Вы находитесь на странице: 1из 25

2/11/2014

2014 Model Comparison

Severe Convective Storms Ohio Valley

Tornado Alley

South East
Salient
1

R i s k

A d v i s o r y

S e r v i c e s

2/11/2014

2014 Model Comparison

Vendor Participation

o Thanks!

Model: v6.1 - June 2008 Platform: Touchstone v 1.5.2 Sims: 10 k years

Model: RQE v14 Aug. 2013 Sims: 300 k years

Model: US SCS Jan. 2014 Platform: RiskLink 13.1 Events: 58 k + high frequency background

Salient

R i s k

A d v i s o r y

S e r v i c e s

2/11/2014

2014 Model Comparison

Comparison Focus & Exposure Portfolios

o 2014 Focus
Frequency Annual Freq. of $1 loss Severity - Expected Loss Location variability Coef. Var. Portfolio Risk - 100 yr. TVar

o 3 Grid Portfolios
~250k grid locations Residential - $ 200 k Bldg.

o 3 Hypothetical Portfolios
Commercial & Residential Varied replacement values Bldg., Contents & BI 300 k locations / $ 60 B TIV - total all 3 portfolios
Salient
R i s k A d v i s o r y S e r v i c e s

Map courtesy of Munich Re

2/11/2014

2014 Model Comparison

Initial Background Questions

US Severe Convective Storm


o Modeling Challenges
What defines an Event? How are models calibrated? What sub-perils are included?

o Risk Management
Given the challenges noted above: How do the models accumulate losses in:
A Single Occurrence? The Annual Aggregate?

o Model Usage
Best practices?
From single location UW to Port. Mgmt.

Whats next?
Any Black Swans?
Salient
R i s k A d v i s o r y S e r v i c e s

Image courtesy of USGS

2/11/2014

2014 Model Comparison


o Annual Frequencies
Of at least $1 loss

Frequency Comparison

o Relative Scale
Left to right - Low to high

o Same
Includes all modeled sub-perils
AIR

o Different!
Expectations Basis Granularity Approach SCALES!
Not possible to map on same scale!

EQE
RMS*

* RMS basis differences


Combined probabilities & different grid sizes Probability of loss causing event in grid AND probability of risk existing in grid
R i s k A d v i s o r y S e r v i c e s

Salient

Maps by Munich Re

2/11/2014

2014 Model Comparison


AL TN GA

Frequency Comparison Southeast & Tornado Alley

AIR

EQE

RMS*

Kansas Oklahoma

Salient

R i s k

A d v i s o r y

S e r v i c e s

2/11/2014

2014 Model Comparison


o Relative Scale
Left to right - Low to high

Ground Up AAL Comparison


o Same
SCALE! Includes all modeled subperils

AIR

o Different!
Granularity Distribution of Expected Losses AIR
Centralized

EQE
Kansas Oklahoma RMS

Population density driven vs. even grid?

EQE
Oklahoma

RMS
43.4 82.0 96.3
EQE AIR RMS

Vertical bands west to east


Maps by Munich Re 10

Ground Up Annual Expected Loss ($ M)

Salient

R i s k

A d v i s o r y

S e r v i c e s

2/11/2014

2014 Model Comparison


AL TN GA

GU AAL ($ m) Comparison South East & Ohio Valley

AIR 23.3

EQE 23.3

RMS 28.6

IN

OH KY

Salient

AIR 22.7 R i s

A d v i s o r y

S e r v i c e s

EQE 24.1

RMS 25.2

11

2/11/2014

2014 Model Comparison

Average Annual Loss General Observations

o General
Totals more similar than distributions of individual location E[L] Greatest $ difference in Tornado Alley

o Frequency vs. AAL


Slightly different distributions / concentrations Implies impact of different peril severities and exposure vulnerabilities
70 60 50 40 30 20 10 0 IN KY OH AL
S e r v i c e s

AIR EQE RMS


10 11 10 5

GU AAL by State ($ m)

58 49 38 33 23 21

11 11 4 5 4

15 4 4 4 3 3 3 5 5 7

GA

MS

TN

KS

OK
12

Salient

R i s k

A d v i s o r y

2/11/2014

2014 Model Comparison

Severe Convective Storms - Sub-Perils

o SCS Sub-perils by Grid


Each model includes components for a variety of sub-perils This years comparison requested location results by sub-perils
Tornado, Hail & Straight Line Wind

o Vendor Submissions
Comparisons difficult More discussions in tomorrows presentations
Tornado Alley
EQE RMS

AIR
All sub-perils

EQE
All sub-perils Tornado Hail

RMS
All sub-perils Tornado Hail SL wind Low Freq. Events

Ohio Valley
EQE RMS

South East
EQE RMS

Hail

Tornado
R i s k

SL wind
A d v i s o r y

Hail
S e r v i c e s

Tornado

SL wind

Hail

Tornado

SL wind 13

Salient

10

2/11/2014

2014 Model Variability of Location E[L] Comparison


E[L] o General
E[L] E[L]

High correlation between location values ($) E[L] and Std. Dev. Selected example using Tornado Alley Same scale used for each vendor (i.e. 1 for E[L], 1 for Coef. Var.) Relative Scale
Left to Right Low to High
AIR EQE Cv RMS Cv

Cv

Kansas Oklahoma

Salient

R i s k

A d v i s o r y

S e r v i c e s

Maps by Munich Re 14

11

2/11/2014

2014 Model Comparison

Loss Accumulation Grid EP curves

o General
SCS peril much more likely to have multiple loss occurrences in a year Risk Metric 100 year Tail Value at Risk (TVaR) (estimated by dots on graph)
300 250 Estimated Losses ($ m)
AIR - AEP EQE - AEP EQE - OEP RMS - AEP RMS - OEP

Ohio Valley Grid All SCS, GU

200
AIR - OEP

150 100 50 0 2 20
R i s k A d v i s o r y S e r v i c e s

Return Period (yrs)

200
15

Salient

12

2/11/2014

2014 Model Comparison

Loss Accumulation Grid EP curves

o General
SCS peril much more likely to have multiple loss occurrences in a year Risk Metric 100 year Tail Value at Risk (TVaR) (estimated by dots on graph)
300 250 Estimated Losses ($ m) 200 150 100 50 0 2 20
R i s k A d v i s o r y S e r v i c e s

Southeast Grid All SCS, GU

AIR - AEP AIR - OEP

EQE - AEP EQE - OEP

RMS - AEP RMS - OEP

Return Period (yrs)

200
16

Salient

13

2/11/2014

2014 Model Comparison

Loss Accumulation Grid EP curves

o General
SCS peril much more likely to have multiple loss occurrences in a year Risk Metric 100 year Tail Value at Risk (TVaR) (estimated by dots on graph)
300 250 Estimated Losses ($ m) 200 150 100 50 0 2 20
R i s k A d v i s o r y S e r v i c e s

Tornado Alley Grid All SCS, GU


AIR - AEP AIR - OEP EQE - AEP EQE - OEP RMS - AEP RMS - OEP

Return Period (yrs)

200
17

Salient

14

2/11/2014

2014 Model Comparison


o General

Allocating Portfolio Risk to Individual Locations

TVaR belongs to a robust and efficient class of risk metrics.


TVaR = Average Loss in the tail of the EP curve beyond a return period

Mapping each grids contribution to the portfolio TVaR is one way to help visualize how locations contributed to tail losses. Contribution TVaR & E[L] can be distributed very differently within any portfolio. Relative Scale
Left to Right Low to High
AIR

EQE

RMS OH

OH
OH IN KY IN KY IN KY

Salient

R i s k

A d v i s o r y

S e r v i c e s

Maps by Munich Re 18

15

2/11/2014

2014 Model Comparison

Next up Impact on Hypothetical Portfolios

o Recap
Frequency Different distributions and granularities E[L] & Variability - Challenges smoothing small footprints, population based grids, event set sizes and sub-peril assumptions Portfolio Risk Multiple occurrences more likely in an give year and each model accumulates loss potential differently

o 3 Hypothetical Portfolios
Commercial & Residential Varied replacement values Bldg., Contents & BI 300 k locations / $ 60 B TIV - total all 3 portfolios Nominal deductibles - $ 1k & $2k options

Salient

R i s k

A d v i s o r y

S e r v i c e s

Map courtesy of Munich Re 19

16

2/11/2014

2014 Model Comparison

Hypothetical Portfolio ALs ($ m)

o Lower Mid West


~73 k locations, $ 8.8 B Bldg. values, $ 11.2 B total replacement cost

o Upper Mid West


~89 k locations, $ 17.8 B Bldg. values, $ 23.5 B total replacement cost

o South East
~155 k locations, $ 19.4 B Bldg. values, $ 26.3 B total replacement cost
45

Lower Mid West


37 33 24

Upper Mid West


AIR 18 6 6 17 19 5 EQE RMS

South East
33 26 24 16 7 19 8

17 11 10

16 9

15

17 17

17

GU

GR_1k

GR_2k

GU

GR_1k

GR_2k

GU

GR_1k

GR_2k

Salient

R i s k

A d v i s o r y

S e r v i c e s

20

17

2/11/2014

2014 Model Comparison


o o GU AAL
Same scale Left to right - Low to high

Lower MW
Look Similar? Geographic Differences
MO OK & N. TX So. TX

Relative Scale

AIR

Sub-Perils
EQE

Different North to South? Terrain Impact?


RMS

Other
Analysis Granularity / smoothing No. of events / simulated years

11

AIR

17
45 Ground Up Annual Expected Loss ($ M)

EQE RMS

Salient

R i s k

A d v i s o r y

S e r v i c e s

Maps by Munich Re 21

18

2/11/2014

2014 Model Comparison


20.0

Lower Mid West AAL Breakdown


17.5

GU AAL by State ($ m)
15.0
10.0 5.0 0.6 1.1 0.0 IA 6.0 5.0 4.0 3.0 2.0 1.0 0.0 Masonry Mobile
A d v i s o r y

AIR EQE RMS 6.4 4.5 2.0 1.0 1.2 KS 4.7 4.0 3.5 6.4 4.1 3.7

15.3

0.8 1.3
MO

2.7

0.7 0.9
NE OK TX

GU Loss Cost (per 1k TIV) by Construction


AIR EQE RMS 4.1

1.7 1.0 0.3 0.5

1.5 0.2

1.4

1.5 0.7 0.1 Steel Unknown 1.0 1.0 1.0

1.7

0.6

Concrete
S e r v i c e s

Wd Frm
22

Salient

R i s k

19

2/11/2014

2014 Model Comparison


o GU AAL
Same scale

Upper MW
Look Similar? Geographic Differences

o Relative Scale
Left to right - Low to high EQE


AIR

Upper WI Upper MI Kentucky SE Ohio

Sub-Perils
RMS

Different North to South? Terrain Impact?

Other
AIR

18
24 Ground Up Annual Expected Loss ($ M)

EQE RMS

Analysis Granularity / smoothing No. of events / simulated years


Maps by Munich Re 23

Salient

R i s k

A d v i s o r y

S e r v i c e s

20

2/11/2014

2014 Model Comparison


8.0 7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0 IL 3.0 2.5 2.0 1.5 1.0 0.5 0.0 Masonry Mobile
A d v i s o r y

Upper Mid West AAL Breakdown


AIR EQE

7.0 5.7 5.0 4.1 3.6

GU AAL by State ($ m)
3.8 3.1 2.5 1.0
1.2 0.5 IN KY MI 1.2 1.1 2.6

RMS

2.1

1.5 1.8 0.6 OH WI

AIR EQE RMS 1.1 0.4 0.5 0.1 0.5

2.0

GU Loss Cost (per 1k TIV) by Construction

1.3 0.9 0.4 0.4 0.1 Concrete


S e r v i c e s

1.2 1.0 0.4

0.5 0.4 0.0 Steel

0.4

Unknown

Wd Frm
24

Salient

R i s k

21

2/11/2014

2014 Model Comparison


o GU AAL
Same scale

South East

o Relative Scale
Left to right - Low to high

Geographic Differences
West of the Appalachians TN, MS & AL Virginia

AIR

Sub-Perils
Different East to West? Terrain Impact?
EQE

Other
Analysis Granularity / smoothing No. of events / simulated years

8 19 33 Ground Up Annual Expected Loss ($ M)

AIR EQE RMS

RMS

Salient

R i s k

A d v i s o r y

S e r v i c e s

Maps by Munich Re 25

22

2/11/2014

2014 Model Comparison


10.0 8.0 6.0 4.0 2.0 0.0 AL 2.0 1.5 1.0 0.5 0.1 0.0 Masonry Mobile
A d v i s o r y

Lower Mid West AAL Breakdown


AIR EQE RMS 5.0 4.8 2.6 4.9

9.0

GU AAL by State ($ m)
6.8

6.1
3.4 2.4 2.3 0.6 GA 1.1

3.5

2.1

1.1

1.3

1.8 0.6

MS

NC

SC

TN

AIR EQE RMS 0.8 0.3 0.5 0.8

GU Loss Cost (per 1k TIV) by Construction


1.6
1.4 1.3 0.8 0.3

0.8

0.5
0.2 0.0 Concrete
S e r v i c e s

0.6 0.4 0.1 Steel Unknown Wd Frm


26

0.3

Salient

R i s k

23

2/11/2014

2014 Model Comparison


o General
The EP curves and AALs tell the same story. The maps show the different distributions of AAL But the underlying assumptions are driving the results:
Smoothing, granularity, number of events, etc.
100 80 60 100 80 60 AIR 40 20 0 2 100 80 60 Estimated Losses ($ m)

Hypothetical Portfolio EP curves

Ohio Valley Hypoth. Port All SCS, GU, AEP


EQE RMS

20 Return Per. (yrs) 200

Tornado Alley Hypoth. Port All SCS, GU, AEP


AIR EQE RMS

Southeast Hypoth. Port All SCS, GU, AEP


AIR EQE RMS

Estimated Losses ($ m)

40 20 0 2 20 Return Per. (yrs) 200


R i s k A d v i s o r y S e r v i c e s

Estimated Losses ($ m)

40 20 0 2 20 Return Per. (yrs) 200 27

Salient

24

2/11/2014

2014 Model Comparison

A big Thank You to


Participating Model Vendors Modeling Teams and Presenters
As well as The Steering Committee & QC Team Special Recognition Mark Bove Mapping Andrew Moore Technical
Salient
R i s k A d v i s o r y S e r v i c e s

28

25

Вам также может понравиться