CAT MODELS RESULTS
Which model should we use ??????
15
CAT MODELS COMPARISON

• Comparison of frequencies of synthetic events
  with similar magnitudes
• Comparison of spatial distribution of synthetic
  events
• Comparison of AAL maps with seismic hazard maps
• Comparison of as-if losses for recent earthquakes
• Comparisons of vulnerabilities for different
  modifiers

16
CAT MODELS HAZARD COMPARISON
GREECE - Return period of events by magnitudes

• Model A in line with historic, slightly
  conservative for low RP
• Model B conservative, for high RP
• Model C seems out of line, far too optimistic !

17
CAT MODELS HAZARD COMPARISON

• Spatial distribution of synthetic event sets

TURKEY Plot of epicenters from model A which
produce losses in Istanbul
18
CAT MODELS HAZARD COMPARISON
The Study on A Disaster Prevention / Mitigation
Basic Plan in Istanbul including Seismic
Microzonation in the Republic of Turkey, Istanbul
Metropolitan Municipality
19
CAT MODELS HAZARD COMPARISON
Historical Earthquakes, 32 A.D. 1896
Seismic model for Istanbul Seismic Mitigation
Plan
20
CAT MODELS HAZARD COMPARISON

• validation against hazard maps

• Hazard map

• AAL map of a flat portfolio from Model A

21
CAT MODELS HAZARD COMPARISON

• Annual Average Loss (AAL) comparison for Athens

• Model B

• Model A

22
CAT MODELS VULNERABILITY COMPARISON

• Vulnerability approaches used in commercial
  models
• Intensity based
• Spectral displacement based

Intensity based damage curves
23
CAT MODELS VULNERABILITY COMPARISON
Spectral displacement based damage curves
24
CAT MODELS VULNERABILITY COMPARISON
As-if losses from recent historical earthquakes

• Izmit 1999

• Insured loss estimate as _at_ 1999
• 500 ml (MunichRe) and 800ml (SwissRe)

25
CAT MODELS COMPARISONVULNERABILITY
As-if losses from recent historical earthquakes

• Athens 1999

• Market loss _at_ 1999 150 ml
• Estimated market growth since 1999 2 3 times

26
CAT MODELS COMPARISONVULNERABILITY
Difference in lines of business Residential,
Commercial, Industrial, etc
27
CAT MODELS VULNERABILITY COMPARISON
Difference in building types
28
CONCLUSIONS

• Components of the cat models are sometimes very
  different from scientific consensus

• Not enough efforts are being made to improve
  Mediterranean Earthquake modelling

• Lack of model credibility is resulting in little
  interest to improve data quality

• Decisions are difficult to make based on cat
  model output only