222 Week 7 F /For WIZARD KIM

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y

P r i c e s : P r e s e n c e , P r o x i m i t y , a n d D e n s i t y

A u t h o r s R a n d y E . D u m m , G . S t a c y S i r m a n s , a n d

G r e g T. S m e r s h

A b s t r a c t Spatial amenities, along with structural characteristics, affect residential property values. Although the bundle of structural characteristics is typically the primary value determinant, studies have shown that externalities and risk factors can adversely affect property values. We use residential property sales data from 2010 to 2014 and sinkhole data from ‘‘sinkhole alley’’ in Florida to examine the effect of sinkhole presence, proximity, and density on the sale price of residential real estate. Using a spatial error regression model, the results show that sinkhole proximity and sinkhole exposure (density) create a negative externality and both have a significant negative effect on house prices. While the results show a negative effect of having a sinkhole on the property, those results are not statistically significant.

‘‘A Florida couple is facing up to 20 years each in prison today for

selling their home without informing the buyers about an enormous

sinkhole that they knew was underneath it.’’ ABCNEWS.go.com, October 9, 2015

Location matters. Spatial amenities, along with structural characteristics, have definite effects on residential property values. Although the bundle of structural characteristics is typically the primary value determinant, studies have shown that externalities and other factors can affect property values.1 Externalities / factors such as golf course, ocean or lake view, proximity to a church, school quality, and presence of trees have a positive effect on house prices while other externalities / factors such as environmental contamination, landfills, proximity to power lines, flood plain, abandoned buildings, and proximity to an interstate have a negative effect.2 In recent studies, Dumm, Smersh, and Sirmans (2016) show that waterfront properties command a price premium and Nyce, Dumm, Smersh, and Sirmans (2015) show that homeowner’s insurance premiums have a negative effect on house prices.

We use Florida data to examine the effect of sinkholes on residential property values. We consider the presence, proximity, and density of sinkholes. As with

4 2 u D u m m , S i r m a n s , a n d S m e r s h

other externalities, a sinkhole may be both a property characteristic and a negative externality since it may affect the value of surrounding properties. Along with hurricane and tropical storm risk, Florida ranks highest in the nation for sinkhole risk. However, sinkhole location is less random and more localized than hurricanes and earthquakes because their occurrence is more likely in areas where the land surface covers the types of rock (limestone, carbonate rock, and salt beds) that are naturally dissolved by groundwater.

Although states located primarily in the eastern United States are the most susceptible to sinkholes, other parts of the country are not exempt. Sinkholes have appeared in Texas and Louisiana as a result of collapsing salt domes. States such as Oklahoma, New Mexico, South Dakota, and Wyoming sit atop gypsum, another type of soluble rock that can dissolve rapidly in water. Although Florida is especially identified with sinkholes, if all underlying water soluble rock (karst, limestone, carbonate rock, gypsum, and salt) across the U.S. is considered, 40% of the ground cover of the U.S. is susceptible to sinkholes.

Our results show that sinkhole exposure has an adverse impact on real estate values. Although the coefficient for the presence of a sinkhole is negative, it is not statistically significant (likely a function of too few observations). However, proximity to and the number of sinkholes within a certain distance (density) have a significantly negative effect on residential real estate selling prices for non- sinkhole properties. These effects decrease with increased distance from a sinkhole and increase as the number of sinkholes increase.

u N e g a t i v e E x t e r n a l i t i e s a n d P r o p e r t y V a l u e There is scant evidence explaining the relation between house prices and sinkholes, either in regard to sinkhole presence, proximity, or density. Only one previous study has examined the effect of sinkholes on house prices. Fleury (2007) applies a hedonic pricing model to 1990 census data and the Florida Geological Survey’s Sinkhole Database to examine the relation between the presence / density of sinkholes to the home values. Using data for the Tampa Bay, Florida area, he estimates OLS and probit models with median home value by census block as the dependent variable. He finds no significant effect of sinkhole presence or density on home values. He posits as possible explanations: homebuyers may not be aware of sinkhole locations and that sinkholes may be viewed as water features where homebuyers do not distinguish between man-made lakes and sinkholes. However, another explanation may be that using census level data, such as median home value by census block, obscures the true price variation across properties that are affected by sinkholes and those that are not. For that reason, we use individual property transaction price data in our study.3

Although the literature on sinkholes is lacking, there is ample literature showing the effects of negative externalities on property values. Simons, Levin, and Sementelli (1997) examine the effect of underground storage tanks on residential

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y P r i c e s u 4 3

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

sales price. They examine three types of tanks: registered non-leaking, nonregistered leaking tanks, and registered leaking tanks and find a 17% reduction in sales price. In a subsequent study, Simons, Bowen, and Sementelli (1999) examine the effects of leaking underground storage tanks (LUSTs) from gas stations on residential and commercial property. Their results show a 14%–16% reduction in sales price for residential properties and a 28%–42% reduction in sales price for commercial properties. In another study of storage tanks, Zabel and Guignet (2012) examine the effect of LUSTs on local soil and surface and groundwater. Their results are interesting. They find that, although the typical LUST has no effect on surrounding property values, more publicized and more severe sites caused a reduction in property values of more than 10%.

Several studies have provided excellent literature reviews of the effects of environmental contamination, air pollution, landfills, etc. Smith and Huang (1995) conduct a meta-analysis of 37 air pollution studies and find that a reduction in air pollution particulate increased property values. Farber (1998) examines the literature on landfills, solid waste facilities, and superfund sites. Variables found to affect property value included type of facility, distance, information relative to opening and closing dates, thin markets, and employment effects. Boyle and Kiel (2001) review 30 papers for air pollution, water quality, undesirable land uses, and pollution sources. They find that air quality features are generally not known to buyers. They did find that water quality variables produced negative effects on price and that undesirable land uses consistently produced negative effects on price. Jackson (2001) examines 45 studies on environmental contamination and finds primarily negative effects on property values. Simons and Saginor (2006) produce a meta-analysis of the effect of environmental contamination on residential real estate. They examine the effects of both environmental contamination and positive amenities on proximate real estate property values in the U.S. The contaminated sites include superfund sites, LUSTs, landfills, pipeline ruptures, and overhead transmission lines. They provide a literature review of 58 peer-reviewed papers on contamination and 17 papers on positive amenities such as parks, beaches, etc. Their model examines dollar property value loss and includes explanatory variables such as distance from source, type of contamination, information, urban or rural environment, local and national market conditions, remediation, etc. All of the contamination effects are negative.

Others have examined the effect of flood hazard on residential property values. Skantz and Strickland (1987) examine the effect of property values on first-time flooding. Interestingly, they find that property values did not fall immediately after the flood. Instead, they find that house prices declined about a year later when increased flood insurance premiums were capitalized into home prices. Harrison, Smersh, and Schwartz (2001) examine changes in value for homes located within 100-year flood plains. They find a negative effect on selling price for homes located within a flood zone. However, this negative price effect is less than the present value of future flood insurance premiums. Bin and Polasky (2004) estimate the effect of flood hazards on residential property value for properties that received

4 4 u D u m m , S i r m a n s , a n d S m e r s h

significant flooding from Hurricane Floyd in 1999. They find that properties located within a floodplain experienced a decrease in property value and that the decrease is significantly larger after Hurricane Floyd than before. Bin and Kruse (2006) examine the effects of flood hazard on residential property values. Using digital flood maps coupled with residential property sales records, they find, on average, a 5%–10% reduction in property values for properties located within a flood zone but not subject to wave action. Interestingly, they find higher values for properties in flood plains subject to wave action. They also find that price differentials were roughly equivalent to the capitalized value of flood insurance premiums.

Examining the effect of earthquake risk, Murdoch, Singh, and Thayer (1993) find an area-wide reduction in property values following the Loma Prieta (World Series) earthquake in the San Francisco Bay area. Interestingly, they also find that apparently some homeowners considered other measures of earthquake risk in their housing purchases, producing an obvious price gradient.4

u W h a t i s a S i n k h o l e ? A sinkhole is a ground depression with no natural external surface drainage capability. Sinkholes are most common in areas of ‘‘karst terrain.’’ The term ‘‘karst’’ is derived from the Slovenian word kras, which refers to the Kras region in Slovenia. The land areas of both Slovenia and Croatia (formerly Yugoslavia) are major sinkhole areas.

A karst terrain is an area where the rocks below the land surface can naturally be dissolved by groundwater circulating through them. Dissolvable types of rocks include salt beds and domes, gypsum, limestone, and other carbonate rocks. As water soaks into the ground, these karst terrain rocks dissolve creating underground caverns and spaces. At some point the underground space becomes too big for the ground cover to support. The surface land collapses and a sinkhole results. About 20% of the U.S. sits atop karst terrain. In fact, most states have some areas with karst terrain. The U.S. Geological Survey map in Exhibit 1 shows the various underlying water-soluble rock across the U.S. and thus the most sinkhole-prone areas. The shaded areas indicate various degrees of karst, limestone, carbonate rock, and gypsum.

Although most sinkholes are ‘‘natural’’ phenomena, some sinkholes are correlated with events such as flooding and land-use practices such as groundwater pumping and construction, which may lower the water table. Real estate development such as buildings and parking lots can divert rainwater runoff and result in concentrated weight that causes the ground cover to collapse. As Sinclair (1982) points out, sinkholes may occur along certain joint patterns in the underlying bedrock. Pumping of large-capacity wells along these joint patterns could increase the probability of sinkhole development.

In theory, the entire state of Florida is susceptible to sinkholes since it sits on porous carbonate rocks such as limestone. Recognizing sinkhole risk, Florida law

S i

n k

h o

l e

s a

n d

R e

s i

d e

n t

i a

l P

r o

p e

r t

y P

r i

c e

s u

4 5

J R

E R

u V

o l

. 4

0 u

N o

. 1

– 2

0 1

8

E x h i b i t 1 u Areas in the U.S. Susceptible to Sinkhole Formation

Map courtesy of the U.S. Geological Survey.

4 6 u D u m m , S i r m a n s , a n d S m e r s h

requires insurers to include sinkhole activity coverage in homeowners’ insurance policies, for which insurers can charge an additional premium. Central Florida’s geology makes that part of the state the most prone to sinkholes. Sinkholes are identified by the Florida Geological Survey, which collects geological data for the state.5

In addition to Florida, sinkholes are also becoming more common in other states. Along with natural formations, sinkholes can also be precipitated by coal mines and aquifer systems. For example, Pennsylvania is in the top seven states most at risk for sinkholes because it has coal mines and an extensive underlay of limestone, especially throughout central Pennsylvania around the State College area and from the Maryland state line up through Harrisburg and Lancaster. Also in Alabama, sinkholes result from the underlying soluble carbonate rocks (especially north Alabama) and abandoned mines.

In the U.S., the most damage from sinkholes tends to occur in seven states: Florida, Texas, Alabama, Missouri, Kentucky, Tennessee, and Pennsylvania.6

Newton (1984) provides a detailed analysis of sinkhole occurrence in the eastern U.S. He identifies 850 sites with an estimated 6,000 sinkholes in 19 states. He confirms that the southern states most impacted are Alabama, Florida, Georgia, and Tennessee with Pennsylvania as the most impacted northeastern state and Missouri the most impacted Midwestern state. He estimates that the cost of damage and preventive measures to minimize severity is about $170 million as of the date of his study.

u S i n k h o l e F o r m a t i o n Sinkholes are divided into three primary types: solution sinkholes, cover-collapse sinkholes, and cover-subsidence sinkholes [see Pappas (2014) and Radazzo (2015) for discussions]. With a solution sinkhole there is very little topsoil over the limestone or other bedrock. Over time, water seeps into the crevices in the bedrock and dissolves it. This results in the gradual formation of a depression. Cover- collapse sinkholes are the most dangerous because they happen suddenly over a short period of time and can range in size. They are the result of erosion of ground with ground sediments forming an underground cavern. The continued erosion eventually causes the surface layer to collapse, resulting in the sudden opening of a sinkhole. Cover-subsidence sinkholes are the most common and form slowly over time as the ground gradually subsides or deflates. These occur as the sand covering the bedrock slowly filters down in the bedrock crevices, eventually causing the land surface to collapse. These sinkhole formations are less dramatic and less noticeable and may go undetected for a time.

The U.S. Geological Survey (USGS) provides a mapping of the U.S. relative to sinkhole risk and likely occurrence. The map in Exhibit 2 shows areas of possible underground cavern formation and resulting sinkholes. The slash covered areas show the existence of evaporites (salt, gypsum, and anhydrite) and carbonates

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y P r i c e s u 4 7

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

E x h i b i t 2 u U.S. Map of Sinkhole Risk

(limestone and dolomite). About 35%–40% of the U.S. is underlain with evaporate rocks, although fortunately for many areas they are buried at great depths.

An interesting question is whether existing sinkholes are predictors of future sinkholes. Some studies have attempted to predict the occurrence and distribution of sinkholes. Gao and Alexander (2003) construct sinkhole probability maps for southeastern Minnesota. Based on the distance to the nearest sinkhole and sinkhole density, the authors create a decision tree to construct maps of relative sinkhole risk. Their model successfully captures the details of the internal structure of high- density areas but is less successful in lower-density areas. Upchurch and Littlefield (1988), using sinkhole data for Hillsborough County, Florida, find that the distribution of ancient sinkholes in bare karst areas successfully predicted the location of modern sinkholes. In areas of covered karst, the location of ancient sinkholes did not predict modern sinkholes. The authors found, in bare karst areas, the sinkholes were localized along major lineaments, consistent with Sinclair (1982).

Gutierrez, Guerrero, and Lucha (2008) argue that two sinkhole components, the probability of occurrence and the severity, must be considered in calculating sinkhole risk. Using a database for Spain that includes location, chronology, size, and subsidence mechanisms, the authors find that trenching (clearing the topsoil in order to observe the abutting relations of faults and fractures) was useful in predicting sinkhole probability of occurrence and severity. Galve et al. (2009) examine sinkhole susceptibility and hazard by mapping three genetic types of sinkholes. Susceptibility models were validated for each of the three types and

4 8 u D u m m , S i r m a n s , a n d S m e r s h

were transformed into a hazard map useful for developing preventive or corrective measures.

NASA has suggested that sinkholes may be spotted as they develop by using Interferometric Synthetic Aperture Radar (InSAR). InSAR provides radar data from satellites and detects tiny movements in the ground. For sinkholes that have some surface deformation prior to collapse, this could be a monitoring system and provide some foresight in the case of some sinkholes. At present, there is no plan to implement such a system country-wide in the U.S.

u T h e ‘ ‘ S i n k h o l e A l l e y ’ ’ o f F l o r i d a Although all states have some areas with karst terrain, unfortunately for Florida, most of the state consists of limestone karst terrain. Florida’s property and casualty insurers have seen an increase in sinkhole claims over the last several years and sinkhole losses have risen. In a 2011 report, the Florida Office of Insurance Regulation (OIR), the number of sinkhole claims increased from 2,360 in 2006 to 6,694 in 2010 (a total of 24,671 claims).7 The approximate dollar amount of these claims was $1.4 billion.8

Within Florida, the ten most sinkhole-prone counties are:

1. Pasco

2. Hernando

3. Hillsborough

4. Marion

5. Pinellas

6. Citrus

7. Polk

8. Orange

9. Seminole

10. Lake

Three of these counties (Hernando, Hillsborough, and Pasco) are known collectively as ‘‘sinkhole alley.’’ Per the Florida Office of Insurance Regulation, over the 2006–2010 period, two-thirds of the reported damage claims came from these three counties. OIR reports that claims from sinkholes have increased in recent years. In 2006, Florida saw just over 2,300 claims. By 2010, that number has risen to almost 6,700. Since there is no geological explanation for the increase, the OIR suspects that some of the claims are questionable. Insurance industry officials say claims are often paid without hard proof of damage.

Exhibit 3 shows the subsurface geology and sinkholes in Florida. Below are examples of the more notorious sinkholes. In late February 2013, a mouth 20 feet (6 meters) wide swallowed 37-year-old Jeff Bush as he slept in Seffner, Florida,

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y P r i c e s u 4 9

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

E x h i b i t 3 u Florida Sinkholes

inhaling his entire bedroom. Five others in the house escaped without injury, including Jeremy Bush, who tried in vain to save his brother. (Jeremy Berlin, National Geographic, August 13, 2013.) One of the most spectacular sinkholes occurred in 1981 in Winter Park, Florida. It was one city block in size and swallowed up a Porsche dealership. These are the examples of the ‘‘classic’’

5 0 u D u m m , S i r m a n s , a n d S m e r s h

sinkholes that result in the complete collapse of the ground. They tend to be different sinkholes than the sinkholes that we are examining in this study. Here, the assumption is that the reported sinkhole is either located away from the house and the house has not been damaged, is located near the house and has caused no or minimal damage, or is located near a house and has been repaired.

u T h e D a t a We analyze single-family detached home sales in Florida for Hillsborough County (Tampa) for the 2010–2014 period.9 Hillsborough County is approximately 1,000 square miles with a population of roughly 1.3 million people and has a population density of around 1,300 persons per square mile. Hillsborough County is the fourth most populated county in Florida, but only the twelfth largest. Tampa is by far the largest city and most of the county is unincorporated.10

The property sale data for this study are provided by the Hillsborough County Property Appraiser’s Office. This includes the property identification number, date of sale, dollar amount of sale, section township range, and the type of legal instrument (i.e., Quit Claim, Warranty Deed, etc.) recorded at the County Clerk’s office. This study includes only single-family detached transactions. All sales in the final sample are verified as qualified sales or ‘‘arms-length’’ transactions. All non-qualified sales (such as purchase by a relative, foreclosure, or any other non- arms-length transaction) were deleted from the sample.11 Binary variables are used to represent the quarters from 2010 to 2014, with the first quarter of 2010 as the base quarter.

The data also include structural characteristics such as heated square footage, year built (used to calculate age), number of bedrooms, flooring, and lot size. Supplementary property characteristics such as fireplace, swimming pool, central air conditioning, and building construction quality are also available. Some of these are converted to binary variables with a value of one to indicate the existence of the specific characteristic or superior condition and zero otherwise. Structural characteristics were linked to the sales database using the unique property identification number for each parcel. During the observation period, some properties sold multiple times and the duplicate sales have been removed from the sample. The year built variable was used to delete any sales where a house had been rebuilt. That is, if the year built is more recent than the sale year, we assume that the original house was torn down. In that case, the observation is deleted as structural data are only available for houses that are currently standing.

Property parcel polygons (or ‘‘shape files’’) were also obtained from the Hillsborough County Property Appraiser’s Office, and imported into a geographic information system (GIS) where the parcel boundary and centroid (latitude / longitude coordinate) were used to create additional variables. For example, a GIS spatial query was performed to identify all the parcels that intersect the boundaries

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y P r i c e s u 5 1

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

E x h i b i t 4 u Location of Hillsborough County in Florida

of the nearby water bodies over 10 acres in size to identify those parcels as being on the water.

Additionally, a file with the location of known sinkholes in the State of Florida was obtained from the Florida Geological Survey (FGS) office. The FGS is the premier state government institution in Florida specializing in geoscience research and assessments to provide objective quality data and interpretations. They maintain and update databases as part of the statewide geomorphic mapping project, and provide a mapping of the state relative to sinkhole risk and likely occurrence. We utilized GIS to identify properties that contain reported sinkhole activity and also to calculate the proximity of properties without sinkholes to the

5 2 u D u m m , S i r m a n s , a n d S m e r s h

nearest property with a sinkhole. GIS spatial queries were performed to calculate sinkhole proximity and density within 1⁄4 mile, within 1⁄2 mile, within 1 mile, and within 2 miles.12

u M e t h o d o l o g y Hedonic regression analysis is commonly used in real estate research to measure the marginal effects of housing characteristics and other factors on house prices. A review by Sirmans, Macpherson, and Zietz (2005) of over 125 real estate studies that have used hedonic pricing models shows that many types of variables, such as square feet, age, number of bedrooms and bathrooms, garage, pool, and fireplace, have been included in these models. This study includes these variables and, additionally, variables to measure the price effect of having a sinkhole on the property or a sinkhole located within a certain proximity.

The general form of the hedonic pricing model is:

P 5 ƒ ( X , L , E ),i j j i

where Pi is the log of the transaction price of house i, Xj is a vector of j structural housing characteristics, Lj is a vector of location variables, and Ei is a vector of externalities affecting the transaction price. The semi-log model is generally preferred since the coefficients provide semi-elasticities, i.e., the coefficients are interpreted as the percentage change in price relative to a one-unit change in the explanatory variable (Sirmans, Macpherson, and Zietz, 2005). The method typically used to measure the marginal effect of the explanatory variables on the house price is OLS regression (Sirmans, Macpherson, and Zietz, 2005), which minimizes the sum of the squared residuals.

Our model includes variables representing several different aspects of sinkholes (presence, proximity, and density) and binary variables for the quarters between 2010 to 2014, with the first quarter of 2010 as the base quarter. With these additional variables, the hedonic model can be written as:

Ln(SP ) 5 a 1 b X 1 b Sinkhole 1 b SinkholeDistancei 0 i ij j ij

1 b SinkholeExposure 1 b Time 1 « ,ij j i

where Ln(SPi ) is the natural log of the selling price for house i, Xij is the matrix of explanatory variables j for house i, Sinkholei is a binary variable for house i with a value of one if a sinkhole is located on the property and zero otherwise.

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y P r i c e s u 5 3

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

SinkholeDistanceij is a set of j variables that capture the proximity of the nearest sinkhole to house i (for properties that do not have sinkholes on them). The variable Distance is the actual distance measure while the variables NearSH1 / 4, NearSH1 / 2, NearSH1m, and NearSH2m are indicator variables that are equal to 1 if the nearest sinkhole falls in that specific distance band.

SinkholeExposureij is a set of j density variables for house i. The density variables are categorized as either low exposure, medium exposure or high exposure based on the number of sinkholes in the 1⁄4, 1⁄2, one or two-mile distance bands. Low exposure is the first decile of sinkhole count for the distance band, medium exposure is between the 10th and 90th decile, while high exposure is at the 90th

decile and above. These exposure variables are cumulative in that each distance band includes the sinkholes from the shorter distance density measure (e.g., the count for the 1⁄2 mile band includes the number of sinkholes reported for low, medium, and high exposures for the 1⁄4, 1⁄2, one, and two mile bands). For properties with sinkholes on them, the Low, Medium, and High counts do not include that sinkhole. Timei is a vector of binary variables indicating the quarter that property i was sold and «i is the error term.

The model is designed to measure three hypotheses. First, the presence of a sinkhole on the property has a negative effect on the selling price of the property. According to this hypothesis, the b on Sinkholei would be , 0. Although Fleury (2007) shows no effect on sinkhole property values, a number of studies on negative externalities provide evidence of a negative effect on property values (e.g., Smith and Huang, 1995; Simons, Levin, and Sementelli, 1997; Farber, 1998; Simons, Bowen, and Sementelli, 1999; Jackson, 2001; Simons and Saginor, 2006; Zabel and Guighet, 2012).

The second hypothesis is that sinkhole proximity has a negative effect on selling price (i.e., the closer the sinkhole, the lower the selling price) and we hypothesize this to be the case whether considering distance on a continuous basis or in distance bands from the property. According to this second hypothesis, the bs on the proximity variables (Distance and NearSH.) would be , 0 with a diminishing negative effect as distance increases. Regarding the effect of proximity to sinkholes, there is no published literature on the effect of sinkhole proximity on property values but there are a number of studies showing a negative effect of proximity to negative externalities. For example, Simons and Saginor (2006) show the negative effect of environmental contamination on proximate real estate [see also Simons, Levin, and Sementelli (1997), Simons, Bowen, and Sementelli (1999), and Zabel and Guignet (2012)].

The third hypothesis is that sinkhole exposure has a negative effect on selling price (i.e., the greater the count or density, the lower the selling price). We expect to see a stronger negative effect for properties with high exposure than for properties with low exposure to sinkholes. Although he found no significant statistical effect, Fleury (2007) predicted that, based on economic theory, sinkhole density should have a negative effect on property values.

5 4 u D u m m , S i r m a n s , a n d S m e r s h

u R e s u l t s The variables included in the regression model are defined in Exhibit 5 and descriptive statistics are provided in Exhibits 6–8. As shown in Exhibit 6, houses that sold between 2010 and 2014 had an average selling price of $218,795 and the average square footage was almost 2,092 square feet. The average age of the house was 26.9 years and the average lot size was 0.29 acres. There were 3.4 bedrooms per house on average and roughly 6% of the properties were waterfront properties. Twenty-seven percent of the houses had a swimming pool, 21% had a fireplace, and 2.4% had a garage. Almost all houses had central heating and cooling and about 50% had superior grade flooring. Just over 1% of the homes were rated as superior construction.

PopDensity captures the census track population densities for Hillsborough County. The average population per census track was just over 3,974 individuals with population density per census track ranging from 1,023 to 9,851. Quarterly property sale activity varied across the five-year period and the results show the strengthening in the Hillsborough housing market across this time period with over 46% of sales occurring in 2013 and 2014.

The focus of this study is measuring three primary effects: the separate price effects of being a sinkhole property, of being within proximity of a sinkhole, and of being exposed to varying densities of surrounding sinkholes. These effects are measured for both sinkhole and non-sinkhole properties’ selling prices. Properties in the sample have an average distance of 1.23 miles from a sinkhole with the maximum distance of just over 6 miles. Considering proximity based on distance bands, 9.8% of the properties had the nearest sinkhole within 1⁄4 mile, 17.9% had the nearest sinkhole with 1⁄2 mile, 27.4% had the nearest sinkhole within 1 mile, while 27.8% had the nearest sinkhole within 2 miles of the property (the remainder were outside of 2 miles).

Exhibit 6 also includes sinkhole exposure (density) by distance bands around the property. As noted above, these bands are cumulative and exposure is classified as low, medium, or high based on the number of sinkholes for a specific property compared to the distribution of sinkholes for the distance bands (i.e., 10th decile, 10%–90%, and the 90th decile).13

Just over 10% of the properties have sinkholes within 1⁄4 mile of the property, with 7.1% classified as low exposure to sinkholes, 1.5% with medium exposure to sinkholes, and 1.2% with high exposure to sinkholes. As distance increases, the low-medium-high distribution changes as well. At 2 miles, 16.1% of properties are classified as having low exposure to sinkholes, 57.2% have a medium exposure, and 8.4% are classified as having a high exposure.

Exhibit 7 provides the means of properties with a sinkhole or near a sinkhole (as reported by Hillsborough County). Although the number of sinkhole properties is limited to only 24, one characteristic of the sinkhole properties is that they tend

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y P r i c e s u 5 5

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

E x h i b i t 5 u Variable Definitions

Variable Definition

Ln(sp) Log of sale price ln(sp) 5 dependent variable.

SqFt The square footage of the house.

Age Age of house at the time of sale.

Beds Number of bathrooms.

Stories Number of stories.

Lotsize The size of the lot in acres.

Pool Binary variable with a value of one if the house has a pool, zero otherwise.

Fireplace Binary variable with a value of one if the house has a fireplace, zero otherwise.

Garage Binary variable with a value of one if the house has a fireplace, zero otherwise.

Aqualcon Binary variable with a value of one for superior, excellent or above average construction rating, zero otherwise.

AqualFloor Binary variable of one if the flooring is tile, hardwood or marble, zero otherwise.

OnWater Binary variable of one for property on the water, zero otherwise.

LnPopDensity Log of the census tract population density.

Sinkhole Binary variable with a value of one if there is a sinkhole on the property, zero otherwise.

Distance Distance from the property to the nearest sinkhole.

Dist SH2m Interaction between Distance and NearSH2m. NearSH1/4m Binary variable of one if the nearest sinkhole is within 1⁄4 mile of the

property, zero otherwise.

NearSH1/2m Binary variable of one if the nearest sinkhole is within 1⁄2 mile of the property, zero otherwise.

NearSH1m Binary variable of one if the nearest sinkhole is within 1 mile of the property, zero otherwise.

NearSH2m Binary variable of one if the nearest sinkhole is within 2 miles of the property, zero otherwise.

LowSHExp Binary variables for each of the 1⁄4, 1⁄2, 1, or 2 distance bands where one indicates a low sinkhole density category, zero otherwise.

MedSHExp Binary variables for each of the 1⁄4, 1⁄2, 1, or 2 distance bands where one indicates a medium sinkhole density category, zero otherwise.

HighShExp Binary variables for each of the 1⁄4, 1⁄2, 1, or 2 distance bands where one indicates a high sinkhole density category, zero otherwise.

Q12010–Q42014 Quarterly time trend variables for the period 2010 to 2014 (Q1:2010 is the omitted quarter).

5 6 u D u m m , S i r m a n s , a n d S m e r s h

E x h i b i t 6 u Descriptive Statistics (N 5 65,205)

Variable Mean Min. Max. Std. Dev.

Price 218795 25000 6500000 194411

SqFt 2091.93 800 9786 865.36

Age 26.86 1 125 22.81

Beds 3.456 0 8 0.864

Stories 1.293 0 6 0.476

Lotsize 0.293 0.01 39.69 0.536

Pool 0.266 0 1 0.442

Fireplace 0.211 0 1 0.408

Garage 0.024 0 1 0.154

Aqualcon 0.012 0 1 0.111

AqualFloor 0.504 0 1 0.500

OnWater 0.061 0 1 0.240

PopDensity 3974.52 1023 9851 2195.64

Sinkhole 0.000 0 1 0.019

Distance 1.233 0.001 6.048 1.017

NearSH1/4m 0.098 0 1 0.298

NearSH1/2m 0.179 0 1 0.383

NearSH1m 0.274 0 1 0.446

NearSH2m 0.278 0 1 0.448

LowSHExp1/4m 0.074 0 1 0.263

MedSHExp1/4m 0.015 0 1 0.122

HighShExp1/4m 0.012 0 1 0.107

LowSHExp1/2m 0.161 0 1 0.368

MedSHExp1/2m 0.078 0 1 0.269

HighShExp1/2m 0.034 0 1 0.182

LowSHExp1m 0.213 0 1 0.409

MedSHExp1m 0.274 0 1 0.446

HighShExp1m 0.058 0 1 0.233

LowSHExp2m 0.161 0 1 0.367

MedSHExp2m 0.572 0 1 0.495

HighShExp2m 0.084 0 1 0.278

Q12010 0.034 0 1 0.182

Q22010 0.051 0 1 0.219

Q32010 0.036 0 1 0.185

Q42010 0.036 0 1 0.187

Q12011 0.036 0 1 0.185

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y P r i c e s u 5 7

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

E x h i b i t 6 u (continued)

Descriptive Statistics (N 5 65,205)

Variable Mean Min. Max. Std. Dev.

Q22011 0.046 0 1 0.209

Q32011 0.045 0 1 0.207

Q42011 0.040 0 1 0.196

Q12012 0.040 0 1 0.197

Q22012 0.054 0 1 0.226

Q32012 0.054 0 1 0.226

Q42012 0.060 0 1 0.237

Q12013 0.051 0 1 0.221

Q22013 0.070 0 1 0.254

Q32013 0.070 0 1 0.254

Q42013 0.061 0 1 0.239

Q12014 0.052 0 1 0.221

Q22014 0.070 0 1 0.255

Q32014 0.067 0 1 0.251

Q42014 0.028 0 1 0.166

E x h i b i t 7 u Sinkhole Activity

OnSinkhole NearSH1/4m NearSH1/2m NearSH1m NearSH2m

Price 207808.3 168470.5 178793.2 200679.3 252358.1

SqFt 2033.5 1844.3 1866.5 1983.8 2236.1

Age 34.29 37.23 35.89 32.16 20.37

LotSize 0.7500 0.3224 0.2911 0.29319 0.2980

Pool 0.5000 0.2679 0.2325 0.2444 0.2994

Garage 0.0833 0.0370 0.0363 0.0299 0.0158

Stories 1.3333 1.1665 1.2046 1.2598 1.3226

PopDensity 3392.9 3881.5 4007.6 4143.3 4131.4

Aqualcon 0.0417 0.0140 0.01664 0.0168 0.0095

N 24 6413 11657 17894 18135

5 8

u D

u m

m ,

S i

r m

a n

s ,

a n

d S

m e

r s

h

E x h i b i t 8 u Spatial Error Regression Model Output

Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

Sqft 0.0003*** 0.0003*** 0.0003*** 0.0003*** 0.0003*** 0.0003*** 0.0003*** 0.0003*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000)

Age 20.0065*** 20.0065*** 20.0064*** 20.0065*** 20.0065*** 20.0065*** 20.0065*** 20.0065*** (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001)

Beds 0.0304*** 0.0302*** 0.0302*** 0.0303*** 0.0303*** 0.0303*** 0.0303*** 0.0302*** (0.0016) (0.0016) (0.0016) (0.0016) (0.0016) (0.0016) (0.0016) (0.0016)

Stories 20.0267*** 20.0270*** 20.0269*** 20.0265*** 20.0268*** 20.0267*** 20.0269*** 20.0270*** (0.0026) (0.0026) (0.0026) (0.0026) (0.0026) (0.0026) (0.0026) (0.0026)

LotSize 0.0626*** 0.0624*** 0.0624*** 0.0626*** 0.0627*** 0.0629*** 0.0628*** 0.0626*** (0.0022) (0.0022) (0.0022) (0.0022) (0.0022) (0.0022) (0.0022) (0.0022)

Pool 0.1222*** 0.1224*** 0.1224*** 0.1222*** 0.1222*** 0.1221*** 0.1226*** 0.1229*** (0.0025) (0.0025) (0.0025) (0.0025) (0.0025) (0.0025) (0.0025) (0.0025)

Fireplace 0.0584*** 0.0590*** 0.0590*** 0.0586*** 0.0584*** 0.0586*** 0.0590*** 0.0593*** (0.0027) (0.0027) (0.0027) (0.0027) (0.0027) (0.0027) (0.0027) (0.0027)

Garage 0.0582*** 0.0578*** 0.0578*** 0.0582*** 0.0582*** 0.0583*** 0.0578*** 0.0582*** (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059)

Aqualcon 0.2835*** 0.2835*** 0.2833*** 0.2838*** 0.2836*** 0.2836*** 0.2836*** 0.2840*** (0.0080) (0.0080) (0.0080) (0.0080) (0.0080) (0.0080) (0.0080) (0.0080)

Aqualfloor 0.0384*** 0.0384*** 0.0385*** 0.0384*** 0.0384*** 0.0383*** 0.0385*** 0.0386*** (0.0018) (0.0018) (0.0018) (0.0018) (0.0018) (0.0018) (0.0018) (0.0018)

OnWater 0.1502*** 0.1494*** 0.1492*** 0.1502*** 0.1502*** 0.1504*** 0.1507*** 0.1500*** (0.0053) (0.0053) (0.0053) (0.0053) (0.0053) (0.0053) (0.0053) (0.0053)

LnDensity 20.0036 0.0004 0.0000 20.0035 20.0035 20.0032 20.0020 20.0006 (0.0050) (0.0049) (0.0049) (0.0049) (0.0050) (0.0049) (0.0049) (0.0049)

Sinkhole 20.0423 20.0356 20.0331 20.0403 20.0406 20.0513 20.0481 20.0462 (0.0446) (0.0447) (0.0447) (0.0447) (0.0447) (0.0447) (0.0447) (0.0447)

S i

n k

h o

l e

s a

n d

R e

s i

d e

n t

i a

l P

r o

p e

r t

y P

r i

c e

s u

5 9

J R

E R

u V

o l

. 4

0 u

N o

. 1

– 2

0 1

8

E x h i b i t 8 u (continued)

Spatial Error Regression Model Output

Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

Distance 0.0605*** 0.0614*** (0.0054) (0.0054)

Dist SH2m 0.0254*** (0.0057)

NearSH1/4m 20.0460*** (0.0088)

NearSh1/2m 20.0358*** (0.0072)

NearSH1m 20.0058 (0.0071)

NearSH2m 0.0295*** (0.0077)

LowSHExp1/4m 20.0076 (0.0074)

MedSHExp1/4m 20.0252* (0.0141)

HighShExp1/4m 20.0760*** (0.0203)

LowSHExp1/2m 20.0448*** (0.0068)

MedSHExp1/2m 20.0537*** (0.0097)

HighShExp1/2m 20.0530*** (0.0158)

6 0

u D

u m

m ,

S i

r m

a n

s ,

a n

d S

m e

r s

h

E x h i b i t 8 u (continued)

Spatial Error Regression Model Output

Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

LowSHExp1m 20.0528*** (0.0079)

MedSHExp1m 20.1225*** (0.0091)

HighShExp1m 20.1134*** (0.0153)

LowSHExp1m 20.0127 (0.0113)

MedSHExp1m 20.1209*** (0.0116)

HighShExp1m 20.1056*** (0.0173)

Q22010 0.0095 0.0093 0.0093 0.0093 0.0094 0.0094 0.0093 0.0094 (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059)

Q32010 20.0211*** 20.0211*** 20.0212*** 20.0208*** 20.0211*** 20.0211*** 20.0210*** 20.0209*** (0.0064) (0.0064) (0.0064) (0.0064) (0.0064) (0.0064) (0.0064) (0.0064)

Q42010 20.0497*** 20.0499*** 20.0500*** 20.0499*** 20.0498*** 20.0498*** 20.0498*** 20.0496*** (0.0064) (0.0064) (0.0064) (0.0064) (0.0064) (0.0064) (0.0064) (0.0064)

Q12011 20.0985*** 20.0986*** 20.0986*** 20.0988*** 20.0984*** 20.0986*** 20.0984*** 20.0983*** (0.0064) (0.0064) (0.0064) (0.0064) (0.0064) (0.0064) (0.0064) (0.0064)

Q22011 20.0875*** 20.0876*** 20.0876*** 20.0873*** 20.0875*** 20.0874*** 20.0872*** 20.0878*** (0.0061) (0.0061) (0.0061) (0.0061) (0.0061) (0.0061) (0.0061) (0.0061)

S i

n k

h o

l e

s a

n d

R e

s i

d e

n t

i a

l P

r o

p e

r t

y P

r i

c e

s u

6 1

J R

E R

u V

o l

. 4

0 u

N o

. 1

– 2

0 1

8

E x h i b i t 8 u (continued)

Spatial Error Regression Model Output

Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

Q32011 20.0878*** 20.0879*** 20.0881*** 20.0880*** 20.0878*** 20.0879*** 20.0878*** 20.0878*** (0.0061) (0.0061) (0.0061) (0.0061) (0.0061) (0.0061) (0.0061) (0.0061)

Q42011 20.0904*** 20.0906*** 20.0906*** 20.0904*** 20.0904*** 20.0903*** 20.0903*** 20.0906*** (0.0063) (0.0063) (0.0063) (0.0063) (0.0063) (0.0063) (0.0063) (0.0063)

Q12012 20.0970*** 20.0973*** 20.0973*** 20.0972*** 20.0970*** 20.0972*** 20.0971*** 20.0972*** (0.0063) (0.0063) (0.0063) (0.0063) (0.0063) (0.0063) (0.0063) (0.0063)

Q22012 20.0650*** 20.0653*** 20.0653*** 20.0652*** 20.0650*** 20.0651*** 20.0651*** 20.0649*** (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059)

Q32012 20.0454*** 20.0457*** 20.0458*** 20.0455*** 20.0455*** 20.0456*** 20.0453*** 20.0457*** (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059)

Q42012 20.0275*** 20.0277*** 20.0278*** 20.0277*** 20.0275*** 20.0277*** 20.0275*** 20.0275*** (0.0058) (0.0058) (0.0058) (0.0058) (0.0058) (0.0058) (0.0058) (0.0058)

Q12013 0.0215*** 0.0213*** 0.0213*** 0.0214*** 0.0214*** 0.0214*** 0.0215*** 0.0217*** (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059) (0.0059)

Q22013 0.0597*** 0.0595*** 0.0595*** 0.0596*** 0.0597*** 0.0596*** 0.0596*** 0.0599*** (0.0056) (0.0056) (0.0056) (0.0056) (0.0056) (0.0056) (0.0056) (0.0056)

Q32013 0.0957*** 0.0955*** 0.0954*** 0.0956*** 0.0956*** 0.0957*** 0.0955*** 0.0959*** (0.0056) (0.0056) (0.0056) (0.0056) (0.0056) (0.0056) (0.0057) (0.0057)

Q42013 0.1060*** 0.1060*** 0.1059*** 0.1060*** 0.1059*** 0.1060*** 0.1064*** 0.1060*** (0.0058) (0.0058) (0.0058) (0.0058) (0.0058) (0.0058) (0.0058) (0.0058)

Q12014 0.1136*** 0.1135*** 0.1135*** 0.1136*** 0.1136*** 0.1137*** 0.1137*** 0.1134*** (0.0060) (0.0060) (0.0060) (0.0060) (0.0060) (0.0060) (0.0060) (0.0060)

6 2

u D

u m

m ,

S i

r m

a n

s ,

a n

d S

m e

r s

h

E x h i b i t 8 u (continued)

Spatial Error Regression Model Output

Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

Q22014 0.1489*** 0.1487*** 0.1487*** 0.1487*** 0.1489*** 0.1488*** 0.1489*** 0.1488*** (0.0056) (0.0056) (0.0056) (0.0056) (0.0056) (0.0056) (0.0056) (0.0056)

Q32014 0.1471*** 0.1470*** 0.1470*** 0.1471*** 0.1471*** 0.1469*** 0.1472*** 0.1470*** (0.0057) (0.0057) (0.0057) (0.0057) (0.0057) (0.0057) (0.0057) (0.0057)

Q42014 0.1449*** 0.1450*** 0.1450*** 0.1449*** 0.1450*** 0.1448*** 0.1453*** 0.1454*** (0.0068) (0.0068) (0.0068) (0.0068) (0.0068) (0.0068) (0.0068) (0.0068)

Constant 11.4790*** 11.3707*** 11.3550*** 11.4815*** 11.4805*** 11.4878*** 11.5147*** 11.5323*** (0.0414) (0.0422) (0.0423) (0.0414) (0.0414) (0.0413) (0.0411) (0.0417)

Lamda 0.8527*** 0.8497*** 0.8489*** 0.8502*** 0.8524*** 0.8507*** 0.8465*** 0.8472*** (0.0025) (0.0025) (0.0025) (0.0025) (0.0025) (0.0025) (0.0025) (0.0025)

R2 .8875 .8875 .8875 .8874 .8874 .8873 .8873 .8874

AIC 3818.0 3397.6 3379.5 3457.3 3509.6 3473.7 3350.1 3367.6

Notes: * Significant at the 10% level. *** Significant at the 1% level.

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y P r i c e s u 6 3

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

to have significantly larger lot sizes. This may be because sinkhole locations are more rural but it also may reflect how parcels with natural sinkholes on them were created. Sinkhole properties tend to be significantly older (over 14 years) than properties that are 2 miles from a sinkhole and are more likely to have a pool, fireplace, or a garage than the rest of the sample. Sinkhole properties are also more likely to be in the less densely populated areas of Hillsborough County.

The results in Exhibit 7 also highlight the property characteristic difference based on the distance to the nearest sinkhole for properties that do not have sinkholes on them. As the distance to the nearest sinkhole increases, price, square footage, and population density increase while the properties where the nearest sinkhole is 2 miles away are substantially newer (age of 20 years versus an average age of 32 for properties where the nearest sinkhole is 1 mile away.

After conducting spatial dependence tests of the dependent variable and error terms for OLS regression models, a spatial error regression model was determined to be necessary.14 The spatial error model results for sinkhole presence, proximity, and exposure (density) are shown in Exhibit 8. Model 1 is the base model containing the property, location, and time variables and the variable of interest: the Sinkhole variable. Models 2, 3, and 4 test sinkhole proximity. Model 2 includes the distance to the nearest sinkhole and Model 3 is expanded to include an interaction term (Dist SH2m) to capture an additional effect of proximity on sale price. Model 4 measures the effect of proximity by using discrete distance bands rather than the continuous distance measure from Models 2 and 3. Models 5, 6, 7, and 8 measure the effect of sinkhole exposure (density) using 1⁄4, 1⁄2, 1 mile, and 2 mile distance bands.

All models include control variables for housing characteristics, location (proximity to externalities), and quarterly binary variables controlling for time. Building characteristics include square footage, age, number of bedrooms, lot size, number of stories, swimming pool, fireplace, garage, quality of construction, and floor quality. The coefficients for all models on square footage, age, lot size, swimming pool, fireplace, garage, quality of construction, floor quality, and central heating and cooling are as expected. Newer construction and waterfront have a positive effect on selling price. The quarterly binary variables for the years show the strengthening in house prices that occurred after the fourth quarter of 2012. The coefficient for the Sinkhole variable is negative, indicating that the presence of a sinkhole on the property has a negative effect on the selling price of the property but it is not statistically significant in any model. However, given that the number of sold sinkhole properties is only 0.04% of the total number of sold properties (24 out of 65,205 properties), a not significant coefficient is unsurprising.

Models 2, 3, and 4 are designed to measure the price effects of proximity to sinkholes. The results in Models 2 and 3 show the sinkhole proximity effect on sales price. The results show that selling price increases as the distance from the nearest sinkhole increases. The inclusion of the interaction term in Model 3 shows

6 4 u D u m m , S i r m a n s , a n d S m e r s h

that this effect is more pronounced for properties that are nearer to sinkholes (i.e., within 2 miles). The results in Model 4 also capture the sinkhole proximity effect using a set of four distance bands. The negative effect on selling price is strongest for the nearest band (within 1⁄4 mile of a sinkhole) with a coefficient of 20.046 and decreases to 20.036 for the 1⁄2 mile band. The coefficient approaches zero for the 1 mile proximity band and then becomes positive for properties for the 2 mile proximity band (where the nearest sinkhole is 2 miles away).

The results for Models 5–8 show a similar pattern for sinkhole exposure (density). As stated, the outer distance bands include the properties in the inner distance bands (i.e., are cumulative) where the 2-mile band includes all properties within 2 miles of a sinkhole. Model 5 measures sinkhole exposure (density) within 1⁄4 mile of the property and the coefficient on the high sinkhole exposure variable is 20.076 and significant (low and medium sinkhole exposures are not significant). Model 6 measures sinkhole exposure (density) within 1⁄2 mile of the property. The results show that all exposures (low, medium, and high) have a negative effect on price, where the medium and high exposures have a greater impact than low exposure to sinkhole. Expanding exposure out to 1 or 2 miles, the results show that low sinkhole exposure does not impact the selling price (the result for low exposure is not significant). On the other hand, medium and high sinkhole exposures (density) have a negative impact on price.

u S u m m a r y a n d C o n c l u s i o n s We examine the effect of the presence of and the proximity to a sinkhole on house prices using geo-coded Florida real estate home sales data for the 2010–2014 period. Florida ranks highest in the nation for sinkhole risk since its primary land surface covers rocks (e.g., limestone, carbonate rock, and salt beds) that can be naturally dissolved by groundwater. Using a hedonic pricing model, the results did not provide support for a negative effect of sinkhole presence on a property (i.e., the null hypothesis could not be rejected). As noted above, this is not a surprising result given the small number of properties that were sold with sinkholes on them (24 out of 65,205 properties).

The results did provide strong support that sinkhole proximity has a negative impact on selling price and this was the case whether evaluating the impact of proximity to a sinkhole as measured to the nearest sinkhole or within distance bands. A property 1 mile from a sinkhole would, on average, have a price that is 6.23% higher than a property next to a sinkhole. This effect is more pronounced when considering properties within 2 miles of a sinkhole. Here, the price for the property 1 mile from sinkhole would enjoy a premium of just over 9%.

The results also provide support that sinkhole exposure (density) also adversely impacts sales price. High exposure to sinkholes had a negative effect on price ranging from 7.4% for the 1⁄4 mile band to just over 10% for the 2 mile band from the property. The findings show, whether the sinkhole is considered a

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y P r i c e s u 6 5

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

negative externality in terms of proximity or exposure, both are capitalized into the selling price of residential real estate.

u E n d n o t e s 1 Sirmans, Macpherson, and Zietz’s (2005) review of 125 hedonic pricing models studies

(examining over 150 variables) shows that 17 of the top 20 variables appearing in these studies are housing characteristics.

2 For a comprehensive review of hedonic pricing models, see Sirmans, Macpherson, and Zietz (2005). They review 125 studies that have used hedonic pricing models in residential real estate. They examine over 150 variables.

3 One important point that Fluery (2007) makes is that sinkholes serve as a signal to the public of possible environmental effects, most of which are negative. The most extreme of these is the possible loss of property and / or life. There may also be negative effects of being located within proximity of a sinkhole(s) since this would presumably increase the possibility of a sinkhole developing on your property. A possible positive effect would be that the sinkhole may have an aesthetic appeal (e.g., has the appearance of a natural or man-made lake).

4 Although some people have expressed concerns about the relation between fracking and sinkholes, there seems to be a dearth of literature connecting the two. For example, writing in a Politics and International Relations blog published on December 10, 2014, Charlie Provah indicates that fracking produces concerns not only about chemical contamination and air pollution but also about sinkholes.

5 The geomorphology of Florida is unique because most of the state is not a sediment ramp off the piedmont or continent to the Atlantic Ocean or the Gulf of Mexico. The Floridian section is part of an enormous carbonate bank, called the Florida Platform, which has existed since at least the Cretaceous Period. Residing entirely within the Coastal Plain, the Florida Platform is covered by a thin layer of younger quartz sand and clay sediments. It is these carbonate rocks that are, in part, responsible for many of the karst features common to Florida, such as springs and sinkholes.

6 Sinkholes are not a problem unique to the U.S. Enormous sinkholes exist in China, Laos, New Guinea, and Venezuela. Sinkholes are also found in Italy, South Africa, Australia, Croatia, Guatemala, Belize, and Mexico.

7 RiskMeter.com from CDS Business Mapping provides a sinkhole database and sinkhole clearinghouse, which contains more than 12,000 sinkholes. It is used by underwriters and agents to determine proximity to natural hazards, including sinkholes.

8 Sources: Florida Department of Environmental Protection and Florida Senate report. 9 A longer time period would have been desirable in order to capture more sinkhole

property sales but prior to 2010 the date of sinkhole occurrence relative to the date of sale could not be verified to a satisfactory level. Thus the property sales prior to 2010 were not included in the analysis.

10 The Tampa-St. Petersburg-Clearwater, Florida MSA has a total population of nearly 3 million, just slightly larger than the Orlando-Kissimmee, Florida MSA and second only to the Miami-Fort Lauderdale-Pompano Beach, Florida MSA with a population approaching 6 million. The State of Florida has a population of roughly 20 million, and is growing by nearly 1,000 people per day, a net population growth rate of 1.8%.

6 6 u D u m m , S i r m a n s , a n d S m e r s h

However, the strongest growth is not in Tampa but rather in north-central Florida, south Florida, southwest Florida, and the panhandle. With median house prices of just under $200,000, Hillsborough and Pinellas counties are in the top third of the state, but far from the most expensive areas to live. For example, areas of Miami-Dade and Monroe County (the Florida Keys) have median house prices that are more than double those in Tampa. Similarly, median household income in the Tampa MSA is very close to both the state and national average.

11 Since many defective properties are foreclosed upon, it would interesting to perform a difference of means test for properties in our sample group (arms-length transactions) relative to foreclosed sinkhole properties. Unfortunately, we do not have the sample of foreclosed sinkhole properties.

12 In general, a seller has a duty to disclose any material defect that may affect the property value, thus proximity to a sinkhole(s) is important. Florida statute requires that sellers and lessors disclose ‘‘known’’ sinkhole activity in an area.

13 For the 1⁄4 mile band, less than 2 sinkholes is low exposure; 2 sinkholes is medium exposure, and more than 2 sinkholes is high exposure. For the 1⁄2-mile band, less and 2 sinkholes is low exposure; between 2 and 5 sinkholes is medium exposure, and more than 5 sinkholes is high exposure. For the 1-mile band, less than 2 sinkholes is low exposure; between 2 and 17 sinkholes is medium exposure, and more than 17 sinkholes is high exposure. For the 2-mile band, less than 4 sinkholes is low exposure, between 4 and 57 sinkholes is medium exposure, and more than 57 sinkholes is high exposure.

14 Spatial autocorrelation is common in hedonic house price models (since the transaction prices of single-family properties tend to be similar across space) although the primary issue is with the residuals. This is an issue because spatial autocorrelation causes the assumption of independence of the residuals to be violated, and can call into question the validity of estimation results. Here, the Moran’s I statistic was 7.8 for the dependent variable, and around 5.6 for the residuals in most of the OLS models. Accordingly, a spatial regression model was incorporated and since spatial autocorrelation in this situation was assumed to be the result of seemingly omitted (yet non-apparent) spatial variables, a spatial error model was deemed most appropriate.

u R e f e r e n c e s Bin, O. and J. Kruse. Real Estate Market Response to Coastal Flood Hazards. Natural Hazards Review, 2006, 7:4, 137–44.

Bin, O. and S. Polasky. Effects of Flood Hazards on Property Values: Evidence Before and After Hurricane Floyd. Land Economics, 2004, 80:4, 490–500.

Boyle, M. and K. Kiel. A Survey of House Price Hedonic Studies of the Impact of Environmental Externalities. Journal of Real Estate Literature, 2001, 9:2, 116–44.

Dumm, R., G. Smersh, and G.S. Sirmans. Price Variation in Waterfront Properties over the Economic Cycle. Journal of Real Estate Research, 2016, 38:1, 1–26.

Farber, S. Undesirable Facilities and Property Values: A Summary of Empirical Studies. Ecological Economics, 1998, 24, 1–14.

Fleury, S. Are Home Values Affected by Sinkhole Proximity? Results of a Hedonic Price Model. The Florida Geographer, 2007, 72–84.

Galve, J.P., F. Gutiérrez, P. Lucha, J. Guerrero, J. Bonachea, J. Remondo, and A. Cendrero. Probabilistic Sinkhole Modelling for Hazard Assessment. Earth Surface Processes and Landforms, 2009, 34:3, 437–52.

S i n k h o l e s a n d R e s i d e n t i a l P r o p e r t y P r i c e s u 6 7

J R E R u V o l . 4 0 u N o . 1 – 2 0 1 8

Gao, Y. and E.C. Alexander. A Mathematical Model for a Map of Relative Sinkhole Risk in Fillmore County, Minnesota. In: Sinkholes and the Engineering and Environmental Impacts of Karst. ASCE, 2003.

Gutiérrez, F., J. Guerrero, and P. Lucha. Quantitative Sinkhole Hazard Assessment. A Case Study from the Ebro Valley Evaporite Alluvial Karst (NE Spain). Natural Hazards, 2008, 45:2, 211–33.

Harrison, D., G.T. Smersh, and A. Schwartz. Environmental Determinants of Housing Prices: The Impact of Flood Zone Status. Journal of Real Estate Research, 2001, 21:1–2, 3–20.

NASA. NASA Radar Demonstrates Ability to Foresee Sinkholes. Retrieved from: https: / / www.nasa.gov / press / 2014 / march / nasa-radar-demonstrates-ability-to-foresee-sinkholes / #.V5eJ5DVEpP0.

Jackson, T. The Effects of Environmental Contamination of Real Estate: A Literature Review. Journal of Real Estate Literature, 2001, 9:2, 93–116.

Murdoch, J.C., H. Singh, and M. Thayer. The Impact of Natural Hazards on Housing Values: The Loma Prieta Earthquake. Real Estate Economics, 1993, 21:2, 167–84.

Newton, J.G. Natural and Induced Sinkhole Development in the Eastern United States. In: Proceedings of the Third International Symposium on Land Subsidence, International

Association of Hydrological Sciences. Wallingford, UK, 1984.

Nyce, C., R. Dumm, G. Smersh, and G.S. Sirmans. The Capitalization of Homeowners’ Insurance Premiums in House Prices. Journal of Risk and Insurance, 2015, 82:4, 891–919.

Pappas, S. What are Sinkholes? Live Science, March 15, 2014. http: / / www. livescience.com / 44123-what-are-sinkholes.html.

Radazzo, A.F. Are There Different Types of Sinkholes? October 15, 2015. http: / / www.geohazardsinc.com / why-do-sinkholes-form-in-florida / .

Simons, R.A., A.M. Bowen, and A.J. Sementelli. The Price and Liquidity Effects of UST Leaks from Gas Stations and Adjacent Contaminated Property. The Appraisal Journal, 1999, 67, 186–94.

Simons, R.A. and J.D. Saginor. A Meta-Analysis of the Effect of Environmental Contamination and Positive Amenities on Residential Real Estate. Journal of Real Estate Research, 2006, 28:1, 71–104.

Simons, R.A., W. Levin, and A. Sementelli. The Effect of Underground Storage Tanks on Residential Property Values in Cuyahoga County, Ohio. Journal of Real Estate Research, 1997, 14:1, 29–42.

Sinclair, W.C. Sinkhole Development Resulting from Ground-water Withdrawal in the Tampa Area, Florida. U.S. Geological Survey, Water Resources Division, 1982.

Sirmans, G.S., D.A. Macpherson, and E. Zietz. The Composition of Hedonic Pricing Models. Journal of Real Estate Literature, 2005, 13:1, 1–44.

Skantz, T. and T. Strickland. House Prices and a Flood Event: An Empirical Investigation of Market Efficiency. Journal of Real Estate Research, 1987, 2:2, 75–83.

Smith, K.V. and J. Huang. Can Markets Value Air Quality? A Meta-Analysis of Hedonic Property Value Models. Journal of Political Economy, 1995, 103:1, 209–27.

Upchurch, S.B. and J.R. Littlefield Jr. Evaluation of Data for Sinkhole-Development Risk Models. Environmental Geology and Water Sciences, 1988, 12:2, 135–40.

Zabel, J.E. and D. Guignet. A Hedonic Analysis of the Impact of LUST Sites on House Prices. Resource and Energy Economics, 2012, 34:4, 549–64.

6 8 u D u m m , S i r m a n s , a n d S m e r s h

The authors would like to thank four anonymous referees for helpful comments. The

authors wish to recognize Mehdi M. Khan, GIS Program Manager, Office of Planning

and Zoning, Anne Arundel County, Maryland for his help with GIS calculations that

were used in the data for this study.

Randy E. Dumm, Temple University, Philadelphia, PA 19122 or randy.dumm@

temple.edu.

G. Stacy Sirmans, Florida State University, Tallahassee, FL 32306-1110 or

[email protected].

Greg T. Smersh, University of South Florida, Tampa, FL 33620 or [email protected].

Reproduced with permission of copyright owner. Further reproduction prohibited without permission.