Connect with us


Unpriced local weather threat and the potential penalties of overvaluation in US housing markets



Property transaction information

We mix the Zillow ZTRAX database49 and the PLACES database50 to entry property transactions, areas, and assessments for all states within the conterminous United States between 1996 and 2021. The ZTRAX database, supplied to us by Zillow, incorporates property transaction and tax evaluation information for about 150 million parcels in over 3,100 counties nationwide (Supplementary Fig. 1). To make sure consistency and accuracy of the info throughout the US, we extensively processed the ZTRAX database primarily based on the suggestions in ref. 26. These information cleansing measures included identification of arms-length gross sales, geolocation of parcels and buildings, temporal linkages between transaction, assessor and parcel information, and the identification of property-types, reminiscent of single-family properties. Extra info on accessing the Zillow ZTRAX database may be discovered at

The PLACES database makes use of assessor parcel numbers to hyperlink ZTRAX information to parcel boundaries utilizing county- and town-specific string sample matching and geographic qc. For roughly one third of United States counties, parcel polygon information in PLACES comes from open-access sources; for the rest, parcel polygon information comes from Regrid by means of their ‘Data with Purpose’ program ( The matching algorithm identifies over 1,000 distinctive mixtures of syntaxes and hyperlinks digital parcel boundaries from 2,951 counties to ZTRAX information with a median county-level success price of 98.2% and a imply of 95.5% (measured as the share of the variety of parcel boundaries matched to a tax assessor report). These linkages had been used to establish the parcels and transactions in our statistical fashions.

Flood hazard information

We evaluated properties’ publicity to fluvial, pluvial and coastal flood hazards within the years 2020 and 2050 utilizing beforehand developed inundation maps29. These maps present estimates of inundation depths at a 30 m spatial decision underneath 5, 20, 100, 250 and 500 yr flood recurrence intervals in each years. The 2020 mannequin outputs had been validated by simulating historic flood occasions and evaluating them to noticed flood extents and depths, discovering 87% similarity between the 251. Additional validation of mannequin outputs is described within the First Avenue Basis technical documentation52. The 2050 estimates are primarily based on downscaled Coupled Mannequin Intercomparison Undertaking 5 information underneath RCP 4.5. These information have been made accessible by the First Avenue Basis and may be discovered on their web site (

Whereas uncertainty and disagreement stay throughout continental-scale flood fashions by way of which areas are most uncovered and the extent of future flooding53, the outputs from ref. 29 present the one peer-reviewed, publicly accessible, climate-adjusted, and traditionally validated US-scale flood mannequin that has a excessive sufficient decision to be property particular. Different high-resolution fashions do exist within the personal sector, however are typically not made accessible for analysis functions or are made accessible with out clear strategies.

Calculating flood losses

We calculated anticipated annual flood losses to single-family properties in the US utilizing the strategies described in ref. 54. Briefly, we overlaid the areas of residential constructions derived from the ZTRAX and PLACES databases with the flood inundation maps and utilized depth-damage features advisable by FEMA55. These features had been developed by means of professional elicitation and are used to estimate the injury to inundated properties as a proportion of their worth primarily based on flood depth relative to first-floor elevation.

Whereas depth-damage features are broadly utilized in flood threat assessments nationally, together with FEMA’s HAZUS-MH software program, these practical relationships are extremely unsure and depend upon quite a few property-specific traits56. Sadly, many of those traits are usually not included within the ZTRAX information, reminiscent of first-floor elevations and construction varieties. To handle the difficulty of lacking first-floor elevations, that are essential for translating inundation depths to depths relative to the primary ground, we utilized changes to inundation depths primarily based on the suggestions in Sections 5.6.1–5.6.3 of the HAZUS Technical Methodology guide57. To beat the limitation of incomplete construction information, we used the proportions of construction varieties discovered within the NFIP insurance policies information, as described in ref. 54.

To calculate AALs for every property in 2020 and 2050, we built-in the estimated damages over the vary of recurrence interval possibilities modelled in ref. 29 utilizing trapezoidal Riemann sums58. In a given yr, the likelihood p of certainly one of these occasions occurring is 1/T, the place T equals the anticipated recurrence interval (for instance, p = 0.05 for a 20 yr flood occasion). These possibilities are impartial of one another, such that a number of flood occasions can happen in a single yr. To estimate AALs for property i in yr t between 2020 and 2050, we used a easy linear interpolation, the place t = 0 is the yr 2020 (equation 1).

$$mathrm{AAL}_{it} = mathrm{AAL}_{i2020} + t instances frac{{left( {mathrm{AAL}_{i2050} – mathrm{AAL}_{i2020}} proper)}}{{30}}$$


We calculated the NPV of AALs for property i over a 30 yr time horizon (that’s, the length of a typical fixed-rate mortgage), the place ρ is the low cost price (equation 2). All outcomes are reported utilizing a 3% low cost price, until said in any other case. The NPVs of AALs by county are proven in Supplementary Fig. 3.

$$mathrm{NPV}_i = mathop {sum }limits_{t = 0}^{30} mathrm{AAL}_{it}(1 + rho )^{ – t}$$


Historic flood insurance coverage price maps

We employed information from all FIRMs launched throughout the US between 2005 and 2019 to find out every property’s FEMA-designated flood zone on the time of sale. We obtained these information by means of Freedom of Data Act requests and one-on-one conferences with former GIS analysts who carried out contractual work for FEMA. Particularly, we used a digital model of the paper-based flood maps that had been efficient earlier than 2005 (referred to as the Q3 information product), mixed with yearly snapshots of the Nationwide Flood Hazard Layer between 2012 and 2020. For every FIRM, we noticed the date when the map grew to become lively (Supplementary Fig. 4), in addition to the spatial extent of the 100 yr and 500 yr floodplains. Evaluating the evolution of the spatial polygons by means of time allowed us to establish adjustments in floodplain boundaries on the property stage. Presently, greater than 115 million or 90% of residential properties are coated by the digital FIRMs. Particulars on the info assortment and cleansing steps are supplied in ref. 27.

Estimating flood threat capitalization

The extent to which publicity to flood threat is capitalized in property values is pushed by many dynamic components, together with latest native expertise with flooding, insurance coverage mandates, the price of insurance coverage (not simply within the NFIP, but in addition personal sector choices), consciousness of non-insurable prices, perceptions of threat and accessible info, not simply from mandated disclosures, but in addition from others concerned available in the market, reminiscent of realtors or neighbours. On this evaluation, we essentially averaged away a few of this heterogeneity and examined capitalization of flood threat amongst properties situated throughout the SFHA. Particularly, we used capitalization of up to date details about flood threat as an imperfect proxy for capitalization of publicity to flood threat (relative to not being uncovered to any threat).

We estimated the empirical flood zone low cost (that’s, noticed capitalization) utilizing a panel mannequin for repeat property gross sales. This methodology has been utilized by a number of different latest research to estimate the results of flood occasions and flood zone remapping on property costs59,60,61,62,63, and is taken into account finest follow on this context25. Just like the strategy utilized in ref. 9, we recognized the impact of flood zone standing on property costs by evaluating single properties to themselves over time, as they’re rezoned from exterior to throughout the SFHA because of FIRM updates. A key benefit of this identification strategy, significantly in contrast with cross-sectional fashions, is that it’s much less weak to the confounding results of time-invariant property-specific attributes, reminiscent of waterfront facilities, which are sometimes spatially correlated with flood threat9,10.

As proven in equation (3), we estimated the impact of location throughout the SFHA (δ) on the sale value (p), the place SFHAit is a binary variable equal to 1 if property i is situated within the SFHA at time t. Property-level fastened results, γi, management for all time-invariant traits of a property. We additionally included county-by-year fastened results, (eta _{ct}), to regulate for native market dynamics over time. These fastened results take in shocks to the housing market brought on by pure hazards, together with previous flood occasions. α is a continuing and εit is an error time period. For a property to be included within the estimation pattern, it should: (1) be exterior of the SFHA within the outdated FIRM, (2) have a recognized floodplain standing within the new FIRM and (3) be bought greater than as soon as whereas its flood zone standing is understood. Gross sales that occurred whereas the flood zone standing was unknown had been omitted from the dataset.

$$mathrm{log}left( {p_{it}} proper) = alpha + delta _gmathrm{SFHA}_{it} + gamma _i + eta _{ct} + varepsilon _{it}$$


We accounted for variation within the empirical flood zone low cost, indicated by the subscript g, pushed by variations amongst state-level flood threat disclosure legal guidelines and particular person perceptions of local weather threat by creating 4 teams of counties. Throughout the US, states fluctuate broadly in what they require sellers to confide in consumers. Based mostly on information compiled by the Pure Useful resource Protection Council, we grouped states by their disclosure legal guidelines or lack thereof. Utilizing the Yale Local weather Survey, we additionally grouped counties primarily based on common responses to the query “Do you think global warming will harm you personally?” (Supplementary Fig. 5). Group A contains counties with no disclosure legal guidelines and beneath median local weather concern; Group B contains counties with not less than one type of disclosure legislation and beneath median local weather concern; Group C contains counties with no disclosure legal guidelines and above median local weather concern; Group D contains counties with not less than one type of disclosure legislation and above median local weather concern. Because of the comparatively small pattern measurement of properties remapped into the 100 yr flood zone, we had been unable to group counties with extra granularity.

This strategy makes two key assumptions in regards to the dynamics of flood threat capitalization amongst rezoned properties and properties situated exterior the SFHA. First, we assume that properties rezoned into the SFHA are consultant of all properties situated within the SFHA. Whereas rezoned properties could also be on the margins of the SFHA and have decrease flood threat than common SFHA properties, rezoned properties may additionally be much less tailored to flood threat than common SFHA properties, because the latter have been topic to floodplain constructing codes for years. Respectively, these two potential variations could concurrently result in under- and overestimation of flood threat capitalization.

Second, we assume that solely properties at present situated throughout the SFHA capitalize flood threat, whereas properties situated exterior the SFHA don’t expertise any low cost regardless of their publicity to flood threat. We make this assumption primarily based on the outcomes of refs. 9,10,24, all of which discover that location within the FEMA-designated 500 yr floodplain (that’s, 0.2–1% annual likelihood of flooding) has no impact on property sale costs. We describe two various analyses that take a look at the sensitivity of this assumption on the finish of the Strategies part.

Estimating property overvaluation

We calculated overvaluation of properties uncovered to flood threat (that’s, each SFHA and non-SFHA properties with flood losses larger than zero) because the distinction between their estimated present truthful market worth and their environment friendly value (equation 4). For properties the place the truthful market worth is lower than the environment friendly value, we calculated overvaluation as zero.

$$mathrm{Overvaluation}_i = mathrm{FairMarketValue}_i – mathrm{EfficientPrice}_i$$


We estimated the present truthful market worth of properties utilizing their most up-to-date transaction value adjusted to the current utilizing the Federal Housing Finance Company’s Home Worth Index (FHFA HPI). For properties with no transaction information, we as a substitute used their assessed worth, equally adjusted to the current utilizing the FHFA HPI, relying on the yr of evaluation. We accounted for variations in evaluation strategies throughout counties by becoming easy linear regression fashions between adjusted transaction costs and adjusted assessed values for all transacted properties in a county. We then utilized the estimated coefficient as a scalar to the assessed values.

We estimated the environment friendly costs of properties uncovered to flood threat because the distinction between their truthful market worth within the absence of any flood threat (RiskFreeMV) and the NPV of AALs between 2020 and 2050 (equation 5). For properties the place the NPV of AALs is larger than the RiskFreeMV, the environment friendly value is zero.

$$mathrm{EfficientPrice}_i = mathrm{RiskFreeMV}_i – mathrm{NPV}_i$$


The usage of AALs to find out environment friendly costs assumes that homebuyers have full details about properties’ publicity to flooding and value that threat in line with the discounted worth of future flood losses. Regardless of uncertainty in projections of future flood losses, this strategy captures the potential value of publicity to flood threat over the lifetime of a property extra precisely than the present value of NFIP premiums. In distinction to our estimates of AALs, NFIP premiums don’t replicate future local weather change, have been traditionally sponsored in lots of areas and have been primarily based on FEMA flood maps, that are inaccurate and lack protection in lots of components of the US. Whereas a few of these deficiencies in NFIP pricing have been addressed by Danger Ranking 2.0, these information are usually not but accessible on the property stage. Additional, on condition that housing markets have the potential to capitalize the full prices of flood threat underneath local weather change, our estimates seize market overvaluation higher than the prices of present NFIP premiums.

To calculate properties’ RiskFreeMV, we eliminated the empirical flood zone reductions (δ) estimated with the hedonic mannequin (equation 6). For instance, if the truthful market worth for property i is US$500,000 and the flood zone low cost for group g is −5%, then the RiskFreeMV for property i is US$526,315. As mentioned within the earlier part, we assume that properties exterior the SFHA don’t capitalize flood threat and that their RiskFreeMV is identical as their present truthful market worth.

$$mathrm{RiskFreeMV}_i = left{ {start{array}{*{20}{l}} {mathrm{FairMarketValue}_i/(1 + delta _g),quad mathrm{SFHA} = 1} hfill {mathrm{FairMarketValue}_i,quad mathrm{SFHA} = 0} hfill finish{array}} proper.$$


Importantly, the usage of AALs to estimate environment friendly property costs inevitably leads to the underestimation of overvaluation. Past structural damages, flooding can also be related to damages to contents and lack of sentimental gadgets, particles clear up, evacuation bills, and unfavorable psychological well being impacts, none of that are included within the NPV of AAL estimates. Whereas troublesome to quantify, together with these different types of injury within the NPV calculation would invariably improve estimates of overvaluation.

Sensitivity and uncertainty analyses

In our major strategies and outcomes, we assume that properties situated exterior of the SFHA don’t capitalize flood threat. Nevertheless, significantly following latest flood occasions, publicity to flood threat could also be quickly capitalized by properties exterior of the SFHA in localized areas64. To check the sensitivity of our outcomes to this preliminary assumption, we reran the evaluation with two various assumptions concerning flood threat capitalization and subsequent overvaluation.

In our first sensitivity evaluation, we assume that each one properties uncovered to flood threat are discounted on the similar price as was estimated for SFHA properties by the panel mannequin described above. This gives an upper-bound estimate of non-SFHA capitalization and a lower-bound estimate of complete overvaluation. Outcomes underneath this various assumption are proven in Supplementary Figs. 11 and 14.

In our second sensitivity evaluation, we carried out a cross-sectional regression mannequin to establish flood threat capitalization amongst properties uncovered to flood threat and which are situated within the FEMA 100 yr flood zone, 500 yr flood zone or exterior of any designated flood zone. Whereas this strategy allowed us to estimate flood threat capitalization amongst properties situated exterior of the SFHA, we want it lower than the panel mannequin due to how difficult it’s to regulate for all of the traits of a property that could be correlated with flood threat and costs. The outcomes from this evaluation are proven in Supplementary Figs. 8, 12 and 15.

For this various evaluation, we estimated the impact of FEMA flood zone standing (δ) on sale value (p) utilizing a 6-level issue variable ((mathrm{FZ}_{it} ast mathrm{Danger}_i)) that mixes flood zone classes (that’s, 100 yr, 500 yr or exterior) and binary publicity to flood threat (equation 7). Properties situated exterior of any flood zone and that aren’t uncovered to flood threat function the reference group. Hit is a vector of property-specific variables that features variety of bedrooms, constructing space and the age of the property on the time of sale; λ is a vector of estimated coefficients for these variables. Following ref. 40, we utilized high-dimensional fastened results, β, to regulate for a collection of location-specific traits (Xi). This time period captures the interplay between a property’s block group, distance to coast bins (0 to 10 m, 10 to 400 m, >400 m), presence of lake and river frontage on the property and elevation bins (0 to five m, 5 to 10 m, >100 m). We additionally utilized fastened results to regulate for seasonal market traits throughout states, (tau _{sq}), and county-level market traits throughout years, (eta _{ct}). We carried out this estimation for gross sales inside every subset of Teams A–D outlined above. To calculate overvaluation, we modified any estimate of δ that’s larger than zero (that’s, a value premium) or not considerably totally different (P ≥ 0.05) to zero.

$$mathrm{log}left( {p_{it}} proper) = alpha + delta _gmathrm{FZ}_{it} ast mathrm{Danger}_i + {uplambda}H_{it} + X_i + tau _{sq} + eta _{ct} + varepsilon_{it}$$


Separate from these two sensitivity analyses, we additionally used a Monte Carlo simulation to guage the uncertainty bounds in our estimates of overvaluation. For every iteration of the simulation (N = 1,000), we randomly sampled regular likelihood distribution features fitted to the empirical flood zone reductions estimated by the panel mannequin (see equation 3). We additionally assessed the sensitivity of overvaluation to the utilized low cost price, evaluating low cost charges of 1, 3, 5 and seven%, and inundation hazard situations. The low, mid and excessive hazard situations signify the twenty fifth, fiftieth and seventy fifth percentile estimates of the flood mannequin simulations, respectively. Variation throughout flood mannequin simulations is pushed by uncertainty in world local weather mannequin outputs. In 2050, the twenty fifth and seventy fifth percentile estimates from the RCP 4.5 mannequin roughly align with the imply estimates from the RCP 2.6 and eight.5 fashions. For extra info on the uncertainty in flood mannequin outputs, please see ref. 29. Outcomes from the uncertainty evaluation are proven in Supplementary Fig. 16.

Reporting abstract

Additional info on analysis design is on the market within the Nature Portfolio Reporting Abstract linked to this text.

Supply hyperlink

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *


Copyright © 2022 - NatureAndSystems - All Rights Reserved