Valuing mitigation: Real estate market response to hurricane loss reduction measures.

Smith, Douglas A.

Kevin M. Simmons (*)

Jamie Brown Kruse (+)

Douglas A. Smith (++)

This paper explores valuation of two measures of windstorm mitigation in a Gulf Coast city. Since the home owner is not able to reduce the probability that a hurricane or tropical storm will occur at the structure's location, any voluntary mitigation intended to fortify the home is a form of self-insurance as defined by Ehrlich and Becker (1972). This distinction is important because market insurance and self-insurance are substitutes and thus subject to the standard moral hazard problem. Using a unique Multiple Listing Service data set with detailed information on several hurricane mitigation features, we construct two models to test the influence of mitigation on resale price. The results of the hedonic study indicate that individuals place a positive value on a self-insurance type of mitigation.

1. Introduction

Insurance companies have been hit hard in the last few years by claims from policyholders because of weather-related hazards such as floods and windstorms. Insured losses from Hurricane Andrew approached $16 billion (IIPLR 1995). However, estimates of potential damage for a storm like Andrew increase to $50 billion if the storm had hit Miami directly. These losses have prompted insurance companies, as well as state insurance boards, to explore incentives to property owners to install better protection for their property and demand construction techniques that minimize damage from storms. The Federal Emergency Management Agency (FEMA) has recently initiated programs to encourage voluntary mitigation by individuals as well as communities.

The general consensus by disaster experts has been that home owners will not voluntarily adopt disaster mitigation measures. (1) An early paper by Ehrlich and Becker (1972) (hereafter EB) provides a theoretical basis for evaluating mitigation decisions. The authors distinguish between two forms of mitigation: self-insurance and self-protection. The difference between these two terms is subtle but important. A self-insurance type of mitigation investment reduces the damage from a disaster but does not affect the probability of the disaster. In contrast, a selfprotection type of mitigation investment reduces the probability that the disaster will occur. Previous hedonic studies of household disaster mitigation, including those cited in this paper, have focused on self-protection. The agent is willing to pay more for a home that is located in a less dangerous area with lower associated probability of a hit. EB show that market insurance and self-insurance logically serve as substitutes. When market insurance is offered at an actuarially fair rate, there is no financial incentive to adopt self-insurance. (2) Fronstin and Holtmann (1994, p. 388) provide evidence that supports this hypothesis. By examining the pattern of damage due to Hurricane Andrew, they find that newer homes suffered proportionately more damage than older homes. As an explanation of their findings, they hypothesize that "consumers have substituted homeowners insurance for structurally sound homes that are built to withstand hurricanes." Our study focuses on the value of a self-insurance type of mitigation in a location with a given historical probability of the disaster. The purpose of this study is to examine market price sensitivity of buyers to mitigation on existing homes. This information can assist policymakers in identifying effective incentives and provide necessary information to buyers in terms of the risk inherent with one property versus another. Our goal is to examine resale market price effects of voluntary mitigation measures taken b y home owners to protect their property using market price data from a coastal city that is vulnerable to tropical storms and hurricanes. This paper is organized as follows. A short review of the relevant literature is outlined in the next section. This is followed by a description of our data set and the construction of a wind engineering-based structural integrity index. Next we specify the two models to be tested, one model with storm blinds as the mitigation focus variable and the second a construction index model that uses the structural integrity index. We then report results for both models and finish with conclusions and proposed extensions.

2. Review of Literature

In a classic paper, EB examine a static model of insurance demand using a state preference approach. They consider market insurance and two forms of risk mitigation. Mitigation efforts can reduce the size of the loss or reduce the probability of the loss. The form that they call self-insurance reduces the size of a loss. Reducing the probability of the loss is referred to as self-protection. Some key results are the following:

(i) The incentive to self-insure (loss-reducing mitigation), "unlike the incentive to use market insurance, is smaller for rare losses. The reason is that the loading factor of self-insurance is larger for rare losses because its price, unlike the price of market insurance, can be presumed to be independent of the probability of the loss" (p. 636).

(ii) Market insurance and self-insurance are substitutes and subject to the standard moral hazard problem.

(iii) Market insurance and self-protection (probability-reducing mitigation) are complements as long as the loading factor reflects the change in loss probability generated by mitigation.

In a residential housing market, both forms of risk mitigation described previously are possible. One way to reduce the probability of the loss is to relocate out of harm's way. Previous hedonic studies attempt to measure the value placed on the vulnerability of a location. The location could be within an earthquake special studies zone, in a floodplain, or not. The third of EB's key results applies to this case. Insurance premiums can easily account for location, and therefore market insurance would not necessarily be subject to moral hazard. The National Flood Insurance Program is an example of insurance designed for homes in locations vulnerable to flooding. Previous hedonic studies of natural hazards have concentrated on the self-protection form of mitigation that complements market insurance. In general, insurance premiums do not differentiate between homes with or without the hurricane mitigation measures we study. A notable exception is a hurricane insurance incentive program offered by the Florida Win dstorm Underwriters' Association (FWUA) (2000). (3) Effective July 1, 2000, the FWUA mitigation discount program now offers insurance discounts of 3% to 18% on an itemized list of mitigation features. The cumulative discount can be as much as 60%. To our knowledge, this is the first incentive program of its kind offered by an insurer.

The low-probability/high-consequence nature of many natural hazards, including hurricanes, has been the subject of discussion by many researchers. Kunreuther (1978) suggests that most individuals ignore this type of risk in decision making. Smith (1986) argues that perceived characteristics of hazards, not statistical estimates of risk, determine individual valuations of safety. Kunreuther (1976) concludes that people underestimate the probability of a disaster. An attitude that "it cannot happen to me" governs the actions of many individuals. Kunreuther (1978), Slovic (1987), and Camerer and Kunreuther (1989) find that individuals buy insurance only when a low-probability/high-consequence risk is above some threshold. McDaniels, Kamlet, and Fischer (1992) examine the relationship between risk perception and willingness to pay for increased safety in a survey using five well-defined hazards and five less defined risks. A hazard is "well defined" if death rates are known and the risk is relatively common. McD aniels, Kamlet, and Fischer conclude that when hazards are well defined, personal exposure has greatest influence on willingness to pay to mitigate.

Another approach to ascertaining preferences for safety is to look at purchase and employment decisions with different levels of risk. These studies calculate a "value of statistical life." Conventional and organic food purchases (Hammitt 1990), driving behavior, smoke detector purchases, cigarette purchases (Fisher, Chesnut, and Violette 1989), diet/activity patterns, alcohol use (Lutter, Morral, and Viscusi 1999), and demand for medical care (Viscusi 1994) have been used to determine an implicit value of life. While hurricanes certainly can be life threatening, emergency management policy has dramatically reduced the loss of life from severe storms. Because of this, we view mitigation as a property protection issue rather than a life safety issue.

There are several hedonic studies that examine the value of reducing the probability of loss from a natural hazard (self-protection). The hazard of living in an earthquake zone was the focus of research by Brookshire et al. (1985). Their study collected real estate transaction data from San Francisco and Los Angeles. One of the variables studied was whether the home was located in a Special Studies Zone (SSZ). Properties located in a designated SSZ are considered at higher risk of earthquake damage, and buyers of property located in one of these zones must be notified prior to closing. The authors found that homes located outside an SSZ commanded a premium over homes located inside an SSZ.

Bernknopf, Brookshire, and Thayer (1990) examined the effect of hazard notices for earthquake and volcano activity in the Mammoth Lakes area of California. The U.S. Geological Survey issued hazard notices for the Mammoth Lakes area in May 1980 for earthquake hazard and in May 1982 for a volcano hazard. These notices were removed in August 1984. The question addressed by this study was what effect these notices had on the selling price of residential real estate. To test the significance of the hazard notices, the authors constructed a hedonic pricing model using standard real estate variables and a variable noting knowledge of the notice and the timing of the purchase as it relates to the hazard notices. Both hazard notice variables were significant and indicated that the hazard notice contributed to a reduction in value for homes in that region. The magnitude of the reduction was 8.2% in one model and 11.4% in the second model.

Three studies evaluate the effect of location in a floodplain on the sales price of a home. One approach to mitigating the risk of flooding is purchase a home outside a floodplain. Shilling, Benjamin, and Sirmans (1985) used data from Baton Rouge, Louisiana, and conclude that homes located in a floodplain sell at a discount to homes outside the flood plain. MacDonald, Murdock, and White (1987) conducted a similar study in Monroe, Louisiana, between January and March 1985. This study differs from Shilling, Benjamin, and Sirmans (1985) by using a maximum likelihood estimator rather than ordinary least squares. The results also show a discount for homes located within a floodplain. Speyrer and Ragas (1991) used data from the New Orleans area and applied a spline regression procedure in order to isolate the effect of flood risk from other location variables. Results from the study confirmed earlier studies on flood and earthquake risk: Homes located in a floodplain sell at a discount. The reduction in selling pri ce was remarkably similar among the three studies and indicated that homes located in a floodplain are discounted 6%.

In contrast to the research described previously, our study examines the self-insurance form of mitigation and is unique for that fact. We examine the value placed on mitigation efforts given that the residential structure is in harm's way. We do not expect that a home owner's efforts will affect the probability that a category 3 hurricane (4) will pass over a particular latitude and longitude. (5) Therefore, according to EB, this type of mitigation is a substitute for market insurance and thus subject to crowding out. The county where the subject homes are located is included in a "wind pool" residual insurance market. (6) If anything, the insurance rates are at least actuarially fair or even underpriced because of the cross subsidization by the state wind pool. It follows that if market insurance rates are subsidized, this creates even less incentive for mitigation. In addition, since the incidence of a damaging hurricane over a specific latitude and longitude is a low-probability occurrence, the first of E B's key results also applies. The testable hypothesis that a self-insurance type of mitigation has zero resale market value would be consistent with EB, consistent with the hedonic studies of self-protection cited previously, and consistent with the general consensus of disaster experts.

3. Data

Data for this study were obtained from the Board of Realtors of a Gulf Coast city. (7) There are approximately 1800 observations for single-family homes used in this study. Our source of data for the study is the MLS (Multiple Listing Service) database covering a span of six years (1992-1997). This Gulf Coast region has a long history of damaging and deadly hurricanes. Consequently, residents are aware of the possible adverse effects of a hurricane on their property and their lives. This particular MLS data set is unique in that listing information is not limited to the standard variables of price, location, and square footage. A detailed list of construction features for each listing is contained in a set of more than 50 "feature codes." We will be able to use the feature codes to identify an obvious hurricane mitigation feature: storm blinds. We will also use the feature codes to construct an overall wind resistance rating that depends on construction, location, and design features. Despite the richness of this data set in other respects, one potential limitation is that we do not have information on the year built. We will discuss this limitation and our efforts to compensate for it in the model specification section.

A storm blind is a visible and popular windstorm mitigation measure. Storm blinds protect the home from damage due to flying debris. (8) The integrity of the structure is maintained, to a large degree, if no openings through the exterior walls occur. For example, once a broken window breaches the building envelope, increased pressure on the roof and walls leads to further structural damage. Additionally, once the building envelope fails, the contents of the home are exposed to the forces of wind and water. Typically, content losses far exceed the structural damage once the envelope is penetrated. Storm blinds protect the glass areas that are most vulnerable to penetration by windblown debris. This mitigation measure can be added to a structure at any time and thus can represent retrofit mitigation.

In this study, we also use the available structural information to construct a measure of storm resistance for each dwelling called a Structural Integrity index. The index value assesses the likelihood that the home may survive a severe windstorm. The formula used in the construction of the index was derived by the Wind Engineering Research Center at Texas Tech University. The functional form reflects the complementarity of a portfolio of protective measures and is nonlinear. The procedure has been widely used in the assessment of storm survivability in previous engineering studies (Mehta and Cheshire 1993; Mehta, Cheshire, and McDonald 1991, 1992; Mehta, Smith, and Cheshire 1993). Homes with higher index values are more likely to survive a storm. The structural integrity index captures mitigation measures that are incorporated in the basic design and construction of the home. Within the insurance industry, there has been interest in creating premium incentives for homes that meet certain engineering requirem ents. For homes that meet these requirements, a "seal of approval" could be issued by the insurance industry (IIPLR 1995). This document would ensure future owners that the home contains construction features that, although unobservable, mitigate windstorm damage. Many mitigation features are not visible once construction is complete, and other construction features may be aesthetically pleasing but increase vulnerability. Consequently, there may be conflicting price effects. Home builders are concerned about buyer response to the increased cost of such fortified homes. This study will help address concerns regarding the market response to more wind-resistant homes.

4. Structural Integrity Index

The Structural Integrity Index (hereafter SII) used in this study grades structures based on topographical location, structural characteristics, and architectural features. A high value indicates that the structure is likely to be wind resistant, whereas a lower number indicates a more vulnerable structure. The knowledge base in the expert system was developed by Kishor Mehta, James McDonald, Douglas Smith, (9) and others using more than 26 years of wind damage documentation. A similar system was developed for the Insurance Institute for Property Loss Reduction (IIPLR; now the Institute for Business and Home Safety, IBHS). (10) The expert system starts with structural characteristics that affect building performance in a windstorm recognizing the complementarities that exist. The system then considers the interaction of the building with environmental factors, such as exposure and topography. The parameters, factors, and the choices for each factor (in parentheses) are shown in Table 1. The environmental para meter characterizes the severe wind environment for the structure. The Frame, Roof, Wall, and other parameters are used to define the wind resistance of each of these components.

For the case of storm blinds, the positive relationship between construction cost and windstorm loss reduction is straightforward. However, for the case of the SII, mitigation features that increase wind resistance can require more, less, or no additional construction investment. Choices of roof geometry certainly do affect construction cost, but the relationship between construction cost and windstorm survivability for different geometries is not consistent. For example, multiple gables are more expensive and less wind resistant than other designs. In addition, skylights and large expanses of glass are costlier to construct and make the structure more vulnerable. Engineers would recognize that a structure built with a positive anchorage system to connect elements of the vertical load to the roof and foundation to form a strong, well-defined, continuous load path are less likely to be structurally damaged in a windstorm. However, the typical home buyer may not know how to evaluate the structure's anchorage sy stem.

The SII of each observation ranged from .99 to 1.81. A positive relationship between the SII and resale value will exist only if potential buyers can discern and care about design, location, and construction features that make the structure more stormworthy. Skylights, large glass areas, and proximity to the water may be aesthetically pleasing and increase market value, but at the same time, they make the structure more vulnerable to wind damage. In addition, the availability of market insurance may crowd Out mitigation as argued by EB. Consequently, we cannot predict, a priori, the relationship between resale price and the SII.

5. Model Specification

We estimate two models: the retrofit (storm blind) model and the construction (SII) model. (11) See Table 2 for definitions of variables and descriptive statistics. Similar to other real estate studies, each model uses a semilog specification.

ln SP = [[beta].sub.0] + [[beta].sub.1](FOCUS) + [[beta].sub.2](ln SF) + [[beta].sub.3](BATH) + [[beta].sub.4](POOL/SPA) + [[beta].sub.5](AC)

+ [[beta].sub.6](HEAT) + [[beta].sub.7](FIREPLACE) + [[beta].sub.8](VIEW)

+ (Vector of NEIGHBORHOOD dummy variab1es)[eta]

+ (Vector of YEAR dummy variables)[omega] + Error.

The dependent variable is the natural log of selling price. Independent variables include traditional real estate modeling variables, such as the natural log of square feet (SF), the number of bathrooms (BATH), dummy variables for pool or spa (POOL/SPA), air conditioning (AC), heating (HEAT), and fireplace (FIRE). Additional variables designate the view (VIEW), neighborhood (NEIGHBORHOOD), and year sold (YEAR). The city is located on a barrier island, which is 1 to 2 miles in width and 25 miles long. Much of the island remains undeveloped or used for nonresidential purposes. Proximity to shopping and other services is about the same for all homes. In this community, there is only one public school district. Further, the school district is limited to one high school; thus, any intradistrict differences would not be expected to affect the results. The focus variable for our retrofit model is a binary variable indicating the presence or absence of storm blinds. Homes that have storm blinds are assigned a value o f 1, and homes that do not have storm blinds are assigned a value of 0. As mentioned earlier, the focus variable for the construction model is a summary index value indicating the stormworthiness of the structure (SII)

The year the home was built was not available. Fronstin and Holtmann (1994) found that subdivisions in southeastern Florida with an average year built of 1980 or later were more likely to sustain damage from Hurricane Andrew than older homes. They also used the average assessed valuation of a subdivision as a proxy for quality and found that subdivisions with a higher average assessed value were less likely to be damaged by the hurricane. Without information on the year built, there is a potential bias in our estimates due to unobserved heterogeneity. The neighborhood variable should, however, pick up both age and some measure of quality. There are 19 neighborhoods with most of the houses within a neighborhood built at about the same time. The Board of Realtors assigns to each listing a code to designate the neighborhood. For most neighborhoods, this reflects the subdivision. As such, because of zoning, deed restrictions, and the development patterns of the community, homes within a neighborhood are similar i n size, price range, and age. Visual inspection of the area confirmed that individual neighborhoods within this community are relatively homogeneous. Still, we interpret the results with the caveat that bias due to unobserved heterogeneity may remain.

To check whether our estimates were sensitive to changes in the specification of the model, we estimated several regression models using different subsets of the variables. One regression eliminated the view (VIEW) variables. These variables were location specific regarding amenities such as beachside property, view of the harbor, view of a golf course, and so on. The changes in the value of the remaining coefficients were minimal. Another model eliminated the view (VIEW) and the pool/spa (POOL/SPA) variables. These variables noted the presence of a swimming pool, outdoor spa, pool/spa combination, indoor whirlpool, sauna, and so on. Again, changes in the value of the coefficients were minimal. (12)

Two other issues regarding certain observations should be discussed. First, 283 of the observations were missing data on total square feet. We used a procedure to recover as many observations as possible. Of the observations that did not report total square feet, most did report the size (dimensions) of each room in the home. To overcome this problem, we ran a simple regression of square feet to the sum of room size for observations that had both square feet and room sizes. This gave us an estimated relationship to infer square footage for observations where room sizes were available. In this way, 205 of the missing observations were recovered. (13) Another issue pertains to 380 homes located in a historical district. The rules of the historical district prohibited home owners from installing new storm blinds to their property. Property owners in these areas were limited to types of storm blinds that were available when the home was built. Most storm blinds currently on the market are constructed with differe nt materials and design and as a result cannot be purchased for these homes. These observations were therefore omitted. The final data set contained 1396 observations.

6. Results

The results of both models indicate that mitigation, both retrofit (storm blinds) and construction (SII), are significant to the resale price of the home. As can be seen in Table 3, all structural variables in the retrofit model are highly significant. The coefficient on storm blinds implies that for an average-priced home of about $80,000, the presence of storm blinds adds more than 5% to the selling price. This is remarkable in that the storm blinds are almost fully valued by potential buyers. According to local dealers, for the average home in our study, installing storm blinds would cost about $4200. This implies that the investment in storm blinds will pay for itself at resale, under very modest assumptions on discounting and depreciation. At least for this community, which has had a long history of hurricanes, storm blinds are a cost-effective mitigation investment in terms of increased protection for the home as well as increased resale value of the home. One could argue that the value placed on storm blinds by the hedonic model captures the present value of the stream of insurance premium savings. However, this could be the case only as an expectation because existing insurance premiums do not provide a discount for storm shutters.

Other variables performed as expected. For example, the signs of features such as air conditioning, heating, and number of baths were positive and highly significant. The signs and significance of the various neighborhoods also performed as expected. Highly desirable neighborhoods had positive signs, and undesirable neighborhoods had negative signs. Coefficients on each year captured general price changes and indicated a gradual rising price trend, with the exception of the change from 1992 to 1993, which was negative. (14)

The structural integrity index is the focus variable of the construction model shown in Table 4. Like the retrofit model, the coefficient on the mitigation focus variable is highly significant and positive. This coefficient is harder to interpret. The complex relationship between the site, the wind environment, design features, and construction of the home cannot be linked directly to the cost of construction. Clearly, a costly structure need not be wind resistant. However, construction and design choices that do contribute to a stronger, more wind resistant home are positively valued by the marketplace. To the extent that buyers are aware of improved structural characteristics, they are willing to pay more for the property.

7. Conclusions

Our results demonstrate that individuals place a positive value on a self-insurance type of mitigation. The availability of (possibly subsidized) market insurance did not crowd out this form of mitigation. One could argue that the positive resale market value of mitigation is a form of loading factor adjustment to make self-insurance mitigation equivalent to market insurance. Thus, the insurance market has not necessarily failed; rather, we are seeing an adjustment that is outside the scope of the EB model. Uninsurable or other intangible losses that are also outside the scope of the EB model may have contributed to our result, too. Our study area has a relatively high annual probability of experiencing a landfalling hurricane (8%) and historical experience with a devastating hurricane. (15) An interesting extension of this study is to examine the value placed on storm blinds as the building site is moved inland. Will the value of mitigation decrease as the perceived risk declines? Simmons and Kruse (2000) fi nd that the coefficient on storm blinds is not significant for inland locations. With the implementation of the FWUA mitigation incentive insurance program, an interesting extension would be to examine the effect of insurance premium discounts on the resale market value of homes with hurricane mitigation features. (16) This is the direction of future research.

In some locations of the United States, hurricanes pose a significant threat to residents. Not only is preparing for the possibility of a natural disaster the responsibility of emergency management agencies and insurance companies, but individuals must protect themselves and their property through voluntary mitigation. This study finds that mitigation measures that are obvious to a potential buyer increase the resale value of a home. In addition, the combination of features that make a structure more wind resistant is recognized with increased resale market value. Public policies that provide for stricter building codes and additional insurance may still be necessary. However, the clear message from this study is that there is room for policymakers to provide incentives for voluntary mitigation as well.

(*.) Department of Economics, Oklahoma City University, 2501 Blackwelder, Oklahoma City, OK 73106-1493, USA; E-mail ksimmons@okcu.edu.

(+.) Department of Economics, Texas Tech University, Box 41014, Lubbock, TX 79409-1014, USA; E-mail Jamie.Kruse@ttu.edu; corresponding author.

(++.) Department of Civil Engineering, Texas Tech University, Box 41023, Lubbock, TX 79409-1023, USA; E-mail doug.smith@wind.ttu.edu.

Partial financial support for this work was provided by the National Science Foundation under the NSF CSU/TTU Cooperative Project in Wind Engineering and the Department of Commerce under the NIST/TTU Cooperative Agreement Award. The authors thank Robert Chapman, Harold Cochrane, Bradley Ewing, Howard Kunreuther, Mike McKee, Kishor Mehta, Tom Steinmeier, and members of the Technology Assessment and Advisory Council of the NSF CSU/TTU Cooperative Project in Wind Engineering for helpful comments and guidance. In addition, we would like to thank two anonymous referees for their constructive comments.

Received December 1999; accepted May 2001.

(1.) For example, Kunreuther (1996, p. 171) notes, "Before a disaster, most homeowners, private businesses, and the public sector do not voluntarily adopt cost-effective measures to reduce their potential losses from future storms."

(2.) Simmons and Kruse (2000) use an expected utility framework to show that intangible losses can create an incentive to self-insure although actuarially fair market insurance is available.

(3.) The FWUA was created by the Florida Legislature in 1970. It provides insurance coverage in areas where private insurance companies refuse to write new policies. Approximately 427,000 homes along the Florida coastline are covered by FWUA.

(4.) The Saffir-Simpson Scale establishcs five categories. A category 1 storm has sustained winds of 74 to 95 miles per hour. Categories 2, 3, and 4 have sustained winds in the range of 96 to 110, 111 to 130, and 131 to 155 miles per hour, respectively. The most severe hurricane with sustained winds above 155 miles per hour that is expected to produce catastrophic damage is a category 5. Hurricane Camille, which hit the Gulf Coast in 1969, was a category 5.

(5.) An anonymous referee pointed out that the adverse event could be defined as wind damage to the home. Under this definition of the hazard, mitigation measures would be self-protection that affects the probability that the bad event occurs. However, the issue that insurance premiums do not reflect the expected value of the reduction in loss due to mitigation remains.

(6.) The homes are not located in Florida and thus are not eligible for the new FWUA mitigation incentive program.

(7.) The data set is proprietary and was provided to us under the stipulation that the name of the city would not be revealed.

(8.) In some areas, what we call storm blinds are called storm shutters. A storm blind can he a metal louvered shutter or a motorized metal curtain. When closed, the storm blind completely covers the window glass with a protective metal shell.

(9.) Kishor C. Mehta is a P. W. Horn Professor of Civil Engineering and director of the Wind Engineering Research Center; James R. McDonald is professor, chairman of the Department of Civil Engineering, and director of the Institute for Disaster Research; Douglas A. Smith is an assistant professor of civil engineering at Texas Tech University.

(10.) This system was peer reviewed by Joe Miner, University of Missouri; Peter Sparks, Clemson University; and George Walker, Commonwealth Scientific and Industrial Research Organization, Australia.

(11.) The SII contains the presence of storm blinds as part of the index. To avoid multicollinearity bias, we used separate regressions.

(12.) We created three models for the storm blind regression and three models for the structural integrity index regression. Model 1 used only physical features and year variables. Model 2 used physical features, year variables, and special amenities. Model 3 included physical features, year variables, special amenities, and neighborhood variables. The resulting coefficient and 1-statistic for the storm blind regression were 0.043, t = 2.17; 0.041, t = 2.13; and 0.053, t = 2.99 for models 1, 2. and 3, respectively. For the structural integrity index regression, coefficients and t-statisties were 0.230, t = 5.78; 0.243, t = 6.21; and t = 0.224, t = 6.23, respectively. Complete results from this test are available from thiseauthors on request.

(13.) A separate regression was performed excluding the missing observations. Changes to the values of the coefficients and standard errors were minimal. For storm blinds; the estimated coefficient and t-statistic were 0.053, t = 2.99, for the model that included observations with estimated square footage and 0.052, t = 2.81, for the model that excluded the observations with estimated square footage. For the structural integrity index, the estimated coefficient and t-statistic were 0.224, t = 6.23, and 0.216, t = 5.81, respectively. The complete results of this test are available from the authors on request.

(14.) A test for heteroskedasticity was performed and indicated the possible presence of heteroskedasticity at an alpha of 10%. We ran regressions that corrected for this, but the change in standard errors was minimal.

(15.) Pielke and Pielke (1997, p. 47).

(16.) Measuring the effect of insurance discounts on the resale market value of mitigation measures was suggested by an anonymous referee.

References

Bernknopf, Richard L., David S. Brookshire, and Mark A. Thayer. 1990. Earthquake and volcano hazard notices: An economic evaluation of changes in risk perceptions. Journal of Environmental Economics and Management 18:35-49.

Brookshire, David S., Mark A. Thayer, John Tschirhart, and William D. Schulze. 1985. A test of the expected utility model: Evidence from earthquake risks. Journal of Political Economy 93:369-89.

Camerer, Colin, and Howard Kunreuther. 1989. Decision processes for low probability events: Policy implications. Journal of Policy Analysis and Management 8:565-92.

Ehrlich. Isaac, and Gary S. Becker. 1972. Market insurance, self-insurance, and self-protection. Journal of Political Economy 80:623-48.

Fischer, G. W., L. G. Chesnut, and D. M. Violette. 1989. The value of reducing risks of death: A note on new evidence. Journal of Policy Analysis and Management 8:88-100.

Florida Windstorm Underwriters' Association. 2000. Save money and make your home safer. Media Release. www.fwua.com/media083100.html.

Fronstin, P., and A. G. Holtmann. 1994. The determinants of residential property damage caused by Hurricane Andrew. Southern Economic Journal 61:387-97.

Hammitt, James K. 1990. Risk perceptions and food choice: An exploratory analysis of organic versus conventional-produce buyers. Risk Analysis 10:367-74.

Insurance Research Council and Insurance Institute for Property Loss Reduction. 1995. Coastal exposure and community protection: Hurricane Andrew's legacy. Wheaton, IL: IRC; Boston: IIPLR.

Kunreuther, Howard. 1976. Limited knowledge and insurance protection. Public Policy 24:227-61.

Kunreuther, Howard. 1978. Disaster insurance protection: Public policy lessons. New York: John Wiley.

Kunreuther, Howard. 1996. Mitigating disaster losses through insurance. Journal of Risk and Uncertainty 12:171-87.

Lutter, R., J. F. Morral III, and W. K. Viscusi. 1999. The cost-per-life-saved cutoff for safety-enhancing regulations. Economic Inquiry 37:599-608.

MacDonald, Don N., James C. Murdock, and Harry L. White. 1987. Uncertain hazards, insurance, and consumer choice: Evidence from housing markets. Land Economics 63:361-71.

McDaniels, T. L., M. S. Kamlet, and G. W. Fischer. 1992. Risk perception and the value of safety. Risk Analysis 12:495-503.

Mehta, K. C., R. H. Cheshire, and J. R. McDonald. 1991. Wind resistance categorization of buildings for insurance. In Proceedings: Eighth International Conference on Wind Engineering, London. Canada, pp. 2010-1.

Mehta, K. C., R. H. Cheshire, and J. R. McDonald. 1992. Wind resistance categorization of buildings for insurance. Journal of Wind Engineering and Industrial Aerodynamics 41:2617-28.

Mehta, K. C., D. A. Smith, and R. H. Cheshire. 1993. Knowledge-based system for wind damage to buildings. Paper presented at the ASCE Conference on Hurricanes of 1992, Miami. FL.

Mehta, K. C., and R. H. Cheshire. 1993. Wind damage expert system for insurance industry. Paper presented at Third Asia-Pacific Symposium on Wind Engineering, Hong Kong.

Pielke, Roger A., Jr., and Roger A. Pielke, Sr. 1997. Hurricanes: Their nature and impacts on society. New York: John Wiley.

Shilling, James D., John D. Benjamin, and C. F. Sirmans. 1985. Adjusting comparable sales for floodplain location. Appraisal Journal 53:429-36.

Simmons, Kevin M., and Jamie Brown Kruse. 2000. Market value and perceived risk: Empirical results. Journal of Economics 26:41-51.

Slovic, Paul, 1987, Perception of risk. Science 236:280-5.

Smith, V. Kerry. 1986. Benefit analysis for natural hazards. Risk Analysis 6:325-34.

Speyrer, Janet Furman, and Wade R. Ragas. 1991. Housing prices and flood risk: An examination using spline regression. Journal of Real Estate Finance and Economics 4:395-407.

Viscusi, W. K. 1994. Mortality effects of regulatory costs and policy evaluation criteria. RAND Journal of Economics 25:94-109.

Table 1

Parameters, Factors, and Factor Choice Used in the Expert
System to Compute the Structural Integrity Index

Parameter Factors (Factor Choices)

Environmental Debris exposure (yes, no)
 Terrain exposure (water, open,
 suburban, and city)
 Topography (flat, hill, promontory)
 Wind speed region (region 1, region
 2, region 3)
 Tornado exposure (yes, no)
Frame Primary roof structure (wood
 truss, glue-laminated beams,
 wood beam/ rafter, unknown/other
 Positive anchorage (yes, no)
 Primary vertical load resisting
 system (reinforced masonry,
 unreinforced masonry, timber/wood
 stud, adobe)
Roof Roof deck (plywood, oriented
 strand board, wood plants, wood
 battens, other/unknown)
 Roof covering (standing seam metal
 roof, tile, slate, metal panel,
 shingles, unknown/other)
 Roof geometry (flat, gable, hip,
 mansard, stepped, complex)
 Skylights (yes, no)
 Roof age (<5 years, 5-10 years, 11-
 15 years, 16-20 years, >20 years)
Wall Percentage glass (0-5%, 6-20%, 21-
 60%, 61-100%).
 Cladding (none, masonry, precast
 concrete, stone panels, metal
 panels, EIFS over studs, EIFS
 over concrete or masonry, adobe,
 siding, unknown/other)
 Impact-resistant glass/shutters
 (yes, engineered; yes, nonengi-
 neered; no)
 Sliding glass doors (none: yes; no
 shutters; yes, engineered
 shutters; yes, nonengineered
 shutters)
 Overhead garage doors (none,
 reinforced single width, single
 width, reinforced doubel width,
 double width)
Other considerations Awnings, canopies (yes, no)
 Prescriptive building
 (prescriptive no/unknown)
 Number of stores (one, two)
Table 2

Variable Definitions and Descriptive Statistics

Variable Definition

BLND Storm blinds
INDEX Structural integrity index

LSF Natural log of total square
 feet
BATH Number of bathrooms
POOLSPA Swimming pool and spa
AC Air conditioning

HEAT Heating system
FIREPLACE Fireplace
VIEW Beachside, view of harbor,
 view of bay, view of the
 golf course
NEIGHBORHOOD Location in area by neighbor-
 hood identifier
SPRICE Natural log of sale price

Variable Descriptive Statistics

BLND Homes with storm blinds: 387
INDEX Range: .99-1.81
 Average: 1.42
LSF Average square feet: 1673

BATH Average number of baths: 2.3
POOLSPA Homes with pool and spa: 27
AC Proportion with air condition-
 ing: 96%
HEAT Proportion with heat: 99.5%
FIREPLACE Homes with fireplace: 410
VIEW Homes with special view: 23


NEIGHBORHOOD Number of distinct neighbor-
 hoods: 19
SPRICE Average price: $81,874
Table 3

Retrofit Model

 Analysis of Variance
Source d.f. Sum of Squares Mcan Square F-Value

Model 42 395.79052 9.42358 121.014
Error 1354 105.43891 0.07787
C total 1396 501.22943
[R.sup.2] 0.7896
Adjusted [R.sup.2] 0.7831

 Analysis
 of
 Variance
Source Prob > F

Model 0.0001
Error
C total
[R.sup.2]
Adjusted [R.sup.2]
 Parameter Estimates
Variable Parameter Estimate Standard Error t-Statistic Prob > \t
BLND 0.053019 0.017743 2.988 0.0029
INTERCEPT 5.002939 0.235720 21.224 0.0001
LSF 0.748537 0.030191 24.793 0.0001
BATH 0.093117 0.009383 9.924 0.0001
POOLSPA 0.177566 0.059395 2.990 0.0028
AC 0.215870 0.044611 4.839 0.0001
HEAT 0.302549 0.113215 2.672 0.0076
FIRE 0.176512 0.020073 8.793 0.0001
Table 4

Construction Model

 Analysis of Variance
Source d.f. Sum of Squares Mean Square

Model 42 398.05219 9.47743
Error 1354 103.17724 0.07620
C total 1396 501.22943
[R.sup.2] 0.7942
Adjusted [R.sup.2] 0.7878

 Analysis of Variance
Source F-Value Prob > F

Model 124.373 0.0001
Error
C total
[R.sup.2]
Adjusted [R.sup.2]
 Parameter Estimates
Variable Parameter Estimate Standard Error t-Statistic

INDEX 0.223651 0.0359032 6.229
INTERCEPT 4.824078 0.235194 20.511
LSF 0.732525 0.029882 24.514
BATH 0.092989 0.009281 10.020
POOL/SPA 0.168253 0.058779 2.862
AC 0.220996 0.044069 5.015
HEAT 0.299516 0.111972 2.675
FIRE 0.161523 0.020037 8.061

 Parameter
 Estimates
Variable Prob > \t
INDEX 0.0001
INTERCEPT 0.0001
LSF 0.0001
BATH 0.0001
POOL/SPA 0.0043
AC 0.0001
HEAT 0.0076
FIRE 0.0001