Dr. Jeffrey Czajkowski (jczaj@wharton.upenn.edu)

Willis Research Network Autumn Seminar – November 1, 2017

Talk Components

• Wharton Risk Center & Research Context

• TC Flood Research Approach

• Freshwater Flood Main Results

Freshwater flooding risk identification & spatial characterization

Statistical results and their utilization

• Claim Damage Analysis

• Concluding comments

Freshwater Flooding from Tropical cyclones continues to be a significant issue in the U.S.

Hurricane Matthew - October 2016, inland flooding losses represented about half of

the insured losses and three quarters of percent of total loss (insured and uninsured)

including both wind and surge loss estimates

http://www.riskmanagementmonitor.com/tag/hurricane-matthew-insured-losses/

Hurricane Harvey – August 2017, inland flooding nearly all of total insured losses

Region Wind Storm Surge Inland Flood Total

Texas 2 - 3 < 0.1 20 - 35 22 - 38

Louisiana - - 1 - 2 1 - 2

Total 2 - 3 < 0.1 21 - 37 22 - 40

https://www.rms.com/newsroom/press-releases/press-detail/2017-09-09/rms-estimates-hurricane-harvey-insured-losses-from-wind-storm-surge-and-inland-flood-

damage-will-be-between-usd-25-and-35-billion

A risk that has been recently recognized by the Industry

“As the private flood insurance market

continues to take shape, enhanced

flood analytics can be used to gain

important insights into potential flood

events before they occur, particularly

along the coast, allowing (re)insurers to

be more proactive than reactive when

developing, managing and growing a

profitable flood portfolio” Exposure,

RMS.com

Risk Center research on this

emerging issue

http://www.nature.com/articles/srep41609/

8

An Integrated Approach

3) Empirically demonstrate that our approach to quantify flood magnitude is a key driver of the

insured economic losses experienced

2) Combine the quantified flood magnitudes with associated inland flood claim information from the

NFIP database

1) Quantify the spatial extent and magnitude of freshwater flooding related to TCs at the regional

scale

Data from 28 North Atlantic TCs with USGS station located within

500 km from the center of the storm over the period 2001-2014

8,030 NFIP

communities are

impacted on

average per TC

(>1/3 of total NFIP

communities)

Defining the Flood Hazard

• normalized 2-year flood peak ratios

• leverage existing NWS flood classifications & interpolate

=

𝒎𝒂𝒙 𝒇𝒍𝒐𝒐𝒅 𝒑𝒆𝒂𝒌 𝒇𝒓𝒐𝒎 𝟐 𝒅𝒂𝒚𝒔 𝒑𝒓𝒊𝒐𝒓 𝒂𝒏𝒅 𝟕 𝒅𝒂𝒚𝒔 𝒂𝒇𝒕𝒆𝒓 𝒑𝒂𝒔𝒔𝒂𝒈𝒆 𝒐𝒇 𝒔𝒕𝒐𝒓𝒎

𝟐 𝒚𝒆𝒂𝒓 𝒓𝒆𝒕𝒖𝒓𝒏 𝒑𝒆𝒓𝒊𝒐𝒅 (𝒎𝒆𝒅𝒊𝒂𝒏𝒐𝒇 𝟐𝟎+ 𝒚𝒆𝒂𝒓 𝒉𝒊𝒔𝒕𝒐𝒓𝒊𝒄𝒂𝒍 𝒓𝒆𝒄𝒐𝒓𝒅)

Maps showing the flood

ratio values stratified

according to the NWS

flags for Hurricane Irene

(2011; left panel) and

Hurricane Sandy (2012;

right panel).

Relating the flood ratio to flood status classification. This figure shows the relationship between flood ratio values and

flood status classification according to the National Weather Service (Action, Bankfull, Minor, Moderate and Major

Flooding) for all the stations in our domain. In each boxplot, the limits of the whiskers represent the 0.1-0.9 quantiles, the

limits of the box the 0.25 and 0.75 quantiles, while the line in the middle of the box the median (0.5 quantile). The colored

horizontal lines represent the limits of the three categories in which we have stratified the flood ratios: a flood ratio below

1 is for bankfull conditions; a flood ratio between 1 and 1.5 represents minor flooding conditions, while values

between 1.5 and 2.2 and larger than 2.2 are indicative of moderate and major flooding, respectively.

Main outcomes of the research are:

1) the identification of the areas that are more at risk

from freshwater flooding from North Atlantic TCs;

2) the characterization of the extent and magnitude of

these events;

3) the development of statistical models relating flood

magnitude to direct economic losses (number of claims)

controlling for exposure and vulnerability aspects

4) the use of the resulting empirical relationships to perform

sensitivity analysis examining potential impacts of interest

such as climate change precipitation.

~ 10 percent of

the impacts NFIP

communities

have experienced

are minor

flooding or

greater

Over 6,000 NFIP

community

experienced

major flooding

from these 28

TCs

Main outcomes of the research are:

1) the identification of the areas that are more at risk from

freshwater flooding from North Atlantic TCs;

2) the characterization of the extent and magnitude of these

events;

3) the development of statistical models relating flood

magnitude to direct economic losses (number of

claims) controlling for exposure and vulnerability

aspects

4) the use of the resulting empirical relationships to perform

sensitivity analysis examining potential impacts of interest

such as climate change precipitation.

443,484 total residential non-surge claims with a median of 3,818

claims per storm

Total claims are

concentrated in

6,600 NFIP

communities

with at least 1

policy in force

Minor flooding or greater, overlaid with claims

Regression Results

• NWS minor, moderate, and major flood classifications

(based upon 2 yr. RP) are positive and statistically

significant drivers of higher flood claims per impacted

community – robust across a number of model runs

• Control for a number of other relevant factors:

Number of housing units & NFIP policies-in-force

Percent of NFIP community classified as high to low flood risk

Percent of NFIP community impervious surface coverage

TC type – tropical storm, minor hurricane, major hurricane

NFIP community distance to coast

Year and geographical fixed effects

Main outcomes of the research are:

1) the identification of the areas that are more at risk

from freshwater flooding from North Atlantic TCs;

2) the characterization of the extent and magnitude of

these events;

3) the development of statistical models relating flood

magnitude to direct economic losses (number of

claims) controlling for exposure and vulnerability

aspects

4) the use of the resulting empirical relationships to

perform sensitivity analysis examining potential

impacts of interest such as climate change

precipitation.

Percentage increase in the

number of flood claims

1% increase in flood ratio 0.7%

5% increase in flood ratio 3.5%

10% increase in flood ratio 8.6%

20% increase in flood ratio 17.1%

Urbanization increase 2.4%

Hurricane Harvey Flood

Type % of

Communities

Mean

Housing

Units

%

Impervious

Surface

% High

Risk

Zone

Bankfull 80% 4,608 10% 39%

Minor 6% 4,170 9% 28%

Moderate 5% 3,571 9% 37%

Major 9% 15,574 16% 31%

ANALYZING THE DIFFERENCES IN

FRESHWATER AND STORM-SURGE

DOLLAR LOSSES FROM THESE EVENTS

Existing evidence On Flood Insurance Claim Losses

Wharton Risk Center analysis (Kousky and Michel-Kerjan, 2015)

35 years of NFIP claim data, Jan 1978-Dec 2012

Over 1 million single-family residential claims across all 50 states

Years Mean

Paid

Claim

Median

Paid Claim

Number of

paid claims

1980-2012 $34,478 $12,555 1,119,274

2000-2009 $54,506 $21,740 456,255

2005 $91,911 $72,887 177,100

2012 $34,080 $20,000 105,434

http://opim.wharton.upenn.edu/risk/library//ZAlliance-why-individuals-lack-flood-protection_IB2015.pdf

But no distinction between coastal and freshwater flooding

Terminology • Surge claim: caused by storm surge/velocity flow – “tidal water

overflow” – (note, excluded from freshwater claim analysis)

• Freshwater claim: claims associated with overflow from a stream, river, or lake and claims associated with accumulation of rainfall or snowmelt. [these can be in coastal areas]

• Flood zones • A: Special Flood Hazard Area (SFHA) – subject to rising waters

• V: Special Flood Hazard Area (SFHA) - subject to wave velocity

• X: outside of the Special Flood Hazard Area (Non-SFHA)

• Closed vs. open claims

• Paid claims vs. damage (replacement cost)

• Note - we decided to exclude Katrina & many of the claims from Sandy were still open

Summary of Paid Claims & Damage

Paid claims Damage

Number of claims

Surge (36%) (32%)

Freshwater (64%) (68%)

Total 231,724 223,183

Amount

Surge (38%) (49%)

Freshwater (62%) (51%)

Total $6.826B $15.919B

Average amount

Surge $31,320 $108,972

Freshwater $28,417 $53,455

Overall $32,978 $71,330

For Hurricane Harvey, the NFIP has received more than 87,000 claims and incurred more

than $819 million in losses as of September 26. [$9,413 per claim] https://www.verisk.com/press-releases/2017/october/pcs-and-nfip-to-report-aggregate-flood-loss-estimates/

28

Surge claims concentrated in a handful of storms

29

Freshwater claims: inland vs. coastal areas

A: Special Flood Hazard Area (SFHA) – subject to rising waters

V: Special Flood Hazard Area (SFHA) - subject to wave velocity

X: outside of the Special Flood Hazard Area (Non-SFHA)

30

Surge claims: SFHA vs. Non-SFHA

A: Special Flood Hazard Area (SFHA) – subject to rising waters

V: Special Flood Hazard Area (SFHA) - subject to wave velocity

X: outside of the Special Flood Hazard Area (Non-SFHA)

Boxplot of

freshwater depth-

damage ratio by

depth, coastal (C)

and inland (I)

Boxplot of surge

depth-damage

ratio by flood

zone

Depth-damage

ratios with USACE

depth-damage curve

Covariate Surge Freshwater

A zone property (+) ***

V zone property (+) ***

PreFIRM property (+) *** (+) ***

Repetitive Loss property (+)* (+) ***

Elevated structure (-) *** (-) ***

Multi-story structure (-) *** (-) ***

Structure with basement (-) *** (-) ***

Structure replacement cost value (ln) (+) *** (+) ***

Community population (in 100,000s) (+) *** (+) ***

Community pct. Impervious surface (-) *** (-) ***

Flood depth (+) *** (+) ***

Property located 0 to 25 mi. from coast (+) *** Property located 100 to 500 mi. from

coast

Year (+) ***

Gulf coast (+) *** (+) ***

Florida (+) *** (-) ***

Southeast (-) *** (-) ***

Northeast (-) *** (-) ***

33

Key Points:

34

• Freshwater flooding a significant component of U.S. TC risk

In both coastal and inland areas

Roughly 2/3 of the number of claims, and ½ of the resulting damage

• Freshwater vs. Surge Dollar Losses

While average paid claim amounts are comparable, average damage

for surge claims is roughly two times greater

Losses increase by flood depth more significantly for storm surge

claims

• Improved risk assessment needs

Storm surge risk – significant outside V-zones

Depth-damage functions may need more calibration at higher flood

depth levels

Research next steps:

35

• Fine-tune modeling for individual events

• Based on the available rainfall forecasts, development of a

forecasting system that would forecast insurance claims

and losses. The lead time would range between 6 hours

and five days.

Last but not least - what can we do to reduce

the risk?

• Flood Resilient Communities

Measuring resilience in over 100 communities globally

Financial, human, natural, physical, and social capital

• Role of mitigation

Community rating system of the NFIP – flood insurance premium rates are discounted to reflect the reduced flood risk resulting from the community actions

Enhanced building codes – demonstrate the economic effectiveness of statewide wind code in Florida

• Improve communication of the risk - make the risks facing

individuals, firms and communities more transparent and to encourage

them to invest in protective measures prior rather than after a disaster

THANK YOU

HTTPS://RISKCENTER.WHARTON.UPENN.EDU/