Measuring Vulnerability: An Application to Hurricane Katrina

Transcription

1 CREATE Research Archive Non-published Research Reports Measuring Vulnerability: An Application to Hurricane Katrina Haydar Kurban Howard University, Mika Kato Howard University, Follow this and additional works at: Recommended Citation Kurban, Haydar and Kato, Mika, "Measuring Vulnerability: An Application to Hurricane Katrina" (2009). Non-published Research Reports. Paper This Article is brought to you for free and open access by CREATE Research Archive. It has been accepted for inclusion in Non-published Research Reports by an authorized administrator of CREATE Research Archive. For more information, please contact

2 Measuring Vulnerability: An Application to Hurricane Katrina Haydar Kurban y and Mika Kato z September 9, 2009 Abstract This paper develops an empirical method to quantify and rank vulnerability of population groups in a disaster. Our measure can be used to e ciently allocate resources among the impacted groups. In our model, the degree of vulnerability depends on the nature of the disaster, e.g., its size, duration and scope, and the household s ability to respond and recover. Vulnerability of various socioeconomic groups is computed on the basis of risk-averse public perceptions. Currently, such methods for assessing and ranking vulnerabilities are lacking. Using public and private data related to Hurricane Katrina, we measure vulnerability for 88 impacted counties. Key Words: vulnerability, disaster loss, recovery, homeowners insurance, Hurricane Katrina JEL Codes: Q56, Q54, Q58, D81 1 Introduction In this paper, we develop an empirical method to measure vulnerability of various population groups during a disaster. We propose a vulnerability measure that can predicts how fast individuals or groups in di erent economic status and geographic locations can recover from losses caused by a disaster. Two points are emphasized in our vulnerability measure. First, our measure is based not only on potential We thank Emin Dinlersoz, Je rey Groen, Joseph Persky, and Emily Blank for comments. We also thank Alexis Miller for excellent research assistance. This research was supported by the United States Department of Homeland Security through the Center for Risk and Economic Analysis of Terrorism Events (CREATE) under grant number 2007-ST However, any opinions, ndings, and conclusions or recommendations in this document are those of the authors and do not necessarily re ect views of the United States Department of Homeland Security. y Department of Economics, Howard University; hkurban@howard.edu z Department of Economics, Howard University; mkato@howard.edu 1

3 loss but also on potential ability to recover. Second, our measure is adjusted by the absolete level of post-recovery welfare such that it emphasizes vulnerability of the individuals if their post-recovery welfare falls below the minimum welfare while it discounts vulnerability of the individuals if their post-recovery welfare exceeds the minimum welfare. Our measure implicitly assumes that policy makers are risk averse. We use Hurricane Katrina as case study and assess, using our measure, vulnerability of the 88 impacted counties. Directly or indirectly Hurricane Katrina caused over 1800 deaths and with $81 billion insured losses it became the costliest natural disaster in U.S. history (Ewing et al., 2009). Lack of housing and disruptions in income and employment opportunities prevented evacuees to return. According to the recent studies that investigated the impact of Katrina on the labor market outcomes of evacuees and the factors that a ected the evacuees return decisions (Groen and Polivka, 2008a, 2000b), age, home ownership, and the severity of damage in an evacuee s county of origin were important determinants of whether an evacuee returned. Older residents and homeowners are considered more closely tied to their neighborhoods. Thus homeowners play an important role in re-stabilizing neighborhoods. Compared to the national average, home ownership rate among lower income populations was higher in New Orleans area. According to Census 2000, 47 percent of lower income residents in Orleans Parish were home owners. Other recent studies explored how race and class (Elliot and Pais, 2006), the e ect of the storm on an individual s ties to the local area (Pakson and Rouse, 2008), and "sense of place" (Falk et al., 2006) in uence the decision to return. Since New Orleans had been in decline for some time, it has been predicted that the post Katrina New Orleans will be smaller and di erent as far as industry and population compositions are concerned (Vigdor, 2008). However, without fully accounting for the resiliency of the local economy to disasters, it is hard to predict the pace of recovery. Rose (2004) emphasizes the importance of market prices to guide local economies to cope with the a ects of disasters. Our vulnerability measure quanti es and ranks vulnerability in terms of uninsured losses that may delay the recovery. In addition to the nature of disaster itself, i.e., the size, the duratio, and the scope, the individual s ability to respond to a disaster plays an important role in explaining vulnerability. As Hurricane Katrina revealed, the lower income groups tend to be more vulnerable because they have limited access to private and public assets to respond to a disaster (Alwang, et al., 2001). The lower income groups are also vulnerable during the post-disaster period as the recovery can be delayed due to lack of economic resources and exclusion from social networks (Holzmann and Jorgensen, 2000). Social policy can reduce or eliminate some constraints by allocating resources according to the relative vulnerability of population groups during a disaster. To make the concept of vulnerability operational and useful, a socially accepted minimum has to be agreed upon for each risk and outcome. Following Luers, et al. (2003), our vulnerability measure distinguishes between variability and vulnerability. Vulnerability of the poor partly results from their closeness 2

4 Initial State of Wealth W0 An Event occurs (Hurricane, earthquake, etc.) Recovery Effort (Insurance pay, public and private assistance, etc.) Current State of Wealth W1 Time Immediate Loss, L, is determined Recovery, R, is made Net Loss, L R, is determined W1 = W0 L + R Figure 1: Order of Events to the socially accepted minimum level of well-being. In fact, given the level of risk, lower income groups, ceteris paribus, tend to fall and stay below the minimum level of well-being because of their inability to respond to welfare loss. On the other hand, while higher income groups may experience larger variability in wealth as a result of a disaster, they are less likely to stay below the minimum level of well-being because they have more resources to recover the loss. 2 De ning the Vulnerability Our interest is to create a vulnerability measure that can capture not only the loss caused by an event but also the potential recovery that can be made in the future. Individuals who have ability to recover most of their loss will be considered less vulnerable in our model regardless of the size of their initial loss. 2.1 The Model We consider an individual (e.g., a household) with initial (pre-disaster) wealth W 0. A simple time-line of events which describes the experience of an a ected individual in shown in Figure 1. When an event occurs, some or all of the initial wealth W 0 may be a ected. We assume, for simplicity, that the rate of loss depends only on the magnitude of an event X. The relationship between and X, however, may vary by physical and geographical conditions. In case of hurricane Katrina, for example, the wind speed measures the magnitude of an event X and we found that coastal areas and non-coastal areas have signi cantly di erent loss structures (as shown later in Section 2). Therefore, we rst divide the impacted areas into smaller groups i so that each group should have a similar loss structure. We then de ne the immediate loss L of an individual who belongs to group i as L = i (X) W 0 ; (1) 3

5 where 0 < i < 1 and 0 i > 0 for any X > 0. An a ected individual is assumed to make recovery e ort to the loss L. Among all resources available to an individual, private insurance is the key factor to predict the individual s ability to recover the loss. Typically, homeowners insurance protects an individual s properties from disasters. To receive a higher coverage limit, however, an individual must pay a higher insurance premium with other conditions the same. The other factors, such as weather and landscape that are speci c to the area, also signi cantly a ect the relationship between insurance premium and the coverage limit. According to the National Association of Insurance Commissioners (NAIC), the national average premium for homeowners insurance in 2005 is $764 and the top three most expensive states are Texas ($1,372), Louisiana ($1,144), and Florida ($1,083). Those states are commonly exposed to severe storms and hurricanes. We therefore de ne the coverage limit function by area. The coverage limit C of an individual who lives in area j is C = C j (p) ; (2) where C 0 j > 0. Assume that the amount of insurance premium that an individual is willing to pay is positively correlated with initial wealth W 0 p = p (W 0 ) ; (3) where p 0 > 0. If one has an insurance policy that pays the coverage limit C, the actual loss covered by this insurance is equal to either the coverage limit C or the loss L, whichever the smallest. If one does not have insurance, then we assume that no recovery is made. Assume that the probability that an individual has insurance is and it depends on the wealth W 0. Then the actual amount of recovery R is calculated as R = (W 0 ) minfc; Lg, (4) where 0 > 0. After all recovery is made, the post-recovery wealth W 1 of an individual with initial wealth W 0 in group i and in area j is W 1ij (X; W 0 ) W 0 L + R (5) = W 0 i (X) W 0 + (W 0 ) minfc j (p (W 0 )) ; i (X) W 0 g. 2.2 The Vulnerability Measure Our vulnerability measure ranks the well-being of various population groups impacted by disasters. It computes marginal loss of an individual with initial wealth W 0 in group i and in area j due to an additional stress, i.e., j@w 1ij =@Xj. We also 4

6 introduce a policy coe cient 1= (W 1ij (X; W 0 ) =W ) 1+ to the measure. This emphasizes the vulnerability of an individual whose post-recovery wealth W 1ij falls below the poverty line of wealth W, i.e., W 1ij =W < 1; it, on the other hand, discounts the vulnerability of an individual whose post-recovery wealth still exceeds the poverty line, i.e., W 1ij =W > 1. De ne the vulnerability of an individual in group i and area j 1ij(X;W 0 V ij (X; W 0 ) = i 1+ ; (6) h W1ij (X;W 0 ) where 0 is a policy parameter. A greater implies that a policymaker puts higher priority on those who fall below W. Notice that the vulnerability depends only on the wind speed X and the pre-disaster wealth W 0. Our vulnerability measure can be seen as a gradient of vulnerability that shows the unit change in vulnerability as housing value and the strength of disaster vary. W 3 Estimation In this section, we attempt to actually estimate the vulnerability (6) that is applicable to the areas impacted by Hurricanes Katrina and Rita. This helps us to predict vulnerability of an individual or a group in di erent locations and in di erent economic status if a hurricane with its size X passes by it. We rst estimate functions i (X), p (W 0 ), C i (p), (W 0 ), and W that t to the impacted areas. 1. Loss Rate We estimate the loss rate i (X) of group i in (1). We found that the loss structure of coastal counties is signi cantly di erent from that of non-coastal counties. This is so as damages in coastal counties are not only from strong wind but also from hurricane tidal surge ooding. Therefore, we divide the 88 impacted counties into two groups, 72 non-coastal counties and 16 coastal counties, and estimate the loss rate function for non-coastal counties nc (X) and the loss rate function for coastal counties c (X) separately. First, we estimate the dollar values of the full extent of housing damage in each of the 88 counties by using a special data set compiled by the Department of Housing and Urban Development (HUD). After Hurricane Katrina, the Federal Emergency Management Agency (FEMA) and the Small Business Administration (SBA) inspected the disaster area and identi ed three damage levels, minor, major and severe and then estimated the dollar value of loss. 1 1 The O ce of the Federal Coordinator for Gulf Coast Rebuilding at the Department of Homeland Security, the Federal Emergency Management Agency, the Small Business Admin- 5

7 Table 1: Veri ed losses We summarize, in Table 1, the mean veri ed loss for each damage level by state. We use Census 2000 data to nd the number of houses in each damage level in each county. We compute each county s loss gure as L = P 3 s=1 d sn s ; (7) where s = 1 (minor), 2 (major), 3 (severe) is the damage levels de ned by FEMA, d s is the veri ed loss at damage level s in the state to which the county belongs (as reported in Table 1), and n s is the number of occupied housing units in damage level s in the county reported in Census Next, we estimate the county s pre-disaster initial wealth W 0. We compute the aggregate house value of the county as W 0 = W 0 N; (8) where W 0 is the pre-disaster median house value of the county 2 and N is the number of total occupied housing units reported in Census Note that P 3 s=1 n s N as istration, and the Department of Housing and Urban Development have created a data set to assess the full extent of housing damage due to Hurricanes Katrina, Rita, and Wilma ( FEMA inspectors classi ed the damage levels as minor, major and severe. A subset of FEMA registrants with real property damage applied to the Small Business Administration for loans to repair their property. SBA inspectors then estimated veri ed loss for units assessed by the FEMA inspector to have either major damage or severe damage. We used SBA median veri ed loss tables, FEMA categories and the number of occupied housing units in each category to estimate total loss for 88 counties impacted by Hurricane Katrina. To estimate the aggregate damage level for each county, we multiplied the number of housing units in each damage category by the median veri ed losses. Property owners recovered some of their uninsured losses, up to $10,500 from federal government. FEMA provided assistance at three levels, i.e., less than $5,200, $5,200, and $10,500 for damage not covered by insurance. For less than 50 percent damage, FEMA provided up to $5,200 in repair assistance for damage not covered by insurance. For damage greater than 50 percent, FEMA provided $10,500 in repair assistance for damage not covered by insurance. 2 Ideally, we want to use the mean house value for W 0. However, only the mean house value 6

8 Rate of Loss (L/W0) Noncoastal Coastal Wind Speed (X) Figure 2: Wind speed in mph and loss rate (actual) Figure 3: Wind speed in mph and loss rate (estimated) 7

9 not all occupied housing units were damaged. From Eqs. (7) and (8), the loss rate of a county can be computed by L W 0 = P 3 s=1 d sn s W 0 N : (9) For the magnitude of an event X, we use maximum wind speeds. The U.S. Geological Survey s Hurricane Katrina Wind Speeds has ve maximum wind-speed categories: Category 1 60 mph, Category 2 > 60 mph, Category 3 > 75 mph, Category 4 > 90 mph, and Category 5 > 100 mph. Based on the data, we assess and assign the 88 counties 3 in Alabama, Mississippi and Louisiana impacted by Hurricanes Katrina, Rita and Wilma the median value of the maximum wind speed of the relevant category, i.e., X = 53, 67, 82, 95, and 107 mph. Table 2 in Appendix shows the assignment of wind speed category to each impacted county. 4 According to Kerry (2006), [T]he amount of damage increases roughly as the square of the intensity of the storms, as measured by their maximum wind speed... However, Nordhaus (2006) argues that this presumption is based on an energywind speed relationship, which is not necessarily applicable to the impact of wind and water on designed structures. Nordhaus, therefore, treats the exponent on wind speed as a behavioral parameter to be estimated and estimates a double-log relationship between normalized damage and maximum wind speed. Following Nordhaus (2006), we rst regress ln on ln X. 5 For a group of non-coastal counties (N = 72), we nd that the relationship is ln nc (X) = 16: : ln X; (10) and for a group of coastal counties (N = 16) we nd that the relationship is Taking antilog of Eqs. (10) and (11) gives ln c (X) = 22: : ln X. (11) nc (X) = 9: X 2: (12) of owner-occupied housing units is available while our population N includes both owner-occpied housing units and renter-occupied housing units. We are concered that we may overestimate the mean house value of all units by assuming that the mean house value of owner-occupied hosing units is the same with that of renter-occupied housing units. We instead use the median house value of all, which is available, for W 0. 3 These counties were designated by the Federal Emergency Management Agency as eligible to receive individual and public assistance as of September 14, For Cameron Parish, LA and Vermilion Parish, LA, we use the maximum wind speed of Hurricane Rita, instead of that of Hurricane Katrina, as most of their losses reported in the Federal Emergency Management Agency (FEMA) and the Small Business Administration (SBA) were caused by Harricane Rita. 5 The loss rate in Nordhaus (2006) is de ned as damage divided by GDP. 8

10 Figure 4: Annual premium and coverage limit and c (X) = 2: X 4: (13) Figure 2 shows the actual average loss rates plotted against wind speeds and Figure 3 shows our estimation, Eqs. (12) and (13). We nd that the loss rate is on average higher in coastal counties at any level of wind speed and that marginal loss from an additional wind speed is much higher in coastal counties, and therefore that di erence in loss between the two groups tends to enlarge as wind speed increases. 2. Coverage Limit We next estimate the relationship between the coverage limit C j and the premium pay p in area j in Eq. (2) for our vulnerability measure as the second major component. We attempt to estimate state-speci c coverage limit functions for three impacted states, j = al (Alabama), ms (Mississippi), and la (Louisiana). However, data on coverage limit and premium are available at the national level only, not at the state level. Thus, we take two steps to derive the state-speci c coverage limit functions. We rst estimate the relationship between the coverage limit and the premium pay at the national level. We use data on real insurance coverage and the national premium compiled from National Association of Insurance Commissioners report (NAIC, 2000). The NAIC report is based on the data collected from insurance regulatory o - cials. Since the NAIC gures are based on actual policy forms, they re ect the actual nationwide coverage limit and premiums paid. The estimated national coverage-limit function is C na (p) = :32p. (14) 9

11 As aforementioned, applying the national coverage limit function (14) to an individual state j may be inappropriate. Insurance premium re ects state-speci c factors such as landscape and weather that determine the likelihood of disaster occurrence. Therefore, the insurance premium for the same type of policy can be very di erent across states. We con rm this point in Table 3 in the Appendix. It shows that the average premium of the most commonly written insurance package HO-3 6 varies signi cantly by state and that states constantly a ected by disasters tend to have higher premiums. Denote the national average premium as p na and denote the state j s average premium as p j. We then assume that the state j s homeowners pay p j =p na of the national premium to purchase any amount of coverage limit. From this assumption and the estimated national coverage-limit function (14), we may estimate the state j s coverage limit function as pna C j (p) = :32 p; (15) where the average premium for an HO-3 insurance package of Alabama is p al = 847, that of Mississippi is p ms = 939, and that of Louisiana is p la = 1144, and the nationwide average is p na = 764. Louisiana is the most expensive state and Alabama is the least expensive state. Figure 4 shows the the estimated coverage limit functions of the three states. 3. Insurance Premium We estimate the premium function p (W 0 ) in Eq. (3). By regressing annual homeowners insurance payments on home value data compiled from American Housing Survey s 1999 National Sample, we nd p j p (W 0 ) = 0:0018W :39: (16) Figure 5 shows this relationship. Not surprisingly, homeowners with higher home values are willing to pay a higher annual premium. For every $10,000 increase in home value, annual insurance payments increase by $ Insured Homeowners Rate We estimate the insured homeowners rate (W 0 ) in Eq. (4). The data for this variable, again, come from American Housing Survey s 1999 National Sample. We constructed this variable by dividing the number of owner-occupied housing units % of homeowners insurance are this type. HO-3 is an open perils policy that covers any direct damage to the house or other structures on the property unless it is speci cally excluded. However the coverage for personal property is for named perils only. 10

12 Figure 5: Home value and annual premium with homeowners insurance by the total number of owner-occupied housing units in the sample. In Figure 6, we observe a discontinuity in purchasing pattern of homeowners insurance at $100,000 home value. The rate of insured homeowners for home value less than $100,000 shows diminishing returns while that for home value $100,000 or more is approximately constant. Therefore, it is reasonable to use di erent functional forms to represent these patterns. For home value less than $100,000, the rate of insured homeowners is (W 0 ) = 0:2327W 0: for 0 W 0 < 100; 000; (17) and for home value equal to or more than $100,000, the rate of insured homeowners is = 0:97 for W 0 100; 000. (18) 5. Poverty Line To estimate poverty line of an individual s wealth W in (6), we use below poverty level median house value as proxy for the socially accepted minimum level of well being. According to American Housing Survey s 1999 National Sample the median housing value of those that earned less than below poverty level income was $86,643. W = (19) 11

13 Figure 6: Home value and insured homeowners 6. Policy Emphasis In actual assessment of vulnerability, federal and local emergency management agencies are often required to determine rank priority of various groups of individuals because they have only limited resources to aid the disaster victims. Our policy parameter in Eq. (6) re ects the public interest and consensus on such priority rank. A larger implies that a greater emphasis is put on vulnerability of individuals with the post-recovery wealth below the poverty level, i.e., W 1ij =W < 1, and that a greater discount is imposed on vulnerability of individuals with the post-recovery wealth above the poverty level, i.e., W 1ij =W < 1. We, in this paper, consider a policy that favors the counties with higher poverty rates. This type of policy may be reasonable when a higher poverty rate gives rise to greater negative externalities and therefore higher poverty areas may need more policy attention. We use the below-poverty percentage rate for. Table 2 in Appendix lists for 88 a ected counties in Alabama, Louisiana, and Mississippi. 4 Simulation Using the estimated loss rate functions, Eqs. (12) and (13), the coverage limit function Eq. (15), the premium function (16), the rate of insured homeowners, Eqs. (17) and (18), and the poverty line (19), we can predict vulnerability of an individual in group i (non-coastal or coastal) and in state j (AL, MS, or LA) with a given initial home value W 0 when wind speed X occurs. 12

14 We picked three example counties in Hurricane zone; Mobile, AL ( = 0:185), Orleans Parish, LA ( = 0:279), and Hancock, MS ( = 0:144). They are all coastal counties. The estimated vulnerability is shown in Figures 8 for a given pre-disaster home value W 0 in the range of $20,000-$500,000 and for a wind speed level in the range of 1-6. Several important implications can be derived from the simulation results. 1. Some parts of the graph show nearly zero vulnerability. This means that loss is almost totally covered by insurance. Vulnerability of less wealthy homeowners is, however, slightly higher because the rate of insured homeowners among less wealthy home owners is smaller. In the other parts of the graph, vulnerability is clearly higher. This happens when loss exceeds the coverage limit. Vulnerability in this case tends to increase as wind speed increases. 2. Wealthier homeowners may have larger uninsured losses as the coverage limit to home value ratio, as shown in Figure 7, tends to decline as home value increases. However, loss to wealthier homeowners is more discounted than loss to poorer homeowners. Although the degree of discount depends on policymaker s preference, vulnerability tends to decline as home value increases. 3. In Orleans Parish, LA, the overall vulnerability is greater than Mobile, AL and Hancock, MS. There are two possible reasons: 1. premium to the same type homeowner s insurance in Louisiana is more expensive than premium in Alabama and Mississippi. Thus, homeowners in Louisiana, as Figure 7 shows, tend to be under insured, and 2. Orleans Parish, LA has a much higher belowpoverty percentage rate than Mobile, AL and Hancock, MS. Orleans Parish, LA therefore draws greater attention by policymakers and its loss tend to be exaggerated. Figure 9 shows a amp of 88 counties impacted by Hurricane Katrina. Using our model, we asses the vulnerability of the household of median house value when a hurricane with wind speed 120 mph hits the region. As expected, coastal counties, especially those in Louisiana, are most vulnerable. The uninsured losses of the household with median house value will be between $4,838 and $5,686. The rest, on the other hand, shows high resilience and thus little vulnerability. 5 Conclusions This paper developed a novel approach to quantify and rank the vulnerability in terms of uninsured losses that may delay the recovery. In our model, the degree of vulnerability depends on the nature of the disaster, i.e., its severity, duration and scope, and the households ability to insure against their losses. For the majority of 13

15 Figure 7: Coverage limit to home value 14

16 Figure 8: Estimated vulnerability 15

17 Figure 9: Estimated vulnerability of 88 counties (if X = 5) 16

18 population home ownership is the most important source of wealth. Thus access to home owner insurance and the coverage rate on losses during a disaster become key factors in disaster recovery. Given that natural catastrophes have been occurring with greater frequency and severity in the last decade, this study provides sound analytical and empirical guidance to decision makers regarding the most e ective and e cient way to allocate resources among the cities in order to minimize social and economic vulnerability. Currently, methods are lacking for assessing and ranking vulnerabilities in a systematic and integrated manner. Our model parameters were estimated based on the socioeconomic and loss data compiled from public and private sources. Our study contributes to the vulnerability literature by developing a vulnerability measure that is consistent with utility maximizing consumer behavior and risk-averse public perceptions. As seen during Hurricane Katrina, a lack of such models can lead to tremendous costs and su ering for vulnerable populations and national economy. Our simulation indicates that less wealthy individuals in non-coastal counties are more vulnerable as there is a negative relationship between home values and rate of insured homeowners. The homeowners with home value above $100,000 face less vulnerability because about 97 percent of homeowners purchase insurance. For noncoastal counties we observe overall higher degrees of vulnerability as loss exceeds the coverage limit. We observe that the inverse relationship between vulnerability and home values becomes stronger when local area s poverty level is taken into account. The reason is that while the coverage limit increases as the home value rises, it does not increase as much as the home value. Of the coastal counties impacted, the most vulnerable counties are in Louisiana and the least vulnerable ones are in Alabama. This re ects the fact that Louisiana is the most expensive state and Alabama is the least expensive state for the HO-3 type homeowners insurance among the three states. This also implies that homeowners in Louisiana tend to be underinsured. The empirical method developed and applied to Hurricane Katrina can easily be extended to other types of disasters in the states other than Alabama, Louisiana, and Mississippi. Our model parameters were estimated based on national level information on insurance coverage level, tenure, homeowners insurance premium and whether homeowners have insurance coverage or not. Our empirical methodology can be extended to other areas face Hurricanes or other types of natural disasters. Given information on pre-disaster level wealth, the type of disaster, its impact area and its severity, one can estimate the vulnerability for various population groups in coastal and non-coastal communities. 17

19 References [1] Alwang, J. P. B. Siegel and S.I. Jorgensen Vulnerability: a new from di erent disciplines Social Protection Discussion Paper No. 0115, World Bank. [2] Congressional Research Service Hurricane Katrina: Social-Demographic Characteristics of Impacted Areas. [3] Elliott, James R. and Jeremy Pais Race, Class, and Hurricane Katrina: Social Di erences in Human Responses to Disaster. Social Science Research 35 (2): [4] Emanuel, Kerry Increasing destructiveness of tropical cyclones over the past 30 years. Nature 436: [5] Emanuel, Kerry Anthropogenic E ects on Tropical Cyclone Activity. [6] Ewing, Bradley T., Jamie Brow Kruse, and Daniel Sutter An Overview of Hurricane Katrina and Economic Loss. Journal of Business Valuation and Economic Loss Analysis 4 (2): [7] Falk, William W., Matthew O. Hunt, and Larry L. Hunt Hurricane Katrina and New Orleanians Sense of Place: Return and Reconstitution or Gone with the Wind? Du Bois Review 3: [8] Groen, Je rey A., and Anne E. Polivka. 2008a. Hurricane Katrina Evacuees: Who They Are, Where They Are, and How They Are Faring. Monthly Labor Review 131 (3): [9] Groen, Je rey A. and Anne E. Polivka. 2008b. The E ect of Hurricane Katrina on the Labor Market Outcomes of Evacuees. American Economic Review 98 (2): [10] Holzmann, Robert; Steen Jorgensen Social risk management: A new conceptual framework for social protection, and beyond. World Bank. [11] Liu, Amy, Matt Fellowes, and Mia Mabanta Katrina Index: Tracking Variables of Post-Katrina Recovery. Brookings Institution. [12] Luers, Amy L., David B. Lobell, Leonard S. Sklar, C. Lee Adams, and Pamela A. Matson A method for quantifying vulnerability, applied to the agricultural system of the Yaqui Valley, Mexico. Global Environmental Change 13:

20 [13] Logan, John R The Impact of Katrina: Race and Class in Storm- Damaged Neighborhoods. [14] Louisiana Department of Health and Hospitals Louisiana Health and Population Survey, Survey Report November 28, [15] National Association of Insurance Commissioners Homeowners Insurance Results, NAIC Research Quarterly, Vol. VI, Issue 2: [16] Newberger, Robin, and Michelle Coussens Insurance and Wealth Building among Lower-income Households. Chicago Fed Letter, June 2008: 1-4. [17] Nordhaus, William D The Economics of Hurricanes in the United States. NBER Working Paper No [18] Paxson, Christina and Cecilia E. Rouse Returning to New Orleans after Hurricane Katrina. American Economic Review 98 (2): [19] Pielke Jr., Roger A., and Christopher W. Landsea Normalized Hurricane Damages in the United States. Weather and Forecasting 13: [20] Rose, Adam De ning and measuring economic resilience to disasters. Disaster Prevention and Management 13(4): [21] Smith, Kerry V., Jared C. Carbone, Jaren C. Pope, Daniel G. Hallstrom, and Michael E. Darden Adjusting to Natural Disasters. Journal of Risk and Uncertainty 33: [22] U.S. Census Bureau Special Population Estimates for Impacted Counties in the Gulf Coat Area. Release/www/emergencies/impacted_gulf_estimates.html. [23] U.S. Department of Housing and Urban Development s Of- ce of Policy Development and Research Current Housing Unit Damage Estimates Hurricanes Katrina, Rita and Wilma. [24] Vigdor, Jacob L "The Economic Aftermath of Hurricane Katrina," Journal of Economic Perspectives 22.4 Fall: [25] Viscusi, W. Kip and Patricia Born The Catastrophic E ects of Natural Disasters on Insurance Markets. NBER Working Paper No. W

21 Appendix: Data county coast =1 wind speed occupied units damage ($) below poverty rate county coast =1 wind speed occupied units damage ($) below poverty rate Baldwin County, AL Choctaw County, MS Choctaw County, AL Claiborne County, MS Clarke County, AL Clarke County, MS Greene County, AL Copiah County, MS Hale County, AL Covington County, MS Mobile County, AL Forrest County, MS Pickens County, AL Franklin County, MS Sumter County, AL George County, MS Tuscaloosa County, AL Greene County, MS Washington County, AL Hancock County, MS E Acadia Parish, LA Harrison County, MS E Ascension Parish, LA Hinds County, MS Assumption Parish, LA Jackson County, MS E Calcasieu Parish, LA Jasper County, MS Cameron Parish, LA Jefferson County, MS East Baton Rouge Parish, LA Jefferson Davis County, MS East Feliciana Parish, LA Jones County, MS Iberia Parish, LA Kemper County, MS Iberville Parish, LA Lamar County, MS Jefferson Parish, LA E Lauderdale County, MS Jefferson Davis Parish, LA Lawrence County, MS Lafayette Parish, LA Leake County, MS Lafourche Parish, LA Lincoln County, MS Livingston Parish, LA Lowndes County, MS Orleans Parish, LA E Madison County, MS Plaquemines Parish, LA Marion County, MS Pointe Coupee Parish, LA Neshoba County, MS St. Bernard Parish, LA E Newton County, MS St. Charles Parish, LA Noxubee County, MS St. Helena Parish, LA Oktibbeha County, MS St. James Parish, LA Pearl River County, MS St. John the Baptist Parish, LA Perry County, MS St. Martin Parish, LA Pike County, MS St. Mary Parish, LA Rankin County, MS St. Tammany Parish, LA E Scott County, MS Tangipahoa Parish, LA Simpson County, MS Terrebonne Parish, LA Smith County, MS Vermilion Parish, LA Stone County, MS Washington Parish, LA Walthall County, MS West Baton Rouge Parish, LA Warren County, MS West Feliciana Parish, LA Wayne County, MS Adams County, MS Wilkinson County, MS Amite County, MS Winston County, MS Sustained Wind Speed (PMH) Category: 1 60; 2 > 60; 2 > 75; 3 > 90; 4 > 100 Table 2: 88 counties impacted by hurricane Katrina 20

22 Table 3: Average premium by state 21