catastrophic tornado losses then and now: are you prepared? · 1896 st. louis – east st. louis...

1CONFIDENTIAL ©2015 AIR WORLDWIDE

Catastrophic Tornado Losses Then and Now: Are You Prepared?

Eric D. Robinson, PhD – Senior Scientist, AIR Worldwide


The Problem: Severe Thunderstorm Losses are GROWING

CAT Losses 1995-2014**

CAT Losses 2014*

CAT Losses 2015*

ST Losses 1980-2014*

* Munich Re, NatCatSERVICE, PCS

**PCS


- More People- In Bigger Houses- With More Stuff

The Problem: Severe Thunderstorm Losses are GROWING

0

0.2

0.4

0.6

0.8

1

1.2

1950 1960 1970 1980 1990 2000 2010

Valu

e re

l. to

201

5 (2

010)

Year

US Growth Trends

Population

Avg. Square Footage

Disposable Income/CapitaExposure can change

SIGNIFICANTLY over a short period of time!!


- The changing landscape of tornado risk is a multi-faceted problem

The Past ≠ The Future

• Population Growth• Urban Sprawl• Biased Observations• Changing Vulnerability

• Data Accuracy• Hazard Uncertainty• Model Exclusions

• Climate Change• Climate Variability• Exposure Growth


The 3 “Whats” That We Want to Answer…

What can the past tell us about tornado loss potential?

What can we do NOW to better quantify this risk?

What does the future have in store for us?


- Tornado activity exhibits significant decadal variability

- Over the entire observational record, tornadoes have only impacted a small fraction of the country


1980-1990

2000-2010

Torn

ado

Occ

urre

nce

Pro

babi

lity

1950-2014: Only 8% of CONUS affected (assuming NO overlap)

How many years of data is “enough”? 10? 20? 30? 1000?


Even the Data is NOT Without its Own Challenges

0

50

100

150

200

250

300

350

400

0

200

400

600

800

1000

1200

1400

1600

1800

1940 1950 1960 1970 1980 1990 2000 2010

Torn

ado

Cou

nt

Year

Historical Tornado CountsAll Tornadoes

EF3+

0

200

400

600

800

1000

1200

1940 1950 1960 1970 1980 1990 2000 2010

Wid

th (y

d)

Year

Average Tornado Width

All Tornadoes

EF3+

Data Quality (and bias)

Matters!!


Catastrophe Models

Engineering

ScienceClaims

- Claims Data• Exact Losses• Sensitive to changing exposures• Short Experience

- Tornado Observations• Longer time frame (1950-present)• Various Biases (eye-witness based)• Inconsistent reporting procedures

- Engineering• Controlled damage studies, but often

scaled• Typically “pristine” (nature rarely is!)

How Do We Best Leverage the Past Data We DO Have?


The Basic CAT Modeling Framework

ExposureInformation

IntensityCalculation

DamageEstimation

PolicyConditions

LossCalculation

Limit

Deductible

EventGeneration


Stochastic Catalog Allows for Plausible, Yet Unrealized Events

Frequency

StochasticCatalog

Location

Length

Width

Clustering

Wind Speed

Day 1

100K Iterations of Plausible Activity

Day 2 Day 3

Year 1

Year 2

Year 3

Year 100K


CAT Models Translate Hazard into Damage

Day 1

100K Iterations of Plausible Activity

Day 2 Day 3

Year 1

Year 2

Year 3

Year 100K

Intensity

Dam

age

WF+SFH

Intensity

Dam

age

MH

Intensity

Dam

age

SF+COM

Ground Up Loss


CAT Models Translate Damage into Probabilistic Loss Metrics

Ground Up Loss

Policy Terms

Limits

Deductibles

Reinsurance

Uncertainty

Gross Loss

Stochastic Catalog

AAL 50% 20% 10% 5% 2% 1% 0.4% 0.2%

Exceedance Probability Curve

Exceedance Probability

Loss


How do we use information about individual events from the distant past to examine tornado loss potential?

CAT Models can Also Tell Us About Specific Events

1925 Tri-State Tornado1896 St. Louis – East St. Louis

Tornado


Loss Trending is Not Always Straight-Forward

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1945 1955 1965 1975 1985 1995 2005 2015

Valu

e R

elat

ive

to 2

014

Year

US Wealth and Inflation 1947-Present

GNP

CPI

Tangible Wealth

~6.3%/year ~3.3%/year

~4.9%/year St. Louis c. 1875

St. Louis – Present Day

Method Rate/yr Loss 2014Inflation Only 3.3% $ 543,653,610 GNP 4.9% $ 3,558,163,038 Tangible Wealth 6.3% $ 15,976,856,168

*Orig. Loss ~ $12,000,000 -- 1896


We Can Estimate the Loss Directly Using the Model

Intensity

Dam

age

WF+SFH

Intensity

Dam

age

MH

Upon reaching King's Highway the tornado was in a fair way to last on a voyage through the city. The scattering clouds on the edges were rapidly closing in on the central mass, and the screw shape was becoming more pronounced. From the direction it was traveling it seemed, when it passed the Insane Asylum, to be bound for Carondelet, but the strategy and ingenuity that actuated its movements came into play and steered it to a path more productive of loss of property and life.

--The Great Cyclone at St. Louis and East St. Louis


We Can Estimate the Loss Directly Using the Model

Gateway Arch

Eads Bridge

Method Rate/yr Loss 2014Inflation Only 3.3% $ 543,653,610 GNP 4.9% $ 3,558,163,038 Tangible Wealth 6.3% $ 15,976,856,168 Modeled N/A $ 7,256,136,150

*Orig. Loss ~ $12,000,000 -- 1896

Busch Stadium

MO Botanical Garden


So…What Can the Past Tell Us?


• A highly biased view if we are not careful• A probabilistic view of loss if we are careful• A view into how bad things could have been

given today’s exposure


- CAT models can be useful tools, but only when designed and used properly.

- What are some “Best Practices” in designing and using CAT models?

Enough About the Past… What About the Present

I bestow upon you, Eric Robinson’s 3 Commandments of CAT modeling…


- In the modern vernacular: “Garbage in, Garbage out”

My Job:• Is my hazard data de-biased? Representative? Correlated?• Does my stochastic model represent “reality”• Are my benchmark events representative?

Your Job:• Is your exposure data: Complete? Accurate? Coded Properly?• Is you hazard validation representative?

1st Commandment of CAT Modeling: Thou Wilt Not Get Gold from Garbage


- Address any problems with completeness and reasonability of exposure data• Primary characteristics • Geographic information• Policy, layer, and location terms

- Has an exposure review ever been done, verifying exposure data with physical policies?

- Issues with exposure data will flow on to all other elements of the model.

The First Step in Model Usage: Exposure Validation

Exposure


- Sometimes seemingly small differences can be important• Ex:

Proper Exposure Classification is VERY important!!

Auto Auto Dealership

Construction Class: 261 (Auto)Occupancy Class: 300 (Unknown)

Relative Vulnerability: 1.81

Construction Class: 261 (Auto)Occupancy Class: 312 (Retail)



Other Examples of Potential Differences

Exposure Type: General Commercial, Steel FrameOcc: 311, Construction: 151

Low RiseRelative Vulnerability: 1.00 Mid Rise

Relative Vulnerability: 0.59High Rise



- Models contain significant uncertainty (even if not documented)• Footprint location• Intensity• Damage Levels

2nd Commandment of CAT Modeling: Thou Shalt Not Ignore Model Uncertainty (and it’s effects)

Low $ 3,650,808,480 Avg $ 5,783,880,182 High $ 6,773,588,918

Losses for the 1925Tri-State Tornado

EF-Scale Degree of Damage – Single Family Residence


- Models ALWAYS have simplifying assumptions and exclusions• Explicit versus Implicit perils• Excluded perils• Footprint Representation

3rd Commandment of CAT Modeling: Thou Shalt Not Expect Apple Juice from an Orange

Excluded Perils Footprint Representation

Implicit Perils


CAT Modeling Best Practices

What can we do NOW to better quantify this risk?

• Data Quality, Data Quality, Data Quality!!!• Sensitivity/Uncertainty Studies• Know your model (and what it doesn’t include)!


How Will Tornado Risk Change in the Future?

Largely Dependent on 2 Main Factors: Exposure Growth

and Changing Tornado Distributions

0

0.2

0.4

0.6

0.8

1

1.2

1950 1970 1990 2010

Valu

e re

l. to

201

5 (2

010)

Year

US Growth Trends

PopulationAvg. Square FootageDisposable Income/Capita

Disposable Income Per Capita: +3.6%Population +0.7%Avg. Square Footage +1.2%Inflation +3.2%

Average Change Per Year, 2014-2024


What About Changing Tornado Distributions?

1980-1990

2000-2010

Torn

ado

Occ

urre

nce

Pro

babi

lity

Decadal Variability is Regionally Dependent

What drives the change in seasonal tornado activity?


Moisture, Instability and Lift Are Key Ingredients for the Development of Severe Thunderstorms and Tornadoes

Humidity is needed for thunderstorm formation is an important source of updraft

energy

The most unstable conditions occur when warm surface air lies beneath

much colder air aloft

These ingredients can be quantified in the atmosphere by a parameter called:

CAPE


Wind Shear Is a Necessary Ingredient for Development and Evolution of Severe Thunderstorms and Tornadoes

Winds that change direction in height further enhance growth of

storms

Wind speeds increasing vertically result in a tilted

updraft, aiding storm longevity and severity

The measure of how winds change with height is called:

Wind Shear


- Where CAPE, wind shear, and lift come together, there is tornado potential

- In the US, this happens most often in the Southern Great Plains:

• Jet Stream (shear), Warm Southerly Flow (CAPE), and the Dry line (lift)

- Understanding how these “ingredients” change is key to understanding ST activity!

To Understand Tornadoes You Must Understand the Environment


How Will These Parameters Change in the (Near) Future?

El Nino La Nina

Tornado and Hail Environment Frequencies over March, April, and May

Global Climate Signals (ENSO, MJO, PDO, etc) have the potential to influence activity

Large Inter-annual variability makes it difficult to determine significant relationships

0

500

1000

1500

1950 1960 1970 1980 1990 2000 2010

Torn

ado

Cou

nt

Year

Tornado Counts

Torn

ado

Hai

lNote: These are favorable “environments”, which still suffer from the “initiation problem”

NOAA Climate.gov


More Broadly, What About Climate CHANGE?

Decreased Equator-Pole Temperature Gradient

Less Shear

Increased Global Temperature and Moisture

More CAPE

Fewer Storms More Storms

Competing Effects… Which one wins out?

Diffenbaugh et. al, 2013


Recent Research Suggests Increases in ST Environments

BUT Remember… this is only “ENVIRONMENTS”


- Energy delivered can be non-linear with “intensity”• Ex: Hail Diameter

- Damage can be non-linear with energy delivered• Ex: Impact-rated shingles

What Would a Modest Increase in Intensity Mean for Losses?

Change in Average Annual Loss for +10% Intensity

USA 48%OK 35%TX 39%MA 56%

0%

50%

100%

150%

200%

250%

10% 20% 30%

Perc

ent A

AL In

crea

se

Percent Intensity Increase

Percent Change USA AAL


The Future and Beyond…

What does the future have in store for us?

• Even Bigger Houses!• Continued Research on Teleconnections• Possibly More Frequent (and more damaging)

Severe Thunderstorms


The 3 “Whats” We Answered Today

What can the past tell us about tornado loss potential?• A highly biased view if we are not careful• A probabilistic view of loss if we are careful• A view into how bad things could have been

What can we do NOW to better quantify this risk?• Data Quality, Data Quality, Data Quality!!!• Sensitivity/Uncertainty Studies• Know your model!

What does the future have in store for us?• Even Bigger Houses!• Continued Research on Teleconnections• Possibly More Severe Thunderstorms


Thank You!

catastrophic tornado losses then and now: are you prepared? · 1896 st. louis – east st. louis...

Documents