Adaptation Infrastructure and Its Effects on Property Values in the Face of Climate Risk
Abstract
We evaluate the effect of climate adaptation infrastructure investments on property transaction prices, using data on over 400,000 property transactions and 162 infrastructure projects in Miami-Dade County, an area that is highly vulnerable to flooding and sea level rise due to climate change. Exploiting the timing and siting of different adaptation projects in Miami-Dade, we are able to identify significant gains in property values after completion of adaptation infrastructure projects. These gains are concentrated in areas close to the project and for projects that are visually identifiable. Our results suggest an aggregate mean benefit, net of adaptation cost, of about $0.68 million per project and almost $300 million in aggregate net benefits for all projects in our sample. Most projects generated positive net benefits, indicating that the vast majority of adaptation efforts are being placed in areas passing a benefit-cost test.
The global mean temperature has already risen by 0.95°C relative to preindustrial levels (NOAA 2020), and adverse impacts from climate change are already occurring in many sectors. Given that even aggressive mitigation efforts are unlikely to lessen such impacts in the near term, households, firms, and governments are considering actions to limit the damage from climate change. Efforts to alleviate adverse climate change impacts are known as adaptation and are a promising approach for reducing the damage from climate change. For example, the Fourth National Climate Assessment estimates that coastal property damage from sea level rise may be reduced by 90% through adaptation (EPA 2017).1 Relative to the inevitability of at least moderate sea level rise, the ability to avoid an overwhelming percentage of damages is remarkable.
Still, for successful adaptation to take place, households, firms, and local governments must evaluate potential adaptation investments that have up-front costs and uncertain future benefits. For instance, local governments must predict sea level rise to separate which properties might be saved by infrastructure adaptation investments at a cost less than the value of the properties and which should be abandoned. Households must also understand climate change and the benefits of adaptation. Otherwise, local governments may not get political approval for such infrastructure investments. Accordingly, successful and efficient adaptation requires adoption of adaptation infrastructure in vulnerable areas for which the marginal adaptation net benefits are large. If adaptation investments are instead solely undertaken by local governments with the most funding or are chosen based on political considerations rather than need, then the benefits of infrastructure adaptation may fall significantly.
This study analyzes this issue by evaluating the effectiveness of adaptation infrastructure investments, as measured by their effect on property values. Our data set consists of all real estate property transactions and 162 adaptation infrastructure projects in the Miami-Dade County (MDC), an area that is highly vulnerable to sea level rise due to climate change (likely the most vulnerable in the continental United States). The analysis provides empirically grounded answers to several policy-relevant questions. First, is flood and storm surge risk a significant factor for buyers and sellers in coastal property markets? If property values appreciate near infrastructure projects after completion, then coastal property buyers and sellers are aware that the area is vulnerable and recognize the benefits of adaptation infrastructure. Second, is it efficient to protect vulnerable coastal properties, or should another option like retreat be considered? If infrastructure benefits are capitalized into property values, then we can calculate the benefits of adaptation infrastructure and, given our data on infrastructure costs, determine whether or not aggregate adaptation infrastructure investments benefits exceed construction costs. Third, a larger positive question exists in the literature as to whether climate change is itself reflected in property values (McAlpine and Porter 2018; Hino and Burke 2021). A significant increase in property values following infrastructure investments to protect against what MDC categorizes as flood and climate risk provides a strong indication that in fact the potential negative effects of climate change are capitalized in property values.
Establishing the effect of infrastructure adaptation investments on property values, however, presents a number of econometric challenges. First, municipalities deploy infrastructure projects at different points in time in different locations. To address this dynamic heterogeneity, we construct a high-resolution repeated cross-section and track individual property values across multiple communities before and after adaptation projects are finalized.
Second, unobserved factors may affect both flood risk and property values. For example, developers may build lower quality housing in low-elevation areas (or higher quality housing in areas with desirable views), creating a correlation between prices and flood risk. Further, macroeconomic shocks such as interest rate changes could result in spurious correlations between flood risk and some neighborhoods’ market trends. To address these concerns, we rely on a difference-in-differences approach and evaluate the differences in property prices following construction of adaptation infrastructure near a transacted parcel vis-à-vis the change farther away. This approach controls for preexisting trends as well as unobserved neighborhood quality differences. Our specification further controls for heterogeneity in housing quality by focusing on properties sold more than once.
Third, adaptation projects are not allocated at random. Guidelines exist for prioritization of adaptation investment deployment, and the characteristics driving investment decisions might be correlated with the changes in property values. To alleviate these problems, we implement a rich set of controls for individual property characteristics, as well as time-specific trends by neighborhood.
Our data come from the Miami-Dade Property Appraiser Office, and the records consist of over 400,000 transactions from 2000 to 2019. Because each transaction is based on buyers and sellers evaluating the property, the transaction price is more likely to reflect individual property characteristics and current value, relative to appraisal data based on nearby transactions that have occurred in the past. Our projects data consist of 162 fixed infrastructure projects categorized by MDC as dealing with storm surge, flooding, and/or sea level rise. These projects include installation and repair of drainage, pumping stations, reservoirs, seawalls, road elevation, shoreline stabilization, and other investments in adaptation capital. The large number of temporally and geographically diverse projects also allows us to estimate the value of projects by project type (e.g., reservoirs, pumping stations).
Consistent with a market that recognizes climate risk and the benefits of adaptation investment, we find statistically significant gains in property values close to infrastructure projects after installation relative to farther away. These gains, however, are concentrated in properties near the project and are largest for more visible projects such as elevating roads. Our results are highly robust to a variety of alternative econometric specifications and provide evidence that projects are being deployed in vulnerable areas and provide protection, as evidenced by transaction prices increasing near the project subsequent to its completion.
We also calculate the return on investment for the adaptation infrastructure projects. Miami-Dade is a dense urban environment, and so most projects are surrounded by hundreds or thousands of properties, which leads to relatively large aggregate net benefits across MDC. As most projects generated positive net benefits, this pattern indicates that the majority of projects are being placed in areas that pass the benefit-cost test.
Combined, these results provide evidence that property buyers, sellers, and local governments not only perceive risks of storm surge, flooding, sea level rise, tropical cyclone activity, and other impacts exacerbated by climate change but also perceive that infrastructure projects alleviate such risks. The implication is that the local community is able to overcome coordination barriers at least enough so that most projects have benefits reflected in property values that exceed construction costs. This result is surprising. While MDC often cites studies showing benefit/cost ratios of up to 9 to 1 (Urban Land Institute 2020), many science and policy groups are skeptical that this sort of coordination is easy to achieve (Shi and Moser 2021). Further, evidence exists that public infrastructure projects are becoming more expensive and provide only limited benefits (e.g., Duranton et al. [2021] find these stylized facts in a review of the transportation infrastructure literature).
The next section provides a summary of the literature on adaptation, property values, and sea level rise. Section 2 provides a theoretical model that underpins our empirical strategy. Section 3 describes the data in detail, and section 4 gives the empirical specification. Section 5 shows the results, and section 6 concludes.
1. Background
An important existing literature connects real estate market dynamics, climate risk, and adaptation. Much of the literature focuses on the loss to property values subsequent to hurricane events.2 Bin and Polasky (2004), Hallstrom and Smith (2005), Bin and Landry (2013), Ortega and Taspınar (2018), Davlasheridze and Fan (2019), and Gibson and Mullins (2020) estimate declines in property values following individual hurricanes.3 Similarly, Shr and Zipp (2019) and Hino and Burke (2021) estimate the effect of flooding risk (defined as being in a Federal Emergency Management Agency [FEMA] flood zone) on property values following changes in flood zone maps. A common result is that prices in well-informed markets decline initially but often the effect fades over time. In addition, in most studies the decline in property values exceeds the cost of insurance, indicating that some flood and hurricane risks which are not insurable, such as job loss risk, affect property values by suppressing demand for properties exposed to these risks. Our results strengthen this literature by showing that coastal infrastructure can mitigate such losses and that mitigated losses exceed insurance costs.
The relationship between adaptation and property values depends on risk perceptions. Using surveys, Ludy and Kondolf (2012) find that households underestimated the residual risk of flooding after a levee was built, and Bakkensen and Barrage (2017) show that coastal flood zone residents have lower flood risk perceptions than their inland counterparts. One can also infer changes in risk perceptions by changes in the purchase of insurance and household adaptation investments following flood or storm events (e.g., Bin and Landry 2013; Gallagher 2014).4 A general result from this literature is that perceptions of risk increase following a storm or flood but often fade thereafter. Our results provide corroborating evidence for this literature, as gains in property value resulting from the addition of adaptation infrastructure imply a change in risk perceptions.
Directly relevant to our results are a small number of studies that estimate the total cost of flooding, storms, and sea level rise to coastal communities, in terms of lost real estate value. Keenan et al. (2018) find a positive association of elevation and property appreciation for most jurisdictions in MDC. McAlpine and Porter (2018) integrate flood, surge, and sea level rise data with real estate property transaction from MDC. They find that lots expected to be affected by tidal flooding in 2032 have already experienced $1,276 declines in value each year between 2005 and 2016. Finally, Bernstein et al. (2019) show that properties that would be under water with 1 foot of sea level rise sell at a 14.7% discount, whereas properties that require 6 feet of sea level rise to be inundated experience only a 4.4% discount.
These studies do not rely on a particular storm or flood event but instead correlate sea level rise risk with property values. This approach allows an estimate of the total cost of sea level rise using property values rather than a change from an unknown baseline. However, this advantage comes with a trade-off. First, the argument for causality is weaker relative to studies that rely on a single exogenous storm or flood map change. Second, it is unclear to what degree property values have priced in future adaptation investments that can reduce the overall cost of sea level rise. Our results take the next step forward in this literature. Because we contrast property values before and after public adaptation infrastructure, causality relies on parallel trends in comparable properties that are sufficiently far from projects. In addition, we can estimate the price effect of the reduction in impacts resulting from adaptation infrastructure by use of the counterfactual when no infrastructure projects are built.
Closely related to our results, there are a number of estimates of the value of adaptation investments, especially infrastructure. For example, Fell and Kousky (2015) find that commercial properties in 100-year floodplains protected by a levee in Missouri sell at an 8% premium relative to unprotected properties in 100-year floodplains. Similarly, Jin et al. (2015) find that coastal properties in Massachusetts with seawalls have property values 10% higher than unprotected properties.5 Walsh et al. (2019) estimate that properties with bulkheads and riprap are associated with price premiums of about $66,000 and $102,000, respectively. Barrage and Furst (2019) show that construction activity is negatively associated with sea level rise exposure in areas in which polls indicate that residents are worried about climate change. These papers take an important first step in showing the association between property values and adaptation investments. Yet, causality is difficult to ascertain since most infrastructure investments have been in place for many decades. Because we study a region with hundreds of recently completed infrastructure projects, we are able to improve on this research by employing methods that leverage before and after completion data to produce plausibly causal estimates.
Finally, Kim (2020) also looks at the effect of infrastructure on property values in MDC using a difference-in-differences framework and finds that infrastructure projects have a very large effect, as much as an 18.1% increase in property values for storm surge projects in one year after completion. Our study improves this work in a number of ways. First we use a much larger data set, which allows us to better control for individual property characteristics by using repeated sales. Second, we implement a difference-in-differences design that corrects for the variation in adaptation infrastructure timing. With these improvements, we find that infrastructure projects generate only about a 5% increase in property values in the year after completion, rising slowly to almost a 10% increase five years after completion. This indicates that controlling for property-level heterogeneity is not only important, but necessary for estimating the effect of infrastructure projects. Finally, we also break down our results by project type, and examine the implications of our estimates in terms of the reduction in climate impacts resulting from adaptation investment.
2. Theory
Here we present a theoretical model of the housing market with adaptation infrastructure. The model shows how the identification assumptions used in the empirical section relate to theoretical assumptions and properties of the housing market. Further, we derive theoretical predictions as to how the empirical results are affected if the identification assumptions are violated. Finally, the theory also implies patterns in the dynamic adjustment of property values after construction of adaptation infrastructure, which we test for empirically.
2.1. Rental Prices
The model determines an equilibrium rental price of housing based on several assumptions. First, households indexed by have Cobb-Douglas preferences for housing service flows from private and public amenities, hit, adaptation infrastructure service flow amenities, ait, and a numeraire consumption good. This assumption implies that households spend a constant fraction of income on adaptation amenity service flows. Households obtain service flows by renting a property. Second, the supply of properties, and thus the supply of service flows, is fixed. This assumption generates equilibrium rental prices Rjt for properties that depend on a property quality index I (Hj, Aj), where Hj and Aj are the housing and adaptation amenity service flows from property j. We assume that changes in public adaptation amenities and property service flows are unanticipated or equivalently that H and A are independent of time. Finally, household income yit grows at average rate g. This assumption implies that housing rental prices also increase at trend rate g. The equations for the quality index and the rental price of property j are (see Kelly and Molina [2022] for detailed derivations of all results in this section):
2.2. Property Price
No-arbitrage requires that the sale price of a property P(j, t) equals the present discounted value of the rental income. Equivalently, buyers in the property market are indifferent between buying and renting. Thus:
Integrating equation (3) reveals that
2.3. Amenity Changes
Consider now an unanticipated public adaptation infrastructure amenity increase that affects a subset of properties. It is likely that such infrastructure provides different levels of protection and thus different amenity values to different properties. For example, without adaptation a once-in-10-year flood might damage properties adjacent to the coast, but not inland properties, but a once-in-100 year flood might damage both. Therefore, a seawall would provide more protection to the properties adjacent to the coast because such properties are more vulnerable. The protection provided by the seawall then declines with distance from the seawall.7
To model this effect, let dj denote the distance between property j and the nearest public adaptation project. Then construction of the public adaptation infrastructure increases the amenity value from Aj to , where the nonincreasing function f governs how amenity value decreases with distance.
Next, consider two properties j and j′. Let denote the construction completion date of the project nearest to property j and the completion date of the project nearest to property j′. Then for transactions that occur at t0 and t1, where , we can derive from equation (4) the price prior to treatment, , the price subsequent to treatment , and the price of the control property both before and after treatment.
These prices may be used to form a difference-in-differences estimator, where the interest is in the difference between the treated property price less the unobserved counterfactual . The difference is identified via the assumption of parallel trends between properties j and j′, or
The identification assumption in (5) embeds two implicit assumptions:
Properties j and j′ have identical trend growth rates.
There exists a property transaction for j′ after completion of the project associated with property j but before completion of the project associated with property j′. That is, there exists a t1 such that .
In addition, if f is nonincreasing in distance as hypothesized, then the difference in log prices is also nonincreasing in distance between the amenity and property j. A property with greater service flows or private adaptation amenities, Hj, sees increases in price both before and after an adaptation infrastructure amenity is constructed, but the difference is a decreasing function of Hj. Because households have diminishing marginal utility, the additional utility and thus the additional willingness to pay for increases in adaptation amenities diminishes in the level of other property service flows. Similarly, as adaptation infrastructure amenities increase, the log price increases by a smaller amount for a property with higher private amenities. Given that adaptation infrastructure has already made the property less affordable, further increases in private amenities can generate only small further increases in prices and so the value of additional private amenities decreases with adaptation infrastructure amenities. Market-wide economic effects, like changes in the market interest rate r or trend growth g, are eliminated through the difference-in-difference.
Importantly, deviations from fundamental value (θ), which vary over time, can be interpreted as a slow adjustment of prices after completion of the adaptation project. For example, suppose that prices always reflect fundamental values (), then the creation of the adaptation amenity results in an instantaneous increase in property values equal to the percentage increase in property quality. Figure 1A graphs this case. Price evolution after adaptation investment. Panel A shows a difference in differences with instantaneous price adjustment, while panel B shows a slow adjustment. The dotted lines represent the price of the control property j′, while the solid line depicts the price of a property j for which the closest project is completed at time t*. For panel B, denotes the end of the linear adjustment in prices after the adaptation project is finalized.
Now consider a second special case, where property prices are at fundamental values prior to completion of the project, and then the value of the project is incorporated into the housing price at a linear rate over the period . In this case, for all t, , and
Therefore, the difference-in-differences estimator picks up the partial price adjustment that has accrued at some arbitrary point in time, t1. Figure 1B graphs this case, which illustrates the importance of examining the results over time, as a simple average of all observations subsequent to t* underestimates the long-run value of the adaptation amenity. We provide the empirical version of figure 1 in section 5, figure 3.
Other predictions are also possible depending on θ. In general, however, if the value of the adaptation amenity is quickly incorporated into the property value, then the value generated from the coefficient on the post completion and post completion-distance interaction terms in the difference-in-differences estimator is close to the long-run benefit. In contrast, if the value of the adaptation amenity is incorporated slowly over time and many of the sales are soon after the date of completion, the long-run benefit exceeds the value generated from the estimated coefficients.
2.4. Econometric Predictions
Here we discuss several possible violations of parallel trends and then develop theoretical predictions on how such violations might be expected to be seen in an empirical analysis.
2.4.1. Staggered Rollout Assumptions
Section 2.3 shows that, given assumptions 1 and 2, the difference-in-differences equals the percentage increase in the property quality index resulting from the adaptation amenity plus changes in deviations from fundamental value. Assumption 1 requires that trend growth in rental prices g is identical across properties j and j′, even though the properties may be in different locations. If the trend growth rate varied across neighborhoods around projects, equation (6) would have an extra term on the right-hand side equal to the difference in trend growth rates. Since we have assumed that the project associated with j′ is completed at a later date, if early projects were completed in areas with faster (slower) growth rates than later projects, the coefficient of the interaction term between distance from the project and a post-completion indicator would be biased upward (downward). This problem has been pointed out by Callaway et al. (2021), Sun and Abraham (2021), and others. To guard against this, we include a battery of controls, including project and zip-by-year fixed effects. The empirical section will show that adding such fixed effects terms has only small effects on the coefficients, indicating that the assumption of identical trends across project areas, while stronger than parallel trends with a single treatment/project, is reasonable.
Next, assumption 2 requires that the time period t1 is such that project j is completed but project j′ is not. That is, the data are a “staggered treatment,” where the later rollout/completion of the project associated with j′ allows identification relying on parallel trends. Of course, the difference-in-differences estimator is in fact a variance weighted average of all possible two by two property/time periods in the data (Goodman-Bacon 2021). Thus, many property/time periods are such that both j and j′ are treated (associated with completed projects at time ). In this case, the last two terms of the difference-in-differences formula (6) change from the percentage increase in the property quality index to the difference in the percentage increases of the property quality indices. Therefore, the difference-in-differences procedure creates a biased-downward estimate of the effect of infrastructure on property values because the difference-in-differences includes differences between two treatments as well as differences between treated and not treated.
The quantitative significance of this bias depends on how the variance weighted least squares estimator weights the difference in treated pairs versus the treated/untreated pairs. The bias is relatively small if the treated/untreated pairs are small in number and/or have low variance weight. In the empirical section, we consider also the Sun and Abraham (2021) procedure which addresses this issue and show that the coefficient estimates and standard errors exhibit only small changes in our study.
2.4.2. Parallel Trends Violation versus Slow Adjustment
One possible concern is that the slowly increasing price difference in figure 1B might be caused by a shift in trends rather than a slow adjustment to a change in fundamental value. Suppose in equation (6) that the trend growth is g0 in both treatment and control properties prior to completion but shifts to in the treatment neighborhood after project completion. For example, the neighborhood j becomes more attractive to wealthier buyers, who also have a higher average income growth rate.
Suppose further that so that there is no mispricing prior to completion, and changes in buyer incomes are unanticipated until completion. Equation (6) becomes
Initially, either a slow price adjustment ( and approaching zero over time) or an increase in the price trend can result in the upward trend seen in figure 1B. However, for the mispricing ends whereas the difference in trends remains. Hence, an empirical price difference that shows an increasing trend after completion that levels off, as in figure 1, indicates a slow price adjustment. Conversely, a continuously increasing price difference after completion indicates a violation of parallel trends, since mispricing that increases over time is inconsistent with rationality.
2.4.3. Parallel Trends Violation via Anticipated Treatment
Consider next a potential violation of parallel trends that occurs if the increase in property value after completion is anticipated before completion at some date . In this case, the rental prices still shift at t*, but the no-arbitrage condition (3) implies that the anticipated price, PA, is
1. : some of the public adaptation amenity value is priced in prior to completion.
2. : even though the market anticipates higher rental rates in the future, the value of the property is lower prior to completion, as a period of time exists when rental rates are lower as the public adaptation is not yet completed.
3. is increasing in t0: the anticipated public adaptation amenity becomes more valuable as completion time nears, and hence prior to completion the difference between treated and control properties widens as completion time nears.
3. Data
Analyzing the influence of adaptation infrastructure on the housing market requires spatial integration of transaction data with local infrastructure projects. We obtain individual property transactions along with specific housing characteristics for MDC from the Miami-Dade County Appraiser’s Office.
The data contain housing characteristics, including number of stories, bedrooms, baths, and the year of construction for each property. For properties sold at least three times, each property record contains the three most recent transactions. Properties sold once or twice contain one or two transactions, respectively. Each transaction is listed as a qualified or nonqualified sale.9 Each transaction is given a unique identifying number, known as a book and page number. All property transaction prices are converted to 2019 dollars using the consumer price index for all urban consumers as reported by the US Bureau of Labor Statistics.
We focus on condominiums and single family residential homes. We exclude commercial property transactions because the incentives propagated by climate adaptation infrastructure likely differ for residential and commercial properties (Fell and Kousky 2015). Further, only qualified sales are included in the final data set. Finally, and to align with completion dates of the projects, we limit the data to the period 2000–2019 and focus on properties sold at least twice.
The data also have a number of sales that combine more than one property (e.g., an entire condominium building). Hereafter, we denote such sales as “wholesale” transactions. The appraiser records the sale price of each individual unit in a wholesale transaction as the sale price for all units in the transaction.10 Wholesale transactions are identified in the data as multiple transactions with the same book and page number. To account for this feature, we first remove any wholesale transactions that include both residential and commercial properties. For wholesale transactions that are strictly residential, we then let the property transaction price equal the total sale price for all properties divided by the number of units sold.
There are two potential problems with this adjustment. First, some units may be larger or of higher quality than others and thus have a value that exceeds the average price. Second, a unit sold as part of a wholesale transaction that is subsequently sold at least three times is not available in the data, as the appraiser keeps only the last three transactions. Thus, the number of units in a wholesale transaction may be underestimated, causing the sale price to be overestimated. For this reason, we explicitly control for these type of properties in the empirical analysis.
Some book and page numbers are also missing in the data. Inspection of the data reveals that some transactions with missing book and page numbers appear to be wholesale transactions. These data points have very large positive (e.g., if the property was sold individually prior to the wholesale transaction) or negative (e.g., if the unit was sold subsequent to the wholesale transaction) appreciation rates. However, there is no clear way to filter these data points. For this reason, in the regressions we will remove outliers in both tails (i.e., top and bottom fifth percentiles of price changes).11
Using spatially designated data for the Florida coastline, we calculate the nearest distance of each property’s centroid to the coast as well as the parcel’s elevation. We assign inherent flood risk for each property using delineated FEMA flood zone designations.12
The next step of our data setup requires obtaining and spatially locating adaptation infrastructure projects in MDC. These projects are referred to as LMS (local mitigation strategy) projects. Data for a total of 1,958 projects were downloaded from MDC’s open source website. The data include the geographic location of the project construction site, title, cost, start and end years, description, and the particular hazard or set of hazards that the project is designed to address. Each data point is potentially recorded by a different administrator, and so the descriptions vary in detail and many observations have missing data.
The LMS includes projects that are completed, projects that are still under construction, and projects that are still in the planning stages or have yet to be funded. The LMS is coordinated by the Whole Community Infrastructure Planner (Planner) of the Miami-Dade Office of Emergency Management.
In the selection process, the Planner writes a letter of support for grant opportunities. Vulnerability assessments are then conducted and hazard mitigation opportunities are identified. Accordingly, priorities are established concerning each proposed project’s impact on (i) life safety, (ii) quality of life, (iii) cost effectiveness, and (iv) value to the overall community. Projects are then prioritized following three criteria: suitability (30%), risk reduction potential (55%), and cost and time (15%). In addition, each agency proposing a project is required to complete a self-prioritization process identifying where they scale among these criteria.13
We restrict the data to projects that report geographic construction location. Many projects are ongoing; we consider only completed projects that report an end year. We further restrict the data to projects specifically addressing sea level rise, flooding, storm surge, and wind. Within these categories, we include projects related to canals and rivers, coastal erosion, drainage, flooding, road protection/elevation, and bridge protections. We exclude projects designed to protect public buildings (e.g., adding hurricane shutters to a police station), projects without a fixed location (e.g., mobile pumping stations), and feasibility studies.14
Most projects are irregular polygons (i.e., point, lines, and polygons), and some are large enough so that some properties are inside project boundaries. Further, some projects consist of multiple geographically unconnected sites. For this reason, we geo-locate the boundary of each project and use the distance from each property to the nearest project boundary.
A majority of the adaptation projects include many different infrastructure items. The mean project cost is $1.13 million, and so many of the projects are relatively small scale. These projects provide some protection for sea levels reasonably close to current levels but are unlikely to protect against very high sea levels that might occur in the long run. As an example, North Bay Village is a township that sits on a manmade island in Biscayne Bay. A recent project consisted of 30 total items, including seawall repair, installing and repairing drainage systems and pumping stations, bay restoration, boardwalk restoration, and moving power lines underground.15
The final data sets report 431,410 property transactions (57% are condo transactions) and 162 adaptation projects in MDC between the years 2000 and 2019. All of the projects are designated as reducing flood risk, and the vast majority are also designated as protection against storm surge. Table 1 gives a summary of the filtering process to construct these data. It is important to note that relative to the literature, this filtering process retains a large share of the unfiltered observations (e.g., Kim 2020).
| Transaction Data | Projects Data | ||
|---|---|---|---|
| Initial transactions | 2,178,033 | Initial projects | 1,958 |
| Filters | |||
| Only condos and single family homes | Only fixed water infrastructure projects | ||
| Only qualified sales | Only completed projects with an end date | ||
| Only data since 1980 | Only projects with location data | ||
| Only properties sold at least twice | Only sea level rise, flooding, surge, and wind | ||
| Only multi-unit sales that are not mixed use | |||
| Top/bottom 5% prices and appreciation removed | |||
| Final transactions | 431,410 | Final projects | 162 |
Figure 2 shows how projects and transactions are distributed over space. Figure 2A shows the geographic location of the projects, which are dispersed throughout MDC, reflecting that flooding and storm surge are problems even in inland locations. Indeed, the average elevation for MDC is only about 1.8 meters, so even most inland properties face significant flood risk. Note that the panel displays the centroid of each project. However, we note that drainage and flood protections around rivers and canals, as well as road modifications, may extend beyond the dots in the figure in a linear way. In addition, some projects are defined by the rectangular grid of roads that border the project area, and these polygons may also extend beyond the plotted dots. Figure 2B shows the intensity of property transactions by location. Most municipalities across MDC have a large number of transactions. However, property transactions were especially common in certain coastal and island locations, such as the island of Miami Beach in the top right. Location of adaptation projects and property transactions in Miami-Dade County. In panel A, the center of each circle is a geographic centroid of a project. Actual project boundaries may extend beyond the circle, especially projects around roads and canals. In panel B, the center of each circle represents a transaction. In both panels, projects and transactions do not typically overlap, but circles are shown larger in the figure to improve visibility. Overlapping circles appear darker.
4. Empirical Approach
To measure the effect of adaptation investment on real estate in MDC, we rely on a difference-in-differences approach. The empirical model is:
The binary indicator Postit takes a value of one if the closest adaptation project to property i has been completed, relative to the year of transaction, at time t. The variable is an interaction between distance and time of completion of the nearest adaptation project. The parameters of interest, β, δ, and γ, denote the approximate elasticity of property transaction prices with respect to distance,16 the change in mean property value after project completion, and the difference-in-differences parameter which gives the change in the distance elasticity following completion of the project, respectively.
For control variables, Sizei is a vector containing the size of the property in square footage, as well the number of bedrooms and bathrooms in the property.17 The variables Wholesaleit and Condoit are dummy variables indicating wholesale and condo transactions, respectively. Further, Locationi is a vector of location observables, in particular, flood zone, elevation, and distance from the coast. Finally, to control for macro and local dynamics such as hurricanes or the Great Recession, we rely on a battery of project, property, and zip code by year fixed effects, Xit. The error term is εit. To control for spatial correlation for properties around a given project, we cluster standard errors by project.
Our hypothesis is that the post-completion parameter δ is positive, and the key interaction coefficient, γ, is negative so that properties closer to the project experience higher property transaction values subsequent to completion of the project. In turn, the hypothesis holds if adaptation projects provide positive benefits at the property level.18 Nonetheless, γ could certainly be positive if the project obstructed views or traffic, or created other negative amenities, or zero if real estate market participants do not observe the infrastructure or do not perceive any risk. In the analysis, we will use project characteristics to shed light on these potential effects.
5. Results
Table 2 summarizes the data used to implement this analysis.19 The vulnerability of MDC is evident in table 2, with the mean elevation of transacted properties equal to about two meters, and an average distance of 3–4 kilometers (km) from water. The table also shows that statistically significant differences exist between condo and non-condo properties. In particular, recorded condo transactions have higher prices on average but appreciate at a slower rate. Since most land designated for single family homes in MDC is already developed, the supply of condos is more elastic, which may temper the price growth path from demand shocks.20 In addition, condos are farther away from adaptation projects, closer to water bodies, built on lower ground, and smaller in overall size.
| Non-Condo | Condo | Difference | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Min | Max | Mean | SD | Min | Max | Mean | SD | Mean Diff. | SE | |
| Response variables: | ||||||||||
| Price ($/foot2) | 65.71 | 514.46 | 194.03 | 84.47 | 65.72 | 514.55 | 230.56 | 107.88 | 36.53*** | .36 |
| Appreciation (%/year) | −12.82 | 49.71 | 5.62 | 9.10 | −12.82 | 49.72 | 4.49 | 8.90 | −1.13*** | .03 |
| Distance to the closest project (km) | .00 | 12.04 | 1.64 | 1.42 | .00 | 8.91 | 1.94 | 1.52 | .30*** | .01 |
| Features: | ||||||||||
| Distance from water (km) | .00 | 16.79 | 3.88 | 3.77 | .00 | 15.12 | 2.64 | 3.75 | −1.25*** | .01 |
| Elevation (m) | −.07 | 11.92 | 2.38 | .65 | −.38 | 30.95 | 2.05 | 1.06 | −.33*** | .00 |
| Wholesale (0/1) | .00 | 1.00 | .00 | .04 | .00 | 1.00 | .01 | .11 | .01*** | .00 |
| Bedrooms (no.) | .00 | 32.00 | 3.33 | .86 | .00 | 12.00 | 1.91 | .90 | −1.42*** | .00 |
| Bathrooms (no.) | .00 | 14.00 | 2.15 | .87 | .00 | 12.00 | 1.67 | .67 | −.48*** | .00 |
| Parcel size (feet2) | 90.00 | 27,055.00 | 2,150.75 | 1,027.97 | 47.00 | 8,668.00 | 1,178.91 | 480.10 | −971.84*** | 2.85 |
5.1. Adaptation Project Analysis
This section shows that adaptation investment has a positive effect on property prices. Table 3 gives the primary results of our analysis under a variety of specifications. The first key parameter of interest, the project completion coefficient, is positive and is significant for all specifications except column 5, which indicates that the completion of a project results in an increase in price. We interpret the positive and significant post coefficient as a “level” effect: completion of the project results in an average increase in price over all properties assigned to the project.21 This could result from direct reduction of flooding across all properties or more likely a spillover effect whereby increases in property transaction prices near the project after completion raise prices for all properties in the neighborhood (Ioannides and Zabel 2003). The level effect in the post coefficient is an average. The benefits to an individual property assigned to the project might be higher than average, lower than average, zero (if a property near the project was not protected), or even negative (for example, if flood waters were diverted from one property to another). Thus, the cost-benefit analysis in section 5.5 correctly sums the heterogeneous benefits to all properties associated with the project through the average.
| asinh[Price] | |||||
|---|---|---|---|---|---|
| (1) | (2) | (3) | (4) | (5) | |
| Post | .053** | .054** | .051*** | .046** | .046 |
| (.024) | (.025) | (.017) | (.022) | (.034) | |
| Post × asinh[Distance] | −.015*** | −.016*** | −.016*** | −.017*** | −.017* |
| (.005) | (.005) | (.004) | (.006) | (.009) | |
| Size controls | X | X | |||
| Condo dummy | X | X | |||
| Location controls | X | X | |||
| Wholesale dummy | X | X | X | X | |
| Project FE | X | X | |||
| Property FE | X | X | |||
| Zip-year FE | X | X | X | X | X |
| R2 | .677 | .689 | .708 | .935 | .935 |
| Observations | 431,410 | 431,410 | 431,410 | 431,410 | 431,410 |
Interpreting asinh as an approximation of log means that property values adjacent to or inside the project boundary rise approximately 4.6%–5.4% post completion. Following Bin and Polasky (2004), we can generate a rough estimate of the cost of flood insurance for comparison. Our data have a median transaction price for a single family residence of $212,000 and an average flood policy premium of $401, or $8,020 in perpetuity at a 5% discount. Thus, average flood insurance premiums are about 3.8% of the median property value. Thus, the value of the projects exceeds the cost of insurance (note that FEMA only insures properties up to $250,000 and household items up to $100,000).
The second key parameter of interest, the interaction term, is negative and is significant for all specifications except column 5. The results indicate that the distance elasticity falls significantly following completion of the project. In particular, after the project is finished, the transaction prices rise by 1.4%–1.7%, for properties 1% closer to the project. Therefore, the positive effect of the infrastructure project on transaction prices is stronger for properties closer to the project. Following equation (12), the estimated effect of a completing a project on the percentage change in the property price per square foot (treating asinh as approximately equal to log) is . Since is positive and significant in table 3, the results provide evidence that projects successfully alleviated some negative flooding amenities, since the property transaction prices rose subsequent to project completion.22
These results are highly robust. Coefficient estimates remain stable in terms of sign, magnitude, and statistical significance across all specifications in table 3, except for the loss of significance in column 5. However, this loss in precision is expected as simultaneously adding project and property (which are observed at most three times) fixed effects averages out a large share of the variation identifying the key coefficients.
The coefficients on the control variables, reported in appendix A, are intuitive and mostly significant. Condos see lower prices per square foot, likely because the conditional condo supply is more elastic, which limits price increases from demand shocks. A negative and significant (at the 10% level) coefficient for distance to the coast indicates that an amenity value appears to exist for properties near the coast, which is consistent with previous literature (Bin et al. 2008). In addition, an additional 1 meter of elevation translates to a statistically significant increase in sale price, which is similar to other estimates in the literature.23 Properties in flood zones are also associated with a statistically significant negative price premium.
Finally, our estimates are far smaller than Kim (2020), who find as high as an 18.1% increase in price one year after completion for properties less than 400 meters from a project. This discrepancy reflects our use of transactions beyond the initial year of completion and our reliance on properties sold at least twice. Prices adjust dynamically over time, and so the long-run effect on prices differs from the short-run effect. These results make sense in that most projects are marginal improvements, such as installing pumping stations and drainage systems. The nature of these improvements results in more protection but not the complete elimination of flood risk. Therefore, we expect the price increases to be correspondingly modest.
5.2. Dynamic Effects of Adaptation Infrastructure
The previous analysis examines the average post-completion effect of the adaptation project. A relevant question raised in section 2, however, is how these capitalization dynamics accrue over time. To answer this question, we conduct the analysis as an event study and track the effect of the adaptation project ±5 years from the completion year.24 We impose further structure in the analysis by subdividing transactions at different distances from the respective project. That is, rather than using the continuous measure of distance in equation (12), we put the distances into four discrete bins: 0–200 meters from the project, greater than 200 and up to 400 meters, greater than 400 and up to 600 meters, as well as greater than 600 meters. The result of this analysis is shown in figure 3.25 Event study on price before and after completion of adaptation infrastructure. Results are relative to a control group of all properties outside the given radius. Solid lines indicate point estimates, while the gray area circumscribes the 95% confidence interval. Percentage changes for both treated and control groups are relative to a reference year of one year prior to the year of completion. Completion year is indicated by the vertical dashed line. Controls, fixed effects, and clustering remain consistent with specification 3 in table 3.
Note that figure 3 sets all percentage changes as relative to the price in period . Thus, is the reference calendar year in which percentage price changes are zero for both properties within the given distance of the project boundary and for the control group of all properties outside the given distance. Further, represents the calendar year of completion.
Figure 3 shows no significant differences in trends prior to completion of infrastructure projects. There is also no significant pattern of increasing relative price differences prior to completion, which the theory shows might indicate that the property market is anticipating future price increases prior to completion. Figure 3 shows further that treated and control properties have similar pre-trends, despite their geographic diversity, after including our large number of property characteristics and control variables.
Post completion, however, figure 3A shows that properties within 200 meters of the adaptation project start to sell at a higher price and achieve an increase of almost 10% five years after the project has been finalized. About half of the effect is present after one year, indicating that the value of projects is incorporated into property values slowly over time. This is intuitive as some infrastructure projects, like drainage and pumping stations, are not easily observed by buyers and sellers in property markets initially but become more apparent after reductions in flooding events occur over time.
Figure 3A corresponds closely to the theoretical model in section 2. In particular, the event study is consistent with figure 1B. Prices increase in the first year after completion, then increase still further in years 3–5, indicating some deviations from fundamental value. In the theory, θ(j, t) represents the percentage difference between the price and fundamental value. In particular, if the price slowly over time incorporates the fundamental value of the infrastructure, then θ(j, t) is negative and converging to zero. If we assume that the fundamental value is reached years after completion, then , the fundamental value is 6%–9% above precompletion prices, and % to −9%.
As noted in section 2.4.2, the increase in prices postcompletion could also result from a change in trends post completion, or convergence to fundamental value could happen in years subsequent to year 5. Given a longer time series, we could more confidently discern between these possibilities, since price deviations would either level off at years 3–5 (indicating convergence within five years), level off at some time period greater than five years, or continue to increase (indicating a change in trends). We cannot reject that prices have leveled off in years 3–5, which is suggestive. However, since many of the projects were completed recently, we cannot yet ascertain whether or not prices fully reflect the fundamental value of the projects in the long run.
Panels B and C in figure 3 show that the effect of infrastructure on property values is largely limited to the first 200 meters from the project.26 This pattern is consistent with the project data in that most of the projects provide only localized protection, such as installation of fixed pumping stations, drainage systems, and raising roads. Such projects are unlikely to provide much protection farther away, especially for macro events such as hurricane-induced storm surge or flooding. Note also that properties within a geographically small effect area are more likely to be similar, supporting our identification assumption that properties at all distances from the project follow a parallel trend.
5.3. Project Categories
Having covered the price effect of the average adaptation project, we now examine how the effect of a project on the property transaction price differs by type of project. Characteristics of each project are not random. Flooding near a canal might entail raising embankments, whereas flooding in a low-elevation area might call for a catch basin. Thus, the change in property values following installation of a particular type of project is conditional on potentially unobserved factors that make the particular project type preferred over others.27 Nonetheless, we might expect large visible projects such as raising streets and stabilizing shorelines to have larger effects than smaller projects, such as installing pumping stations (which can be hard to see) and improving storm drains.
Using the available description of projects, we create seven inclusive project adaptation categories. The adaptation categories are drainage, including culverts and exfiltration pipes; road/bridge elevation or improvements; pumping stations, including improvements of existing ones; shoreline stabilization, including berms, seawalls, and embankments; small water-holding infrastructure, including catch basins, swales, and French drains; larger water-holding infrastructure, including holding ponds and reservoirs; and aesthetic improvements such as improving a park.28 Many projects include several categories. Only two projects were large water-holding infrastructure, and only one project mentioned aesthetic improvements, so we focus on the first five categories.
We implement the difference-in-differences analysis by subsampling transactions that are near projects of each category. Table 4 gives the results.29 For properties at the project boundary, drainage, road/bridge elevation, and water-holding infrastructure projects have large and significant effects, and shoreline stabilization projects have large but insignificant point estimates. These results are intuitive, as pumping stations are the least visible, and all project types provide the most protection close to the project boundary. At 200 meters from the boundary (note that ), road/bridge elevation projects have the largest point effects. This is also intuitive, as road elevation is not only visible at a distance but is also helpful even for properties not adjacent to the road. Note that a loss of precision exists for some project categories, especially pumping stations and shoreline stabilization, which each affect a relatively small (5%) share of properties.
| asinh[Price] | ||||||
|---|---|---|---|---|---|---|
| (1) | (2) | (3) | (4) | (5) | (6) | |
| Post | .051*** | .082** | .176*** | .017 | .091 | .107** |
| (.017) | (.031) | (.036) | (.098) | (.073) | (.051) | |
| Post × asinh[Distance] | −.016*** | −.017** | −.041*** | .008 | −.044* | −.041** |
| (.004) | (.007) | (.012) | (.027) | (.022) | (.018) | |
| Subsample | All | Drainage | Elevation | Pump | Shore | Small W. |
| Size controls | X | X | X | X | X | X |
| Condo dummy | X | X | X | X | X | X |
| Location controls | X | X | X | X | X | X |
| Wholesale dummy | X | X | X | X | X | X |
| Project FE | X | |||||
| Zip-year FE | X | X | X | X | X | X |
| R2 | .708 | .717 | .645 | .713 | .663 | .613 |
| Observations | 431,410 | 173,137 | 110,391 | 20,605 | 20,766 | 87,390 |
5.4. Robustness to Threats to Identification
In this section, we explore more deeply our identification assumptions and show that the results are robust to a number of potential identification issues.
5.4.1. Matching Estimator
As noted in section 4, the data consist of a staggered rollout of projects, and all properties are eventually treated when the nearest project is completed. It follows that in a given time period, when a subset of projects is completed, property transactions near projects that are not finished yet serve as controls for property transactions for which the closest project is completed. This is advantageous in that the entire data set is used, and a large number of control transactions exist. However, control properties far away from completed projects might differ in unobserved ways from properties near completed projects. To account for this issue, the main analysis includes property-level amenity controls (e.g., number of bedrooms and elevation) and property fixed effects in specifications 4 and 5 in table 3. An alternative is to use a matching estimator so that transactions near completed projects are matched with property transactions near a project that are not completed but are otherwise similar on observable property characteristics. We execute this analysis in appendix C.3 and demonstrate that the same patterns and magnitudes in the main analysis remain relatively robust to this alternative approach.
5.4.2. Coastal Properties
Column 3 of table 3 controls for distance to the coastline. Still, the effect of distance to the coastline on property transaction prices might not be linear, and coastal properties and projects might differ in ways that might affect our identification strategy, which relies on using properties near projects that are not yet completed as controls for property transactions near completed projects. That is, the control and treated properties are a mix of coastal and inland properties. To address this issue, we restrict the sample to properties near the coast. Thus risks and environmental amenities are likely more similar among treated and control properties. The results of this analysis are reported in appendix C.4 and show that the signs and significance of the estimates remain robust, but the estimates for Post and become larger in magnitude.
There are at least two potential explanations for the interesting “intensification” of the effect near the coast. One explanation is that coastal properties may be fundamentally different from inland properties, and our estimates are confounding the timing and distance from the project with these characteristics. The other explanation is that the hazards mitigated near coastal properties are potentially more damaging (because coastal properties are either more valuable or more exposed), and hence the benefit from adaptation is greater and dissipates to zero at a slightly faster rate than in the regressions of table 3.
Figure 6 in appendix C.4 provides some evidence with respect to these competing explanations. In particular, figure 6 shows that most of the property features observed in the data (elevation, number of bedrooms, bathrooms, parcel size, and even wholesale transactions) do not exhibit clear patterns as distance to the coast increases. However, the relative share of condos and the price both decrease with distance to the coast. We interpret these patterns as evidence that the changes in estimates are not associated with the characteristics of the parcels but with the amenity value of living near the coast, as well as being protected from the risks at those locations. These patterns highlight the importance of controlling for distance to the coastline in the main analysis.
5.4.3. Imposing a Maximum Treatment Distance
The data count each property transaction as treated when the nearest project is completed, even if the nearest project is far from the property. But if beyond a certain distance the projects have no effect on property transaction prices, then the Post coefficient is biased toward zero as close properties with positive treatment effects are averaged with distant properties with zero treatment effects, and the interaction term is also biased toward zero, as the negative distance slope of the treatment effect for properties near the project is averaged with a zero slope for distant properties.
We tackle this issue in appendix C.5 by setting up the analysis such that the Post coefficient takes a value of one if and only if the closest project has been completed and the property is within a certain distance from this project. In this way, there now exists a large number of “never treated” properties. Table 13 shows that the effects are generally larger in magnitude and more significant. We can explain this result by noting that the Post coefficient no longer averages in small treatment effects far from the project, and the interaction term no longer averages in zero slopes far from the project. However, in this case the cutoff is arbitrary, and it is difficult to resolve whether or not a maximum distance exists for the effect or if there are just very small treatment effects far from the project.
5.4.4. Multiple Treatments
About 5% of the property transactions are such that the second closest project is 600 meters or less from the property. Figure 3 shows that the significant results are largely confined to the first 200 meters. Nonetheless, “multi-treatments” effects can still affect the results in a number of ways. To shed some light on this issue, we replicate our analysis for a subsample of transactions at least a given distance from the second closest project. The results are reported in appendix C.6, table 14, and show no meaningful differences with our main set of results.
5.4.5. Other Robustness Checks
The analysis is robust to several other threats to identification. First, a possible concern is that subsequent to completion of adaptation infrastructure, households undertake remodeling that improves the property value. If so, the infrastructure value is confounded by the remodeling value. Section 2.3 shows, however, that the incentive to undertake remodeling or other private amenities actually decreases after completion. Since the property is already less affordable, additional private amenities are rewarded with smaller further increases in property values. Although our data on remodeling are limited, we do have data on the date of a large remodel (defined as costing at least one-half of the value of the home). Controlling for major remodel dates leaves the regression coefficients and significance levels virtually unchanged (results available on request).
Similarly, public infrastructure investment might crowd out private adaptation investments (e.g., sandbags). Section 2.3 shows that indeed properties protected with public infrastructure value additional private adaptation investments less. This would bias our coefficients toward zero, as the increase in property values relative to control properties is a mix of more public and less private adaptation. Thus, our results are conservative with respect to possible crowding out.
Finally, our results include a control for condo transactions, but it might be insufficient in that condos are more prevalent near the coast, where our effects are strongest. Further, some projects protect mainly condos. Our results are robust to restricting the data set to only single family homes near the coast, for the subset of projects that protect a sufficiently large number of single family homes (results available on request).
5.5. Cost-Benefit Analysis
We next conduct cost-benefit analysis for all of the projects. Consistent with the results of section 5.2, we define the set of properties affected by an individual project as all properties within 200 meters. For an individual property i within 200 meters of a given project, solving equation (12) reveals that , where r.h.s. is the right-hand side of (12). Let denote the fitted value of Pit using the parameter estimates from our preferred specification, column 3 of table 3.30 The estimated benefit to property i from the adaptation amenity at time t is thus:
The estimated total project value grows over time, as shown in figure 3. Thus, the true value of a project depends on how such time variation is interpreted. Consider the two cases illustrated in section 2. Since θ(j, t) represents temporary deviations from the fundamental value of the property, if , then no mispricing exists and all transactions over time are estimates of the fundamental value of the property. This case is illustrated in figure 1A, and amounts to averaging the upward-trending line in figure 3A, which varies from 3% to 9%, to get an average increase. However, if instead we assume initially that some mispricing exists and the price converges to the fundamental value as in figure 1B, then the later periods in figure 3A are more accurate representations of the fundamental value and should receive correspondingly higher weight, generating an average value closer to the 9% increase at the end of five years post completion. Here we take a conservative approach and weight all time periods equally. Therefore,
We compute the estimated total project value (15) for each project using our preferred specification and give the summary statistics in table 5. Table 5 shows that the average project generated about $0.68 million in net benefits, measured as the change in property values less project cost. Indeed, 117 of the 158 projects analyzed had positive net benefits. Assuming that all benefits and costs are realized after five years, the internal rate of return for the average project is 19% per year.
| Total Benefits (USD × 106) | Project Cost (USD × 106) | Total Net Benefits (USD × 106) | IRR (%/year) | |
|---|---|---|---|---|
| Min | $.00 | $.01 | −$23.12 | −100 |
| Max | $17.74 | $28.00 | $16.82 | 173 |
| Mean | $1.80 | $1.13 | $.68 | 19 |
| Median | $.63 | $.27 | $.31 | 27 |
| SD | $2.99 | $3.34 | $4.02 | 48 |
| Total, all projects | $285.04 | $178.35 | $107.70 | 10 |
| Total, per household ($K) | $.40 | $.25 | $.15 | … |
Municipalities are often concerned with how much tax revenue is generated from the project through higher property values, which offsets the cost of the project to the city or county. Using the average property tax rate of 2.1% for Miami-Dade municipalities, the average project generates about $0.04 million in tax revenue, which is enough to offset the average cost in a little under 30 years.32
Our analysis suggests that the total benefits of all projects are on the order of $285 million. This is indeed relatively large in magnitude for a metropolitan county, but we note that these are still below previous projections using the results found by MDC (Urban Land Institute 2020) or previous studies (Kim 2020). Further, these estimated benefits are also much smaller than the estimated cost of a direct hit by a hurricane or the projected aggregated cost of sea level rise in MDC (EPA 2017). This misalignment highlights the need for careful dynamic causal assessments of the benefits, as captured by property value increases, for future evaluation of adaptation policies.
6. Conclusions
This study analyzes the effect of adaptation infrastructure projects on property values. We find that property values increase after completion. In particular, project completion results in an increase in price of about 10% after five years for properties adjacent to or inside the adaptation project boundary. The completion effect dissipates as transactions take place farther from the adaptation site boundary.
Although the gains are relatively small per property, many projects protect hundreds or thousands of properties. Thus, the total net benefits for all 158 projects is about $285 million. Together, these results provide evidence that property buyers and sellers are both aware of the cost of flooding and storm surge resulting from climate risk but are also aware about the benefits from adaptation.
While our results are highly robust to specifications and assumptions, they come with several caveats. First, we are only able to measure how property buyers and sellers perceive risk, but the actual risk to properties may be higher or lower (health and employment benefits might be especially difficult to perceive). Second, most adaptation projects provide only limited protection. More extreme sea level rise requires more costly adaptation infrastructure, which in turn requires a different cost-benefit analysis. Thus, it is unclear whether long-run risk of extreme sea level rise is priced in. Third, Miami-Dade County contains a variety of real estate markets that vary based on price, distance to the coast, urban/suburban/rural, and vulnerability. Still, it is unclear whether the results completely generalize to markets with characteristics outside our sample (for example, infrastructure projects or property elevations not in our sample). Fourth, public adaptation expenditures may affect private adaptation expenditures at the property level as public adaptation may replace private adaptation via crowding out. If borrowing constraints are binding, government resources may be more efficiently spent subsidizing private adaptation investment. These possibilities are difficult to test for, given that we lack data on private adaptation.
We also note that our results estimate the benefits of adaptation infrastructure but not the distribution of said benefits. An interesting question for future research is whether or not higher property values caused by adaptation infrastructure cause low-income households to relocate to less protected areas with lower rental costs.
Sea level rise, and its associated flooding impacts, is an important problem for coastal communities across the world. Even if carbon emissions are drastically reduced, inertia in the climate system will result in continued sea level rise for many years. Thus, coastal communities must adapt. The Fourth National Climate Assessment argues that adaptation can significantly reduce the impacts of sea level rise and, indeed, that the benefits of increased protection currently outweigh the costs in many locations. In our case study, we demonstrate that these benefits are partly captured by the real estate market and, indeed, exceed project costs.
Coastal adaptation is an important tool for mitigating the impacts associated with climate change. Our results provide evidence that coastal communities can overcome barriers such as lack of funding, difficulty coordinating, and imperfect information to provide valuable adaptation infrastructure. Whether or not coastal communities realize these additional gains from adaptation in the future will only become an increasingly important question.
Notes
David L. Kelly is in the Department of Economics, University of Miami ([email protected]). Renato Molina is in the Department of Environmental Science and Policy and the Department of Economics, University of Miami ([email protected]). We would like to thank Matthew Varkony, Steven Koller, and Colin Servoss for research assistance. We thank Andrew Plantinga, two anonymous referees, Esber Andiroglu, Joel Lamere, Kathleen Sealey, Christopher Parmeter, James Sobczak, Prannoy Suraneni, and seminar participants at the 2020 AERE summer conference, the Triangle Resources and Environmental Economics Seminar Series, the University of Miami, and the 2020 Climate Adaptation Symposium at the University of California Los Angeles for helpful comments and suggestions. This work was supported by the Laboratory of Integrative Knowledge at the University of Miami (U-LINK).
1. From 2015 to 2099, under the high emissions, business as usual scenario (RCP8.5), and including adaptation costs. This result varies between approximately 82% and 94% depending on the emissions scenario and time horizon. See the report’s technical appendix, p. 216, for details.
2. More broadly, an expansive and growing literature uses hedonic methods to estimate the effect of other natural hazards, such as earthquakes, volcanoes, and algae blooms, on property values.
3. Davlasheridze and Fan (2019) in particular analyze properties protected by a seawall vs. unprotected properties.
4. Other methods include the use of prediction markets (Kelly et al. 2012) and property prices (e.g., Kousky 2010).
5. A number of papers also examine noninfrastructure hurricane adaptations. For example, Davlasheridze et al. (2017) find that a 1% increase in ex ante spending by FEMA reduces hurricane damages by 0.21%. Dundas (2017) and Gopalakrishnan et al. (2011) study the effect of natural capital such as dunes and beach nourishment on property values.
6. Market prices may deviate from fundamental values for a variety of reasons, including difference in tax treatment between housing and market interest, borrowing constraints, and property price inflation. These factors may also vary over time. This assumption allows for the theoretical possibility that changes in amenities may be incorporated into housing prices slowly over time rather than instantaneously.
7. Even very localized protections such as drainage may still increase property values at a declining rate with distance from the project as property values increase due to spillover effects from increases in neighboring property values (Ioannides and Zabel 2003).
8. The empirical results show that the treatment effect is significant for only 200 m around the project. Since the treated area is relatively small, this assumption is likely reasonable.
9. If the appraiser’s office determines that a sale is an arm’s length market transaction, then the transaction is labeled qualified. Nonqualified transactions include changes of ownership such as inheritances and usually have zero or very low prices.
10. For example, a 20-unit development that sold for $5 million is recorded as $5 million for each unit in the development.
11. Appendix C.2 shows that the results are not sensitive to including outliers and that many outliers are part of wholesale transactions.
12. Miami-Dade’s website on Flood Insurance Rate Maps (FIRM) states that the current maps were adopted in July 2009 and became effective in September 2009. Individual property owners can amend a property’s location in relation to the Special Flood Hazard Area (SFHA) via a letter of map amendment (LOMA). However, communication with MDC Department of Regulatory and Economic Resources (DRER) reveals that the most common removal of properties from the SFHA in MDC is by the placement of fill, which is not performed by MDC and not considered an infrastructure adaptation project.
13. Additional background on LMS and funded projects can be found at https://www.miamidade.gov/global/emergency/projects-that-protect.page (accessed July 25, 2022).
14. One possible threat to identification is the possibility that some of the 26 projects designed to protect public buildings also cause property values to rise and are completed relatively close temporally and spatially to other adaptation projects. The average completion date for public building projects is earlier than adaptation projects, although some are completed during the 2013–16 time period, which is when multiple adaptation projects are also completed.
15. Only one project in the project data mentions amenity benefits independent of public adaptation.
16. Strictly speaking, the use of asinh means that β and are approximations of the price elasticity of distance before and after completion, respectively. Using the results of Bellemare and Wichman (2020), the approximation error is less than 1% for properties at least 10 meters from the project boundary.
17. Although the price is already per square foot, larger homes typically sell at a higher price per square foot.
18. Note that the tax cost of an adaptation project is spread throughout MDC and so is removed by the difference-in-differences specification.
19. Five condo transactions (two properties) were found to have an excessive number of bedrooms relative to their transaction price (1,694 and 930 bedrooms), indicating a probable data entry error by the appraiser. We substituted the excessive values by the overall average before filters, 2.62 bedrooms, for these transactions.
20. Parton and Dundas (2020) show that building permits for new residential construction significantly increased in North Carolina, an area with more undeveloped land, following a report recommending accounting for sea level rise in land use decisions.
21. Other explanations are possible but appear less plausible. For example, projects could potentially be assigned to neighborhoods that MDC expects will appreciate faster post completion. However, given that projects are planned many years in advance, it is unclear how MDC would have such knowledge. The positive post coefficient is also unlikely to result from macro trends (e.g., the Great Recession), as the post variable reflects projects that are completed in a variety of years (from 2000 to 2016) and we use zip-by-year fixed effects that control for such macro trends in a flexible way.
22. For example, if the projects were built in areas that were depressed due to the low cost of obtaining land for the project, rather than due to vulnerability, we would not expect a price premium subsequent to completion.
23. See, for example, Keenan et al. (2018), although the comparison is not exact since that paper reports neighborhood-level estimates.
24. Since the vast majority of projects were completed in the 2014–16 period, few data points exist more than five years after completion.
25. The full regression results are given in table 8. Additional results implementing the Sun and Abraham (2021) methodology for staggered treatment assignments are also provided in table 4. Results remain robust to this refinement.
26. Note that 200 meters corresponds to the distance to the boundary of the project, with properties inside the boundary counted as zero. Hence the geographic area of properties that benefit from the project is larger than the area of the radius 200 meter circle around the centroid of the project.
27. Note that because projects are not assigned at random, it is possible that projects are assigned based on unobserved property characteristics. If similar projects with similar unobserved characteristics are completed at similar times, and such properties experienced a trend change after completion, then a violation of parallel trends occurs. To guard against this, we include a large number of geographic controls, including project fixed effects, elevation, and distance to the coast.
28. Such an aesthetic improvement would only be included as water infrastructure if the project also added infrastructure in at least one of the other categories.
30. Our preferred specification does not include property fixed effects, which are determined from heterogeneous property-level amenities, including private adaptation amenities. Thus, if property fixed effects were included, then the estimated price given in eq. (12) is consistent with the theoretical prediction of eq. (6), in that the price is increasing in a private adaptation amenity both before and after project completion, but the price change is decreasing in the adaptation amenity. Still, given that our data have at most three transactions per property, the fixed effects coefficients are imprecisely estimated. We therefore use col. 3 for the cost-benefit calculation, noting that the coefficients of interest are nearly identical with or without property fixed effects.
31. We also note that all properties within 200 meters of all projects have been historically transacted at least once.
32. Note that Miami-Dade’s Homestead Exemption law prevents increases in appraised values until a property is sold. Therefore, these gains in tax revenue will accrue only in the long run.
References
Bakkensen, Laura A., and Lint Barrage. 2017. Flood risk belief heterogeneity and coastal home price dynamics: Going under water? Technical report, NBER Working paper 23854, National Bureau of Economic Research, Cambridge, MA. Barrage, Lint, and Jacob Furst. 2019. Housing investment, sea level rise, and climate change beliefs. Economics Letters 177:105–8. Bellemare, Marc F., and Casey J. Wichman. 2020. Elasticities and the inverse hyperbolic sine transformation. Oxford Bulletin of Economics and Statistics 82 (1): 50–61. Bernstein, Asaf, Matthew T. Gustafson, and Ryan Lewis. 2019. Disaster on the horizon: The price effect of sea level rise. Journal of Financial Economics 134 (2): 253–72. Bin, O., and S. Polasky. 2004. Effect of flood hazards on property values: Evidence before and after Hurricane Floyd. Land Economics 73:490–500. Bin, Okmyung, Thomas W. Crawford, Jamie Brown Kruse, and Craig E. Landry. 2008. Viewscapes and flood hazard: Coastal housing market response to amenities and risk. Land Economics 84 (3): 434–48. Bin, Okmyung, and Craig E. Landry. 2013. Changes in implicit flood risk premiums: Empirical evidence from the housing market. Journal of Environmental Economics and Management 65 (3): 361–76. Callaway, Brantly, Andrew Goodman-Bacon, and Pedro H. C. Sant’Anna. 2021. Difference-in-differences with a continuous treatment. ArXiv preprint arXiv:2107.02637. Case, K. E., and R. J. Shiller. 1989. The efficiency of the market for single-family homes. American Economic Review 79:125–37. Davlasheridze, Meri, and Qin Fan. 2019. Valuing seawall protection in the wake of Hurricane Ike. Economics of Disasters and Climate Change 3:257–79. Davlasheridze, Meri, Karen Fisher-Vanden, and H. Alen Klaiber. 2017. The effects of adaptation measures on hurricane induced property losses: Which FEMA investments have the highest returns? Journal of Environmental Economics and Management 81:93–114. Dundas, Steven J. 2017. Benefits and ancillary costs of natural infrastructure: Evidence from the New Jersey coast. Journal of Environmental Economics and Management 85:62–80. Duranton, Gilles, Geetika Nagpal, and Matthew A. Turner. 2021. Transportation infrastructure in the US, chap. 3, 165–218. Chicago: University of Chicago Press. EPA (Environmental Protection Agency). 2017. Multi-model framework for quantitative sectoral impacts analysis: A technical report for the Fourth National Climate Assessment. Fell, Harrison, and Carolyn Kousky. 2015. The value of levee protection to commercial properties. Ecological Economics 119:181–88. Gallagher, J. 2014. Learning about an infrequent event: Evidence from flood insurance take-up in the United States. American Economic Journal: Applied Economics 6 (3): 206–33. Gibson, Matthew, and Jamie T. Mullins. 2020. Climate risk and beliefs in New York floodplains. Journal of the Association of Environmental and Resource Economists 7 (6): 1069–1111. Goodman-Bacon, Andrew. 2021. Difference-in-differences with variation in treatment timing. Journal of Econometrics 225:254277. Gopalakrishnan, Sathya, Martin D. Smith, Jordan M. Slott, and A. Brad Murray. 2011. The value of disappearing beaches: A hedonic pricing model with endogenous beach width. Journal of Environmental Economics and Management 61 (3): 297–310. Hallstrom, D., and V. Smith. 2005. Market response to hurricanes. Journal of Environmental Economics and Management 50:541–61. Harding, John P., Stuart S. Rosenthal, and C. F. Sirmans. 2007. Depreciation of housing capital, maintenance, and house price inflation: Estimates from a repeat sales model. Journal of Urban Economics 61 (2): 193–217. Hino, Miyuki, and Marshall Burke. 2021. The effect of information about climate risk on property values. Proceedings of the National Academy of Sciences 118 (17): e2003374118. Ioannides, Yannis M., and Jeffrey E. Zabel. 2003. Neighbourhood effects and housing demand. Journal of Applied Econometrics 18:563–84. Jin, Di, Porter Hoagland, Donna K. Au, and Jun Qiu. 2015. Shoreline change, seawalls, and coastal property values. Ocean and Coastal Management 114:185–93. Keenan, Jesse M., Thomas Hill, and Anurag Gumber. 2018. Climate gentrification: From theory to empiricism in Miami-Dade County, Florida. Environmental Research Letters 13 (5): 054001. Kelly, David L., David Letson, Forrest Nelson, David Nolan, and Daniel Solís. 2012. Evolution of subjective hurricane risk perceptions: A Bayesian approach. Journal of Economic Behavior and Organization 81 (2): 644–63. Kelly, David L., and Renato Molina. 2022. Adaptation infrastructure and its effect on property values: Expanded version. University of Miami Working paper. Kim, Seung Kyum. 2020. The economic effects of climate change adaptation measures: Evidence from Miami-Dade County and New York City. Sustainability 12 (3): 1097. Kousky, Carolyn. 2010. Learning from extreme events: Risk perceptions after the flood. Land Economics 86 (3): 395–422. Ludy, Jessica, and Matt Kondolf. 2012. Flood risk perception in lands “protected” by 100-year levees. Natural Hazards 61:829–42. McAlpine, Steven A., and Jeremy R. Porter. 2018. Estimating recent local impacts of sea-level rise on current real-estate losses: A housing market case study in Miami-Dade, Florida. Population Research and Policy Review 37 (6): 871–95. NOAA (National Oceanic and Atmospheric Administration). 2020. Climate at a glance. Technical report, NOAA, Washington, DC. Ortega, Francesc, and Süleyman Taspınar. 2018. Rising sea levels and sinking property values: Hurricane Sandy and New York’s housing market. Journal of Urban Economics 106:81–100. Parton, Lee C., and Steven J. Dundas. 2020. A green paradox in climate adaptation for coastal housing markets. Journal of Environmental Economics and Management 104:102381. Shi, Linda, and Susanne Moser. 2021. Transformative climate adaptation in the United States: Trends and prospects. Science 372 (6549): eabc8054. Shr, Yau-Huo Jimmy, and Katherine Y. Zipp. 2019. The aftermath of flood zone remapping: The asymmetric impact of flood maps on housing prices. Land Economics 95 (2): 174–92. Sun, Liyang, and Sarah Abraham. 2021. Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. Journal of Econometrics 225 (2): 175–99. Urban Land Institute. 2020. The business case for resilience in southeast Florida: Regional economic benefits of climate adaptation. Technical report, Urban Land Institute, Washington, DC. Walsh, Patrick, Charles Griffiths, Dennis Guignet, and Heather Klemick. 2019. Adaptation, sea level rise, and property prices in the Chesapeake Bay watershed. Land Economics 95 (1): 19–34.
