Graduated Flood Risks and Property Prices in Galveston County
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Graduated Flood Risks and Property Prices in Galveston County Ajita Atreya Jeffrey Czajkowski Wharton Risk Center Wharton Risk Center University of Pennsylvania University of Pennsylvania July 1, 2015 Working Paper # 2015-09 _____________________________________________________________________ Risk Management and Decision Processes Center The Wharton School, University of Pennsylvania 3730 Walnut Street, Jon Huntsman Hall, Suite 500 Philadelphia, PA, 19104 USA Phone: 215‐898‐5688 Fax: 215‐573‐2130 http://www.wharton.upenn.edu/riskcenter ___________________________________________________________________________
THE WHARTON RISK MANAGEMENT AND DECISION PROCESSES CENTER Established in 1984, the Wharton Risk Management and Decision Processes Center develops and promotes effective corporate and public policies for low‐probability events with potentially catastrophic consequences through the integration of risk assessment, and risk perception with risk management strategies. Natural disasters, technological hazards, and national and international security issues (e.g., terrorism risk insurance markets, protection of critical infrastructure, global security) are among the extreme events that are the focus of the Center’s research. The Risk Center’s neutrality allows it to undertake large‐scale projects in conjunction with other researchers and organizations in the public and private sectors. Building on the disciplines of economics, decision sciences, finance, insurance, marketing and psychology, the Center supports and undertakes field and experimental studies of risk and uncertainty to better understand how individuals and organizations make choices under conditions of risk and uncertainty. Risk Center research also investigates the effectiveness of strategies such as risk communication, information sharing, incentive systems, insurance, regulation and public‐private collaborations at a national and international scale. From these findings, the Wharton Risk Center’s research team – over 50 faculty, fellows and doctoral students – is able to design new approaches to enable individuals and organizations to make better decisions regarding risk under various regulatory and market conditions. The Center is also concerned with training leading decision makers. It actively engages multiple viewpoints, including top‐level representatives from industry, government, international organizations, interest groups and academics through its research and policy publications, and through sponsored seminars, roundtables and forums. More information is available at http://wharton.upenn.edu/riskcenter .
Graduated Flood Risks and Property Prices in Galveston County Abstract A number of hedonic property pricing studies of flood risk indicate that properties within a designated higher flood risk zone sell for a lower price than an equivalent property outside of it. However, often the homes most at flood risk are also the most desirable in terms of their proximity to the water, and this concurrent existence of positive water-related amenities and negative flood risk may be problematic to parse out in a hedonic estimation. Our hedonic property analysis approach in Galveston County, Texas aims at estimating the impacts of flood risk and water-related amenities in a more systematic way by interacting distance to the nearest coastline and flood risk in order to account for these impacts acting together on housing sales prices in our coastal community. Further, we employ a more granular view of the flood risk split by varying flood risk return periods to allow for a more meaningful interaction between the negative and positive amenities related to proximity to the water. Results show that the hedonic price effect is dependent upon the distance to the nearest coastline, and as expected the distance effect varies by flood risk type. We find that in this coastal housing market properties located in the highest risk flood area, for up to nearly a quarter mile from the nearest coastline, actually command a price premium. A recent movement toward risk-based flood insurance premiums in the United States was deeply opposed by the real estate sector for fear of causing property values to steeply decline. This analysis sheds some further light on this depressed property value assertion highlighting its sensitivity to distance to the water. Key words: hedonic housing prices, flood risk, return periods, water amenity, Galveston Texas, 1
I. Introduction Since 1968 homeowners’ flood insurance in the United States has been mainly provided through the federally-run National Flood Insurance Program (NFIP), which as of 2014 has 5.34 million NFIP policies-in-force nationwide with a total of $1.27 trillion of insured coverage (FEMA, 2014). To set premiums, the NFIP maps participating communities via Flood Insurance Rate Maps (FIRMs), designating flood risks through different flood zones. A number of hedonic property pricing studies of flood risk indicate that properties within a designated higher flood risk zone sell for a lower price than an equivalent property outside of it, typically on the order of 4 to 12 percent (Bin and Polasky 2004; Bin, et al. 2008a; Kousky 2010; Posey and Rogers, 2010; Bin and Landry 2012). This negative flood risk price differential is characteristically attributed to higher flood insurance rates (the discounted sum of future flood insurance payments) being capitalized into housing sales prices in the higher flood risk zones (Bin et al. 2008a; Bin and Landry 2012).1 However, an estimated hedonic price discount for location in a higher flood risk zone does not always hold (USACE, 1998; Bin and Kruse, 2006; Morgan, 2007; Daniel et al., 2009). For example, Daniel et al. (2009) in their meta-analysis of 19 studies and 117 point estimates of the implicit price for location in the 100 year flood plain find that the estimates vary considerably, anywhere from -52% to +58%. Furthermore, with any estimated hedonic flood risk price differential two complications potentially arise (Daniel et al. 2009): subjective bias of the objective flood risk including a complete lack of awareness of the risk; and the concurrent existence of positive water-related amenities and negative flood risk inherent in living near to the water. In this study in addition to using the traditional flood zones as a measure of objective flood risk we incorporate a more comprehensive measure of flood risk through identified flood return periods 2
and aim to determine if the potential effects of higher flood risks is offset by coastal amenities conferred on a property from being located close to water using hedonic property analysis. We allow for the flood risk and the coastal amenity to act jointly on housing sales price in Galveston County, a coastal community in Texas. . As other studies (Kousky 2010; Bin and Landry 2012; and Atreya et. al. 2013 ) find no evidence of flood risk impacting housing prices until the occurrence of a major event we also demonstrate the robustness of our results to the potential subjective bias associated with the occurrence of Hurricane Ike in 2008 which made landfall in Galveston. 2 Often the homes most at flood risk are also the most desirable in terms of their proximity to the water. Hence, by only capturing the location of a home in or outside of the floodplain through an indicator variable in the hedonic estimation (where 1= location within the high risk floodplain such as whether a property lies in the 100 or 500 year return periods, and 0 = location outside of it) may bias the estimation of flood risk. Bin et al. (2008a) and Daniel et al. (2009) explicitly discuss the importance of controlling for the positive amenity values related to water proximity, and utilize distance measures of proximity to the water in their estimations. Others also control for positive water-related amenities through distance measures in their hedonic flood risk studies (Kousky, 2010; Conroy and Milosch, 2011; Bin and Landry 2012; Atreya et al., 2013). 3 Although the distance to water measure is incorporated to account for the positive water amenity effects, these studies do not vary the amenity effects by flood risk type in their estimations. In other words, any change in distance to or from the water has an analogous impact across high to low flood risk types. But intuitively this risk-amenity trade-off would seem to vary by flood risk type, where a high flood risk area would be penalized more for the loss of the positive amenity given that the high flood risk remains. Here we capture these distance varying risk-amenity trade- 3
offs through interaction terms of flood risk and distance that have rarely been employed to our knowledge.4 Regardless of controlling for water-related positive amenity values, the hedonic literature’s inconclusiveness concerning a home price discount stemming from a negative flood risk may be partially attributed to the tendency to capture flood risk in the empirical analysis through the use of the aforementioned indicator dummy variable where 1= location within the high risk floodplain, and 0 = location outside of it (Bin and Polasky 2004; Bin and Kruse 2006; Morgan, 2007; Bin et al. 2008a; Kousky 2010; Posey and Rogers, 2010; Bin and Landry 2012; Atreya et al., 2013). This dummy indicator structure implies that the hedonic price flood risk discount is constant across the floodplain (USACE, 1998) despite the fact that the flood risk clearly is not. For example, within an identified 100-year flood zone the flood risk can vary from a 10 year return period (10% probability of occurrence in any given year) to a 100 year return period (1% probability of occurrence in any given year). And NFIP insurance premiums also vary within a particular aggregate flood zone via the elevation of the first floor of the dwelling in relation to the 100-year return period, indicating hedonic price discounts (capitalized insurance payments) should vary by varying risk within a flood zone. Czajkowski et al. (2013) and Michel-Kerjan et al. (2014) use flood catastrophe models in two Texas communities – including Galveston County - to show not only how much flood risk varies within a single NFIP designated flood risk zone, but by how much corresponding non-NFIP probabilistically derived localized risk-based flood insurance rates would vary as well.5 In order to capture this inherent varying flood risk within a given flood map zone Griffith (1994) included a flood frequency (i.e., return period) variable in her analysis and found that there is a hedonic price discount only for those homes deep in the 100 year floodplain due to their higher 4
annual probability of occurrence. A number of other studies have utilized measures of elevation or flood depth in lieu of, or in addition to, a flood risk zone indicator variable to account for the spatially inherent varying flood risk (Barnard, 1978; Tobin and Montz, 1994; Kriesel and Friedman, 2002; Zhai et al., 2003; Kousky, 2010; McKenzie and Levendis, 2010; and Atreya et al., 2013), and typically find a statistically significant relationship in the hedonic estimation. However, often the employed measure of elevation does not necessarily best convey the flood risk which would be relative to the base flood elevation, i.e., the computed elevation to which floodwater is anticipated to rise during the flood having a one percent chance of being equaled or exceeded in any given year (FEMA, 2014). For example, McKenzie and Levendis (2010) utilize elevation in relation to the mean sea-level, whereas Kousky (2010) and Atreya et al. (2013) simply control for the elevation of the ground, not in relation to the base flood elevation level. Here in addition to using the traditional flood zone indicator dummy variables, we utilize data provided to us by CoreLogic, a large real estate data provider that identifies the varying flood risk return periods within the classified FIRM flood zones. Overall then, our hedonic property analysis approach in Galveston County, Texas aims to determine countervailing impacts of flood risk and water-related amenities in a more systematic way than has previously been employed (Daniel et al., 2009). We first control for positive amenities associated with water proximity by using spatial analysis in Arc-GIS to calculate for each property their distance to the nearest coastline. We then interact distance and flood risk in our hedonic estimations in order to account for these impacts acting together on housing sales prices in our coastal community. Finally we employ a more granular view of flood risk split by varying flood risk return periods to allow for a more meaningful interaction between the negative and positive amenities related to proximity to the water than would otherwise be achieved with an 5
aggregate flood zone indicator variable. We feel it is reasonable to believe that homeowners in a 100 year floodplain are able to discern the varying levels of flood risk within a particular flood zone, especially in a frequently flooded area such as Galveston County. This belief is further supported by the statistical significance of elevation or flood depth variables utilized in the existing hedonic literature already cited. Our results show that properties located in the high risk flood areas command a price premium compared to those located outside. But this hedonic price premium is dependent upon the distance to the coast, and as expected the distance effect varies by flood risk type. This illustrates the importance of capturing these two variables acting together through the interaction terms, otherwise the flood risk is misestimated. We show that the misestimation is more pronounced when we utilize a more granular flood risk and distance interaction with the varying return period risk information inherent to each zone, highlighting the need for using more detailed risk information when available. Overall, we find that properties located in the highest risk flood area, for up to nearly a quarter mile from the nearest coastline, actually command a price premium. Importantly results are also robust to the potential subjective bias associated with the occurrence of Hurricane Ike in 2008. In 2012, Congress passed the Biggert-Waters Flood Insurance Reform Act (BW-12) in order to address a number of the well-documented structural and fiscal issues of the program, including key provisions of the bill that would increase existing discounted premiums to full-risk levels. However, BW-12 was itself reformed in March 2014 with the passage of Homeowner Flood Insurance Affordability Act (HFIAA-14) that importantly curbed many of the planned BW- 12 rate increases. Realtors, homebuilders, and lenders had provided steep opposition to BW-12 (WSJ, 2013) decrying the movement toward risk-based premiums as causing “property values to 6
steeply decline and made many homes unsellable, hurting the real estate market” (Insurance Journal, March 2014). Importantly, our results shed some further light on this depressed property value assertion highlighting the sensitivity of the assertion as being dependent upon the sufficient disentanglement of the coincident flood risk and water amenity effects. That is, even despite our systematic interaction approach coupled with more detailed flood risk data, we find that in the absence of a significant flood event it is difficult for the negative flood amenity value based upon the objective flood risk to sufficiently counteract a homeowner’s strong desire to live near the water. The remainder of the paper proceeds as follows: section two provides an overview of the Galveston County study area as well as the details of the data utilized in our hedonic analysis; section three lays out the methods we employ while the corresponding main results are presented in section four; robustness of these results are presented in section five; and finally section six concludes. II. Study Area and Data We focus our study in Galveston County, Texas exposed to both riverine and storm-surge flooding. Property transaction data for single family homes in Galveston County was provided by CoreLogic. After cleaning the dataset6 we retained 35,586 property sales for our analysis between the years 2001 and 2010, including only the most recent sale in the analysis. Figure 1 illustrates the location of these sales by our aggregated FEMA designated flood zones V, A, X500, and X. 7 V and A zones represent high risk - 1 percent or greater annual chance of flooding – in coastal and non-coastal areas respectively. Whereas X500/B and X/C zones represent moderate to minimal flood risk areas. While the number of sales in any one year is largest in the X/C zone, i.e., minimal flood risk areas, on average there were 381 sales per year in the high risk flood areas (V and A zones) 7
and 627 home sales per year in the moderate flood risk area (X500/B) during this timeframe (Figure 2). We adjusted all sales prices to 2010 values utilizing the housing price index for Houston–The Woodlands-Sugar Land, Texas metropolitan statistical area from the office of Federal Housing Finance Agency (FHFA, 2014). In a coastal community such as Galveston, an important amenity measure that affects the property price is proximity to nearest coastline (Bin et al., 2008a; Daniel et al. 2009). Using spatial analysis in Arc-GIS we calculated for each property their Euclidean distance to the nearest coastline. As would be expected given their relative proximity to the coastal waterfront as shown in Figure 1 above (and from Table I), homes in the V and A zones sell for more on average after accounting for the size of the home with sales prices per square foot of $198.51, $115.92, $86.72, and $92.75 for the V, A, X500/B, and X/C zones respectively. However, while homes located in V and A zones are in the aggregate relatively closer to the water as compared to those located in X500 and X zones (approximately 98 and 66 percent of V and A zone homes sales are within 1 mile of the nearest coastline as compared to 31 and 3 percent of X/500 and X zone home sales), there is still a significant amount of variation of distance to the nearest coast within each zone. Figure 3 provides the average sales price per square foot split by distance to the nearest coastline in 500 to 1000 foot increments overlaid with an A zone linear trendline. From this view of our sales price data we see that each flood zone has homes relatively close the water, but as distance from the nearest coastline increases, sales price per square foot within each flood zone generally declines as expected. Furthermore, not only does the distance to the nearest coastline vary within a flood zone, i.e., the positive amenity value, but so does the negative amenity flood risk value vary within an aggregated zone. CoreLogic determines the associated relative flood risk (i.e., flood return 8
periods) for each home based upon a proprietary scoring of a property’s elevation variance (EV) and distance to the most immediate water flood hazard. Where EV is the difference between the elevation of the ground upon which the immediate structure rests and the elevation of the flood plain that presents the greatest flood risk such as the base flood elevation for homes in the A zone. The specific discretized flood return periods (RPs) determined by CoreLogic that we use for our hedonic analysis are: RP ≤ 10; 10 < RP ≤ 25; 25 < RP ≤ 50; 50 < RP ≤ 100; 100 < RP ≤ 250; 250 < RP ≤ 500; 500 < RP ≤ 1000; 1000 < RP ≤ 5000; and RP > 5000. As an example of just how much the RPs vary within a particular flood zone, we highlight the distribution of the Galveston County A zone home sales by the various return periods within this 100 year flood zone. While all of these homes are located within the 100 year flood plain equating to a 1 percent annual chance flood event, some of these homes are at a higher risk of flooding based upon their individual location within the zone. For example, homes subject to at least a 10 percent chance of a flood event, or a 10 year return period comprise 14 percent of the total sales from 2001 to 2010. Overall, nearly 80 percent of the home sales from 2001 to 2010 were at a 50 year return period or less (10 < RP ≤ 25 = 37 percent and 25 < RP ≤ 50 = 27 percent). The V, X500 and X zones have similar varying home sale flood risk (RP) distributions. We present in Figure 4 sales price per square foot by a combined view of flood risk return periods up to 100 years and distance to the coast from 0 to 5000 feet. From here we can see that riskier homes (RP 25 or less) in general have higher prices per square foot compared to less risky homes across all of the defined distance bands. However, where this difference is greatest is for homes directly on the water (0 to 500 feet), but the price differences become less as distance to the coast increases (from Table I we see that homes in RP 25 or less are on average closer to the water, nearly all within a half a mile distance on average). In fact, we begin to see an uptick in the sales 9
price per square foot in the 50 to 100 RP area as the distance to coast increases. Clearly then there is a variation in price per square foot depending upon the interaction of distance to the nearest coast and flood risk, or the varying trade-off between positive and negative water-related amenities dependent upon the particular flood risk.8 And it is precisely this interaction we isolate in our hedonic framework. Beyond the square footage of the home, flood risk return period, and distance from the nearest coastline there are a number of other relevant housing attributes – structural, location/neighborhood, etc - that consistently impact property sales prices and we include these in our hedonic framework. Specifically we include the following structural attributes in the estimated hedonic price functions: land square footage, building square footage, the number of stories, exterior wall type with brick = 1 and 0 = otherwise (aluminum, asbestos, brick veneer, brick/wood, concrete, concrete block, frame wood, metal, stone, stucco, tilt-up, wood frame) type of foundation with slab-on-grade = 1 and 0 = otherwise (wood, concrete, concrete block, pier, pipe/iron) , a dummy indicator variable for the condition of the home with 1 = excellent and 0 = otherwise (Poor, Average, Fair, Good, Very Good) and age of the property at the time of sale. In addition to the distance to the coast we also use spatial analysis in Arc-GIS to calculate for each property their Euclidean distance to the nearest park, bus route, railroad and school. The demographic characteristics, such as the median household income and the percent of nonwhite population was determined at the census tract level using 2000 census data. As properties that are built after a community joins the NFIP require the lowest floor of the residential building to be elevated above the base flood elevation, we include a dummy variable, NFIP=1 if the property was built after 1974 (i.e. after the communities in Galveston County joined NFIP) and 0 otherwise. Finally, we include 10
another dummy variable seawall=1 if the properties were protected by the Galveston seawall and 0 otherwise.9 Table I provides the summary statistics of the variables used in the analysis. The average selling price in our sample is $181,690 with a typical home about 20 years old and 2,150 square feet. About 5.5 percent of the homes sold are located in V zone and 16 percent of the homes sold are located in A zone. 6 percent of the homes sold are located in the flood return period less than 10 years. On average the distance to coast is 25,694 feet and 2.4 percent of the houses are located within 500 feet of the coast. 82 percent of the houses in our sample were built after the community joined NFIP and 11 percent of the houses are protected by seawall. As illustrated earlier, there is much variation in the price of the properties within the FEMA designated flood zones and the price variation also depends on the proximity to the coast. We further split the flood zone and return period by mean price as well as mean coastal distance at the end of Table I. We also present the average price of the properties split by their distance to the coast. III. Methods This study aims at disentangling the countervailing impacts of flood risk and coastal water amenities as reflected in the property prices in Galveston County TX. We employ hedonic models (Rosen 1974; Freeman 2003) that have been extensively used in the past to partition out the value of an environmental amenity or disamenity using the actual property transaction data based on the notion that the component values of various attributes of heterogeneous goods are reflected in price differentials. The standard hedonic model can be represented as: P=f (S, L, R, C). In a hedonic model, price of the property (P) is modeled as a function of its structural attributes (S) such as building square feet, age, number of bathrooms; location attributes 11
(L) such as distance to road, distance to park; and the environmental variables of interest which in our case is coastal amenity as measured by distance to the nearest coast (C) and flood risk (R). The first order differentiation of price, (P), with respect to the housing attributes provide the marginal implicit price which can be interpreted as the marginal willingness to pay for that attribute. For flood risk (R) we use both the flood zones as given by FEMA’s flood insurance rate maps as well as a more granular measure of flood risk through specific flood return periods Use of FEMA designated flood risk zones in hedonic model First, we incorporate flood risk (R) in our model using traditional FEMA designated flood hazard zones: the V, A, X500, and X zones. Distance to coast is an important factor that affects the price of a property (Bin et al. 2008a; Daniel et al. 2009; Kousky, 2010; Conroy and Milosch, 2011; Bin and Landry 2012; Atreya et al., 2013). To allow the implicit price of coastal amenity to vary between properties locating in various flood risk zones, the distance to coast variable enters the hedonic model in three ways. First, in model (1), we include the natural log of distance to the nearest coastline to capture the diminishing marginal returns as the distance to the coast increases. The model is as follows: J K L log( Pit ) o j S ijt k Likt l Ril m ln(C ) i i t it (1) j 1 k 1 l 1 where, log(Pit) is the natural log of the price for observation i in time t, Sijt is the jth structural attribute for observation i in time t, Likt is the kth locational attribute for observation i in time t, Ril is the lth risk variable for observation i which in our case include dummy variables for V, A and X500 zones, i.e. incorporates three dummy variables equal to 1 if the property falls within the designated V, A and X500 zones and 0 otherwise. The properties that fall in X zone are the control 12
groups. Ln(C)i is the distance-to-coast variable for observation i. The estimate from the natural log of coastal distance (C) is an average premium effect measured at the mean. We expect the effect of distance-to-coast to be non-linear and thus secondly, we construct distance dummy variables to capture discrete distance effects as follows: J K L M log( Pit ) o j S ijt k Likt l Ril m .Cim i t it (2) j 1 k 1 l 1 m 1 In model (2), variable Cim a dummy variable denoting a specific distance from the coast for each observation i. The distance dummy variable are adjusted to capture smaller distance effects, i.e., within 500 feet, between 500 and 1000 feet, between 1000 and 2000 feet and so on. While model (2) takes into account the non-linearity, we believe that this decay will likely differ conditional upon the varying flood risk types. To test the conditional impact of distance to coast and flood risk zones, in model (3), we interact the continuous natural log of distance to coast ln(C)i variable with the flood risk variables Ril to account for the positive and negative amenity variables acting together on housing values as follows: J K L N L log( Pit ) o j S ijt k Likt l Ril m ln(C ) i n (ln(C ) i * Ril ) i t it j 1 k 1 l 1 n 1 l 1 (3) In model (1), the expected marginal value of being located in the flood risk zone is equivalent to βl and the decay in premium as the distance to coast increases is constant at βm for all the flood risk zones. However, in equation (3), the expected marginal value is a function of risk and amenity acting together and varies by the flood risk zone and distance to the coast. The marginal value of being located in the flood zone (V zone, for example) is equivalent to βl *Vzone +βn*Vzone*ln 13
(coast) where Vzone is equal to 1. The total value of being located in the flood zone is however, equal to βl + βm +βn. In all the three models, we include fixed effect dummies for zip code and year denoted by γi and δt respectively to control for potential sub-market housing effect within Galveston County. As we adjusted all sales prices to 2010 values utilizing the FHFA housing price index for Houston– The Woodlands-Sugar Land, TX MSA, we utilize this index to identify the fixed effect time segmentations in our data (δt). Specifically we apply a segmented regression methodology to the quarterly Houston–The Woodlands-Sugar Land FHFA HPI values from 2001 to 2010 in order to identify the unknown structural breakpoints in time for this housing market (Figure A.1, appendix). Segmented or piecewise regression allows the detection of single or multiple change points at unknown points in time (Muggeo, 2003). We detect five change points in the HPI data as illustrated in the appendix: 1) 2001 between the 2nd and 3rd quarters; 2) 2004, between the 3rd quarter and the 1st quarter of 2005; 3) 2007 between the 2nd and 3rd quarters; 4) 2009 between the 1st and 2nd quarters; and 5) 2009 between the 3rd quarter and the 1st quarter of 2010. Given these identified housing market breakpoints we account for their potential time effect in our estimations by creating time interval dummy variables (δt) to represent each of them.10 Use of flood risk return periods (RP) in hedonic model In order to account for the inherent variation in the flood risk within any designated flood zone we ran separate regressions using the return periods. The specific discretized flood return periods (RPs) that we use for our hedonic analysis are: RP ≤ 10; 10 < RP ≤ 25; 25 < RP ≤ 50; 50 < RP ≤ 100; 100 < RP ≤ 250; 250 < RP ≤ 500; and RP > 500 (omitted category equating to X zone). We ran three variations of models (1), (2), and (3) as explained above by replacing the FEMA designated flood hazard zones with the RP dummy variables, i.e. a dummy equal to 1 if property i 14
fall in RP ≤ 10 and 0 otherwise and so on. We note that according to the CoreLogic data there are more than 13,000 properties within the Galveston County X zone that have a return period less than 500 years when in theory zone X is the area determined to be outside the 500 year flood. This discrepancy is probably due to FEMA flood hazard maps not always being completely in sync with the actual flood return period (Czajkowski et al. 2013). However, for other zones (V /A /X500) the return periods are comparable. For the robustness of our above methods we perform two further analyses: 1) we examine the possibility of subjective bias in our estimates using a Difference-in-Difference (DD) model related to the occurrence of Hurricane Ike that made landfall in Galveston in September 2008 as category 2 hurricane ; and 2) we estimate spatial hedonic models to account for the possible spatial dependence among the neighboring properties. A Difference-in-Difference (DD) Model to account for the impact of Hurricane Ike The other main complication in the hedonic pricing of flood risk is that subjective bias of the objective flood risk may exist in the community and thus flood risk will be not be a significant attribute of housing prices (Chivers and Flores, 2002; Daniel et al., 2009; Bin and Landry 2012). Often this bias is enhanced (reduced) by the occurrence (non-occurrence) of a significant flood event (Kousky, 2010; Bin and Landry, 2012; Atreya et al. 2013). To examine the impact of Hurricane Ike, if any, which made landfall in Galveston in September 2008 and caused extensive flooding damage in Galveston County we estimate; i) a Difference-in-Difference (DD) model; and ii) a separate model for only those home sales pre-Ike from 2001 to 2008. A DD design allows us to isolate the effect attributable to flood from other contemporaneous variables since the control group -which in or case will be the properties in the low flood risk zone/high return periods - will 15
experience most of the contemporaneous impacts but offer lower flood risk. We use the following specification for the DD model: J K L N L log( Pit ) o j S ijt k Likt l Ril m ln(C ) i n (ln(C ) i * Ril ) j 1 k 1 l 1 n 1 l 1 P L (4) ike p (ike * Ril ) i t it p 1 l 1 where, variable ike is equal to 1 if the property was sold after Hurricane Ike made landfall (2009 onward) and 0 otherwise. In the above model the interaction coefficient between the variable Ike and the risk variables (βp ) show how Hurricane Ike might have affected the prices of the properties that are in the risk zones. For the flood risk variables we use both the FEMA flood zones and the more granular return periods. Spatial Hedonic Model accounting for the spatial dependence in the property prices One of the econometric concerns in using a hedonic model is the presence of spatial dependence among neighboring properties. Spatial dependence in property values can arise due to neighboring properties sharing common features such as similar location amenities, similar structural attributes due to common timing of construction. Recent critiques by McMillen (2010), Pinske and Slade (2010) and Gibbons and Overman (2012) suggest that spatial models do not provide a valid approach to causal identification, however, it is also argued that the use of spatial hedonic models is appropriate since ignoring the spatial dependence in a hedonic analysis lead to an inefficient or even inconsistent estimates. Therefore, to check for the robustness of our results we utilize the spatial hedonic model allowing for spatial interactions in the dependent variable and the disturbances. 16
More formally, a spatial autoregressive model with autoregressive disturbance (SARAR) is employed following Anselin and Bera, (1988) and Kelejian and Prucha (2010). The SARAR model corresponding to model (3) above can be written as: J K L log( Pit ) o W ln( Pjt ) j S ijt k Likt l Ril m ln(C ) i j 1 k 1 l 1 (5) N L n (ln(C ) i * Ril ) i t it n 1 l 1 Where, it M jt it ; it is i.id (assumed to be independent and identically distributed) The spatial weights matrix W and M (W=M) are taken to be known and stochastic. The lambda (λ) and rho (ρ) are the spatial lag parameter and spatial autocorrelation coefficient respectively. Again we use both FEMA flood zone and return periods for the flood risk variables. In spatial models, one of the challenges lies in defining an exogenous weights matrix (W) that captures the relationship between the spatial units. In general, there is no consensus on appropriate spatial weights matrix (Anselin and Bera, 1998). Queen Contiguity matrix and inverse distance matrix are the most commonly used matrices in spatial models. Queen Contiguity matrix is structured so that if the ith and jth properties share a common border or vertex, the elements of the spatial weights matrix Wij receive a value of 1, 0 otherwise. The inverse distance matrix is structured in such a way that the elements of the spatial weights matrix Wij receive a value equal to inverse of Euclidian distance between the ith and jth properties. In our case we use a hybrid matrix combining the queen contiguity and inverse distance matrix where distance decay was allowed in the queen contiguity matrix. The hybrid matrix was min-max normalized.11 We employ a generalized spatial two-stage least square (GS2SLS) estimator.12 17
IV. Main Results Table II presents the results of standard hedonic model that uses the FEMA flood zone classifications of V, A, and X500 zones with the low risk X zone as the control group. We estimate three different models as discussed in the methods section where model (1) includes the log of coastal distance; in model (2), we control for distance to coast using coastal distance dummies with 500 to 1000 feet increments for distance within a mile; and in model (3), we interacted the flood zones with the log of coastal distance. Across all three models, we find that the properties located in high-risk areas such as V and A zones command a price premium (statistically significant at the 1 percent level in all three models) suggesting that the associated positive amenity values of living in the 100 year flood zone (V and A zone) in Galveston county outweigh the negative flood risk. This result is similar to other coastal community estimates as shown by Bin and Kruse (2006) and Daniel et al. (2009), where the positive amenity impacts of a coastal community has a strong effect. 13 The price premium for V zone properties is 40.9 percent and 40.1 percent in model (1) and (2) respectively14 which means that a property in V zone sells for $74,311 and $72,676 more than an equivalent property in the X zone (the control group) when evaluated at an average priced home ($181,690). Likewise for the A zone, the price premium is equivalent to 8.12% and 9.6% in model (1) and (2) respectively which means that a property in A zone sells for $14,739 and $17,309 than an equivalent property in the X zone when evaluated at an average priced home ($181,690). The variable ln_coast is negative and statistically significant in model (1), implying that proximity to coast is highly desirable and increasing distance from the coast has strong negative impact on the property prices.15 To put this in perspective, for an average priced home ($181,690), moving away from the coastline 25,694 feet (average distance) results in a decrease in property 18
values by $12,718 (7%). In model (2), we find that there is a monotonic decline in coastal premium, from a 36% premium for properties located within 500 feet, to 12% for those between 500 feet and 1000 feet, to 7.6% for those between 1000 and 2000 going down to 3.9% for those between 4000 and 5000 feet. Using the model (1) coefficient results, Figure 5 illustrates the decay in the hedonic price flood zone premium given increasing distance. For each (natural log) foot decreased from the nearest coastline, sales prices decline by the same amount for all flood zones, 7 percent. Given this equivalent distance decay rate for all flood zones, the V zone estimated hedonic premium disappears after approximately 100 feet from the nearest coastline, whereas for the A zone it disappears at approximately 10 feet. To test the conditional impact of distance to coast and flood risk zones, in model (3) we interact the negative amenity flood zone risk variable with the proximity to coast to estimate these two variables acting together jointly on housing prices. We find a marked increase in the price premium for the V zone properties of almost 146%, which is equivalent to $266,537 when evaluated at an average priced home (Figure 6).16 However, note that this high premium is for the properties in the V zone that are right on the coast and the premium decays as the distance from the coast increases as suggested by negative and statistically significant coastdxVzone interaction. For example, the premium decreases to almost 72.09 percent from 146 percent in V zone as the distance from the coast increases to 100 feet which is equivalent to a decrease of $130,989 when calculated for average priced home. 17 Similarly, compared to X zone properties the A zone properties also command a marked price premium of 28% for the A zone properties that are right on the coast which is equivalent to almost $52,000 when evaluated at an average priced home. We 19
also find that the premium decays (negative and statistically significant coastdxAzone interaction) as the distance to coast increases for A zone properties. The interaction results illustrate the importance of accounting for these values acting jointly on housing prices. While model (1) showed that the average rate of distance decay for all the zones is constant at 7 percent, which is clearly not the case as demonstrated by model (3). The marginal rate of distance decay for V zone property is 7.8 percent and the for A zone properties it is 1.8 percent. 18 That is, as the positive amenity value decreases with increasing distance from the nearest coast, the remaining designated high flood risk is discounted more severely. Model (3) results also indicate a negative price premium for X500 zone properties which is the opposite result from the previous positive coefficients in models (1) and (2), although not statistically significant at any meaningful level. We ran a likelihood ratio test to see if model (3) fitted the data better than model (1). We find a χ2 value of 54.69 with a significant p-value suggesting that the difference between the two models is significant and model (3) fits the data significantly better than model (1). Figure 6 shows the distance decay in the premiums as calculated using coefficients from model (3) and thus accounting for the varying distance-amenity trade-offs by flood risk. In comparison to Figure 5 (not accounting for this variation) we see now that hedonic premiums for the V zone while decaying faster actually remain positive for nearly a quarter of a mile from the nearest coastline. This impacts approximately 745 properties in the V zone with an estimated price premium, whereas in model (1) this information was confounded and showed that the premium existed for 44 properties within 100 feet of the nearest coastline. We see a similar result for the A zone where in model (1) the hedonic price premium decayed away by 10 feet, has now expanded to 100 feet impacting an additional 60 homes. 20
Regarding the structural variables, across all three models in Table II the coefficient are significant at one percent level and have expected signs except for variable stories which is insignificant. As per the location variables, coefficient estimates indicate that being farther from a bus route or park decreases property prices, whereas being nearby a school or railroad increases property prices. The median household income have an expected positive sign. The NFIP dummy is positive and significant suggesting that the properties that are built after the communities joined NFIP in Galveston County are worth more, ceteris paribus. Also, the properties that are protected by seawall are priced higher as suggested by positive and significant seawall dummy (Seawall), likely due to the sense of safety that the seawall provides in the risky zones. In all the models, we have included the time segment and the zip code fixed effects. Accounting for varying flood risk return periods To capture the effect of the inherent varying flood risk within any flood zone we replaced the FEMA designated flood risk zones with the more granular return periods (RP) in our hedonic estimations and estimated model (1), model (2) and model (3) with RP > 500 years as the omitted categories. Table III presents these regression results. From models (1) and (2) we see hedonic price premiums that are statistically significant up to the 100 year RP and thus comparable to the V and A zone results in Table II. From model (3), however, we see negative coefficient signs for all return periods greater than ten years with statistical significance for RPs 25 to 50 and beyond 100 years, the X500 zone. Now, the only flood risk area with a positive hedonic price premium from model (3) is the RP
However, the hedonic price premium for RP
A Difference-in-Difference (DD) Model accounting for the impact of hurricane Ike Table IV and V report the results from the DD models using aggregate flood zones as well as the return periods respectively where column 1 in both tables gives the results of the DD model as in equation (4), and column 2 gives the result of a hedonic model as in equation (3) using only the pre-Ike sales data (2001-2008).19 From Table IV we find that there was no significant impact on the property prices in the V zone due to the occurrence of Hurricane Ike as suggested by an insignificant coefficient in the (Ike*V zone) interaction term. However, the statistically significant (Ike*A zone) and (Ike*X500 zone) interaction terms indicate that property prices in A zone and X500 zone decreased after Ike made landfall. V and A zone premiums in the DD model are also lower in comparison to those from the 2001 to 2008 model. Importantly, in both columns statistical significance and coefficient values/negative signs on the log of coastal distance and the flood zone and distance interaction terms are comparable to the Table II results. From Table V we also find that there was no significant impact on the property prices in the RP
the return periods respectively. Consistent with the results in Table II that uses the aggregate flood zones, Table VI presents a positive and significant premium associated with the V zone and A zone properties when also accounting for the spatial dependence. We also find that the premium decays as the distance from the coast increases as suggested by negative and significant coefficient of the distance and flood zone interaction term (coastxVzone & coastxAzone). All the other variables such as structural attributes, location attributes, additional dummies (NFIP and Seawall), time fixed effects and zip code fixed effects are included in all the models20 Regarding the spatial parameters, we find that the spatial lag parameter (λ) is not significant suggesting that there is no significant adjacency effect however, there is presence of spatial autocorrelation as suggested by a significant spatial error parameter (ρ). In table VII, we present results using the return periods instead of the aggregate flood zones (comparable to table III results). Consistent with table III results, we find a significant premium associated to the properties located in return period less than 10 years, the most risky zone. However, after controlling for the distance to the coast conditional upon the risk return period in model (3), we see negative and insignificant price premium for RP between 10 and 250 while negative and significant premium for properties located in return period greater 250 and less than 500. In addition to these reported robustness tests we ran a series of other analyses not reported here including: 1) separate spatial models for each aggregate flood zone (V zone and A zone) 2) spatial hedonic model in DD framework for V and A zone separately 3) spatial hedonic model using different spatial weighting matrices 4) separate models for mainland and island properties and 5) models using the yearly time dummies. We were able to further verify the robustness of our main results in each of the above cases. 24
VI. Conclusions We have attempted to estimate the effect on housing prices of locating near the coast through a hedonic property analysis in Galveston County, TX allowing for the flood risk and coastal amenity of living close to the water to act jointly on housing sales prices. We estimate these effects not only through the traditional use of flood zones as a measure of the flood risk, but also incorporate a more comprehensive measure of graduated flood risk through identified flood return periods. Our results show that properties located in the high risk flood areas command a price premium compared to those located outside. For example, the properties located in high-risk areas such as V and A zones command a price premium of up to 146%. But this hedonic price premium is dependent upon the distance to the coast, and as expected the distance effect varies by flood risk type with found premiums to higher risk homes decaying at a faster rate the further one moves away from the water. This illustrates the importance of capturing these two variables acting together on sales prices. These results are also robust to the potential subjective bias associated with the occurrence of Hurricane Ike in 2008. Given the varying flood risk and distance tradeoffs captured in the interaction terms, we show that properties located in the highest risk flood areas, for up to nearly a quarter mile from the nearest coastline, command a sales price premium. These results directly contrast to some of the previous hedonic property analyses findings that homes at the highest flood risk will sell at a discount to account for higher flood insurance rates being capitalized into housing sales prices. We acknowledge that our measure of positive water amenities – distance to the nearest coast – may not fully represent the variety of positive amenities inherent to living near the water. Thus, we additionally controlled for the impact of other possible coastal amenities such as including waterfront properties (dummy=1 if property is within 500ft from the coast) as well as adding 25
distance to the nearest beach access points in our models. In both cases, we still find a premium associated with V and A zone properties, although of a lower magnitude, with the coefficient on distance to beach access points being relatively small in magnitude and statistically insignificant at the 10 percent level.21While another coastal amenity of interest could be the measure of view, Daniel et al. (2009) find that distance is a more meaningful control for positive water amenities than view. Other measures of positive water amenities warrant more attention in future flood risk and housing price research. Therefore, the assertion by realtors, homebuilders, and lenders in opposition to BW-12 that moving to risk-based premiums will cause property values to steeply decline and make homes unsellable is not universally true in our study area. We show that this depressed property value assertion is sensitive to the distance to water. The powerful amenity value provided by the nearby coastal water shadows the flood risk and therefore masks the influence of increased flood insurance premiums on property prices. Notably, even with our systematic interaction approach coupled with more detailed flood risk return period data, we find that in the absence of a major flood event it is difficult for the negative flood amenity value based upon the objective flood risk to sufficiently counteract a homeowner’s strong desire to live near the water. 26
References: Arraiz, I., Drukker, D. M., Kelejian, H.H., and Prucha I. R. 2010. “A Spatial Cliff-Ord Type Model with Heteroskedastic Innovations: Small and Large Sample Results.” Journal of Regional Science 50 (2): 592–614. Anselin, L., and Bera A.1998. “Spatial Dependence in Linear Regression Models with an Introduction to Spatial Econometrics.” In Handbook of Applied Economic Statistics, eds A. Ullah and D. Giles. Atreya, A., Ferreira, S., and Kriesel, W.P. 2013. “Forgetting the Flood? An analysis of the Flood Risk Discount over Time.” Land Economics 89 (4): 577-596 Barnard, J.R., 1978. Externalities from urban growth: the case of increased storm runoff Bin, O., and Polasky, S. 2004. “Effects of Flood Hazards on Property Values: Evidence before and after Hurricane Floyd.” Land Economics 80 (4): 490–500. Bin, O., and Kruse, J.B. 2006. “Real Estate Market Response to Coastal Flood Hazards.” Natural Hazards Review 7 (4): 137–44. Bin, O., Kruse, J. B, and Landry, C. E. 2008a. “Flood Hazards, Insurance Rates, and Amenities: Evidence from the Coastal Housing Market.” Journal of Risk and Insurance 75 (1): 63–82. Bin O., Crawford, T. W., Kruse, J. B., and Landry, C. E. 2008b. Viewscapes and flood hazard: Coastal housing market response to amenities and risk. Land Economics 84(3): 434–48 Bin, O., and Landry, C. E. 2012. “Changes in Implicit Flood Risk Premiums: Empirical Evidence from the Housing Market.” Journal of Environmental Economics and Management 65 (3): 361–76. Boyle, K., Lewis L., Pope J., and Zabel. 2012. “Valuation in a bubble: Hedonic modeling pre and post-housing market collapse.” AERE Newsletter, 32(2). Conroy, S. J., & Milosch, J. L. 2011. “An estimation of the coastal premium for residential housing prices in San Diego County.” The Journal of Real Estate Finance and Economics, 42(2), 211-228. Czajkowski, J., Kunreuther, H., & Michel‐Kerjan, E. 2013. “Quantifying Riverine and Storm‐ Surge Flood Risk by Single‐Family Residence: Application to Texas.” Risk Analysis 33(12):2092-2110 Chivers, J. and Flores, N. E. 2002. “Market Failure Information: The National Flood Insurance Program”. Land Economics, 78(4):515-521 27
Daniel, V., Florax, R., Rietveld, P., 2009. Flooding Risk and Housing Values: An Economic Assessment of Environmental Hazard. Ecological Economics 69(2):355-365 Drukker, D. M., Egger, P. and Prucha, I. R. 2009. On Single Equation GMM Estimation of a Spatial Autoregressive Model with Spatially Autoregressive Disturbance. Technical report, Department of Economics, University of Maryland. FEMA 2014. http://www.fema.gov/national-flood-insurance-program/base-flood-elevation FEMA 2014. http://www.fema.gov/policy-claim-statistics-flood-insurance/policy-claim- statistics-flood-insurance/policy-claim-13 FHFA 2014, http://research.stlouisfed.org/fred2/series/ATNHPIUS26420Q Freeman, A. 2003. The measurement of environmental and resource values: theory and methods: RFF press. Gibbons, S. and Overman, H. G.. 2012. “Mostly Pointless Spatial Econometrics?” Journal of RegionalScience 52(2):172 – 191 Griffith, R. S.1994. The Impact of Mandatory Purchase Requirements for Flood Insurance on Real Estate Markets, Doctoral Dissertation, University of Texas at Arlington, August. Insurance Journal, 2014. House Passes Flood Insurance Bill; Key Senators Sign On Available at http://www.insurancejournal.com/news/national/2014/03/04/322194.htm Kelejian, H.H., and Prucha, I.R. 2010. "Specification and estimation of spatial autoregressive models with autoregressive and heteroskedastic disturbances." Journal of Econometrics 157(1):53-67. Kousky, C. 2010. “Learning from Extreme Events: Risk Perceptions after the Flood.” Land Economics 86 (3): 395–422. Kriesel, W., Friedman, R. 2002. Coastal hazards and economic externality: implications for beach management policies in the American South East. H. John Heinz III Center for Science, Economics and the Environment, Washington DC. McKenzie, R., Levendis, J. 2010. Flood Hazards and Urban Housing Markets: The Effects of Katrina on New Orleans. Journal of Real Estate Finance and Economics 40:1:62-76 McMillen, Daniel P. 2010. “Issues in Spatial Data Analysis,” Journal of Regional Science, 50(1): 119–141. Michel-Kerjan, E., Czajkowski, J., Kunreuther, H. 2014. Could Flood Insurance Be Privatized in the United States? A Primer, The Geneva Papers, forthcoming. 28
Morgan, A. 2007. The impact of Hurricane Ivan on expected flood losses, perceived flood risk, and property values. Journal of housing research, 16(1), 47-60. Muggeo, V.M.R., 2003. “Estimating regression models with unknown break-points” Statistics in Medicine, 22: 3055-3071. Pinske, J. and Slade, M.E. 2010. “The Future of Spatial Econometrics,” Journal of Regional Science, 50(1), 103–117. Posey, J., & Rogers, W. H. 2010. The impact of Special flood Hazard Area designation on residential property values. Public Works management & Policy, 15(2), 81-90. Rosen, S. 1974. “Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition,” Journal of Political Economy 82(1): 34-55. Tobin, G.A., Montz, B.E., 1994. The flood hazard and dynamics of the urban residential land market. Water Resources Bulletin 30 (4), 673–685. US Army Corps of Engineers (USACE), 1998. Empirical studies of the effect of flood risk on housing prices. Alexandria, Virginia, Water Resources Support Center Institute for Water Resources. Wall Street Journal (WSJ) 2013. Flood Program Puts Industries at Odds. http://online.wsj.com/news/articles/SB10001424052702304773104579268620558111400 Zhai, G., Fukuzono, T., Ikeda, S., 2003. Effect of flooding on megalopolitan land prices: a case study of the 2000 Tokai flood in Japan. Journal of Natural Disaster Science 25 (1), 23–36. 29
APPENDIX: Definitions of FEMA Flood Zone Designations Flood zones are geographic areas that the FEMA has defined according to varying levels of flood risk. These zones are depicted on a community's Flood Insurance Rate Map (FIRM) or Flood Hazard Boundary Map. Each zone reflects the severity or type of flooding in the area. Moderate to Low Risk Areas In communities that participate in the NFIP, flood insurance is available to all property owners and renters in these zones: High Risk Areas In communities that participate in the NFIP, mandatory flood insurance purchase requirements apply to all of these zones: High Risk - Coastal Areas In communities that participate in the NFIP, mandatory flood insurance purchase requirements apply to all of these zones: 30
You can also read