Severe Convective Storms: A Reinsurer’s Perspective - Tornado … · 2019-09-09 · 14% May Jun...
Transcript of Severe Convective Storms: A Reinsurer’s Perspective - Tornado … · 2019-09-09 · 14% May Jun...
Severe Convective Storms: A Reinsurer’s Perspective
Kirsten Orwig, Ph.D.Atmospheric Perils Specialist
Kenneth SlackSenior Treaty Underwriter
Tornado Summit, March 2 , 2 016
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Severe Convective Storm Risk
• Risk dependent on:
– Frequency and severity of hazard
– Location re lative to hazard
– Structural vulnerability
– Portfolio s tructure and conditions
2
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
0
200
400
600
800
1000
1200
1400
1600
1800
2000
Num
ber o
f Occ
urre
nces
per
Yea
r
F/EF5
F/EF4
F/EF3
F/EF2
F/EF1
F/EF0
US Severe Convective StormsTornado Statis tics
Doppler radar era
12 27
Jan3% Feb
3%Mar6%
Apr14%
May22%Jun
19%
Jul10%
Aug6%
Sep6%
Oct5%
Nov4%
Dec2%
3
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
US Convective StormsHail Statis tics
0
50
100
150
200
250
300
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
Num
ber o
f Day
s
<1 inch 1-2 inches 2-3 inches 3-4 inches 4-5 inches >5 inchesObserved max hail size
Feb2%
Mar6%
Apr13%
May19%
Jun20%
Jul16%
Aug12%
Sep6%
Oct3%
Nov – Jan3%
236
4
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
US Annual Insured Losses
$-
$5
$10
$15
$20
$25
$3019
91
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
Billi
ons
(USD
201
5)
Annual Losses
$ 0 .5 B increase per yearon average
5
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
US Average Annual Insured Loss by State19 9 0 -2 015
Millions (2 0 1 5 USD)
6
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Tornadoes and HailWorldwide Statis tics
0
300
600
900
1200
US Canada
0
300
600
900
1200
SouthAmerica
0
300
600
900
1200
Europe
0
300
600
900
1200
Africa
0
300
600
900
1200
Asia
0
300
600
900
1200
Australia/New Zealand
CombinedTornadoesHailSource : Swiss Re CatNet®.
7
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Worldwide Insured Losses2 014 Swiss Re Sigma Catastrophes
2 015 Sigma Reportcoming soon!
Source: Swiss Re http:/ / media.swissre .com/ documents/ sigma2 _2 0 1 5 _en_final.pdf
8
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Severe Convective Storm Risk ModelingWhy is it so difficult?
9
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Basic Cat Modeling MethodologyThe four box model approach
Hazard Exposure Vulnerability Economics
Where, how often and how strong?
What damage degree? What is covered?
• Property value• Contents• Coverage type• Expected loss
Where, what and characteristics?
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Likelihood of tornado hitting any one point is extremely remote due to localized nature
• Hail swath and tornado path characteris tics
• Hail and tornado his torical records contain many biases
Hazard Considerations
11
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Spatial corre lation of damage
• Large wind speed gradient from outer to inner circulation
• Building orientation
Exposure ConsiderationsDamage re lative to location and time
Lower Damage Zone
Moderate Damage Zone
Peak Damage Zone
Moderate Damage Zone
Lower Damage Zone
Building
1 km
12
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Example: Joplin Tornado Source : US Army Corps of Engineers
13
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• 2 5 0 m wide tornado
• EF2 —winds to 13 5 mph (6 0 m/ s)
• Moving from WSW to ENE
• Debris impact would be low-moderate
Hypothetical Scenario
~265m
14
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• The most vulnerable areas for this s ite :
– upwind-facing windows coinciding with peak pressures on roof
– lee-s ide overhead doors
• If envelope compromised then:
– uplift on roof from inside and out
– significant content damage
– greater like lihood of total loss
windows
peak negative pressures
on roof
overheaddoors
+
+ windows
15
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Exposure ConsiderationsBuilding Contents
Source : NIST (2 0 1 4 ) NIST Technical Inves tigation of the May 2 2 2 0 1 1 Tornado in Joplin, MO
16
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Exposure ConsiderationsBusiness Interruption
Emergency Generator Building
Source : NIST (2 0 1 4 ) NIST Technical Inves tigation of the May 2 2 2 0 1 1 Tornado in Joplin, MO
Source : NOAA"Ten days after …restored power to all customers who were able to receive service ."
• 2 substations , 4 ,0 0 0 dis tribution poles and transmission towers , 1,5 0 0 transformers , and 110 miles of transmission/ dis tribution lines
• 5 0 ce ll towers downed or destroyed• 4 ,0 0 0 leaking water lines , 2 5 broken fire-service lines• 3 ,5 0 0 gas meters and 5 5 ,0 0 0 ft of gas main damaged
17
2016 Tornado Summit | Dr. Kirsten Orwig and Ken SlackSource : Orwig PhD Dissertation,
Examining Strong Winds from a Time-Varying Perspective
• Building codes and s tandards do not include loads from tornado winds
• Stationary vs non-stationary
• Strength of wall-to-roof and wall-to-floor connections
Vulnerability ConsiderationsTornadoes vs Straight-line Winds
Source : NIST report on Joplin tornadohttp:/ / nvlpubs .nis t.gov/ nis tpubs/ NCSTAR/ NIST.NCSTAR.3 .pdf
Modeled Tornadowind speed and direction
Win
d sp
eed
(m/s
)W
ind
Dire
ctio
n
0 6 0 0 1 2 0 0 1 8 0 0 2 4 0 0 3 0 0 0 3 6 0 0Time (s)
0 6 0 0 1 2 0 0 1 8 0 0 2 4 0 0 3 0 0 0 3 6 0 0Time (s)
Measured in Hurricane Katrina
18
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Not only vulnerable in EQ
• FEMA observations:
– Catastrophic collapse in several locations– Some failures from overload on long span roof system– Most failures occurred at wall-to-roof connections
Structural Performance in Tornado WindsTilt-Up Panel Walls
Puddle weld failures common
Source : FEMA (2 0 1 2 ) http:/ / www.fema.gov/ media-library-data/ 2 0 1 3 0 7 2 6 -1 8 2 7 -2 5 0 4 5 -3 8 3 0 / tornado_mat_chapter5 _5 0 8 .pdf
19
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Tilt-Up Concrete Association (TCA) task force report after Joplin
– Initial failures occurred in s tee l jois t roof system
– Panels themselves were very robust (were generally intact despite collapse)
– "Tilt-Up construction played no role in the failure ."
– Recommendations:
o Develop procedure for more predictable collapse performance
o Establish roof system design criteria "that has ultimate failure capacities…similar to over-s trength requirements for certain e lements in the se ismic design…"
o Send recommendations to ICC, FEMA, Steel Deck Institute , Stee l Jois t Institute , and others to develop s tandard codes, procedures , and products
Structural Performance in Tornado WindsTilt-Up Panel Walls
Source : TCA (2 0 1 2 ) http:/ / tilt-up.org/ tilt-uptoday/ wp-content/ uploads/ 2 0 1 2 / 0 3 / TCA-Report-2 0 1 -1 2 -0 1 .pdf
2 0
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Source : NOAA/ NWS
“Missiles”
21
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Highly localized, short-lived phenomena
• Large model domain coupled with high frequency makes probabilis tic modeling computationally intensive
– Most of southern Canadian provinces and entire eastern 2 / 3 of the United States is vulnerable to severe thunderstorms
– Millions of years of s imulation required for s ingle location convergence
• Hail data are based on subjective reports (e .g. pea-, penny-, baseball-s ized), and are mostly along roadways
• Tornado wind measurements are scarce , damage serves as a proxy for wind speed
– Rating based on maximum damage observed, which is often a small fraction of total path area– Under-classification common in rural areas– Original intensity scale was the Fujita scale , Enhanced Fujita scale adopted in 2 0 07 in US and
2 013 in Canada
Severe Convective Storm ModelingChallenges and Obstacles
2 2
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Treaty perspective
– Tropical Cyclones are major driver of loss in United States for national accounts
– Tornado/ Hail exposure drives losses for regional accounts
– Only large outbreaks or severe urban tornadoes penetrate cat treaties
– Cedent loss experience common
• Single risk perspective
– Damage paths of severe thunderstorms are very localized
– Probability of direct s trike by tornado VERY low
– Direct hit by s ingle tornado can result in a large loss , regardless of industry impact
– Highly corre lated losses over small areas
– Cedent loss experience is scarce
Tornado/ Hail RiskReinsurer’s Perspective
RISK PERSPECTIVE MATTERS!
2 3
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Modelling results
• What results to the models produce
• How do we interpret them
• Art over Science?
24
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Subject income US$ 12 5 m +
• State exposure Gulf
• Line of Business 10 0 % Homeowner
Sample company 1
2 5
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Model 1 Model 2
2 5 0 yrs 2 9 ,4 34 ,411 13 9 ,13 6 ,5 0 6
10 0 yrs 21,511,4 8 3 9 2 ,2 0 3 ,74 3
5 0 yrs 16 ,4 3 0 ,8 0 0 76 ,2 9 0 ,5 3 2
2 5 yrs 12 ,8 8 6 ,6 2 0 6 6 ,2 3 6 ,4 9 9
10 yrs 9 ,316 ,34 5 2 8 ,16 2 ,14 6
5 yrs 6 ,9 5 0 ,018 17,5 91,7 0 6
AAL 16 ,5 3 8 ,5 5 3 3 0 ,175 ,0 0 9
• Actual Tornado Hail Loss experience
• 2 015 18 .0 4 m, 5 .47 m, 3 .24 m
• 2 014 16 .01 m, 10 .16 m, 8 .4 9 m
• 2 013 16 .87 m, 10 .01 m, 9 .15 m
• 2 012 2 0 .2 8 m, 5 .3 5 m, 2 .3 6 m
• 2 011 13 .8 0 m, 3 .4 5 m, 2 .5 5 m
• 2 010 3 .64 m, 2 .5 9 m, 1.67 m
• 2 0 0 9 10 .8 0 m, 6 .8 2 m, 4 .3 3 m
• 2 0 0 8 6 .012 m, 4 .24 m 2 .4 4 m
What results do the models produce?
What does a Tornado Hail Exceedance Probability curve actually look like
2 6
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Model 1
• looks very light throughout curve
• 10 yr return period is just 9 .31 m yet company has incurred 9 events exceeding that number in past 8 years
• Model 2
• looks appropriate up to the 10 year period - 10 yr return period is 2 8 .16 m and largest loss in period is 2 0 .2 8 m
• How credible is curve excess of 10 years?? – the 2 5 year return number infers a very s teep increase in expected loss at $ 6 6 .2 3 m
• Much higher losses reflected at higher return periods
Observations
27
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Blend the models?
• Might make the upper return period numbers reflect a reasonable number
• 2 5 0 year return period number would be $ 8 4 .5 0 m
• However - 10 year RP number would be just $ 18 .5 m – still looks light
-------------------------------------------------------------------------------------------
• Ignore model 2 and load model 1 by how much?
• Load the whole curve by 2 5 0 % ?
• 10 year return period loss is $ 2 3 .2 9 m
• 2 5 0 year return period number $ 73 .5 8 m
Possible actions
2 8
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Loading the whole curve produced by model 1 by a factor of 2 5 0 % gives us a viewpoint which seems to match our knowledge of the company and their recent experience.
• Company purchases 16 0 m of vertical protection
– close to a 2 5 0 return period event for Hurricane
– excess of a 2 5 0 return period (no matter what model) event for TH
Conclusion
2 9
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Subject income US$ 275 m +
• State exposure Upper Mid West
• Line of Business 8 5 % Homeowner
10 % Commercial,
5 % Auto
Sample company 2
3 0
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Model 1 Model 2
2 5 0 yrs 94 ,6 8 5 ,0 8 5 14 0 ,074 ,3 97
10 0 yrs 6 6 ,19 2 ,9 6 0 10 5 ,011,78 3
5 0 yrs 4 8 ,131,517 72 ,191,2 6 0
2 5 yrs 3 3 ,17 9 ,415 5 5 ,47 0 ,6 0 9
10 yrs 2 0 ,711,94 8 3 8 ,0 4 9 ,5 5 0
5 yrs 14 ,3 2 2 ,4 57 2 8 ,2 9 5 ,2 8 2
AAL 2 0 ,572 ,3 6 5 2 6 ,16 5 ,19 2
• Actual Tornado Hail Loss experience
• 2 015 12 .3 5 m, 8 .84 m, 8 .0 6 m
• 2 014 24 .9 6 m, 17.9 9 m, 9 .2 2 m
• 2 013 10 .74 m, 6 .61 m, 4 .9 5 m
• 2 012 8 .71 m, 4 .41 m
• 2 011 10 .0 0 m, 4 .12 m
• 2 010 18 .6 5 m, 6 .51 m,
• 2 0 0 9 nil
• 2 0 0 8 11.34 m, 8 .2 9 m,
What results do the models produce?
What does a Tornado Hail Exceedance Probability curve actually look like
31
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Model 1
• Once again lighter through curve
• closer to experience -10 yr return period is 2 0 .7 0 m with only one loss greater than in within the period.
• Model 2
• Appears penal when compared to experience - 10 yr return period is 3 8 .0 4 m and largest loss in period is 24 .9 8 m
• Once again the curve accelerates s teeply - the 2 5 year return number infers an RP loss of 5 5 .4 5 m
Observations
3 2
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Load a model or blend them?
• Blending 5 0 / 5 0 might make sense – or you may wish to load model 1 by 5 0 %
5 0 / 5 0 Model 1 * 5 0 %
2 5 0 117.37 m 14 2 .0 2 m
10 0 8 5 .5 0 m 9 9 .2 8 m
5 0 6 0 .15 m 72 .19 m
2 5 4 4 .3 2 m 4 9 .75 m
10 2 9 .3 8 m 31.0 6 m
5 21.3 0 m 21.4 9 m
Possible actions
3 3
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Either approach produces a return period result up to 10 years which makes sense but blending does provide some re lief to the higher return period loss estimates .
• Company purchases 13 5 m of vertical protection
– just in excess of a 2 5 0 return period event for TH on a blended approach
– Just within the 2 5 0 return period for TH on a loaded approach
In this instance a blended approach may be considered appropriate
Conclusion
34
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Subject income US$ 7 0 0 m
• State exposure Broad Central and Mid West Footprint
• Line of Business 9 0 % Homeowner
2 % Commercial,
8 % Auto
Sample company 3
3 5
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Model 1 Model 2
2 5 0 yrs 278 ,412 ,24 5 3 3 9 ,9 91,24 9
10 0 yrs 2 0 9 ,2 3 5 ,5 2 0 2 51,9 8 6 ,847
5 0 yrs 16 9 ,94 5 ,778 19 6 ,8 37,3 94
2 5 yrs 13 5 ,2 61,74 9 14 4 ,6 61,3 0 8
10 yrs 9 3 ,78 0 ,417 9 3 ,74 5 ,5 6 2
5 yrs 6 6 ,2 5 5 ,19 9 6 5 ,8 6 6 ,2 01
AAL 16 2 ,3 3 6 ,2 6 9 14 3 ,5 8 3 ,3 67
• Actual Tornado Hail Loss experience
• 2 015 12 .3 5 m, 8 .84 m, 8 .0 6 m
• 2 014 2 2 .41 m, 19 .37 m, 16 .19 m
• 2 013 67.3 2 m, 27.9 6 m, 2 2 .8 6 m
• 2 012 41.67 m, 3 6 .9 6 m 2 9 .0 0 m
• 2 011 101.27 m, 3 6 .6 3 m, 3 0 .5 2 m
• 2 010 41.5 6 m, 2 0 .71 m,
• 2 0 0 9 2 9 .6 0 m, 2 3 .17 m, 2 3 .0 9 m
• 2 0 0 8 6 0 .61 m, 17.0 8 m, 16 .5 6 m
What results do the models produce?
What does a Tornado Hail Exceedance Probability curve actually look like
3 6
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Both models far closer together
• Both models reflect reasonable losses vs experience for period
• 5 yr losses of $ 6 5 m and 10 yr loss of $ 9 3 m
• Two 6 0 m losses and one loss Xs 10 0 m within period
• Standard loadings for growth and ALAE should be applied, no need to further manipulate model curves
• Slight difference at high return periods
Observations
37
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
• Overall both models produce credible curves
• You may wish to blend the two to get a consensus view at the higher return periods.
• Selecting one curve over another is justifiable
• The much larger footprint and exposure base produced results far closer in agreement from the two models .
Conclusion
3 8
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack 3 9
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Thank you
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
2016 Tornado Summit | Dr. Kirsten Orwig and Ken Slack
Legal notice
©2016 Swiss Re. All rights reserved. You are not permitted to create any modifications or derivatives of this presentation or to use it for commercial or other public purposes without the prior written permission of Swiss Re.
This presentation is for information purposes only and contains non-binding indications as well as personal judgement. It does not contain any recommendation, advice , solicitation, offer or commitment to effect any transaction or to conclude any legal act. Swiss Re makes no warranties or representations as to this presentation’s accuracy, completeness, timeliness or suitability for a particular purpose. Anyone who interprets and employs this presentation shall do so at his or her own risk without re lying on it in isolation.
In no event shall Swiss Re or any of its affiliates be liable for any loss or damages of any kind, including any direct, indirect or consequential damages, aris ing out of or in connection with the use of this presentation.
41