Printer Friendly

Flood insurance coverage in the coastal zone.

ABSTRACT

We explore determinants of flood insurance demand in the coastal zone using micro-data for nine Southeastern counties. Overall estimates indicate price inelastic demand, though subsidized policyholders have greater coverage and are more price sensitive. Mortgage borrowers exhibit no greater coverage; only 12 percent in 100-year flood zone indicate flood insurance was required by their lender. Flood insurance demand is increasing in the levels of flood and erosion risk. We find a positive correlation between household income and coverage, but the effect is not monotonic. Community-level erosion hazard mitigation projects influence flood insurance coverage, with beach replenishment acting as a complement.

INTRODUCTION

Over the past 50 years, coastal areas in the United States have witnessed a growing populace and increased economic activity. On the East and Gulf coasts, the burgeoning population faces considerable risk from coastal storms (hurricanes and nor'easters) that periodically cause extensive flooding, wind, and erosion damage. Increasing coastal populations, development in hazard-prone areas, rising construction costs, and increased value at risk have contributed to rising monetary losses due to these natural hazards (Kunreuther, 1998a; Wharton, 2008). Nordhaus (2006) estimates the value of capital stock in low-lying coastal areas vulnerable to natural hazards at $1.2 trillion (about 3 percent of GDP (in 2005 dollars)). Analysis of climate data suggests that we are entering a period of increased storm activity and intensity (Goldenberg et al., 2001; Webster et al., 2005) that could exacerbate coastal risk.

The National Flood Insurance Program (NFIP) offers indemnification from flood hazard (and some erosion hazards) in communities that agree to regulate development in the floodplain. In this article, we analyze flood insurance coverage choice in the coastal zone, utilizing household micro-data from 6,074 parcels in nine Southeastern U.S. counties. These data were collected by the H.J. Heinz III Center for Science, Economics, and Environment, under the direction of Federal Emergency Management Agency (FEMA), pursuant to addressing questions regarding the impact of shoreline erosion on coastal communities and the NFIP (Heinz, 2000). The data were utilized by Kriesel and Landry (2004) to examine participation in the NFIP. We expand upon their analysis in a number of ways: (1) our empirical model considers not only participation but also the level of coverage elected, (2) we improve upon the insurance premium covariate by employing NFIP rate schedules to determine marginal price measures that reflect specific property risk attributes (rather than average imputed prices as employed by Kriesel and Landry), and (3) we explore a greater array of specifications and covariates in our analysis in order to test economic and behavioral determinants of flood insurance demand. Many of our findings are based on analysis of a much larger data set than that of Kriesel and Landry--observations in their regression model were limited by missing covariates from a household survey.

Consistent with previous research (U.S. GAO, 1983; Browne and Hoyt, 2000; Kriesel and Landry, 2004; Dixon et al., 2006), we find evidence of price inelastic demand for flood insurance. Unlike previous research, we explore price elasticity across subsidized and nonsubsidized insurance policies, finding that subsidized policyholders exhibit greater coverage and elastic demand. (1) Despite federal regulations, only 12 percent of households in the 100-year floodplain claim they were required to hold flood insurance by their lender in 1998. In contrast to Kriesel and Landry (2004), we find that mortgage borrowers hold no greater flood insurance coverage, but they do exhibit less elastic flood insurance demand. While we make no attempt to discern between different models of individual choice under uncertainty, some findings are consistent with expected utility (EU) theory: coverage demand is greater in the highest risk (V-zone) areas and lower in the least risk (B/C/X-zones) areas relative to more moderate risk (A-zone) areas. We find evidence of a positive correlation between income and flood insurance coverage. Our dependent variable controls for the value of the asset at risk, so this result should not reflect the fact that higher income households tend to live in more expensive housing. The estimated effect of income, however, is not monotonic, which could indicate a tendency for self-insurance at high levels of household income.

We provide initial evidence of the relationship between erosion risk and flood insurance demand, finding that coverage is positively correlated with the erosion rate at the nearest shoreline for properties in actively eroding zones. This suggests that erosion risk may induce flood insurance purchase. (2) Surprisingly, flood insurance coverage is also greater along shorelines that are actively accreting, but the effect is very small. Further, we consider the relationship between flood insurance and other forms of erosion risk management. Building upon Kriesel and Landry (2004), we analyze the effects of large-scale erosion control projects on insurance demand but find a difference between structural fortification (i.e., seawalls) and beach replenishment (i.e., adding sand to the beach). We find evidence that coverage is higher in areas that manage erosion through beach replenishment and lower in areas that are structurally fortified, though the latter effect is not statistically significant. Nonetheless, the findings suggest a difference in the way households view community protection policies vis-A-vis formal insurance. Interpretation suggests that coastal households may perceive shoreline armoring as "self-insurance," a form of community protection that decreases the expected loss conditional on a flood or storm (Ehrlich and Becket, 1972). Self-insurance is a substitute for formal insurance. Beach replenishment, on the other hand, appears to be a complement with flood insurance. Since flood insurance rates do not explicitly reflect beach replenishment, we speculate that these types of projects may change local residents' perceptions of coastal risk.

NATIONAL FLOOD INSURANCE PROGRAM

Historically, problems related to adverse selection, the catastrophic nature of flooding, and government's predilection for disaster aid has precluded private insurers from voluntarily offering coverage for flood hazard (Anderson, 1974; Kunreuther, 1998b). Since the late 1960s, the U.S. federal government has played an expanded role in providing protection from floods and other coastal hazards. The National Flood Insurance Act of 1968 made federal flood insurance available, (3) through the NFIP, to communities that agreed to manage development in floodplains, with subsequent legislation (Flood Disaster Protection Act of 1973, National Flood Insurance Reform Act of 1994) designed to augment incentives for insurance purchase and hazard mitigation projects (Pasterick, 1998). The National Flood Insurance Reform Act of 1994 charged the FEMA with evaluating the effects of coastal erosion on coastal communities and the NFIP (Heinz Center, 2000). In light of increasing coastal populations and predictions of increasing coastal storm intensity, there is heightened concern about natural hazard exposure in coastal areas and the viability of NFIE Understanding household demand for coverage is a key element in assessing the viability of the market for flood insurance and the role of market insurance in conjunction with other forms of indemnification from coastal hazards.

Due to the large number and diversity of affected communities, delineation of flood risk under the Flood Disaster Protection Act of 1973 proved a laborious task, leading to the development of the NFIP in phases. The "Emergency Phase" of the program offered insurance at subsidized rates to households in communities that agreed to adopt floodplain management ordinances. Subsidized insurance rates applied only until detailed Flood Insurance Rate Maps (FIRMs) could be produced, after which the community enters the "Regular Phase" of the program. New construction in the regular phase is required to meet more stringent building standards designed to make structures more flood resistant and receives "actuarial" insurance rates determined by flood zone, structural characteristics, and the existence of community hazard mitigation projects. (4) As of 1997, 35 percent of properties in the flood zone nationwide were eligible for explicitly subsidized insurance, paying approximately 37 percent of the actuarial premium (Burby, 2001). (5)

Since its inception, the NFIP has suffered from low levels of participation among residential homeowners. The Flood Disaster Protection Act of 1973 required communities be enrolled in NFIP in order to qualify for certain types of federal disaster assistance and required flood insurance purchase for mortgages in the 100-year floodplain (Special Flood Hazard Area or "SFHA") made by federally regulated lenders (Pasterick, 1998). Mandatory purchase requirements were strengthened under the National Flood Insurance Reform Act of 1994, and programs were expanded to encourage local hazard mitigation projects. Nonetheless, evidence suggests that mandatory purchase requirements are not aggressively enforced after the initial year of a mortgage contract (Kunreuther, 1996; Palm, 1998; Tobin and Calfee, 2005), so that after a loan is secured participation becomes de facto voluntary. (6) In 1997, market penetration for the NFIP across the United States was estimated at 26 percent of eligible parcels (Price-waterhouseCoopers, 1999). Explanations for low market penetration have included ignorance of and lack of experience with flood hazard, subjective misperceptions of the likelihood of flooding and magnitude of loss, lack of awareness of the availability of flood insurance or belief that the price is too high, and "charity hazard"--a reliance on assistance from others (e.g., government) in the event of disaster (Kunreuther, 1984, 1996; Lewis and Nickerson, 1989; Browne and Hoyt, 2000).

FLOOD INSURANCE COVERAGE: THEORY AND EMPIRICS

Optimal insurance coverage has been analyzed within an EU maximization framework by Smith (1968) and Mossin (1968). We briefly sketch a simple version of the model in the context of flood insurance. Let utility U(*) be defined over individual wealth, Y = A + L, with A representing endowed wealth and L the value of property exposed to risk. Assume risk aversion: U'(Y) > 0 and U"(Y) < 0. The probability of loss L is [PI]. The individual may purchase insurance coverage C, providing indemnity under the loss scenario, with 0 [less than or equal to] C [less than or equal to] L. The insurance premium is proportional to C, given by pC. The individual purchase decision problem is

[max.sub.C] = E[U(Y)] = [PI]U(A + (1 - p)C) + (1 - [PI])U(A+L - pC). (1)

It is widely recognized that maximization of (1) implies full coverage (C = L) if insurance is actuarially fair (p = [PI]) and less than full coverage if the premium includes a loading factor (p = (1+[lambda])[PI] for 0 < [lambda] < 1). (7) Introducing an exogenous constant deductible to the loss state increases optimal coverage. Flood insurance coverage on structure is capped at $250,000, which restricts the range of C, though this limit is not binding for the majority of properties (Michel-Kerjan and Kousky, 2010). Under common assumptions, demand for insurance coverage is decreasing in price and increasing in risk factors ([PI] and L). For actuarially unfair insurance (p > [PI]), the relationship between demand and wealth (A) depends upon the nature of risk aversion (see, e.g., Schlesinger, 1981; Cummins and Mahul, 2004).

A number of plausible decision-making heuristics give rise to what are considered behavioral anomalies in the context of EU and lead to systematic errors in optimization; behavioral anomalies include subjective misperception of risk (probability or loss), optimism bias (i.e., "it can't happen to me"), desire to reduce anxiety about risk, wanting to legitimize a decision in the eyes of others, a tendency to behave as do one's peers, and an inclination to ignore low-probability events (Camerer and Kunreuther, 1989; McClelland, Schulze, and Coursey, 1993; Kunreuther, 1996; Palm, 1998; Krantz and Kunreuther, 2007). Lack of information on probabilities and magnitudes of loss may invalidate the EU framework in (1). Saliency of accurate risk information may vary over time and by context (Kunreuther, Sanderson, and Vetschera, 1985; Krantz and Kunreuther, 2007). On the other hand, if full insurance is legally required with strictly enforced provisions, insurance coverage may not be an object of discretionary choice. Variations in optimal coverage choice can also be explored through the introduction of different forms of utility in (1) (e.g., Braun and Muermann, 2004; Lee, 2007) or through defining utility over changes in wealth (Kahneman and Tversky, 1979). Though not explicitly considered in model (1), the likelihood and expected magnitude of disaster assistance may affect the demand for flood insurance. An expectation of unconditional assistance in the event of a disaster would serve as a substitute to formal flood insurance.

There exists little empirical work on demand for flood insurance coverage. Baumann and Sims (1978) find evidence that past experience with disasters motivates insurance adoption, as do social class and personality. Survey research suggests that lower income and non-white households, women, and elderly all tend to exhibit greater fear of disasters, though it is unclear whether this fear translates into insurance purchase or other types of mitigation and protective behavior (Palm, 1998). Browne and Hoyt (2000) use state-level panel data to estimate a flood insurance demand model. They find a negative price effect (inelastic in a market penetration model and approximately unitary in a coverage level model) and positive income effect on flood insurance demand. Consistent with previous findings, their results suggest that demand is increasing in flood damages of the prior year. Contrary to expectations, they find that insurance demand is decreasing in the number of mortgages by federally regulated lenders and increasing in the amount of federal disaster assistance.

National data gathered by Dixon et al. (2006) support the finding that market penetration rates are not sensitive to price and further suggest that penetration is significantly higher in SFHA and higher for communities with a larger number of parcels in SFHA.

The authors attribute the latter finding to more aggressive marketing of and more familiarity with flood insurance on the part of insurance agents in such communities. Dixon et al. find that the probability of purchasing insurance is substantially higher in communities subject to coastal flooding than in communities that are not--63 percent versus 35 percent. They speculate that demand for flood insurance may be lower in communities not subject to coastal flooding because there is less appreciation for flood risk or because the type of coverage offered by flood insurance policies is less attractive in inland areas.

Michel-Kerjan and Kousky (2010) examine county-level panel data and individual-level policy data to explore characteristics of the flood insurance market in Florida (which represents approximately 40 percent of policies-in-force and total dollars of coverage). They find that the overwhelming majority of policyholders elect the lowest level of deductible ($500) and that coverage levels have increased in reaction to the floods of 2004, while deductibles have decreased. For most policyholders, the $250,000 limit on structure coverage is not binding, as their replacement value is less than this limit. Further, they find that Florida's average flood insurance premium is the lowest in the nation, and surprisingly, the average Florida premium level has decreased in the most recent year of their data.

Kriesel and Landry (2004) use household-level data from the coastal zone to examine participation in NFIP. They find price inelastic demand for flood insurance and a positive income effect. Consistent with federal rules, their results suggest that mortgage requirements result in a much higher likelihood of participation in NFIP. Further, they find that insurance participation is higher in coastal areas that are fortified with artificial erosion protection (shoreline armoring and/or beach replenishment), lower for properties located farther back from the shoreline, and lower for geographical areas that have a higher hurricane return period (lower hurricane risk).

We expand upon the analysis of Kriesel and Landry (2004) by considering both participation and coverage level in our empirical model, employing different measures of flood insurance premiums and exploring a greater array of specifications and covariates in our analysis in order to test economic and behavioral determinants of flood insurance coverage. Our baseline specifications allow for an analysis of the entire data set rather than relying upon survey returns as in Kriesel and Landry. Our approach is more similar to the analysis of Guiso and Jappelli (1998), which examines casualty insurance in Italy and how coverage is influenced by uninsurable household wealth risk and other factors.

FLOOD INSURANCE COVERAGE DATA

Our data were gathered by the H.J. Heinz III Center for Science, Economics, and the Environment, under the direction of FEMA, to address issues of flood insurance and coastal erosion. The sampling frame is residential parcels in the near-shore zone (8) of nine coastal counties in Delaware, North Carolina, South Carolina, Georgia, Florida, and Texas. A stratified random sample of the near-shore zone was selected across the nine counties using a T-shaped sampling frame within each county in order to ensure adequate coverage on the oceanfront; weights are used to adjust all reported statistics for representation of the near-shore zone.

Table 1 displays a breakdown of the 6,074 parcels that were randomly selected for the study. Galveston County, Texas and Dare County, North Carolina provide the most observations (18.5 percent and 17.6 percent of the sample, respectively), while Lee County, Florida and Glynn County, Georgia provide the fewest (7.5 percent and 5.4 percent of the sample, respectively). For each parcel, contractors made onsite visits to collect information, such as structure elevation, foundation type, presence of basement or other obstruction below the main floor, ocean frontage, etc. Geographic information systems were employed to estimate distance from the shoreline, distance from the central business district, and flood zone. Parcel and structure characteristics from the county tax assessor database were appended to the data. (9) The sample was then merged by address with the Federal Insurance Administration's policies-in-force database in order to provide accurate information on market penetration and coverage levels. Of the 6,074 parcels, 52 percent of property owners were identified as holders of flood insurance. Lastly, the data set was complemented with information from a survey questionnaire sent to the home address of all parcel owners in the sample during 1998. The response rates, indicated in the last column of Table 1, vary significantly across counties, with a high of 53 percent in Dare County, North Carolina and a low of 19 percent in Sussex County, Delaware. The overall survey response rate was 34 percent.

Table 2 reports weighted descriptive statistics on insurance, parcel, and structure characteristics for the entire sample. The average flood insurance coverage for structure in the sample (obtained from both policies-in-force data and mail survey) was $71,600 (1998 dollars), with a minimum of zero and a maximum of $250,000. (10) Average coverage for NFIP participants was $142,431. The next two rows of Table 2 indicate measures of marginal flood insurance premium expressed in dollars per $100 coverage. Marginal premiums were calculated using descriptive information on the property, the level of elected coverage, and detailed NFIP rate tables from 2004 (adjusted back to 1998 levels). (11) At the parcel level, flood insurance premiums depend upon a number of factors, including: flood zone, year of construction relative to publication of FIRM, presence of basement or obstruction below a property, type of structure, elevation above base flood elevation (BFE--estimated height of the 100-year flood), Community Ratings System (CRS) score, the level of coverage, and chosen deductible. (12) We discuss each of these factors in turn.

Most of the properties in our data (50 percent) are located in the V-zone, 100-year flood zone with additional risk due to high-velocity waves associated with storm surge. Forty-one percent are located in the SFHA or A-zone (100-year flood zone), and 9 percent are located in the B/C/X-zones (500-year flood or lower risk zones). Houses built before the publication of FIRMs in their community and those in the V-zone built between 1975 and 1981 (13) are "grandfathered" in the NFIP and pay explicitly subsidized insurance rates. Fifty-seven percent of the parcels in our data set qualified for subsidized insurance under these guidelines. Subsidized and regular flood insurance premiums vary by flood zone, with structures in the V-zone paying the highest rates and structures in the X-zone paying the lowest rates. Subsidized rates vary according to whether a basement or other obstruction is present and by type of structure (single or multiple family). Regular rates vary by number of building stories, presence of basement or obstruction, structure type, and elevation above BFE. Post-FIRM structures with greater elevation pay lower rates. Almost 70 percent of structures in our data set are elevated on piles, and 18 percent have obstructions below the property. Average elevation above BFE was 3.3 feet, with a high of 97 feet and a low of-12.5 feet (i.e., 12.5 feet below BFE).

The National Flood Insurance Reform Act of 1994 established the CRS to evaluate and summarize mitigation projects in a community. The CRS score ranges from 1 (many mitigation projects, low flood risk) to 10 (little or no mitigation projects, baseline flood risk); a lower CRS score decreases flood insurance premiums. It is worth noting that beach replenishment is not a flood hazard mitigation strategy recognized by the CRS, and while CRS does offer credits for levee maintenance, it apparently does not offer similar credits for construction of sea walls or other forms of armoring along ocean coastlines. The average CRS score for our sample was 8.3 with a low of 5 and a high of 10. All premiums are adjusted to reflect the CRS score for the community, with discounts ranging from 0 percent (for a score of 10) to 25 percent (for a score of 5).

Premiums also vary by amount of coverage. A basic lower rate applies to the first $50,000 of coverage on structure, while a higher rate applies to additional coverage up to the $250,000 limit on structure. (14) Knowing coverage level, we are able to apply the marginal rate in our empirical analysis. The marginal rate should affect decision making via the theoretical model in (1). Previous research (Browne and Hoyt, 2000; Kriesel and Landry, 2004) has employed an estimate of the average insurance price.

The standard deductible for NFIP structure coverage is $500. Reduced premiums are awarded for those opting for a higher deductible, up to $5,000 deductible on single-family structures. Premiums for post-FIRM structures in the V-zone built after 1981 (approximately 14 percent of our data) depend upon the ratio of coverage level to replacement value (replacement cost ratio). Unfortunately, our data contain limited information (N = 1,668 for policyholders) on deductible level (15) and no information on replacement value.16 To make full use of the available data, we consider two measures of marginal premium--a high and a low estimate--in order to assess the responsiveness of coverage demand to premium level. The high-premium model assumes that all households elect the standard $500 deductible and that post-FIRM structures in the V-zone built after 1981 select a level of coverage that is less than 50 percent of the structure replacement cost. The data of Michel-Kerjan and Kousky (2010) suggest that 98 percent of Florida policyholders select a deductible less than the maximum and 80 percent choose the lowest deductible of $500. Thus, the high-premium assumptions probably provide the most accurate results. The average high marginal premium is $1.01 per $100 coverage with a minimum of $0.06 and a maximum of $5.97. The price elasticity from the coverage model that employs the high marginal premium will be a lower bound on the true value. The low-premium model assumes all households elect a $1,000 deductible and that post-FIRM structures in the V-zone built after 1981 select a level of coverage that is greater than or equal to 75 percent of the replacement cost. The average low marginal premium is $0.87 per $100 coverage with a minimum of $0.06 and a maximum of $3.90. The price elasticity from the coverage model that employs the low marginal premium will be an upper bound on the true value.

The average historical beach erosion rate is 2.7 feet per year for those properties in an actively eroding zone (71 percent of the sample). A much smaller proportion (6.5 percent) of parcels are in accreting zones, with an average accretion rate of 0.2 feet per year. The remaining 22.5 percent of parcels are classified as being in neither an erosion or accretion zone. (17) Kriesel, Randall, and Lichtkoppler (1993) use a variable transformation, geotime, to measure erosive pressure on a parcel. Geotime is defined as the ratio of setback (or distance from the shoreline) to historical erosion rate, providing an estimate of the number of years a parcel is expected to remain in the face of constant, deterministic shoreline erosion. Average geotime in our sample is 787 years, but approximately 30 percent of the parcels exhibit geotime less than 10 years. Hurricane return period, the mean number of years expected to elapse between landfall of major hurricanes in an area, was calculated at the county level from summary information provided by FEMA. The average is 47 years, with a low of 16 years and a high of 190 years. The average distance from the shore is 318 feet and 42 percent of properties are oceanfront.

The tax assessor's database provides information on assessed building and land values, recent sales price, year of construction, year of sale, and other structural variables. Building and land assessed values are unreliable measures of value for our analysis due to differences in assessment and updating across municipalities. Since information on sales price is limited (N = 2,844), we employ hedonic price regression to produce imputed current property values. (18) Sixty percent of the imputed value is taken to represent the value of housing structure (net of land value). The average property sales price is $187,177 (1997 dollars), and the average imputed housing asset value is $143,683. The average ratio of flood insurance coverage to asset value is 0.651. Year of construction is used to determine whether the structure was built after the publication of a FIRM in the community; post-FIRM buildings are required to meet more stringent building standards and pay "actuarial" flood insurance rates.

We turn next to survey data gathered from the mail questionnaire. We find evidence of response bias in these data, as 82 percent of respondents are identified as NFIP participants. This is considerably greater than the overall sample average of 52 percent. As such, analysis of the raw data could lead to erroneous inference on price responsiveness and other effects. We reweight observations in this data set to correct for the overrepresentation of NFIP participants (in addition to the T-shaped sampling frame, discussed earlier). Nonetheless, there could be other unobservable sources of response bias, and thus the descriptive statistics and regression results associated with these data should be interpreted and applied with caution.

Table 3 presents the weighted descriptive statistics for the survey data. Household income is measured by a nominal response to eight income categories, with the midpoint utilized as an estimate. The average income is over $100,000. The distribution of income is depicted in Figure 1. Twenty percent of respondents have high school as the highest level of educational attainment, 43 percent are college graduates, and 36 percent have at least some graduate school training. Forty-five percent are retired, and 5 percent work part-time. The average age is 61 years, and the average household has 0.46 children.

[FIGURE 1 OMITTED]

Sixty-eight percent of respondents indicated that they would have purchased their coastal home regardless of whether flood insurance was available. This suggests that for almost one-third of households, availability of flood insurance played some role in the decision to purchase a beach house. There is some evidence of a problem with retention of flood insurance, as 11 percent indicate that they have allowed their policy to lapse at some time in the past. Ten percent indicate that they have submitted an insurance claim for flood damages in the past.

A number of survey items inquired about awareness of and community response to erosion hazard. Only 28 percent of respondents claimed to be aware of the erosion rate at the nearest shore at the time of purchase. This is not surprising, as there are no disclosure provisions in relation to erosion hazard. Nineteen percent indicated that shoreline armoring was being used to combat erosion at the shoreline nearest their property, while 35 percent indicated that beach replenishment was being utilized at the nearest shoreline. These policy options are sometimes used in conjunction, and 10 percent of respondents claimed both policies were being pursued at the nearest shore. (19)

A relatively low proportion of homeowners (39 percent) hold a mortgage on their beach house. Focusing on homes in the 100-year floodplain, the proportion of mortgages drops to 34 percent, and surprising only 12 percent of those with mortgages claim that they were required to purchase flood insurance by their mortgage lender. The majority of respondents (35 percent) utilize their property as a vacation home, allowing it to remain vacant when they are not using it. We conjecture that this pattern of usage could be a proxy for wealth, as these owners choose to forego rental income on their beach house. Thirty percent of respondents use the property as part-time rental and part-time vacation home. Almost a quarter utilize their property as their primary residence, and 10 percent offer their beach house as a full-time rental.

A subset of respondents (N = 292) provided information regarding why they did not hold flood insurance. The majority (30 percent) indicated that flood insurance was too expensive. A quarter indicated that they perceived the risk of flooding as very low, while 20 percent claimed they were not required to purchase flood insurance. Nine percent indicated that flood insurance was unavailable.

ECONOMETRIC MODELS OF FLOOD INSURANCE COVERAGE DEMAND

We employ multiple regression analysis to explore determinants of flood insurance coverage choice for residential building structures in the near-shore coastal zone. Following EU theory of insurance choice under uncertainty (Smith, 1968; Mossin, 1968), insurance price, the probability and size of loss, and endowed wealth are primary demand parameters. We use NFIP rate tables and detailed property characteristics to determine each household's marginal price of flood insurance (the amount charged for additional $100 coverage). Risk factors include presence in a flood zone and elevation above BFE; we hypothesize that households in higher risk zones and with lower elevation will demand greater coverage, but the higher cost of insurance in these zones makes the effect uncertain. Likewise, the correlation between price and risk factors complicate the estimation of pure price elasticity. Income, only available in the survey data, serves as imperfect proxy for wealth. In addition, we record whether the beach home is a vacation home and conjecture that this could also proxy for wealth. (20) We also explore erosion hazard factors, such as the erosion/accretion rate and the presence of erosion mitigation projects (shoreline armoring or beach replenishment) in the nearby area. Households may view such projects as substitutes or complements to formal flood insurance depending upon their own assessment of the protection offered. Other covariates in the model include household demographic factors.

Size of potential loss should be related to the value of the asset at risk. Unfortunately, we have limited information on property replacement values. We employ hedonic price regression analysis to produce imputed current property values and take a proportion of the imputed value as an estimate of the replacement value of the structure at risk (net of land value). The hedonic price regression parameters for the entire sample are displayed in Table 4. The estimation utilizes a semi-log functional form and includes housing sales between 1980 and 1997. Due to missing data, the specification is fairly restrictive, including only square footage and lot size (both in quadratic form), dummy variables for missing information on square footage or lot size, the age of the structure at time of sale, dummy variables for oceanfront and vacant lots at time of sale, and distance to the central business district (CBD). Year and county fixed effects are included. (21) The estimated model is used to impute housing sales price in 1997, and 60 percent of the estimated sales price provides a proxy for the structure asset value. (22)

As the asset value is imputed, we transform the dependent variable of our model to the ratio of coverage to imputed asset value in order to avoid the generated regressor problem (Murphy and Topel, 1986). Thus, [y.sub.i] = cov/[??], where cov is chosen coverage level and [??] is imputed asset value. This specification is a minor alteration of our theoretical framework (1) and does not change the fundamental nature of the decision problem. As a robustness check, we also estimate regression models with raw coverage level as dependent variable and include imputed asset value as an explanatory variable, employing bootstrapping to produce reliable standard errors (Efron and Tibshirani, 1986; Shao and Sitter, 1996). While interpretation of coefficients changes, results are generally consistent across both specifications. (23)

We employ the Tobit model (Tobin, 1958; Wooldridge, 2001), which assumes that flood insurance coverage is censored at zero. The dependent variable for a Tobit model is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2)

where [Y.sub.i] is the observed response variable, [[Y.sup.*].sub.i] is the latent response variable, and LL is the lower limit ($0). Due to the presence of the error term in the hedonic regression, there is no upper bound on [y.sub.i]. The Tobit model assumes the continuous portion of the error distribution is reasonably approximated by a Gaussian probability density, while the censored value is represented by cumulative Gaussian probability masses. The log-likelihood function for the Tobit model is

LF = [summation over (i[member of] {[y.sub.i])=LL}] ln [PHI] (LL - [x'.sub.i][beta]/[sigma]]) + [summation over (i[member of] {LL < [y.sub.i]] (3)

where [empty set](*) represents the standard normal probability density function, [PHI](*) represents the standard normal cumulative distribution function, x is a vector of covariates hypothesized to effect demand for flood insurance coverage, including county fixed effects, and [beta] and [sigma] parameters to be estimated. Due to differences in subsample sizes across counties (see Table 1), we cluster standard errors at the county level.

Method of estimation is quasi-maximum likelihood, as weights ([[omega].sub.i]) are applied to each observation of the log-likelihood function to correct for the T-shaped sampling frame (and underrepresentation of flood insurance nonparticipants in the case of models 3 and 4 that rely on survey data). A modified Newton-Raphson algorithm is used to obtain parameter estimates (Amemiya, 1973, 1985). Marginal effects are transformations of (3) that provide an estimate of the effect that a unit change in an element of the vector x have upon the response variable. Marginal effects for the Tobit model are calculated as:

[partial derivative]E(y|x)/[partial derivative][x.sub.j] = [PHI] (x'[beta]/[sigma])[[beta].sub.j] (4)

for each continuous element j of the vector x, where E(*) is the expectations operator. Marginal effects for discrete covariates h are calculated as (24)

[DELTA]E (y|x)/[DELTA][x.sub.h] = E(y|[x.sub.-h], [x.sub.h] = 1) - E (y|[x.sub.-h], [x.sub.h]= 0). (5)

Elasticities transform marginal effects into unit-free, percentage change effects, and are calculated as

[[epsilon].sub.j]= [partial derivative]E(y|x)/[DELTA][x.sub.j] x [[bar.x].sub.j]/[bar.y] or [[epsilon].sub.h] = [DELTA]E(y|x)/[DELTA][x.sub.h]x 1/[bar.y'] (6)

where [bar.x] and [bar.y] are weighted means of the independent and response variables, respectively, and the latter discrete measure effect is a half-elasticity.

RESULTS

Tables 5 and 6 report parameter estimates for various specifications of the Tobit models (3), employing high and low estimates of marginal insurance premium, respectively. Each table includes four models. The first model serves as a baseline and includes marginal flood insurance premium, indicators for the V and B/C/X flood zones (excluded category is A-zone), the hurricane return interval (return), and the historical average erosion rate (er) or accretion rate (ar). An additional risk factor, elevation above BFE, exhibited statistically insignificant parameter estimates in all models and was subsequently dropped from the analysis. Model 2 explores differences in coverage for subsidized policyholders. We note that both models I and 2 make use of a much larger data set than was utilized in the analysis of Kriesel and Landry (2004), because these specifications do not rely on survey data. Due to possible response bias in the survey data, we view the results in models 1 and 2 as more reliable. Models 3 and 4 explore additional covariates that are only available in the survey data, specifically the influence of local hazard mitigation projects and household level variables, respectively. Asterisks indicate covariates that are statistically significant at various levels.

All specifications exhibit a negative flood insurance price coefficient. Estimates of price elasticity of demand are [[epsilon].sub.p] = -0.620 for the high-premium model and [[epsilon].sup.p] = -0.870 for the low-premium model. Thus, both models indicate inelastic demand, with the high premium providing an arguably better estimate and a lower bound on the responsiveness of flood insurance demand to price. Model 2 explores the variability in coverage by subsidy class. Subsidized policyholders exhibit greater demand for flood insurance, with a marginal effect ranging from $0.289 to $0.745 per $1 imputed asset value, depending upon which price specification is used. Subsidized policyholders are also much more price sensitive than nonsubsidized policyholders; price elasticity is [[epsilon].sub.p] = -1.550 for subsidized policyholders compared to [[epsilon].sub.p] = -0.133 for nonsubsidized in the high-premium model, or [epsilon].sup.p] = -4.478 for subsidized compared to [[epsilon].sub.p] = -0.502 for the low-premium model.

All models indicate higher insurance coverage in the V-zone and lower insurance coverage in the B/C/X-zones relative to the A-zone. For the high-premium models, the marginal effect for V-zone is $0.26 to $0.30 per dollar of asset value, while the marginal effect for presence in B/C/X zones is -$0.05 to -$0.06 per dollar of asset value. For the low-premium model, the marginal effect for V-zone is $0.33 to $0.70 per dollar of asset value, while the marginal effect for presence in B/C/X zones is -$0.05 to -$0.20 per dollar of asset value. Results from model 1 suggest that those households facing higher erosion hazard demand greater flood insurance coverage. The elasticity of the rate of shoreline erosion is 0.03 to 0.06 in the high-premium models, and 0.06 to 0.11 in the low-premium models. Somewhat surprisingly, the rate of shoreline accretion also exhibits a positive elasticity, but the estimated effects are very small--0.001 to 0.004 in the high-premium models and 0.003 to 0.004 in the low-premium models. The coefficient for hurricane return interval has the expected negative sign in model 1 for the high-premium specification, but an unexpected positive sign in all other models. This likely reflects the poor nature of this proxy for hurricane risk.

Turning to model 3, we find that flood insurance holdings are greater in locations that manage coastal erosion through beach replenishment. The marginal effect is $0.17 to $0.18 per dollar of asset value. Parameter estimates suggest that flood insurance holdings are lower in locations that employ coastal armoring, but the effect is not statistically significant. Results of model 4 explore the effect of mortgage status on demand and price elasticity. Those households that hold a mortgage have more flood insurance coverage, but the effect is not statistically significant. As expected, mortgage holders exhibit much lower price elasticity than those that do not hold a mortgage. For the high-price model, [[epsilon].sub.p] = -0.035 for mortgage holders compared to [[epsilon].sub.p] = -0.370 for owners of nonmortgaged property. For the low-price model [[epsilon].sub.p] = -0.113 under a mortgage versus [[epsilon].sub.p] = -0.720 for those without a mortgage. Flood insurance demand is lower for those with high school as their highest level of educational attainment (relative to those with graduate training)--marginal effect is -$0.11 to --$0.12 per dollar of asset value. Retired individuals exhibit lower demand in the high-price model (about $0.06 per dollar of asset value), but this effect is not statistically significant in the low-price model.

The income parameter was not statistically significant in either of the initial specifications, nor was our wealth proxy--a dummy variable identifying properties that are vacation homes. The latter variable was dropped from the analysis. Average income varies considerably across counties, from a low of $58,000 in Brazoria, Texas to a high of $147,000 in Lee, Florida. Thus, the effect of income may be reflected in the fixed county effect parameters. At the suggestion of an anonymous reviewer, we estimate additional models that include categorical dummy variables for income class (rather than using the midpoint of the interval as an estimate of household income). Results for the high and low-premium models are displayed in Table 7. The omitted category is the lowest level of household income, $0-$29,999. With this specification, we do find income effects; higher income households generally display greater flood insurance coverage relative to the lowest income category. This pattern holds for five of the seven income coefficients in the high-premium model and six of the seven coefficients in the low-premium model.

For the high-premium model, households in the second income category ($30,000-$49,999) exhibit greater flood insurance coverage than those in the lowest income category, with a marginal effect of $0.14 per dollar of asset value. The coefficient estimate for category 5 ($100,000-$149,999) is significantly greater than that of category 2, with a marginal effect of $0.26 per dollar of asset value. Comparing other coefficients for the high-price model, the effects of income categories 4 ($75,000-$99,999) and 5 ($100,000-$149,999) are significantly different and greater than that of category 3 ($50,000-$74,999), with an estimated marginal effect for category 4 of $0.20 per dollar of asset value. While not statistically significant on its own, the coefficient for income greater than or equal to $250,000 is significantly less than some of the lower income categories (5 and 6 ($150,000- $199,999), in particular). A similar pattern holds for the low-premium model, with households in the second income category ($30,000-$49,999) exhibiting greater flood insurance coverage of about $0.12 per dollar of asset value (relative to the first income category), and households in category 5 ($100,000-$149,999) having significantly greater coverage (marginal effects of $0.17 per dollar asset value). For the low-premium model, however, households in the highest income categories are significantly different from the excluded (lowest) category, but not significantly different from other income categories.

DISCUSSION

Consistent with previous research, we find overall inelastic demand for flood insurance (U.S. GAO, 1983; Browne and Hoyt, 2000; Kriesel and Landry, 2004; Dixon et al., 2006). We believe that our estimates may be more accurate than previous estimates due to the fact that we employ marginal measures of insurance premium and utilize household-level micro-data. Our results, however, are limited to coastal properties in the southeast. Moreover, since NFIP premiums are set at the national level and reflect property risk characteristics, we are not able to isolate a pure price effect using Tobit regression analysis. Due to the lack of information on deductible and replacement value, we estimate dual models employing high and low estimates of marginal premium for all specifications. Despite this limitation, our price elasticity estimates are rather tight; the overall estimates range from --0.620 to --0.870. As existing data suggest that most flood insurance policyholders elect for low deductibles (Michel-Kerjan and Kousky, 2010), the former estimate is arguably better.

Owners of houses built before the production of FIRMs or built in the V-zone before building codes were adjusted to take account of storm surge receive explicitly subsidized flood insurance rates; 57 percent of the parcels in our data set qualify for such a subsidy. Examining demand by subsidy class, we find that explicitly subsidized policyholders exhibit greater demand for coverage (ranging from $0.289 to $0.745 per $1 asset value) than nonsubsidized policyholders. Accordingly, we find greater price elasticity of demand for subsidized policyholders ranging from --1.550 to --4.478 (compared to -0.133 to -0.502 for nonsubsidized policyholders). Again, we deem the former as likely better estimates in both cases. These are the first flood insurance price elasticity estimates, to our knowledge, that distinguish between subsidy classes.

Using the subset of survey data, we are also able to analyze demand and price elasticity by mortgage status. Those households that hold a mortgage (somewhat surprisingly, only 39 percent of survey respondents) exhibit no statistically significant difference in flood insurance demand, (25) but do exhibit a much lower price elasticity (ranging from --0.035 to --0.113) than those that do not hold a mortgage ([[epsilon].sub.p] ranging from -0.270 to -0.370). Given the potential for response bias in the survey data, these differences must be interpreted with caution. Nonetheless, the ability to examine price elasticity in the absence of explicit controls on demand via federal policy makes this result noteworthy. Our estimates suggest that unregulated demand is price inelastic.

The Congressional Budget Office estimates that NFIP "collects about 60 percent of the premiums needed for actuarial balance, leaving a cost to taxpayers estimated at about $1.3 billion per year" (Marron, 2006, p. 1). This shortfall reflects both the explicit subsidy for certain classes of parcels and the reputed lack of actuarial rates for other parcels. Our results support the contention that moderate increases in flood insurance premiums will probably not induce wholesale cancellation of policies, but the reduction in demand is likely to be significantly greater for subsidized policyholders.

To the extent that mortgage requirements mandate a specified level of flood insurance coverage, price increases will have little effect on demand. Our data suggest that mortgage contracts are not as common in the coastal zone as in other housing markets. This likely reflects the long tenure of many families in the coastal zone and significant wealth of many recent in-migrants. Further, of those mortgage borrowers located in the SFHA, only 12 percent claim they were required to purchase flood insurance by their mortgage lender in 1998. Our data also suggest that 11 percent of respondents have allowed their flood insurance coverage to lapse as some time in the past. These results are consistent with the suggestion that lenders have not been especially zealous in enforcing insurance purchase requirements as required by law (Kunreuther, 1984, 1996; Pasterick, 1998), but anecdotal evidence suggests that more recent data for the coastal zone may not show a similar tendency. The potential lack of enforcement of federal provisions for flood insurance purchase is a problem that requires further analysis. Moreover, since mortgages appear to be somewhat rare in the coastal zone (caveats on potential response bias in our survey data notwithstanding), mortgage requirements will have limited impact on flood insurance coverage.

We find evidence of significantly higher insurance coverage in the V-zone (ranging from $0.26 to $0.70 per $1 asset value) and lower insurance coverage in the B/C/X-zones (ranging from --$0.05 to -$0.20 per $1 asset value) relative to the A-zone. This pattern of results suggests that, conditional on the price of flood insurance and the value of the asset at risk, homeowners anticipate higher damage and thus purchase greater coverage in the 100-year flood zone with high-velocity waves relative to the standard 100-year flood zone, and that anticipation of damage and purchase of insurance coverage is lower in flood zones with less risk. Nonetheless, optimistic subjective risk perceptions do seem to influence flood insurance purchase for some survey respondents. A subset of survey data provided information on perspectives of NFIP nonparticipants. Sixty-two percent of nonparticipants in the V- and A-flood zones indicated that the price of insurance was too high or that the risk of flooding was very low.

We employ housing use data to test for wealth effects on flood insurance demand. Our survey data identify those households that use their coastal property as a vacation home. The rental market for housing in coastal areas is typically very active. Those households that own multiple homes (at least one in the coastal zone) and choose to forego rental income on their coastal property are likely wealthier than other households. Our vacation-home dummy variable was not statistically significant in our regression models, however. Thus, our findings are not particularly insightful regarding wealth, and this remains a difficult topic to explore empirically.

We find that households with greater income have greater flood insurance coverage, but the effect does not appear to be monotonic. Results suggest that households with moderate income ($30,000-$49,999) exhibit greater flood insurance coverage than those in the lowest income category ($0-$29,999), with a marginal effect of $0.12 to $0.14 per dollar of asset value. Households with greater income (ranging from $75,000 to $149,999) also exhibit greater coverage than those in the lowest income category, with marginal effects ranging between $0.17 and $0.26 per dollar asset value. Flood insurance coverage for households in the highest income categories (ranging from $200,000 to $249,999 and those with income greater than or equal to $250,000), however, were no different from those moderate income ($30,000-$49,999). In fact, results from one model indicate that those households with the greatest income hold no greater coverage than those with the lowest income. This pattern of results could indicate a positive income effect for moderate to high levels of income, but a tendency toward self-insurance at the highest income levels.

The FEMA project for which these data were collected sought to explore the effect of coastal erosion on the NFIP. The Flood Disaster Protection Act of 1973 made explicit the terms under which damages due to coastal erosion would be indemnified under flood insurance provisions. In particular, erosion losses must be associated with a flooding event in order to be covered by flood insurance. It is unclear, however, to what extent erosion risk affects expected loss and flood insurance demand. The data provide some insight regarding the latter. We find that households facing higher erosion hazard demand greater flood insurance coverage. The estimated elasticity is 0.03 to 0.11, suggesting a I percent increase in the erosion rate raises flood insurance demand by 0.03 to 0.11 percent. Surprisingly, location on an accreting shoreline also has a positive effect on flood insurance demand, but the elasticity is very small, ranging from 0.001 to 0.004. Both of these results are relative to locations that are classified as "neither eroding nor accreting." These results suggest that some homeowners view flood insurance as a form of partial protection from erosion hazard.

Lastly, consistent with the findings of Kriesel and Landry (2004), we find evidence that community-level erosion hazard mitigation projects influence flood insurance holdings. In contrast to the results of Kriesel and Landry, however, we find a possible asymmetry across the types of projects. Beach replenishment appears to be complementary with flood insurance (increasing coverage by around $0.18 per $1 asset value), while the estimated effect of shoreline armoring is negative but statistically insignificant. (26) Interpretation suggests that shoreline armoring may be seen as "self-insurance," a form of community hazard mitigation that decreases the expected loss conditional on a flood or storm (Ehrlich and Becker, 1972). Since it reduces expected loss while not affecting insurance price, self-insurance is a substitute for formal insurance. Seawalls buffer the hinterland against erosion and can act as a levee in providing protection from storm surge. If households believe seawalls are effective in this manner, they may feel protected against hurricane flooding and erosion risk, thus purchasing less flood insurance. As such, expansion of shoreline armoring in coastal areas could discourage flood insurance purchase. More exploration of this relationship is warranted, however, as parameter estimates on shoreline armoring are not statistically significant.

Beach replenishment, on the other hand, appears to be complementary to flood insurance purchase. "Self-protection" projects that lower the probability of loss, and thus the price of insurance, are complements to formal insurance (Ehrlich and Becker, 1972). To our knowledge, however, beach replenishment activities are not explicitly recognized under the CRS or other flood insurance rate-setting provisions. Like shoreline armoring, beach replenishment also buffers against erosion and can absorb some of the energy associated with storm surge, but our results suggest that households see this form of protection as complementary to flood insurance. Households may see beach replenishment as an inferior form of protection but may support it in order to maintain recreation potential of the beach. (Structurally fortified coastlines often exhibit poor beach quality.) It is also possible that beach replenishment serves as a reminder of coastal risks, as beach sand has to be replenished at regular intervals to compensate for erosion. Loss of beach may heighten perception of erosion and flood risk.

CONCLUSIONS

We use Tobit regression models to explore behavior and test theory regarding the determinants of flood insurance coverage in the coastal zone using micro-level data for nine southeastern U.S. counties. Unlike previous research, we incorporate both the extensive and intensive margin of demand and employ measures of marginal insurance premium to assess price elasticity. Overall estimates indicate price inelastic demand, though subsidized policyholders are more sensitive to price and hold greater flood insurance coverage. Despite federal regulations, mortgage borrowers do not exhibit higher flood insurance coverage but do exhibit significantly lower price elasticity. The extent of mortgage borrowing appears to be low in the coastal zone, and even for those households that are borrowers, enforcement of flood insurance requirements appears to be low in 1998.

The accuracy of our price elasticity estimates is compromised by correlation between insurance price and property risk factors; the ability to control for risk factors in our analysis is limited. Thus, since premiums reflect riskiness of the asset, it is difficult to identify the component of demand that is purely driven by price. Future research should focus on attempting to use thresholds in the price schedule that exhibit minimal differences in risk to identify a pure price effect. Such estimates would be very useful for policy analysis of potential price changes for NFIP.

While we do not attempt to discern between competing models of individual choice under uncertainty, we find some support for rational choice in the coastal zone, with flood insurance coverage correlated in the level of flood risk, controlling for insurance price and value of the threatened asset. We find a counterintuitive sign, however, for hurricane return interval, likely reflecting error in this county-level proxy for hurricane risk. We attempt to proxy for household wealth, using a dummy variable indicating vacation homes, but results are statistically insignificant. We find evidence of a positive correlation between income and flood insurance coverage (normalized by asset value), but the effect is not monotonic, which could indicate self-insurance at high-income levels.

We find evidence that erosion risk does affect flood insurance demand, as households facing higher erosion hazard demand greater insurance coverage. Surprising, demand is also increasing in the rate of shoreline accretion for areas that are accreting, but the effect is extremely small. Lastly, we find evidence that community-level erosion hazard mitigation projects influence flood insurance holdings, with beach replenishment appearing to act as a complement for flood insurance and shoreline armoring possibly acting as a substitute. Unfortunately, we are unable to address the importance of "charity hazard," or a reliance on third-party assistance in the event of natural disaster. Finding data that will allow for an assessment of charity hazard vis-a-vis other determinants of flood insurance demand remains an important topic for future research.

REFERENCES

Amemiya, T., 1973, Regression Analysis When the Dependent Variable Is Truncated Normal, Econometrica, 41: 997-1016.

Amemiya, T., 1985, Advanced Econometrics (Cambridge, MA: Harvard University Press).

Anderson, D. R., 1974, The National Flood Insurance Program--Problems and Potential, Journal of Risk and Insurance, 41: 579-600.

Baumann, D. D., and J. H. Sims, 1978, Flood Insurance: Some Determinants of Adoption, Economic Geography, 54(3): 189-196.

Braun, M., and A. Muermann, 2004, The Impact of Regret on the Demand for Insurance, Journal of Risk and Insurance 71: 737-767.

Browne, M. J., and R. E. Hoyt, 2000, The Demand for Flood Insurance: Empirical Evidence, Journal of Risk and Uncertainty, 20(3): 291-306.

Burby, R. J., 2001, Flood Insurance and Floodplain Management: The U.S. Experience, Environmental Hazards, 3: 111-122.

Camerer, C. F., and H. K. Kunreuther, 1989, Experimental Markets for Insurance, Journal of Risk and Uncertainty, 2: 265-300.

Cummins, J. D., and O. Mahul, 2004, The Demand for Insurance With an Upper Limit on Coverage, Journal of Risk and Insurance, 71(2): 253-264.

Dixon, L., N. Clancy, S. A. Seabury, and A. Overton, 2006, The National Flood Insurance Program's Market Penetration Rate: Estimates and Policy Implications, Report for the Federal Emergency Management Agency (Contract #282-98-0029), RAND Corporation, Santa Monica, CA.

Efron, B., and R. Tibshirani, 1986, Bootstrap Methods for Standard Errors, Confidence Intervals, and Other Measures of Statistical Accuracy, Statistical Science, 1(1): 54-77.

Ehrlich, I., and G. S. Becket, 1972, Market Insurance, Self-Insurance, and Self-Protection, Journal of Political Economy, 80(3): 628-648.

Goldenberg, S. B., C. W. Landsea, A. M. Mestas-Nunez, and W. M. Gray, 2001, The Recent Increase in Atlantic Hurricane Activity: Causes and Implications, Science, 293: 474-478.

Grace, M. F., R. W. Klein, and P. R. Kleindorfer, 2004, Homeowners Insurance With Bundled Catastrophe Coverage, Journal of Risk and Insurance, 71(3): 351-379.

Guiso, L., and T. Jappelli, 1998, Background Uncertainty and the Demand for Insurance Against Insurable Risks, The Geneva Papers on Risk and Insurance Theory, 23: 7-27.

Heinz Center for Science, Economics and the Environment, 2000, Evaluation of Erosion Hazards, Final report to FEMA (Contract EMW-97-CO-0375). World Wide Web: www.heinzctr.org.

Kahneman, D., and A. Tversky, 1979, Prospect Theory: An Analysis of Decision Making Under Risk, Econometrica, 47(2): 263-292.

Krantz, D. H., and H. C. Kunreuther, 2007, Goals and Plans in Decision Making, Judgment and Decision Making, 2(3): 137-168.

Kriesel, W., and C. Landry, 2004, Participation in the National Flood Insurance Program: An Empirical Analysis for Coastal Properties, Journal of Risk and Insurance, 71(3): 405-420.

Kriesel, W., A. Randall, and F. Lichtkoppler, 1993, Estimating the Benefits of Shore Erosion Protection in Ohio's Lake Erie Housing Market, Water Resources Research, 29(4): 795-801.

Kunreuther, H., 1984, Causes of Underinsurance Against Natural Disasters, The Geneva Papers on Risk and Insurance, 31: 206-220.

Kunreuther, H., 1996, Mitigating Disaster Losses Through Insurance, Journal of Risk and Insurance, 12: 171-187.

Kunreuther, H., 1998a, Introduction, in: H. Kunreuther and R. Roth, eds., Paying the Price: The Status and Role of Insurance Against Natural Disasters in the United States (Washington, DC: Joseph Henry Press).

Kunreuther, H., 1998b, Insurability Conditions and the Supply of Coverage, in: H. Kunreuther and R. Roth, eds., Paying the Price: The Status and Role of Insurance Against Natural Disasters in the United States (Washington, DC: Joseph Henry Press).

Kunreuther, H., W. Sanderson, and R. Vetschera, 1985, A Behavioral Model of the Adoption of Protective Activities, Journal of Economic Behavior and Organization, 6: 1-15.

Lee, K., 2007, Wealth, Income, and Optimal Insurance, Journal of Risk and Insurance, 74(1): 175-184.

Lewis, T., and D. Nickerson, 1989, Self-Insurance Against Natural Disasters, Journal of Environmental Economics and Management, 16: 209-223.

Marron, D. B., 2006, The Budgetary Treatment of Subsidies in the National Flood Insurance Program: Testimony before the Senate Committee on Banking, Housing, and Urban Affairs (January 25).

McClelland, G. H., W. D. Schulze, and D. L. Coursey, 1993, Insurance for Low-Probability Hazards: A Bimodal Response to Unlikely Events, Journal of Risk and Uncertainty, 7(1): 95-116.

Michel-Kerjan, E., and C. Kousky, 2010, Come Rain or Shine: Evidence on Flood Insurance Purchases in Flordia, Journal of Risk and Insurance, 77(2): 369-397.

Mossin, J., 1968, Aspects of Rational Insurance Purchasing, Journal of Political Economy, 76(4): 552-568.

Murphy, K. M., and R. H. Topel, 1986, Estimation and Inference in Two-Step Econometric Models, Journal of Business and Economic Statistics, 3(4): 370-379.

Nordhaus, W. D., 2006, The Economics of Hurricanes in the United States, American Economic Association Annual Meetings, Boston, MA, January 6.

Palm, R., 1998, Demand for Disaster Insurance: Residential Coverage, in: H. Kunreuther and R. Roth, eds., Paying the Price: The Status and Role of Insurance Against Natural Disasters in the United States (Washington, DC: Joseph Henry Press).

Pasterick, E. T., 1998, The National Flood Insurance Program, in: H. Kunreuther and R. Roth, eds., Paying the Price: The Status and Role of Insurance Against Natural Disasters in the United States (Washington, DC: Joseph Henry Press).

Pricewaterhouse Coopers, 1999, Study of the Economic Effects of Charging Actuarial Based Premium Rates for Pre-FIRM structures I. Executive Summary. Pricewaterhouse Coopers, Washington, DC, May 14.

Shao, J., and R. Sitter, 1996, Bootstrap for Imputed Survey Data, Journal of the American Statistical Association, 91(435): 1278-1288.

Schlesinger, H., 1981, The Optimal Level of Deductibility in Insurance Contracts, Journal of Risk and Insurance, 48: 465-481.

Smith, V. L., 1968, Optimal Insurance Coverage, Journal of Political Economy, 76(1): 68-77.

Tobin, J., 1958, Estimation for Relationships With Limited Dependent Variables, Econometrica, 26(1): 24-36.

Tobin, S., and C. Calfee, 2005, The National Flood Insurance Program's Mandatory Purchase Requirement: Policies, Processes, and Stakeholders (Washington, DC: American Institute for Research).

U.S. General Accounting Office (GAO), 1983, National Flood Insurance Program: Major Changes Needed if It Is to Operate Without a Federal Subsidy, RCED-83-53, Washington, DC, January.

Webster, P. J., G. J. Holland, J. A. Curry, and H.-R. Chang, 2005, Changes in Tropical Cyclone Number, Duration, and Intensity in a Warming Environment, Science, 309: 1844-1846.

Wharton Risk Management and Decision Processes Center, 2008, Managing Large-Scale Risks in a New Era of Catastrophes, University of Pennsylvania, Philadelphia, PA.

Wooldridge, J., 2001, Econometric Analysis of Cross Section and Panel Data (Cambridge, MA: MIT Press).

(1) This finding is consistent with Grace, Klein, and Kleindorfer (2004), who find a positive effect of subsidy upon demand for catastrophe and noncatastrophe home insurance in Florida and New York (the latter effect not significant for New York).

(2) The Flood Disaster Protection Act of 1973 clarified terms under which coastal erosion losses would be considered indemnified under flood insurance provisions. Erosion losses must be associated with flooding conditions in order to be covered.

(3) The NFIP is actually a cooperative venture of federal, state, and local governments and private insurers. The federal government sets flood insurance premiums, stipulates building standards, designates flood hazard areas, and certifies hazard mitigation projects. State and local governments can augment building standards, enforce building codes, and administer hazard mitigation projects. Private insurance companies sell and service flood insurance policies under the Write-Your-Own (WYO) program (Burby, 2001).

(4) There exists skepticism over whether the actuarial NFIP rate schedules accurately reflect expected loss; prior to the 2005 hurricane season (a record loss year), the NFIP exhibited a cumulative deficit of $3 billion after 37 years of operation (Wharton, 2008).

(5) The Congressional Budget Office estimates that the extent of subsidy has dropped to 25 percent of policies as of 2005 (Marron, 2006).

(6) Recent anecdotal evidence suggests that enforcement of mandatory purchase provisions in the coastal zone has improved. For the period over which we have data, however, mandatory purchase provisions were apparently not aggressively enforced.

(7) Inclusion of a loading factor in the premium to cover administrative, marketing, and capital costs is standard practice in private insurance markets. Differential loading factors across policies may also reflect an attempt to alleviate adverse selection.

(8) For the purposes of this study, the near-shore zone is defined as parcels within approximately 1,000 feet of the ocean.

(9) Details of the data collection effort are available in Heinz Center (2000).

(10) Almost 50 percent of the respondents in our data set hold no flood insurance, while consistent with the findings of Michel-Kerjan and Kousky (2010), only 7.5 percent elect for the maximum coverage of $250,000.

(11) Flood insurance rates have been generally increasing over time. Between 1998 and 2004 there were three targeted rate increases that we had to factor into our marginal premium calculations.

(12) Total premium also includes a $30 Federal Policy Fee that applies to high-risk areas, an Increased Cost of Compliance coverage premium, and a Probation Surcharge (if applicable). These additional fees do not affect the marginal premium but induce price differences on the extensive margin.

(13) Post-FIRM structures in the V-zone built between 1975 and 1981 are "grandfathered" because building standards did not take account of damage due to wave heights. The level of the subsidy is different for pre-FIRM structures and these "grandfathered" V-zone structures.

(14) Basic coverage rates on building contents apply to the first $20,000 in insurance, with higher rates applying to additional coverage up to the $100,000 limit on contents. We do not consider contents coverage in this article.

(15) Of these data, 50 percent claim structure deductible of $500 and 80 percent claim deductible of $1000 or less.

(16) Building assessed values are often outdated and procedures for reassessment vary across counties, while housing sales prices reflect both structure and land values.

(17) The erosion rates were calculated by state coastal zone managers. In some cases, managers set the erosion rate to zero if structural fortification (i.e., seawalls) were in place.

(18) The hedonic price regression results are presented in Table 4. The estimated model is used to impute housing sales price in 1997.

(19) Unfortunately, presence of shoreline protection projects was not verified by onsite inspection. Nonetheless, it is likely perception of shoreline protection that influences flood insurance decisions, so our data are appropriate in this regard.

(20) At the suggestion of an anonymous reviewer, we explored using the University of Michigan's Panel Study of Income Dynamics data set to estimate a wealth imputation equation. Having obtained data for the six states of interest in 1999, we are able to estimate OLS models for household wealth. The models exhibit [R.sup.2] of less than 10 percent, with 1999 household wealth (including and excluding real estate equity) as dependent variables and income, employment status, retirement status, education attainment, location, age of the primary owner, existence of a mortgage on the property, the number of children in the household, and state fixed effects as independent variables. Results indicate that employment status, retirement status, education level, and mortgage status are statistically significant predictors of household wealth. When we estimate Tobit regression models that includes generated wealth (with bootstrapped standard errors (Efron and Tibshirani, 1986; Shao and Sitter, 1996), we are not able to make any inference on wealth effects because the parameter estimates are not statistically significant. As such, we have chosen not to include these results, but results are available from the authors upon request.

(21) The [R.sup.2] indicates that the covariates explain 51 percent of the variation in log of housing sales prices, and the F-statistic for the model is statistically significant at the 1 percent level. All parameters have the expected sign and all are statistically significant at the 5 percent level for a Type I error, except for missing lot size, and Glynn County and Sussex County dummy variables.

(22) Sixty percent is the average value of the ratio of building assessed value to total assessed value in our data set.

(23) Results for the bootstrapped coverage models are available from authors on request.

(24) For the Tobit model, E(y|x) = [PHI] (x'[beta]/[sigma]) x'[beta] + [sigma][empty set] (x'[beta]/[sigma]).

(25) In addition, an interaction term for mortgage status and presence in the SFHA was statistically insignificant.

(26) At the suggestion of an anonymous reviewer, we explored an additional specification that included an interaction term for armor and replenish, thus identifying areas where both erosion management policies were pursued in conjunction. The estimated parameter for this term was not statistically significant.

DOI: 10.1111/j.1539-6975.2010.01380.x

Craig E. Landry is Associate Professor, Department of Economics, Interim Director, Center for Natural Hazards Research, East Carolina University. Mohammad R. Jahan-Parvar is Assistant Professor, Department of Economics, East Carolina University. The authors can be contacted via e-mail: landryc@ecu.edu and jahanparvarm@ecu.edu. We are thankful to Robert Hoyt, Andrew Keeler, Warren Kriesel, Lynne Lewis, Tom Mroz, and seminar participants at the University of Georgia and the 78th Southern Economics Association Annual Conference for their input, and to the H.J. Heinz III Center for Science, Economics, and the Environment and the Federal Emergency Management Agency for their roles in supporting and assisting in data collection. We also thank Haiyong Liu and an anonymous reviewer for their comments. This research was partially supported by a North Carolina research competitiveness grant. Craig Landry gratefully acknowledges support from the ECU Faculty Senate 2007-2008 Research/Creative Activity Grant.
TABLE 1
Coastal Counties Included in Study

                             Number     Percentage        Survey
County                      in Sample    of Total    Responses (Rate)

Brazoria, Texas                488        0.080            121
                                                         (0.248)
Brevard,Florida                547        0.090            134
                                                         (0.245)
Brunswick, North Carolina      623        0.103            282
                                                         (0.453)
Dare, North Carolina          1069        0.176            564
                                                         (0.528)
Galveston, Texas              1124        0.185            423
                                                          0.376
Georgetown, South Carolina     493        0.081            193
                                                         (0.391)
Glynn, Georgia                 326        0.054             68
                                                         (0.209)
Lee, Florida                   455        0.075            129
                                                         (0.283)
Sussex, Delaware               949        0.156            178
                                                         (0.188)
Total (Average)               6074        1.000           0.344

TABLE 2
Insurance, Parcel, and Structure Descriptive Statistics

Variable                           Definition

part (a)       NFIP participation indicator
flcov          NFIP flood insurance coverage ($100)
flratio        NFIP flood insurance coverage/asset value
prom_hi        Marginal flood insurance premium (high)
prom_lo        Marginal flood insurance premium (low)
vzone (a)      V flood zone indicator
azone (a)      A flood zone indicator
xzone (a)      X flood zone indicator
postfirm (a)   Indicator for structure built after FIRM
subsidy (a)    Indicator for subsidized insurance
elev           Elevation above BFE
brkaway (a)    Indicator for breakaway walls present below structure
obstct (a)     Indicator for obstructions present below structure
piles (a)      Indicator for structure on piles
crs            Community Ratings System classification (1998)
er             Erosion rate (feet/year)
ar             Accretion rate (feet/year)
geotime        Number of years expected before erosion reduces setback
                 to zero
return         Hurricane return interval (years)
cbra (a)       CBRA indicator
distance       Distance from the shore (feet)
ocean (a)      Oceanfront property indicator
hp             Housing sales price (1000s current $)
asset_val      Generated asset value (1000s current $); 60 percent of
                 imputed price
yearbuilt      Year structure built
yearsold       Year parcel sold
age_at_sale    Age of structure when sold
sqft           Square footage
vacant (a)     Indicator for vacant lot when sold
dcbdm          Distance from CBD (m)

Variable         N       Mean      Std. Dev.     Min.      Max.

part (a)       6,074       0.521       0.569      0          1
flcov          5,834     716.653     987.352      0       2,500
flratio        5,773       0.651       1.052      0       10.24
prom_hi        6,072       1.014       1.167    0.060     5.700
prom_lo        6,072       0.869       1.023    0.056     3.897
vzone (a)      6,074       0.505       0.570      0          1
azone (a)      6,074       0.410       0.561      0          1
xzone (a)      6,074       0.085       0.318      0          1
postfirm (a)   6,074       0.627       0.551      0          1
subsidy (a)    6,074       0.571       0.564      0          1
elev           5,881       3.305      15.751   -12.420    97.260
brkaway (a)    6,074       0.065       0.280      0          1
obstct (a)     6,074       0.184       0.442      0          1
piles (a)      6,074       0.689       0.528      0          1
crs            6,072       8.346       1.475      5         10
er             6,069       2.713       3.720      0       21.467
ar             6,074       0.191       2.068      0       29.850
geotime        6,074     787.924    8,751.90      0       287,280
return         6,074      47.110      61.779      16         190
cbra (a)       6,074       0.056       0.263      0          1
distance       6,074     318.416     270.492      0      1,593.200
ocean (a)      6,074       0.421       0.563      0          1
hp             2,844     187.177     669.815     1.000   10,000.00
asset_val      6,010     143.683     220.613     0.002    5,146.20
yearbuilt      4,632   1,973.50       19.007     1892       1998
yearsold       3,740   1,986.74       13.545     1900       1998
age_at_sale    6,074       8.207      16.024      0         94
sqft           3,930   2,276.67    3,142.42      120      20,000
vacant (a)     6,074       0.540       0.568      0          1
dcbdm          6,074   4,342.15    5,801.51     32.186   29,628.02

(a) Dummy variable; descriptive statistics are weighted to correct
for T-scale sampling scheme.

TABLE 3
Household Descriptive Statistics From Mail Questionnaire

Variable        Definition

incom           Categorical income variable
gradsch (a)     Graduate school indicator
college         College graduate indicator
hschoola        High school graduate indicator
parttime (a)    Part-time employed indicator
retired (a)     Retired indicator
age             Age of respondent
children        Number of children in the household
no_ins (a)      Indicates the individual would have purchased the
                  property regardless of whether flood insurance
                  was available.
lapse_ins (a)   Indicates flood insurance coverage has lapsed in
                  the past
claim (a)       Indicates previous flood insurance claim has been
                  submitted and settled
mort (a)        Indicates property is mortgaged
requ (a)        Indicates mortgage lender required flood insurance
                  purchase in SFHA
ero_know (a)    Indicates respondent has seen information on the
                  erosion rate at the nearest shore
armor (a)       Indicates shoreline armoring employed at the nearest
                  shore
replenish (a)   Indicates beach replenishment employed at the nearest
                  shore
primary (a)     Indicates coastal property is primary residence
vacation (a)    Indicates coastal property is vacation home
pt_rent (a)     Indicates coastal property is part-time rental
rental (a)      Indicates coastal property is full-time rental

                Explanations for not holding flood insurance (subset)

norisk (a)      Indicates respondent thinks the risk of flooding is
                  very low
notreq (a)      Indicates flood insurance not required
too_exp (a)     Indicates respondent thinks flood insurance is too
                  expensive
notavail (a)    Indicates that flood insurance is perceived as not
                  available

                                   Std.
Variable          N      Mean      Dev.     Min.   Max.

incom           1,711   101.431   105.838    20    250
gradsch (a)     1,798     0.357     0.675     0      1
college         1,798     0.436     0.699     0      1
hschoola        1,798     0.206     0.570     0      1
parttime (a)    1,789     0.049     0.305     0      1
retired (a)     1,789     0.456     0.702     0      1
age             1,775    61.208    17.323    25     95
children        1,899     0.462     1.738     0     13
no_ins (a)      1,715     0.681     0.665     0      1
lapse_ins (a)   1,643     0.111     0.451     0      1
claim (a)       1,899     0.102     0.437     0      1
mort (a)        1,825     0.390     0.690     0      1
requ (a)          629     0.114     0.378     0      1
ero_know (a)     ,899     0.281     0.649     0      1
armor (a)       1,899     0.192     0.569     0      1
replenish (a)   1,899     0.349     0.688     0      1
primary (a)     1,814     0.240     0.600     0      1
vacation (a)    1,814     0.350     0.671     0      1
pt_rent (a)     1,814     0.307     0.648     0      1
rental (a)      1,814     0.101     0.424     0      1
norisk (a)        292     0.248     1.058     0      1
notreq (a)        292     0.200     0.980     0      1
too_exp (a)       292     0.300     1.123     0      1
notavail (a)      292     0.088     0.696     0      1

(a) Dummy variable; descriptive statistics are weighted to correct for
T-scale sampling scheme and overrepresentation of flood insurance
participants.

TABLE 4
Hedonic Price Regression Model

Variable                  Coefficient       Standard Error

sqft                       2.53E-4 ***          2.26E-5
[sqft.sup.2]              -7.55E-9 ***          1.04E-9
no_sqft                    0.1901 **            0.0854
lotsize                    1.58E-5 ***          1.96E-6
[lotsize.sup.2]           -3.29E-11 ***         6.96E-12
no_lotsize                 0.0828               0.1784
age_at_sale               -0.0076 ***           0.0013
vacant                    -0.4956 ***           0.0506
ocean                      0.4642 ***           0.0375
distance_CBD              -2.354E-5 ***         3.87E-6
glyn_GA                    0.2220               0.1211
suss_DE                    0.1536               0.0822
dare_NC                   -0.4872 ***           0.0669
brev_FL                   -0.6178 ***           0.0664
geor_SC                   -0.3683 ***           0.0959
brun_NC                   -0.4929 ***           0.0729
galv_TX                   -0.7462 ***           0.0690
braz_TX                   -1.2398 ***           0.2004
Constant                  12.1179 ***           0.0942

Year dummy variables         Yes
N                           2002
[R.sup.2]                  0.5163
F (p-value)            59.97 (p < 0.0001)

*** Statistically significant for 1 percent probability of Type I
error; ** Statistically significant for 5 percent probability of
Type I error; excluded county dummy variable is Lee County, FL.

TABLE 5
Tobit Ratio Model Results (High Premium)

                                Model 1

Variable               Coefficient   Standard Error

Premium                -0.773 ***        0.314
Subsidy
Prem x sub
vzone                   0.604 ***        0.040
xzone                  -0.119 *          0.062
Return                 -0.038 ***        0.014
er                      0.031 ***        0.006
ar                      0.027 ***        0.010
Armor
Replenish
Mort
Prem x mort
In(income)
Retired
College
hschool
Constant                1.042 ***        0.235
Sigma                   0.663 ***        0.026

County fixed effects              Yes
Weight                          T-scale
lnL                             -8,593
LRT (df) p                1,242 (14) < 0.0001
N                                5,766

                                Model 2

Variable               Coefficient   Standard Error

Premium                -0.392 ***        0.029
Subsidy                 1.001 ***        0.046
Prem x sub             -1.926 ***        0.063
vzone                   0.826 ***        0.038
xzone                  -0.251 ***        0.056
Return                  0.038 **         0.015
er                      0.029 ***        0.006
ar                      0.017 *          0.009
Armor
Replenish
Mort
Prem x mort
In(income)
Retired
College
hschool
Constant               -0.405 *          0.244
Sigma                   0.524 ***        0.020

County fixed effects               Yes
Weight                           T-scale
lnL                              -7,953
LRT (df) p                 2,522 (16) < 0.0001
N                                 5,766

                                Model 3

Variable               Coefficient   Standard Error

Premium                 -0.187 ***       0.027
Subsidy
Prem x sub
vzone                   -0.053           0.042
xzone                   -0.188 **        0.074
Return                   0.019 *         0.010
er
ar
Armor                   -0.055           0.042
Replenish                0.184 ***       0.037
Mort
Prem x mort
In(income)
Retired
College
hschool
Constant                 0.304 *         0.178
Sigma                    0.392 ***       0.025

County fixed effects               Yes
Weight                  T-scale and nonparticipant
lnL                               -2,629
LRT (df) p                  424 (14) < 0.0001
N                                 1,668

                                Model 4

Variable               Coefficient   Standard Error

Premium                -0.188 ***        0.030
Subsidy
Prem x sub
vzone                  -0.058            0.042
xzone                  -0.176 **         0.074
Return                  0.014            0.010
er
ar
Armor
Replenish
Mort                    0.065            0.043
Prem x mort             0.106 ***        0.039
In(income)              0.037            0.023
Retired                -0.060 *          0.035
College                 0.010            0.034
hschool                -0.128 ***        0.049
Constant                0.024            0.337
Sigma                   0.378 ***        0.025

County fixed effects                Yes
Weight                   T-scale and nonparticipant
lnL                               -2,200
LRT (df) p                   424 (18) < 0.0001
N                                  1,469

*** Statistically significant for 1 percent probability of Type I
error; ** Statistically significant for 5 percent probability of Type
I error; * Statistically significant for 10 percent probability of Type
I error.

TABLE 6
Tobit Ratio Model Results (Low Premium)

                                Model 1

Variable               Coefficient   Standard Error

Premium                -1.392 ***        0.045
Subsidy
Prem x sub
vzone                   0.743 ***        0.039
xzone                  -0.134 **         0.058
Return                  0.014            0.014
er                      0.032 ***        0.006
ar                      0.025 ***        0.009
Armor
Replenish
Mort
Prem x mort
ln(income)
Retired
College
hschool
Constant                0.398 *          0.229
Sigma                   0.619 ***        0.023

County fixed effects              Yes
Weight                          T-scale
lnL                             -8,247
LRT (df) p                1,934 (14) < 0.0001
N                                5,766

                                 Model 2

Variable                Coefficient   Standard Error

Premium                 -0.716 ***        0.045
Subsidy                  0.873 ***        0.045
Prem x sub              -1.697 ***        0.070
vzone                    0.792 ***        0.036
xzone                   -0.227 ***        0.055
Return                   0.050 ***        0.015
er                       0.028 ***        0.006
ar                       0.016 *          0.009
Armor
Replenish
Mort
Prem x mort
ln(income)
Retired
College
hschool
Constant                -0.520 **         0.243
Sigma                    0.523 ***        0.020

County fixed effects               Yes
Weight                           T-scale
lnL                               -7,925
LRT (df) p                 2,578 (16) < 0.0001
N                                 5,766

                                  Model 3

Variable                Coefficient   Standard Error

Premium                 -0.456 ***        0.040
Subsidy
Prem x sub
vzone                    0.043            0.041
xzone                   -0.176 **         0.072
Return                   0.027 ***        0.010
er
ar
Armor                   -0.039            0.041
Replenish                0.180 ***        0.036
Mort
Prem x mort
ln(income)
Retired
College
hschool
Constant                 0.232            0.177
Sigma                    0.390 ***        0.025

County fixed effects               Yes
Weight                  T-scale and nonparticipant
lnL                              -2,585
LRT (df) p                  512 (14) < 0.0001
N                                 1,668

                                 Model 4

Variable               Coefficient   Standard Error

Premium                -0.429 ***        0.044
Subsidy
Prem x sub
vzone                   0.021            0.042
xzone                  -0.166 **         0.072
Return                  0.020 *          0.010
er
ar
Armor
Replenish
Mort                    0.042            0.044
Prem x mort             0.167 ***        0.057
ln(income)              0.026            0.022
Retired                 0.056            0.035
College                 0.021            0.033
hschool                -0.114 **         0.049
Constant                0.112            0.334
Sigma                   0.376 ***        0.025

County fixed effects              Yes
Weight                 T-scale and nonparticipant
lnL                             -2,169
LRT (df) p                 486 (18) < 0.0001
N                                1,469

*** Statistically significant for 1 percent probability of Type I
error; ** Statistically significant for 5 percent probability of Type
I error; * Statistically significant for 10 percent probability of
Type I error.

TABLE 7
Tobit Ratio Model Results (Categorical Income Variables)

                              High Premium

Variable               Coefficient   Standard Error

Premium                 -0.195 ***       0.029
vzone                   -0.050           0.041
xzone                   -0.183 **        0.073
Return                   0.011           0.010
Mort                     0.052           0.042
Prem x mort              0.105 ***       0.038
Inc_30-49.9              0.139 **        0.059
Inc_50-74.9              0.090           0.058
Inc_75-99.9              0.205 ***       0.061
Inc_100-149.9            0.266 ***       0.058
Inc_150-199.9            0.178 **        0.069
Inc_200-249.9            0.187 **        0.080
Inc_gte250               0.090           0.059
Retired                 -0.063 *         0.034
College                  0.012           0.032
hschool                 -0.119 **        0.048
Constant                 0.358 *         0.185
Sigma                    0.362 ***       0.024

County fixed effects              Yes
Weight                 T-scale and nonparticipant
InL                             -2,297
LRT (df) p                 230 (24) < 0.0001
N                                1,469

                               Low Premium

Variable               Coefficient   Standard Error

Premium                 -0.447 ***       0.044
vzone                    0.032           0.040
xzone                   -0.127           0.100
Return                   0.016 *         0.010
Mort                    -0.055           0.127
Prem x mort              0.085           0.131
Inc_30-49.9              0.173 ***       0.056
Inc_50-74.9              0.124 **        0.058
Inc_75-99.9              0.090           0.057
Inc_100-149.9            0.192 ***       0.060
Inc_150-199.9            0.242 ***       0.057
Inc_200-249.9            0.158 **        0.068
Inc_gte250               0.161 **        0.079
Retired                  0.065           0.058
College                 -0.057 *         0.033
hschool                  0.025           0.032
Constant                 0.338 *         0.185
Sigma                    0.362 ***       0.024

County fixed effects              Yes
Weight                 T-scale and nonparticipant
InL                              -2,261
LRT (df) p                 302 (24) < 0.0001
N                                1,469

*** Statistically significant for 1 percent probability of Type I
error; ** Statistically significant for 5 percent probability of
Type I error; * Statistically significant for 10 percent probability
of Type I error.
COPYRIGHT 2011 American Risk and Insurance Association, Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2011 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Landry, Craig E.; Jahan-Parvar, Mohammad R.
Publication:Journal of Risk and Insurance
Article Type:Survey
Geographic Code:1U5FL
Date:Jun 1, 2011
Words:13881
Previous Article:Forecasting in an extended chain-ladder-type model.
Next Article:Consumer response to a national marketplace for individual health insurance.
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters