severe convective storm - raaresources.com logic raa 2018... · © 2018 corelogic, inc. [nyse:clgx]...

34
© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary. Severe Convective Storm February 2018 1 Weather Science to Better Manage Risk

Upload: trannhan

Post on 20-Sep-2018

214 views

Category:

Documents


0 download

TRANSCRIPT

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Severe

Convective

Storm

February 2018

1

Weather Science

to Better

Manage Risk

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary. 2

David Smith

Senior Leader,

Science & Analytics

Tom Larsen

Principal,

Industry Solutions

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Large but very sparse footprint

Steep hazard gradients within the footprint

Tornado, hail, and wind components of the

footprint are spatially different

Historic data is incomplete and has trends

– observational and climatological

Wide range of intensities

Insurance definition of an event is an entire

storm system

▪ Multiple days

▪ Wide geographic range

▪ Multiple tornado touchdowns, hail swaths, wind

swaths

Spatial and temporal clustering

SCS Modeling Challenges

3

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Tornado Seasons 2005 – 2018

Year to Year Volatility

4

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Aggregate losses exceeded $25 B in

2011, of which:

▪ 6 events had losses > $1 B

▪ Alabama et al. $7.3 B (southern states,

Ohio Valley)

▪ Missouri et. al. $7.1 B (Ohio Valley &

Upper Midwest)

For Example 2011

Single Years – Even Single Events – Can Be Large

5

Updated Feb 4, 2012

20111 Record(1950-2011)

Tornado Days 179 211 (2000)

Tornadoes 1692 (2nd) 1817 (2004)

Most in single day 200 (1st) (27 Apr)

128 (3 Apr, 1974)

Fatalities 550 (4rd) ~700 (Tri-State, 1925)

Longest Track 132 miles(AL-TN) 235 miles (LA-MS, 1953)

# EF4-EF5 22 (4th) 36 (1974)

# EF5 6 (2nd) 7 (1974)

http://www.spc.noaa.gov/wcm/2011-NOAA-NWS-tornado-facts.pdf1

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Hazard Model

▪ Brand new, comprehensive, stochastic event set for tornado, hail, and straight-line wind

▪ High-resolution modeling based on proprietary radar-based weather forensic algorithms from

CoreLogic®

▪ Environment-Conforming Smoothing

▪ Scenario events

Additional Risk Perspectives Considering the Impact of ENSO

Real-time Event Management

▪ Allows users to import real-time event data into the scenario storm set to access the loss

impacts for actual events

Vulnerability Model

▪ Component-based vulnerability, validated with comprehensive claims and exposure data

▪ Incorporation of CoreLogic property characteristic data and reconstruction cost algorithms

▪ Handling of ACV vs RV policy considerations

Scheduled for Release in Summer 2018, in RQE v.18

U.S. Severe Convective Storm Model Updates

6

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Brand new, comprehensive, stochastic event

set for tornado, hail, and straight-line wind

Tornado, hail, and straight-line wind modeled

individually, but as spatially coupled events,

according to climatology and insurers’ event

definition practices

Realistic, high-resolution hail and wind

footprints and climatology derived from

proprietary radar-based weather forensic

algorithms from CoreLogic

High-resolution tornado footprint modeling

Radar data incorporated to augment the

historical/ observed data for hail and wind

300,000-year simulation provides spatially

smooth and stable results

Stochastic Events

Hazard Model Updates

7

Source: NWS Damage Survey of May 22nd 2011: Joplin Tornado

Hail Size Map – Dallas, TX 3/26/17

Source: Core Logic's Reactor Product

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

SPC (Storm Prediction Center) tornado

path data, hail reports, wind reports,

1950-present

NARR (North American Regional

Reanalysis) daily historical environmental

data at high resolution (32km), 1979-

present

Environment-Conforming Smoothing:

Identifies regions with strong but physical

gradients in storm behavior, while also

sufficiently smoothing in regions with

naturally high variability

Event Frequencies Based on Comprehensive Data and the Latest Science

Hazard Model Updates

8

The model accounts for observational and climatological trends in the historical

data

300,000-year simulation captures spatial and temporal clustering, and extreme

events/ seasons

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Two views in addition to the

standard view in the model:

▪ La Niña phase

▪ El Niño phase

Additional Risk Perspectives: Impact of ENSO

9

Active SCS seasons and historic tornado

outbreaks in 1974, 2008, and 2011

followed La Niña conditions during the

previous winter

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary. 10

N. Great Plains

S. Great Plains

Midwest

Southeast

Tornado HailSeasonality

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Broad set of hail events that

occurred in the last 10 years

Several significant tornado

events:

▪ May 22-27, 2008 - Midwest

Tornadoes and Severe Weather

▪ May 1-12, 2010 - Oklahoma,

Kansas, and Texas Tornadoes

and Severe Weather

▪ April 25-28, 2011 - Southeast/

Ohio Valley/ Midwest Tornadoes

▪ May 22-27, 2011 - Midwest/

Southeast Tornadoes

▪ May 18-23, 2014 - Rockies/

Midwest/ Eastern Severe Weather

Scenario Events

11

Source: 2011 –

Joplin Tornado

event

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Allows users to import real-time

event data into the scenario

storm set to access the loss

impacts for actual events

First roll out for hail, and later

extend to tornado and straight-

line wind

Real-Time Event Management

12

Hail Size Map – Dallas, TX 3/26/17

Source: Core Logic's Reactor Product

Forensic Hail Verification Model

What’s in it for you?

▪ Claims management process

▪ Capital outlays

▪ Quick access to loss impacts on the business

and enable timely reporting to users groups

and personas

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Reactor is an interactive, geospatial

mapping solution that enables users to

visualize and query hail and wind storms

that may impact policyholders.

By geocoding policies using PxPoint™,

our parcel-level geocoding engine,

claims actuaries and managers can:

▪ Visualize and map storm impacts at the

parcel level

▪ Query which policyholders were impacted by

the storm

▪ Analyze storm impacts to estimate Incurred

But Not Reported (IBNR) losses and reserves

▪ Export storm impact data for use in other

workflows

Leveraging Cutting Edge Technology

Reactor™ by CoreLogic

13

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Our proprietary hail and wind data set was developed to more accurately verify a

storm’s impact at the property level for the insurance industry

Up to four times more accurate than Hail Detection Algorithm-based (HDA-based)

products

The Science Powering Hail and Wind Verification

Why Reactor?

14

Benefits

Identify storm impact

Anticipate claim activity and location

Speed up the claims process

Improve catastrophe response

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Competitive

HDA-Based Hail Data

CoreLogic

Hail Verification Data

Not Appropriate for Insurance,

Contractors, Engineers, etc.

Recommended for Insurance,

Contractors, Engineers, etc.

Hail Verification Comparison

15

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Vulnerability Updates

16

Component-based vulnerability, validated with

comprehensive claims and exposure data

Incorporation of CoreLogic property

characteristic data and reconstruction cost

algorithms

Secondary structural characteristics such as

roof profile, roof age, etc., with smart defaults

based on year of construction, locality, etc.,

taking into account building codes and

enforcement as well as construction practices

Handling of ACV vs RV policy considerations

with respect to roof damage

Vulnerability functions for auto – personal,

commercial (heavy and light)

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary. 17

Tornado and Straight-Line Wind VulnerabilityRegional Differentiation

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Component-Based Hail Vulnerability

Roof

ComponentWall

Component

Opening

Component

Brown, T.M., Pogorzelski, W.H., and Giammanco, I.M., 2015:

Evaluating Hail Damage Using Property Insurance Claims

Data. Weather, Climate, And Society, 7, 197-210.

Insurance Institute for Business & Home Safety, 2013: Claims

Analysis Study: May 24, 2011 Hailstorms In Dallas-Fort Worth

18

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Proposed pricing model that can appropriately price risk based on the underwriting

terms and conditions

Vulnerability Model – ACV vs RV

19

ACV – RV Model

ACV for

Roof?

- Cov A

- # of stories

- Roof type

- Roof age

Determine the % of

value of Roof from

Cov A

Depreciation

Damage Calculation

CoreLogic Proprietary

Property Data

Option 1

Option 2

Determine the ACV for Roof

Damage Calculation based

on RV

No/Default - RV

Yes

Damage calculated on

the Roof actual cash

value

Damage = min (value of

damage to roof, Roof

actual cash value)

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Attritional loss vs non-attritional loss available via SQL query

These will be generated via the YLT

Users can define the threshold for attritional loss, and the reports will be

generated to reflect that

▪ Based on industry loss

▪ Based on portfolio loss

Reporting Enhancement

Attritional Loss vs Non-Attritional Loss

20

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

You don’t get the same level of information from aggregated data.

Your model results are only as good as the weakest component.

What do you want to rely on to manage your risk?

Close enough is not good enough

Modeling Severe Convective Storm

21

Fire image from CoreLogic 2017 NHRS Report: https://www.corelogic.com/about-us/news/wildfires-and-hurricane-related-floods-were-most-destructive-natural-hazards-in-2017.aspx

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Aggregate models can

work on average, but

will never be able to

identify the highs and

lows

The model can be

tuned with a specific

portfolio to achieve a

single correct point on

the EP curve, but never

for every portfolio or

every return period

You won’t hit your target with an aggregate model

When Aggregate Data Steers You Wrong

22

Aggregate data never steers you right

Chromatic PDF of the last slide

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Modeled events look like real events

Model output that matches historical losses

Using historic storms to provide insights into risk increase expectations

Model fidelity is a term that describes the faithfulness of the model to the

physical world which it represents.

▪ For a probabilistic risk model, the ability to accurately represent the variability observed in

nature is the foundation of a model that can accurately project what can occur in the future.

Delivering a Fit-For-Purpose Model

Managing SCS Risk with Models

23

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Fort Worth | March 17, 2016

Modeling Events

24

The weather service describes this day like

this:

SPC report:

http://www.spc.noaa.gov/exper/archive/event.php?date=2

0180207

http://www.corelogic.com/about-

us/researchtrends/everything-is-bigger-in-texas-hail-

events-2016.aspx#.Wn88enxryUm

Property counts and losses are for residential

properties of 1 to 4 units

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Plano, Texas | March 23, 2016

Modeling Events

25

http://www.corelogic.com/about-us/researchtrends/everything-is-bigger-in-texas-hail-events-2016.aspx#.Wn88enxryUm

Property counts and losses are for residential properties of 1 to 4 units

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Wylie, Texas | April 11, 2016

Modeling Events

26

http://www.corelogic.com/about-us/researchtrends/everything-is-bigger-in-texas-hail-events-2016.aspx#.Wn88enxryUm

Property counts and losses are for residential properties of 1 to 4 units

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

San Antonio, Texas | April 12, 2016

Modeling Events

27

http://www.corelogic.com/about-us/researchtrends/everything-is-bigger-in-texas-hail-events-2016.aspx#.Wn88enxryUm

Property counts and losses are for residential properties of 1 to 4 units

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

In comparison to hurricane, there are two key differences

▪ Physical scale (images below approximately to scale). Severe Convective storm events

produce damage footprints much more compact than hurricanes

▪ Severe convective events can occur anywhere in the mainland, whereas hurricanes are

constrained to coast

To produce a risk map that shows a rational relationship to risk, many more

simulations are required

A Better Model EP Curve

28

http://www.corelogic.com/about-us/researchtrends/everything-is-bigger-in-texas-hail-events-2016.aspx#.Wn88enxryUm

Wind map from RQE (source: CoreLogic)

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Aggregate Loss exceeded $20 Billion

6 events had losses > $1 Billion (2 were $5+ B)

2011 – What Happened?

29

http://www.spc.noaa.gov/wcm/2011-jan-oct_sm.png

2011 Maximum Observed

Tornado Days 179 211 (2000)

Tornadoes 1700 1817 (2004)

Most in single day 200

(27 Apr)

Was 128 (1974)

Fatalities 551 (3rd) ~700 (1925)

Longest Track 132 miles

(AL-TN)

235 miles

(LA-MS, 1953)

# EF4-EF5 22 (4th) 36 (1974)

# EF5 6 (2nd) 7 (1974)

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

“Modeled Market” Loss Curves

30

1 10 100 1,000

Ind

ust

ry L

oss

Return Period (yr)

2011 was an exceptional year, with several very large events

OEP (worst event in year) and

AEP (sum of all losses in year) are

significantly different, due to very

large overall frequency of events

OEP Return

Period

Annual

Probability

$10 B event 30 – 50 yr 2 – 3 % p.a.

$5 B event 15 – 25 yr 4 – 6% p.a.

AEP Return

Period

Annual

Probability

$25 B season 120 -180 yr ½ – 1 % p.a.

$20 B season 50 – 75 yr 1½ – 2% p.a.

$15 B season 15 – 25 yr 4 – 6 % p.a.

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

A large tornado at any one

location has a very low

probability of occurrence

Working layer probabilities

are summations of the

probability of a extreme loss

somewhere in the portfolio

Larger Simulation Model Sets Enable Better Identification of Hits and Misses

Why Does Size of Simulation Matter?

31

Chromatic PDF

Quantitative Risk Analytics Empower Better Decisions

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

$2 Billion in storm losses▪ Source:

https://www.weather.gov/fwd/mayfest15

What does it take to model

it today?

The Mayfest Hail storm (May 5-6, 1995) | Fort Worth, Texas

A Model that Looks Like Your World

32

Source: CoreLogic Weather Verification

http://www.corelogic.com/solutions/weather-verification-services.aspx

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Losses to homes if storm were

to recur in 2018:

▪ ~$700 Million

▪ (excludes personal automobiles

and all commercial lines)

Interesting Sensitivity Analysis

▪ Exclude all properties built since

1995 to develop proxy portfolio for

1995

▪ Ignoring reconstruction cost

inflation, increased urbanization

increases losses by about 15%

Shown: Zoom in on Fort Worth, showing street maps

Mayfest Hail Storm, May 1995

33

Source: CoreLogic Weather Verification

http://www.corelogic.com/solutions/weather-verification-services.aspx

Analytics from RQE (source: CoreLogic)

What do you do to increase risk awareness in your organization?

Granular data and analytics empowers decision making.

© 2018 CoreLogic, Inc. [NYSE:CLGX] All Rights Reserved. Proprietary.

Hazard Model

▪ A brand new probabilistic model, leveraging historical data for – tornadoes, hail and straight

line winds

▪ Include scenario hail events in the model

Provide additional risk perspectives, considering the impact of ENSO cycles

Real Time Event Management

▪ The feature will allow users to import real time event data in the scenario storm set to access

the loss and damage impacts for an actual event

Vulnerability Model

▪ Complete review and update of the vulnerability model

▪ Allow users to model Actual Cash Value and Replacement Values for roofs to appropriately

price risk based on the underwriting terms and conditions

Scheduled for Release in Summer 2018, in RQE® (Risk Quantification &

Engineering) v.18

US Severe Convective Storm Model Updates

34