Jeffrey Czajkowski Wharton Risk Management and Decision Processes Center

Willis Re Research Fellow

Workshop on Statistical Applications to Climate ExtremesOctober 29, 2012

The Economics of Extremes: Toward an Integrated Management Framework

and Some Research Applications

Traditional Modeling of Catastrophic (Extreme) Risks – Risk Assessment

Source: Grossi and Kunreuther 20052

Hazard

Loss

Exposure

Vulnerability

The Role of Economics

1) Proper (economic) loss accounting: direct, indirect, geographical & time aspect

2) Given 1), empirically estimating key drivers of loss: hazard, exposure, vulnerability.

3) Modeling and understanding decision making in this extreme weather context: rational and “irrational”

4) Implementation and evaluation of appropriate risk management strategies to ultimately reduce risk: information (e.g., forecasting), economic incentives, regulation & standards, insurance, disaster assistance, etc.

3

1) Hurricane Loss Estimation (joint with NCAR)

4

Hurricane Damage = f(wind speed[3-9])• Damage typically direct property losses in impacted coastal

counties- insured and uninsured • When damage normalized, account for wealth (national),

population and/or housing unit (coastal county) increases• Estimates used as a basis to project damages into the future

under climate change scenarios Integrated case study approach to Hurricane Damages:

• For example, Hurricane Ivan (2004) vs. Hurricane Dennis (2005)• Similar landfall maximum wind speed at same general landfall

location, but dissimilar damages - $14.2 vs. $2.2 billion – why?

Defining the Hurricane Impacted Areas

5

Size matters – 228 vs. 21 hurricane force impacted census tracts Location matters – hurricane force impacts not limited to coastal

counties, nor is there 100% impact within in each affected county

Assessing Losses – Access to Data Critical

6

Property losses are not limited to residential structures - personal (65-70%), commercial (25-30%), auto (5-10%) – is this variation accounted for in normalization or estimation?

Property losses are not limited to hurricane impacted areas, nor general landfall states:

For example in Ivan, while 85% of total damage in FL AL, a billion dollars in damage outside of these two states – should we account for this in the modeling?

Is it really all about wind speed?

7

Impacts of Tropical Cyclones not limited to wind: 2) Flood Ratio as Hazard Proxy (Princeton & Iowa)

Hurricane Ivan Rainfall Data Hurricane Ivan Flood Ratio Data

8

Translation of quantified flood magnitudes into economic losses: non-surge NFIP claims

9

1,241 census tracts highlighted in pink had at least 1 NFIP claim:

inland areas match relatively well to the regions with a large flood ratio, in particular along the Appalachian Mountains and in Pennsylvania

A total of 19,273 claims with $800.9 million in flood damages related to inland flooding losses – or two-thirds of the total NFIP flood residential insurance claims and more than half (54 percent) of the total residential flood damage

Census Tracts with no NFIP policies in-force

10

6,940 census tracts highlighted in pink had no 2004 NFIP policies in-force

498 (7%) of these tracts had a flood ratio greater than the 10-year flood peak value – majority in PA, TN, and NC (85% of the 498)

Quantification of Flood Ratio to Loss – number of claims incurred

11

The raw data illustrates an upward trend in the average number of claims per census tract for higher flood ratio values

The preliminary empirical results indicate flood ratio is a statistically significant and positive driver of not only the probability of a claim occurring, but also the number of claims an average tract incurs

Earl (2010) Irene (2011)

12

3) Earl, Irene, and Isaac: a Natural Decision Making Experiment (joint with FSU)

Isaac (2012)

Survey residents (by phone) in areas threatened by hurricanes 3-4 days before the storm arrives

Hope: To understand what drives decisions to invest in protection from storm threats (one way to reduce losses) as they are being made

Isaac – Some preparation good news

Food Water Ice Batteries0

10

20

30

40

50

60

70

80

90

100

Percent with 3-days supply on hand before the threat

However … most short-term preparation tended to that requiring limited effort

Supplies Gas Generator Shutters Furniture Evac Plans No Prep0

10

20

30

40

50

60

70

80

90

Percent Taking Different Preventive Actions

More disturbing: evacuation compliance low even among those who said there was a high

chance of surge flooding (Isaac)

<50% >50%0

10

20

30

40

50

60

70

80

Evacuation Intentions by Judged Likelihood of Storm Surge Flood-ing

YesNo

Why? For all the hype, not much worry

Won't hit Will hit, no danger Will hit, danger Not sure0

10

20

30

40

50

60

Beliefs About Personal Imacts By State

LAMSALFL

17

Preliminary Decision to

Evacuate Multivariate

Results:

Irene & Earl

Variable Full Samples NC MA/NY

Physical Vulnerability

Storm & Forecast Awareness

Perceived Vulnerability

Feeling More Safe (-) (-) (-)

Preparation

Existing window protection (+) (+)

Evacuation Plan (+) (+) (+)

Still need supplies (+)

Homeowners insurance (-)

Previous home mitigation (+)

Evacuation Related

Believe live in evacuation Zone (+) (+) (+)

Aware of local evacuation order

Storm Information

Rely greatly on friends (+)

4) U.S. Hail Risk & Losses

18

From the Storm Prediction Center (SPC) data archive – 203,665 hail storms from 1996 to 2011 – or, approximately 13,500 hail storms per year

According to a CDS/Risk Meter report, 44% of the country is at average hail risk or above (2-3 hailstorms per year on average)

Losses can be significant• U.S. property insurers pay out an average of $1.5 billion each

year for hail-related claims• The Kansas City hail storm on April 10, 2001 was the costliest

hail storm in the U.S. which caused damages of an estimated $2 billion

What can be done to reduce these hail losses?

19

One well-encouraged notion is to vigilantly and vigorously promote and support advanced building codes

Unfortunately, many states - such as Missouri - have no statewide building code in place, i.e., the codes are left up to individual municipalities in that state

Further, even if a statewide code did exist, not all jurisdictions equally enforce their codes once they have been adopted

Building Code Effectiveness Grading Schedule (BCEGS) – lower rating = better enforcement

20

MO Average Zipcode BCEGS Rating Geographic Distribution

BCEGS < = 4

BCEGS > = 5

BCEGS not rated

SPC Hail Observed data with Constructed Buffer

21

22

Variable 2008-2010 2008 2009 2010

Low_BCEGS -0.2034*** -0.13888 -0.277*** -0.24866** Hi_BCEGS -0.05205 0.03687 -0.14803* -0.06542

Adj R2 0.47 0.49 0.47 0.49

Results Summary – Dependent variable natural log of damage per zip code

Better BCEGS ratings do in fact reduce hail damage• “Low” BCEGS ratings are statistically significant estimators of reduced hail

losses in comparison to high and unclassified BCEGS

• Still, it is better in general to have some rating than no rating

Can begin to quantify the value of obtaining an improved rating

“The concept is simple: municipalities with effective, well-enforced codes should demonstrate better (all peril) loss experience” (ISO BCEG Summary)

Traditional Modeling of Catastrophic (Extreme) Risks – Risk Assessment

Source: Grossi and Kunreuther 200523

Hazard

Loss

Exposure

Vulnerability

Comprehensive and Integrated Framework? One-way relationships Non-distinguishable scales

• Spatial• Time

Role of:• Behavior / Risk Perception• Uncertainty

Primarily utilized for industry purposes

Integrated Modeling and Management of Extreme Risks

Source: Morss et al., 2011

24BEFORE DURING AFTER

Dotted arrows and boxes indicate relationships or concepts for which knowledge or data are lacking, uncertainty is high, or key issues need be addressed

Multiple boxes in coping and adaptation represent the need for diverse strategies, flexibility, and possible mid-course adjustments given uncertainty

Over the long term, vulnerability is also influenced by the outcomes from previous weather extremes, and coping and adaptation can influence macroscale drivers

Scientific information is influenced by macroscale drivers, influences coping & adaptation strategies

Thank You – Questions?

For more information on theWharton Risk Management & Decision Processes Center

http://www.wharton.upenn.edu/riskcenter/