Estimating the Flood Risk Discount: Evidence From a One-off National Information Shock
Tom Gillespie; Ronan C. Lyons; Thomas K. J. McDermott
Environmental and Resource Economics · 2025-03-10
CC BY 4.0 · https://creativecommons.org/licenses/by/4.0/
Source: https://link.springer.com/article/10.1007/s10640-025-00963-5

Abstract

Flood risk is a highly pervasive and costly natural hazard globally. With significant increases in flood risk expected over coming decades, future exposure to flood risk and associated costs will depend heavily on how private consumption decisions respond to new information about risk. We exploit a one-off national information treatment in the form of the release in 2011, for the first time, of detailed flood risk maps for Ireland, to test the effect of new information about flood risk on housing prices across a national housing market. We combine rich dwelling-level information on over 475,000 dwellings for the period 2006-2015 with detailed official data relating to flood risk, events and defences. Our core finding is that information matters. The price of housing responded clearly to the release of flood risk maps at the end of 2011, with the emergence of a 4% price discount for dwellings at risk of flooding.

1 Introduction

Flood risk is a highly pervasive and costly natural hazard, with an estimated one billion people in 155 countries exposed worldwide (European Commission Joint Research Centre 2017 ). With the prospect of rising sea levels and more intense rainfall events due to climate change, flood risk is expected to increase in many locations over coming decades. Projections of the future costs of flooding depend not only on the risk of flood events but also on societies’ exposure to those events (Hallegatte et al. 2013 ). This underlines the importance of the extent to which flood risk is taken into account in private decisions, especially where costs are borne, at least in part, by taxpayers (Lin et al. 2021 ). Due to the immobile nature of real estate and its prevalence in the typical household’s balance sheet, the housing market represents a unique and important window into how private actions reflect flood risk. Theory would suggest a price discount for dwellings at risk of flooding, all else equal, given the associated costs. However, individual households may lack good information on flood risk (Bakkensen and Acirc; A., Ma, L. 2020 ), and there is also the issue of moral hazard resulting from various forms of government interventions to protect households from disaster (Kydland and Acirc; E., Prescott, E., Acirc; C. 1977 ). In short, where market signals are weak, there may be a tendency towards over-exposure to flood risk.

There is a long literature estimating the effect of flood risk on housing prices, hereafter the flood risk discount . A recent review finds widely varying results, with estimates of the price effect ranging from −75% to \(+\) 61% (Beltrán et al. 2018 ). One common approach is to compare the value of dwellings within flood risk zones to those elsewhere, controlling for a range of dwelling attributes (e.g. MacDonald et al. 1990 ; Bin et al. 2008a , 2008 ; Meldrum and Acirc; R. 2016 ). However, interpreting the results of such studies as causal depends on the strong assumption that hazard risk is exogenous, conditional on other observable determinants of housing prices. As an alternative, many recent papers in this literature have exploited the more plausibly exogenous timing of particular flood events to estimate how flood discounts vary in response to the occurrence of flooding; see, for example, Bin and Polasky ( 2004 ), Hallstrom et al. ( 2005 ), Atreya et al. ( 2013 ), Gibson et al. ( 2017 ), Timar et al. ( 2018 ), Ortega and Taspınar ( 2018 ), Beltrán et al. ( 2019 ), Petkov ( 2022 ). A common finding in these studies is that there are significant discounts after flood events, which fade over time. As noted by Bosker et al. ( 2019 ), this strand of the literature estimates changes in households’ risk perceptions following a recent flood event, rather than directly identifying their level of risk perceptions. There is also the possibility that these estimates reflect, at least partly, the damages from flooding, or indeed other impacts of flood events on local property markets.

A number of recent papers have turned to information treatments to recover flood discount estimates (e.g. Votsis and Perrels 2016 ; Shr and Acirc; J., Zipp, K., Acirc; Y. 2019 ; Hino and Burke 2020 ; Gibson et al. 2020 ; Hsieh and Acirc; C. 2021 ; Seo et al. 2021 ). Several of these papers use updates to the U.S. Federal Emergency Management Agency (FEMA) flood risk maps as the basis of their estimates, but the FEMA maps are problematic on a number of levels. First, the rolling nature of updates to FEMA maps potentially creates a staggered treatment problem. It is now well-known in the applied econometrics literature that two-way fixed-effects (TWFE) estimations of difference-in-differences coefficients can lead to substantial biases when there are staggered treatment timing and heterogeneous or time-varying treatment effects (e.g. Baker and Acirc; C., Larcker, D., Acirc; F., Wang, C., Acirc; C. 2022 ; Borusyak et al. 2021 ; Callaway et al. 2021 ; De Chaisemartin and d’Haultfoeuille 2020 ; Goodman-Bacon 2021 ; Sun and Abraham 2021 ).

Secondly, the process by which FEMA maps are updated creates endogeneity concerns, given that map updates are explicitly prioritised for areas with rapid housing development (Davlasheridze et al. 2017 ). Concerns have also been raised that the mapping process is amenable to local pressures and political influence (see, for example, Pralle 2019 ). Papers exploiting map updates from other jurisdictions have typically been based on relatively small samples (e.g. Votsis and Perrels 2016 ; Lautrup et al. 2023 , using Danish and Finnish data, respectively) and/or found no information effect (e.g. Hsieh and Acirc 2021 ; Seo et al. 2021 , for Taiwan and South Korea, respectively).

Our paper exploits a one-off national information treatment in the form of the release in late 2011, for the first time, of detailed flood risk maps for Ireland. Our baseline specification, using a dataset of almost half a million real estate listings, has a rich set of control variables, including a suite of property-level characteristics, including property type and size, as well as its Building Energy Rating (BER), as well as highly localised spatial (and spatio-temporal) fixed effects. Nonetheless, one may still be concerned that estimates of the flood risk discount could be confounded by unobservables that happen to correlate with flood risk in the cross-section. To address this concern, we employ a difference-in-differences (DiD) style analysis exploiting the release in 2011 of detailed flood risk maps in Ireland. Our estimate of the flood risk discount, based on the DiD estimator, is 4%. Footnote 1 We show that this estimate is robust to varying the set of included covariates and to varying the level of the spatial fixed effects. We also show that the flood discount only appears after the release of the risk information in 2011 and that, prior to the information treatment, properties at risk of flooding and properties not at risk followed similar price trends in our data. Lastly, we show that our results are not driven by the use of listed prices: for a matched sample, switching between transaction and listing prices as the dependent variable has no significant effect on the estimated flood discount.

In addition to the hedonic DiD estimates, we also leverage the richness of our data to implement a repeat-sales estimator, including property-level fixed effects, for properties that were listed both before and after the information treatment in 2011. This version of the analysis returns a very similar estimate of the flood risk discount, 3.4%. Finally, we also implement a border-discontinuity style analysis and explore how our estimate of the flood risk discount varies with distance from the boundary on either side. Hedonic price estimates reveal the marginal willingness to pay (WTP) of the marginal buyer (Kuminoff and Acirc et al. 2014 ; Greenstone 2017 ; Pinchbeck et al. 2021 ). Given the potential for selection into risk based on idiosyncratic household preferences and budget constraints (sorting), we interpret our estimates cautiously as “local” treatment effects. We return to this point in our discussion in Section 6 . Most closely related to our paper in setting is Pilla et al. ( 2019 ), who compare directly the effects of assessed risk and a large flood event for the case of Dublin, Ireland, after severe flooding in 2011. They find evidence that flood events had a bigger impact than assessed flood risk on housing prices. This reinforces the idea that actors in housing markets are not always well informed about flood risk. Pilla et al. ( 2019 ), however, is limited to cross-sectional analysis and uses a more limited set of control variables, which presents challenges to a causal interpretation of their results.

We see our paper as making a number of useful contributions to the growing literature on flood risk and housing markets. It is the first paper to use a difference-in-differences (DiD) set-up for an information shock on the Irish housing market. The analysis is based on a large and unusually rich dataset, with nationwide coverage of the Irish property market over a decade and the ability to control for important covariates that are typically excluded from similar studies, including a sophisticated measure of sea views, independently assessed building quality, and time-varying and spatially explicit measures of flood events and areas protected by flood defences. In addition to these distinguishing features, our paper makes four further substantive contributions. It is the first in the literature to estimate flood risk discounts while controlling not only for flood defences and past flood events, but also for sea-views, a potentially valuable amenity that is plausibly correlated with flood risk. Secondly, our paper is the first to compare estimates of flood risk directly across three estimation strategies: DiD, cross-section and repeat sales. Thirdly, the sensitivity analysis at the boundary of assessed flood risks, in particular the finding that discounts ‘spill over’ the boundaries, suggests that individuals contextualize information provided on ‘objective’ flood risk and thus they interpret risk as a continuous variable that extends beyond the policy boundary. Finally, our paper is also the first to compare results across both listed and transaction prices. Given the scarcity of accurate administrative data in many developing country settings, and the prevalence of online listings in those same markets, this gives the paper a wide policy relevance.

The rest of the paper is structured as follows. The next two sections describe the local context and the data used in our analysis, in particular the data relating to flood risk, flood defences and flood events. Section 4 outlines our empirical framework and our rationale for interpreting the analysis as causal. The results of our empirical analysis are presented in Sect. 5 , before the final section concludes.

2 Context

Flood risk is an important public policy issue in Ireland, and a cause for concern amongst the general public. Footnote 2 Most of Ireland’s urban areas are coastal, and the country regularly experiences flooding events along major rivers and in coastal areas, as a result of heavy rainfall and winter storms. These floods have been costly, with roughly €1bn, or close to €800 per household, in insured losses over the period 2000–2014. Footnote 3 Moreover, the Irish government has committed to spending large sums on flood relief schemes: 68 flood relief schemes included in our analysis cost €226.6 million in total, with an additional €1 billion of planned public expenditure, or roughly 0.5% of annual national income, on flood relief schemes over the next 10 years (Office of Public Works 2018 ). Footnote 4

Prior to 2011, there were no detailed maps available on flood risk in Ireland. This changed with the release in August 2011 of Preliminary Flood Risk Assessment (PFRA) maps for the entire country. This followed a period of investment in detailed flood risk assessment by the Irish government, to comply with EU Directive 2007/60/EC, which requires all member states to assess and manage flood risk. The PFRA maps were published in August 2011 and made widely available by early 2012 with the launch of myplan.ie, a central repository for spatial information related to planning.

Importantly for our purposes, these new risk maps represent a relatively clean information treatment. In particular, the maps are not used to inform flood insurance decisions. In Ireland, flood insurance is provided by the private insurance industry without any government involvement and is usually bundled with general household insurance. As part of the government’s investment in flood risk mapping, an agreement was made with the insurance industry that the new risk maps would not be used as the basis for assessing risk for insurance purposes. Private insurers use their own independent risk assessments to price flood risk, and these did not change with the information treatment. Footnote 5

The maps that we use are the official primary source of flood risk information for the entire country and have been made readily accessible to the general public via online interactive mapping tools. Given their role in helping prioritize investment in flood defences, we control in our analysis for flood defences using detailed time-varying and spatially precise data on properties protected. The official risk maps were updated with more detailed risk maps for 300 Areas of Further Assessment starting in 2016; for this reason, we restrict our analysis to the period up to the end of 2015.

One additional piece of context is in relation to the Irish housing market during the period of our study. We start our analysis in 2006, when national online listings become available– but it is worth noting the dramatic change in housing market conditions between 2007 and 2012, when prices fell by over 50% on average, followed by growth again in prices from 2013. Our method focuses on differences, not levels, and in all our analysis we control for detailed region-specific trends, as detailed further in Section 4 below.

3 Data

3.1 Flood Data

The principal source of information related to flooding in Ireland is the Office of Public Works (OPW), the agency with responsibility for flood risk management in Ireland. The OPW provided us with data on scientifically assessed risk and on flood defences, as detailed below.

Flood risk The flood risk maps that we use in our analysis come in the form of high-resolution polygons depicting both fluvial and coastal flood risk zones for the entire country of Ireland. The maps show areas with at least a 0.1% risk of flooding per year, for both fluvial flood risk and for coastal flood risk. Footnote 6 In most of our analysis, we combine the coastal and fluvial maps together and generate an indicator variable for whether or not a property is inside either of these risk categories, which we label Risk . However, we also present separate estimates of the flood risk discount for fluvial and coastal risk types.

Flood defences The OPW provided us with polygon data related to 68 existing flood defence schemes completed between 1996 and 2017. Footnote 7 Attributes include the date of completion, spatial extent of protection, whether the defence was permanent or demountable, and the cost of each scheme. To ensure that our estimate of the flood discount is based on dwellings at risk of flooding and not protected by flood defences at the time of their listing or sale, we distinguish dwellings at risk of flooding and protected by flood defences from dwellings at risk of flooding but not protected by flood defences.

Flood events We also control for past flood events, using an extensive archive of historical flooding in Ireland, compiled by the OPW. Information in the archive is drawn from various sources, such as reports by local authorities, engineers’ reports, newspaper articles, and photos. We extract location and timing information from this archive. The dataset contains a total of 1,947 dated flood points and 84 dated flood polygons dating from 1763 to 2015. We construct indicators for dwellings affected based on the distance of the dwelling from the most recent flood event: dwellings within 100m of a flood event (or within a flood event polygon), and dwellings between 100m and 250m from a flood event. The time since the most recent flood within a 100 (or 250) meter radius of a dwelling was modelled as a categorical variable of which there were four categories: >30 years since a flood, 10–30 years, 2–10 years, and <2 years, with a base of no flood event recorded.

3.2 Dwelling & Location Data

Our baseline housing dataset is drawn from the Daft.ie National Listings Database, the archive of sale listings from the leading real estate website in Ireland. The daft.ie listings dataset is both long, covering the period from 2006Q1, at the height of a real estate boom, through to 2015Q4, and broad, covering the national market in its entirety. It is also deep, with an estimated coverage of over 95% of all listings in the Irish market, and rich, in terms of the information available for each listing. This includes both structured information on dwelling attributes (including: property type; bedrooms; bathrooms; and building energy rating) and the text of the ad, which is mined for a range of variables, including site size and orientation.

In addition to dwelling attributes, the presence of information on dwelling location allows for individual dwellings to be assigned to “at risk” of flooding, as described above, but also for the calculation of a variety of other location-specific amenities. These include nearest city centre, transport facilities, schools, and natural amenities. Based on Gillespie et al. ( 2023 ), we also include a location-specific measure of the sea-view using 3D simulation, often lacking in the flood discount literature, the omission of which may be a source of downward bias in existing flood discount estimates.

We exclude dwellings with prices above €2,000,000 or below €30,000, as these are either atypical properties or an error in the data. We also exclude dwellings listed as having fewer than one or more than five bedrooms, and fewer than one or more than eight bathrooms, for similar reasons. After applying size and price filters, we have a sample of 475,436 residential sale listings between 2006Q1 and 2015Q4. The data are illustrated in Appendix Figures A1 and A2. The breakdown of observations by risk and timing (pre- and post-information treatment) is provided in Table 1 . Further detail on the data are included in the Data Appendix and in Tables A1 and A2.

Additional analysis is performed on a dataset of transactions matched to listings, using Ireland’s Property Price Register (PPR), the official listing of residential real estate transactions in Ireland. This dataset has several major drawbacks for our purposes: it begins in January 2010, shortly before the information treatment, restricting the set of pre-treatment observations; the PPR data contain no dwelling attributes other than price, address and date of transaction; and finally, there is the issue of non-unique addresses in Ireland (prior to the introduction of postcodes, known in Ireland as Eircodes, in 2015). For these reasons, we use the PPR data as an additional robustness check, rather than as our main source of housing data. To do so, we match observations in the PPR data with listings where possible. The total number of transactions matched to listings was 45,161. Footnote 8

4 Empirical Strategy

The standard methodology in this literature builds on Rosen ’s ( 1974 ) theoretical framework of hedonic prices; a recent review of best practice in using hedonic property value models is given in Bishop et al. ( 2019 ). Conceptually, the value of a dwelling takes the following form:

where S are structural or dwelling characteristics (bedrooms, bathrooms, garden etc.), L are location and environmental characteristics (amenities, neighborhood etc.), and F are flood-related variables. The dwelling price is thus a function of all the attributes relating to the dwelling and the resulting coefficients are the implicit marginal prices of the attributes.

In our context, the key identification challenge with estimating such a model is that, even with a rich set of controls to capture S and L , flood risk is not randomly assigned and therefore may be correlated with unobservables. To counter this concern we employ a difference-in-difference (DiD) style analysis exploiting the release in 2011 of detailed flood risk maps in Ireland, as follows: each listing is assigned to one of two categories for Risk , Treatment , and Defences . We then estimate the following regressions:

where \(Risk_{it}\) is a dummy variable \(=1\) for properties located inside a flood risk zone (for fluvial or coastal flood risk or both); \(\text {Treatment}_{it}\) is a dummy variable \(=1\) for properties listed after the information treatment; \(\text {Defences}_{it}\) is a dummy variable \(=1\) for properties protected by flood defences and listed after the construction of the flood defence and \(=0\) for properties not defended at time of listing. Footnote 9 Equation ( 3 ) is the same as Eq. ( 2 ) but where \(\text {Risk}_{it}\) is split into \(\text {CoastalRisk}_{it}\) and \(\text {FluvialRisk}_{it}\) , where each dummy variable \(=1\) if a property is in a coastal flood risk zone, or a fluvial flood risk zone (allowing for overlaps), respectively. The estimate of the flood risk discount, conditional on information, for both specifications, is then the difference-in-differences: \(\beta _{DiD}\) for Eq. ( 2 ) and \(\beta _{DiD\,Coastal}\) / \(\beta _{DiD\,Fluvial}\) for Eq. ( 3 ).

\(X'_{i}\) is a vector of dwelling-specific attributes, and \(Z'_{i}\) are location-specific amenities. The \(\lambda _{dt}\) are location fixed effects at the level of Ireland’s official electoral districts (EDs), which represents a very fine spatial disaggregation of the data. Footnote 10 In our main specifications, these are included flexibly as ED-by-year fixed effects. The \(\mu _{qt}\) are national quarterly fixed effects and account for any common trends or shocks at the national level, to capture variation in the Irish housing market over time. Robust standard errors \(\epsilon _{i}\) in all specifications are clustered at the ED level.

A number of identifying assumptions are required in interpreting the estimates of Eq. ( 2 ). First, the treatment should not be correlated with unobservables. While flood risk is potentially correlated with many location-based factors that affect property values, e.g. coastal amenities, these do not change with the treatment in our context, given that our treatment refers to both the timing and spatial extent of risk information. Our estimate is still subject to bias, however, if treated and control dwellings experience differential observable or unobservable trends. To test for this, as part of robustness tests, we estimate a version of Eq. ( 2 ) with the binary treatment variable ( \(Treatment_{it}\) ) replaced by a set of year dummies, and omitting quarterly fixed effects ( \(\mu _{qt}\) ). The results are displayed in Fig. 1 and show that the estimated discount is indistinguishable from zero in the years prior to the release of the risk maps.

Further reassurance on the exogeneity of our treatment is provided by additional analysis that shows robustness of our estimate of the flood risk discount to varying the set of included covariates and to variations in the level of the spatial fixed effects (reported in Section 5 below). Our baseline specification also allows for unobservable trends at a local level: In all specifications, we include very local trends ( \(\lambda _{d,t}\) ). While many papers assume a national or city-level trend– we allow for each of over 3,000 EDs to have their own path in housing prices (or experience their own individual annual shocks). We also separately control for within-year variation in prices at the national level (by including quarterly fixed effects). Footnote 11

A further identifying assumption required in our context is that the treatment is relevant– both in the sense that buyers and sellers are aware of the risk information and use it in their decisions (once the maps become available), and that property owners are not somehow insulated from this risk. In terms of the former assumption– awareness– we are confident that market participants had access and motivation to use the risk information, as appears to be borne out in our empirical results. Footnote 12 As for the latter assumption, it is of course possible that households are insulated from the consequences of flood risk, for example through various forms of government intervention, including subsidised insurance, compensation schemes and flood protection measures, as is common in the context of flood risk around the world. In our setting, flood insurance is provided entirely by the private sector in Ireland with no government involvement. Subsidised insurance is therefore not relevant. We also control for existing flood defence investments in our analysis. However, we cannot entirely rule out the possibility that households anticipate future flood defence investments or compensation from government in the event of experiencing flooding. To the extent that either of these are prevalent in our context, they would represent an attenuation bias of our estimates of the flood risk discount, meaning our results understate the true effect.

Finally, intrepreting our estimates as average WTP would require further assumptions– in particular that there is no sorting based on risk (Kuminoff and Acirc et al. 2014 ; Greenstone 2017 ; Pinchbeck et al. 2021 ). We return to this point in the discussion of our findings in Sect. 6 .

5 Results

The results of our baseline estimates of the flood risk discount are reported in Table 2 . The specification estimated includes the full suite of controls– dwelling-specific attributes and location-based amenities, as well as ED-by-year fixed effects and national-by-quarter fixed effects, and is estimated on the full listings sample 2006–2015. Pooling both risk types (coastal and fluvial) together, in column (1), we estimate a flood risk discount of 4% for properties with at least a 0.1% risk of flooding (from either source) per year. Looking at fluvial and coastal flood risk separately, in columns (2)-(4), we see a slightly larger discount for coastal flood risk, at 4.9%, although the estimated discounts for the two sources of flood risk are not statistically different. In column (4) we include both coastal and fluvial categories together and add a separate category for properties identified as at risk from both fluvial and coastal flooding. The results indicate that there is no additional discount for properties in both. Footnote 13

Taken together, the results of Table 2 indicate that the introduction of detailed flood risk maps in Ireland in 2011 led to the emergence of a significant flood risk discount for properties within the flood risk zones defined by the new maps. While there does appear to be a slightly heavier discount for coastal flood risk, we find no strong evidence that the two risk types (fluvial and coastal) are treated differently in the housing market. Tables A3 and A4 present the results of a set of robustness tests. In Table A3 we vary the set of included covariates– removing, in turn, sets of control variables that might be expected to correlate with flood risk in the cross section; flood defences, flood events, blue space variables, and amenities. In each case the estimated flood risk discount, based on the difference-in-difference style estimator, remains remarkably insensitive to the inclusion or exclusion of these groups of control variables. This exercise provides further reassurance that the empirical setup is capturing the effect of flood risk on property values, conditional on availability of flood risk maps, and not the effects of some other unobservables.

Similarly, in Table A4 we test the robustness of our findings to varying the level of the spatial fixed effects. In our baseline specification we include ED-by-year fixed effects. There are over 3,000 EDs in our data. With 10 years of data, this results in the inclusion of over 30,000 fixed effect units in our baseline specification. In Table A4, we experiment with specifications using both more and less granular definitions of the location fixed effects: Markets (of which there are 54 in the data), micro-markets (nearly 400) and Census Small Areas (over 18,000). Again the results show that our estimate of the flood risk discount is remarkably robust to the level of these spatio-temporal controls. Unsurprisingly, the estimated discount is attenuated (and less precise) in Column (4) of Table A4, where we use the most granular version of spatial fixed effects at the Census Small Area (SA) level. With over 18,000 SAs in our data, some of these are geographically very small (in some cases individual apartment blocks), meaning that the unit of the fixed effect and treatment may be correlated: for fixed effects to aid identification, it is required that there is variation within geographical units. Additionally, the specification in Column (4) includes over 120,000 fixed effects units, leaving relatively few degrees of freedom to estimate the coefficients on the Risk categories. Again, the robustness of the estimated flood risk discount to these variations provides additional reassurance that location-based unobservables– potentially correlated with flood risk– are not affecting our estimates. Further reassurance on this point is offered by a repeat sales specification, reported in column 3 of Table A5, which produces a qualitatively very similar estimate of the flood risk discount, after stripping out all time-invariant property characteristics.

One might wonder if our estimate of the discount reflects risk perceptions or simply changes in the salience of flood risk– for example, due to media coverage of the new flood risk maps, or the occurrence of flood events that happen to coincide with the release of the risk maps. Indeed, a common finding in the existing literature is a temporary spike in flood discounts following flood events, that fades over time. Firstly, our empirical specification includes past flood events and, due to flexible functional form, allows more recent events (and those closer to a property) to have a greater impact on prices than more distant ones. To the extent that any concerns about salience of flood risk beyond events is important, we estimate a version of Eq. ( 2 ) with the binary treatment variable ( \(Treatment_{it}\) ) replaced by a set of year dummies, and omitting quarterly fixed effects ( \(\mu _{qt}\) ). The coefficients (and 95% confidence intervals) from the \(Risk*year\) interactions are displayed in Figure 1 . As the figure shows, the estimated discount is indistinguishable from zero in the years prior to the release of the risk maps, turns negative and significant with the release of the maps (in 2011), and remains consistently negative and of a similar magnitude for the remainder of our sample period (to the end of 2015). While we cannot rule out salience as a potential mechanism, the results of this exercise suggest that what we observe in the data is more than a transient spike in the salience of flood risk.

It is certainly plausible that major flood events raise the salience of flood risk to home purchasers. As noted, in all our baseline specifications we control for past flood events, and the flood risk discount we estimate is therefore conditional on controlling for previous flood events in that location. To further allay any concerns that past flood events may be driving our main findings, we present an additional robustness specification in Table A5 where we exclude from the sample any property within 5km of a major flood that occurred in Dublin in 2011. The estimated discount is, if anything, slightly larger than our baseline estimate at 4.2% (significant at 1%; t-stat of 4.89).

In all specifications reported up to now, the Risk variable is defined such that properties are categorised as either at risk or not at risk, according to their location relative to the flood risk zones defined by the official PFRA risk maps released in 2011 (i.e. a binary definition of flood risk). However, it is possible that the information treatment could also have effects on neighboring properties. For example, if demand is displaced by the new risk information, it’s possible that this could push up prices for neighboring properties. On the other hand, the risk information might spill over to neighboring properties if buyers are concerned about being located near to places with flood risk.

In additional specifications, reported in full in Table A6, we test for these neighbor effects by restricting the sample based on a set of buffer zones either side of the flood risk boundaries, as illustrated in Figure 2 . We report separate versions of the analysis that sets these buffer zones at 100m, 250m, 500m and 1,000m, respectively. Footnote 14

The results, presented in Table A6 and in Fig. 2 , show: First, in Panel (b), that restricting the sample to properties just either side of the flood risk boundary, the estimated flood risk discount remains unchanged relative to our baseline estimate for a buffer distance of 1,000m, but declines in absolute magnitude as the buffer distance is reduced. For a buffer distance of 100m, the estimated discount is not quite statistically significant at conventional levels. Second, in Panel (c), where we exclude properties just outside the flood risk boundary from the control group, the results show an increasingly large discount in absolute terms as the buffer distance is reduced. These first two sets of results suggest that our baseline findings are not being driven by locally displaced demand.

Finally, in Panel (d), we test directly the hypothesis that risk information might spill over into neighboring areas. The results show significant flood risk discounts, of 1.1% for properties within 250m of a flood risk zone, and 2.1% for properties within 100m of a flood risk zone. This final set of results– with significant estimated discounts for properties just outside flood risk zones, that decline as the distance from the boundary is increased– is consistent with market participants treating flood risk as continuous across space, rather than a spatially discrete phenomenon defined by the boundaries of the mapped flood risk zones.

Finally, readers might be concerned about the use of list prices to estimate the flood risk discount. It could be, for example, that sellers of risky homes are optimistic about their property’s value and list it for a price that does not reflect flood risk. Alternatively, it could be that agents are motivated to advise owners of risky properties to list for a relatively low price in order to achieve a quicker sale. In columns 4 and 5 of Table A5, we report the results of estimates based on a matched sample of listings and transactions. The two sets of results are based on an identical specification, estimated on the same set of dwellings, but swapping out (log) list prices in Column (4) for (log) transaction prices in Column (5). Given the relatively short pre-treatment period in the transaction price data (the PPR dataset, which only starts in 2010, with just 411 observations in flood risk zones pre-treatment), the coefficients reported here are the simple cross-sectional coefficients on Risk in the post-information treatment period. The estimated flood risk discounts reported here are somewhat smaller than in our preferred specifications (column 1), likely reflecting an urban bias in the (smaller) matched transactions sample. More importantly, the results suggest that moving from list to transaction prices has little or no bearing on our estimate of the flood risk discount: the estimated flood risk discount using list prices (2.9%) is statistically indistinguishable from that using transacted prices (2.4%), where the sample is the same.

6 Discussion and Conclusions

In this paper, we examine the relationship between flood risk and housing market outcomes, using the case of Ireland in the decade around the release of flood risk information in 2011. In particular, we exploit rich housing data– including a dataset of over 475,000 sale listings 2006–2015, supplemented by an additional set of 45,000 transactions 2010–2015– and detailed official data relating to flood risk, previous flood events, and completed flood defences. Our estimation strategy exploits a one-off national information treatment to estimate the flood risk discount. Our baseline estimate, using a difference-in-differences style estimator is a discount of 4%. We show that this estimate is robust to varying the set of included controls and to varying the level of our spatio-temporal fixed effects. We also find that the discount for at risk dwellings is not present prior to the release of the risk maps in 2011, emerges quickly after the information treatment and persists for the remainder of our sample period.

Reviews of the literature on flood risk and house prices have found widely varying results, with estimates of the flood risk premium ranging from −75% to +61% (Beltrán et al. 2018 ). A meta-analysis in that paper found a point estimate for inland flooding of −4.6% for properties located in the 100-year floodplain (Beltrán et al. 2018 ). A more recent meta-analysis study found a point estimate of −5.2% from 11 papers that ranged from −12% to +28% (Contat et al. 2024 ). Those authors highlight the relative lack of studies outside of the US, the need to control adequately for (coastal) amenities including sea views, as well as the need for more studies employing causal estimation strategies such as difference-in-differences; all issues that are addressed in our analysis.

Overall, we believe that our findings represent the causal effect of flood risk on housing prices, conditional on the availability of flood risk information. Nonetheless, our results have limitations. First, given the potential for sorting into risk, as has been observed in other contexts (e.g. Bakkensen and Acirc; A., Ma, L. 2020 ; Bakkensen and Acirc; A., Barrage, L. 2022 ), we interpret our finding as a local treatment effect. The estimated flood risk discount represents the market price of flood risk, given the prevailing market conditions and population preferences at the time of the information treatment (Pinchbeck et al. 2021 ). In our setting, the lack of any difference in prices between homes later assessed as at risk and those not at risk before 2011 suggests that flood risk (beyond past flood events, a control variable in our specification) was not salient to homebuyers before the information shock.

In terms of the magnitude of our estimate, a 4% discount would be consistent with damages per flood event of between €24,000 and €60,000, for reasonable parameter values. Footnote 15 Those values would be in line with or somewhat higher than the (limited) available evidence on average flood costs for residential property in Ireland. Our estimate is perhaps best thought of as an estimate of the welfare costs of living with flood risk– including potential damages to property, as well as more hidden costs of flooding, such as disruptions to daily life and mental health costs– for the marginal household that moves in to a flood risk zone.

A reduction in housing prices could also be expected to have an impact on housing supply, in one of two ways. This includes the volume of homes actively traded on the market or the construction of new homes. We see no impact on the volume of online sale listings of the release of flood risk information. Meanwhile the magnitude of the aggregate effect is likely to have been too small to change the volume of homes built. Günnewig-Mönert et al. ( 2024 ) estimate the elasticity of supply in Ireland during this period to be close to unitary (+0.9). Where 8% of homes are subject to a 4% negative shock to their value, this implies the impact on the housing supply due to areas at risk to be roughly 0.25% (0.32% times 0.9).

It is also possible that the maps we use do not reflect the actual risk experienced by households (or market participants) for example to the extent that they are shielded from the consequences of flooding by government compensation schemes or investments in flood protection (including in the future). Our estimates also exclude other non-residential costs of flooding such as damage to public infrastructure or commercial real estate. On this basis, the true costs of flooding to society are likely larger than what we estimate here. While our discussion and our estimation strategy emphasises the information treatment, we cannot fully rule out salience as an alternative mechanism, for example due to media coverage of the new flood risk maps. The evidence we present shows the information treatment coincided with the emergence for the first time of a flood risk discount, which persists for the remainder of our sample period. This suggests that the effect we observe is more than a transient spike in flood risk salience.

Our findings have important policy implications for flood risk management, insurance and flood defences, as well as for projections of future flood losses in a world of increasing flood risk. Perhaps most importantly for policymakers, our results present compelling evidence on the effectiveness of public investments in flood risk information provision. Better information results in more awareness and a clear price signal, which should translate into less exposure to flooding in future. Finally, we show that, on a like-for-like basis, list prices act as a good proxy for transaction prices, where those are unavailable (Lyons and Acirc; C. 2019 ). Given the prevalence of flood risk in lower-income settings, where formal housing statistics are typically weaker, this is a useful finding for both researchers and policymakers.

Notes

In our analysis, we treat all at-risk properties together in the baseline. We also separate between risk of fluvial flooding, where the main measure of risk is 1% per year, and coastal flooding, where it is 0.5% per year. A small share (roughly one sixth) of at-risk properties are assessed as at 0.1% risk.

As part of this research, we also conducted an online survey of public attitudes to flood risk, which found that the general public in Ireland is concerned about flooding, that those concerns have increased for many over the last 10 years, and that a large majority of people expect the problem to get worse in the coming decades. Further details about the survey available on request.

Information provided directly from Insurance Ireland (the representative body for private insurance companies operating in Ireland).

The stated intention is that these schemes will provide protection to 80% of the 34,500 dwellings in Ireland assessed as having a 1% chance of experiencing a significant flood event in any year. In scaling by national income, the measure used in Ireland is modified Gross National Income (GNI*), which was valued at €197.5 billion in 2018, according to data from the Central Statistics Office (CSO), available here https://www.cso.ie/en/releasesandpublications/ep/p-nie/nie2018/ (last accessed in June 2023).

Our understanding is that, given the relatively modest size of the Irish market, for commercial reasons private insurers have not carried out detailed (hydrological or property level) flood risk assessments in Ireland. Instead, their assessment is largely based on (recent) historical flood events and/or the proximity of water sources. At the time that these maps first appeared in 2011, there was no legal requirement on property sellers to disclose previous flooding or flood risk (Murphy and Whittaker 2012 ). The flood risk assessment carried out by the Irish government, and represented in the maps that we use here, thus likely represented a significant information update relative to what was available previously, including information implicit in insurance price signals.

Most of the at-risk properties in our data are classed as at medium risk, which the OPW defines as 0.5% for coastal flood risk and 1% for fluvial risk. Fewer than one-sixth of at-risk properties in our data are classed as at low risk, defined as 0.1% for either flood risk type.

Pre-existing flood defence schemes are accounted for in the risk maps that we use.

Further details on our housing data and the matching process between transactions and listings are given in the Data Appendix.

Our definition of ‘defended’ only applies to at-risk properties, such that there are no properties in the sample that are ‘defended’ and outside of flood risk zones.

There are over 3,000 EDs in our data.

A related source of concern might be if the supply of properties (listings) responded to the information treatment - for example if owners of risky properties anticipated the release of the new risk information and tried to sell before the information could affect prices, or if owners of risky properties postponed listing their properties immediately after the release of the new risk maps. We find no evidence of such supply effects on our listings sample. The proportion of risky properties listed before and after the information treatment in 2011 is almost identical at 7.7% pre-treatment and 7.8% post-treatment. We are grateful to one of the anonymous referees for suggesting this supply-side story.

Access to the risk maps was made relatively easy with their inclusion in the online platform myplan.ie.

In fact the discount appears to be slightly smaller for properties at risk of both fluvial and coastal flooding. Summing the three coefficients reported in Column (4) of Table 2 , the estimated flood risk discount for properties in the ‘both’ category is 3%, but not statistically significant.

For urban areas, a 100m buffer corresponds roughly to including dwellings on the next street adjacent to a flood risk zone, while a 250m buffer corresponds to including entire city blocks that are adjacent to flood risk zones. See the data illustrations presented in Figures A1 and A2.

The flood risk discount represents the capitalized value of future expected flood damages. For a €300,000 dwelling (close to the average list price in our data) with a 1% annual flood risk, a 30-year horizon and a 2% discount rate, a 4% flood risk discount equates to damages per flood of €50,000. Calculated in perpetuity, the same discount implies damages of about €24,000 per event. For fluvial flook risk, our sample includes properties with 0.1% and 1% annual flood risk, with a weighted average risk across the sample of 0.84%. For this level of risk, a 4% flood risk discount corresponds to damages per event of between €28,000 (in perpetuity) and €60,000 (30-year time horizon). This is in line with private sector estimates cited in the Irish parliament, which found that 70,000 properties are exposed to flood risk in Ireland with an estimated cost of EUR 2 billion, or €28,000 per flood event (Gamma Intelligence Services 2019 ).

References

Atreya A, Ferreira S, Kriesel W (2013) Forgetting the flood? an analysis of the flood risk discount over time. Land Economics 89(4):577–596

Baker A, Acirc; C., Larcker, D., Acirc; F., Wang, C., Acirc; C. (2022) How much should we trust staggered difference-in-differences estimates? Journal of Financial Economics 144(2):370ΓÇô395

Bakkensen L, Acirc; A., Barrage, L. (2022) Going underwater? flood risk belief heterogeneity and coastal home price dynamics. The Review of Financial Studies 35(8):3666ΓÇô3709

Bakkensen L, Acirc; A., Ma, L. (2020) Sorting over flood risk and implications for policy reform. Journal of Environmental Economics and Management 104:102362

Beltrán A, Maddison D, Elliott R (2018) Is flood risk capitalised into property values? Ecological Economics 146(100):668–685

Beltrán A, Maddison D, Elliott R (2019) The impact of flooding on property prices: A repeat-sales approach. Journal of Environmental Economics and Management 95:62–86

Bin O, Crawford T, Acirc; W., Kruse, J., Acirc; B., Landry, C., Acirc; E. (2008) Viewscapes and flood hazard: Coastal housing market response to amenities and risk. Land economics 84(3):434–448

Bin O, Kruse J, Acirc; B., Landry, C., Acirc; E. (2008) Flood hazards, insurance rates, and amenities: Evidence from the coastal housing market. Journal of Risk and Insurance 75(1):63–82

Bin O, Polasky S (2004) Effects of flood hazards on property values: Evidence before and after hurricane floyd. Land Economics 80(4):490–500

Bishop K, Acirc C, Kuminoff N, Acirc V, Banzhaf H, Acirc S, Boyle K, von Gravenitz K, Pope J, Acirc C, Smith V, Acirc K, Timmins C (2019) Best Practices in Using Hedonic Property Value Models for Welfare Measurement

Borusyak K, Jaravel X, Spiess J (2021) Revisiting event study designs: Robust and efficient estimation. arXiv preprint [SPACE] arXiv:2108.12419

Bosker M, Garretsen H, Marlet G, van Woerkens C (2019) Nether lands: Evidence on the price and perception of rare natural disasters. Journal of the European Economic Association 17(2):413–453

Callaway B, Sant’Anna P, Acirc; H. (2021) Difference-in-differences with multiple time periods. Journal of Econometrics 225(2):200–230

Contat J, Hopkins C, Mejia L, Suandi M (2024) When climate meets real estate: A survey of the literature. Real Estate Economics 52(3):618–659

Davlasheridze M, Fisher-Vanden K, Klaiber H, Acirc; A. (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

De Chaisemartin C, d’Haultfoeuille X (2020) Two-way fixed effects estimators with heterogeneous treatment effects. American Economic Review 110(9):2964–2996

European Commission Joint Research Centre. (2017). Atlas of the human planet 2017: Global exposure to natural hazards . Publications Office. https://doi.org/10.2760/932505

Gamma Intelligence Services. (2019). Coastal flooding caused by climate change to impact 70,000+ irish addresses by 2050 . Retrieved September 16, 2024, from https://irishtechnews.ie/coastal-flooding-caused-by-climate-change-to-impact-70000-irish-addresses-by-2050/

Gibson M, Mullins J, Acirc; T. (2020) Climate risk and beliefs in new york floodplains. Journal of the Association of Environmental and Resource Economists 7(6):1069–1111

Gibson M, Mullins J, Acirc T, Hill A (2017) Climate change, flood risk, and property values: Evidence from new york city. Work. Pap., Dep. Econ., Williams Coll., Williamstown, MA

Gillespie T, Hynes S, Lyons R, Acirc C (2023) The value of a sea view: Hedonic estimates using 3d simulation and natural language processing. Land Economics

Goodman-Bacon A (2021) Difference-in-differences with variation in treatment timing. Journal of Econometrics 225(2):254–277

Greenstone M (2017) The continuing impact of sherwin rosen’s ‘hedonic prices and implicit markets: Product differentiation in pure competition. Journal of Political Economy 125(6):1891–1902

Günnewig-Mönert M, Lyons R, Acirc C (2024) Housing prices, costs, and policy: The housing supply equation in ireland since 1970. Real Estate Economics

Hallegatte S, Green C, Nicholls R, Acirc; J., Corfee-Morlot, J. (2013) Future flood losses in major coastal cities. Nature climate change 3(9):802–806

Hallstrom D, Acirc G, Smith V, Acirc K (2005) Market responses to hurricanes. Journal of Environmental Economics and Management , 50 (3), 541–561. https://doi.org/10.1016/j.jeem.2005.05.002

Hino M, Burke M (2020) Does information about climate risk affect property values? (Tech. rep.). National Bureau of Economic Research

Hsieh L-H, Acirc; C. (2021) Is it the flood, or the disclosure? an inquiry to the impact of flood risk on residential housing prices. Land Use Policy 106:105443

Kuminoff N, Acirc; V., Pope, J., Acirc; C. (2014) Do ‘capitalization effects’ for public goods reveal the public’s willingness to pay? International Economic Review 55(4):1227–1250

Kydland F, Acirc; E., Prescott, E., Acirc; C. (1977) Rules rather than discretion: The inconsistency of optimal plans. Journal of political economy 85(3):473–491

Lautrup M, Matthiesen L, Acirc; L., Jacobsen, J., Acirc; B., Panduro, T., Acirc; E. (2023) Welfare effects and the immaterial costs of coastal flooding. Environmental and Resource Economics 85(2):415–441

Lin Y, McDermott T, Acirc K, Michaels G (2021) Cities and the sea level

Lyons R, Acirc; C. (2019) Can list prices accurately capture housing price trends? insights from extreme markets conditions. Finance Research Letters 30:228–232

MacDonald D, Acirc N, White H, Acirc L, Taube P, Acirc M, Huth W, Acirc L (1990) Flood hazard pricing and insurance premium differentials: Evidence from the housing market. Journal of Risk and Insurance, 654–663

Meldrum J, Acirc; R. (2016) Floodplain price impacts by property type in boulder county, colorado: Condominiums versus standalone properties. Environmental and Resource Economics 64:725–750

Murphy H, Whittaker A (2012) Floods, sweat and tears. Law Society Gazette 106:18

Office of Public Works. (2018). Kevin ’boxer’ moran publishes flood risk management plans and announces, iexcl;spicurrencydollar, iquest;257 million for 50 new flood relief schemes, as part of a ten-year, iexcl;spicurrencydollar, iquest;1 billion programme of investment in flood relief measures? Retrieved June 2, 2023, from https://www.gov.ie/en/press-release/5f18ca-kevin-boxer-moran-publishes-flood-risk-management-plans-and-announce/

Ortega F, Taspınar S (2018) Rising sea levels and sinking property values: Hurricane sandy and new york’s housing market. Journal of Urban Economics 106:81–100

Petkov I (2022) Weather shocks, population, and housing prices: The role of expectation revisions. Economics of Disasters and Climate Change 6(3):495–540

Pilla F, Gharbia S, Acirc; S., Lyons, R. (2019) How do households perceive flood-risk? the impact of flooding on the cost of accommodation in dublin, ireland. Science of The Total Environment 650:144–154

Pinchbeck E, Acirc W, Roth S, Szumilo N, Vanino E (2021) The price of indoor air pollution: Evidence from risk maps and the housing market

Pralle S (2019) Drawing lines: Fema and the politics of mapping flood zones. Climatic Change 152(2):227–237

Rosen S (1974) Hedonic prices and implicit markets: Product differentiation in pure competition. Journal of political economy 82(1):34–55

Seo J, Oh J, Kim J (2021) Flood risk awareness and property values: Evidences from seoul, south korea. International Journal of Urban Sciences 25(2):233–251

Shr Y-H, Acirc; J., Zipp, K., Acirc; Y. (2019) The aftermath of flood zone remapping: The asymmetric impact of flood maps on housing prices. Land Economics 95(2):174–192

Sun L, Abraham S (2021) Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. Journal of Econometrics 225(2):175–199

Timar L, Grimes A, Fabling R (2018) That sinking feeling: The changing price of urban disaster risk following an earthquake. International Journal of Disaster Risk Reduction 31:1326–1336

Votsis A, Perrels A (2016) Housing prices and the public disclosure of flood risk: A difference-in-differences analysis in finland. J. Real Estate Fin. Econ. 53(4):450–471

Download references

Rights: © The Author(s) 2025.
Rights and permissions Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . Reprints and permissions
Edition changes: Licensed narrative text edition. Original headings, paragraphs and references are preserved with normalized whitespace and layout. Figures, tables, display equations and separately licensed third-party material are not reproduced; use the linked original full text for those elements and definitive pagination. This is an attributed republication, not newly conducted research.