# RECORD 6e4be5daf59355440452370cc31893fbd8320793 # TITLE: Assessing the Economic Benefits of Flood Defenses: A Repeat‐Sales Approach # YEAR: 2018.0 VENUE: Risk Analysis DOI: 10.1111/risa.13136 # AUTHORS: Allan Beltrán; D. Maddison; R. Elliott # PAGES: 54 CANDIDATE TABLE PAGES (1-based): [30, 31, 34, 35, 37, 42, 43, 53, 54] # ABSTRACT: In this article we identify the impact of the construction of flood defenses on property prices using a difference‐in‐differences repeat‐sales methodology. Our data set contains information on over 12 million individual property transactions, which is merged with GIS data identifying the spatial location and main characteristics of 1,666 flood defenses built in England between 1995 and 2014. Results suggest that at the finer 6‐digit postcode level the construction of flood defenses raises urban house prices by 12.6% to 16.7%. However, for rural properties at the slightly coarser 5‐digit postcode level the construction of defenses reduces house prices by 0.8% to 5.0%. This suggests that in certain locations the disamenity impact of flood defenses and the perceived threat of redirected flooding outweigh the benefits of reduced flood risk. ===== PAGE 1 (score 3, numeric-paren 0, table-words 0, var-words 3) ===== Allan Beltrán, David Maddison and Robert J.R. Elliott Assessing the economic benefits of flood defenses: a repeat-sales approach Article (Accepted version) (Refereed) Original citation: Beltrán, Allan and Maddison, David and Elliott, Robert J. R. (2018) Assessing the economic benefits of flood defenses: a repeat-sales approach. Risk Analysis. ISSN 0272-4332 DOI: 10.1111/risa.13136 © 2018 Society for Risk Analysis This version available at: http://eprints.lse.ac.uk/90207/ Available in LSE Research Online: September 2018 LSE has developed LSE Research Online so that users may access research output of the School. Copyright © and Moral Rights for the papers on this site are retained by the individual authors and/or other copyright owners. Users may download and/or print one copy of any article(s) in LSE Research Online to facilitate their private study or for non-commercial research. You may not engage in further distribution of the material or use it for any profit-making activities or any commercial gain. You may freely distribute the URL (http://eprints.lse.ac.uk) of the LSE Research Online website. This document is the author’s final accepted version of the journal article. There may be differences between this version and the published version. You are advised to consult the publisher’s version if you wish to cite from it.” For more research by LSE authors go to LSE Research Online ===== PAGE 30 (score 93, numeric-paren 73, table-words 1, var-words 8) ===== 29 Table 3. Repeat-sales model: The effect of flood defences on property prices (Area impacted: 6-digit postcode) 1 (1) (2) Variables All Flood Plain Bracket sample Bracket (B) 0.147*** 0.185*** (0.00153) (0.00424) (A) Type of property (see note 2) B*sdetached -0.0109*** -0.0286*** (0.000639) (0.00167) B*terraced -0.0299*** -0.0488*** (0.000673) (0.00173) B*flat -0.0313*** -0.0647*** (0.00142) (0.00392) B*free 0.0510*** 0.0419*** (0.00126) (0.00355) B*rural 0.0241*** 0.0444*** (0.000564) (0.00133) B*coastal 0.0172*** 0.0189*** (0.000442) (0.00104) B*quartile -0.0633*** -0.0702*** (0.000276) (0.000704) Bracket-defence sample B*Defence (D) 0.195*** 0.172*** (0.0409) (0.0419) (B) Type of property (see note 2) B*D*sdetached -0.0293 -0.0100 (0.0211) (0.0217) B*D*terraced -0.0662*** -0.0472** (0.0231) (0.0237) B*D*flat -0.198*** -0.170*** (0.0337) (0.0340) B*D*free -0.0226 -0.0205 (0.0273) (0.0279) B*D*rural -0.0141 -0.0423*** (0.0152) (0.0155) B*D*coastal 0.0144 0.0119 (0.0163) (0.0166) B*D*quartile -0.0356*** -0.0303*** (0.0101) (0.0103) Defence design B*D*sop 8.39e-05 7.93e-05 (6.91e-05) (6.92e-05) B*D*length -2.03e-05 -1.76e-05 (2.45e-05) (2.46e-05) B*D*(sop*length) 7.00e-09 1.99e-08 (2.28e-07) (2.29e-07) Type of defence (see note 3) B*D*embankment -0.00786 -0.00370 (0.0156) (0.0159) B*D*bridgeabt - - B*D*highground -0.0288 -0.0274 (0.0258) (0.0269) B*D*demount 0.0246 0.0299 (0.0487) (0.0485) B*D*floodgate 0.0366 0.0382 (0.0742) (0.0743) Flood perception B*D*months(sqrt) 4 -0.00282*** -0.00310*** (0.000947) (0.000962) B*D*duration 0.000234** 0.000215* (0.000111) (0.000112) B*D*months(sqrt)4*duration 3.79e-06 5.49e-06 (8.37e-06) (8.44e-06) (Continued) ===== PAGE 31 (score 23, numeric-paren 3, table-words 4, var-words 11) ===== 30 Table 3.Continued Lyear (s) -0.000426*** -9.99e-05 (2.98e-05) (7.68e-05) Year (t) 0.000478*** 0.000132* (2.97e-05) (7.66e-05) Observations 7,217,966 1,135,690 Treated Obs. 1,824 1,794 County FE YES YES R-squared 0.112 0.111 Notes: Robust standard errors in parentheses. *, ** and *** means rejection of the null hypothesis at the 10%, 5% and 1% significance level. 1 Properties impacted by the construction of a flood defence are identified using the 6-digit postcode area where the defence was constructed. 2 Omitted categories are dummy variables for detached property, urban location and fluvial flood risk. 3 The omitted category is a dummy variable for floodwall. 4 Square root of the number of months since the previous flood with respect to the second sale. Column (1) indicates that for the case of detached houses in urban areas the construction of flood walls increases prices by 19.5% before accounting for those factors that modify the impact of the construction of flood defences. This seems to be the same for semi-detached properties. For terraced properties the capitalisation is somewhat smaller of around 12.5%. For the case of flats, however, following an examination of the sizes of the coefficients, it appears that there are no significant benefits from the construction of flood defences. Results in column (2) where the control group is confined to those properties located in the floodplain are similar except for rural areas where in column (2) the benefits from flood defences are around 4% smaller. From the coefficient on B*D*quartile it appears that the benefits of flood defences are capitalised mainly in the prices of properties in the lowest price quartile. This is plausible because the value of more expensive properties frequently derives not from the value of the structure itself but rather because of the value of the land upon which it is built. Only the structure is damaged through flooding.15 Examination of the defence design variables shows that the standard of protection is statistically insignificant; even in the immediate vicinity of the flood defence, the area most likely to experience a benefit in terms of a reduction in the risk of flooding, there is no 15 All properties include land but sometimes this land is held leasehold rather than freehold. ===== PAGE 32 (score 13, numeric-paren 0, table-words 5, var-words 3) ===== 31 evidence that defences offering a higher standard of protection results in greater price increases. The coefficient estimates only hint at possibility that an increase in the standard of protection might increase property prices i.e. the coefficients on B*D*sop and B*D*(sop*length) are both positive but insignificant. Likewise, there is little evidence that the type of defence has a significant impact on prices. Finally, contained among the flood perception variables there is evidence to suggest that the benefits of flood defences are greater when there is a more recent history of flooding. Benefits are also greater when the most recent episode of flooding was severe. Both of these are as expected. We now consider a number of changes to gauge the plausibility of the underlying assumptions. One of these is that in the repeat sales model the only characteristic of properties that possibly changes between sales is the construction of a flood defence. This assumption is tested in Table 4 where, instead of making use of all repeat sales that bracket the construction of a flood defence, we use only those properties that are sold no more than 5 years prior to the construction of the flood defence and sold again no more than 5 years after the construction of the flood defence. The results contained in column (1) of Table 4 based on the narrower time-window are very similar to those contained in column (1) of Table 3. More specifically, the coefficient indicating the impact of the construction of the flood defence on the benchmark property is still positive and statistically significant at the 1% level of confidence. There also continues to be strong evidence that the impact on flats is much less than that on other types of property, and that the impact is more pronounced for properties in the lowest price quartile. None of these coefficients moreover, appears dramatically different to what they were previously. We also investigate the consequences of the assumption of a unified market for across different property types and locations (Michaels and Smith, 1990). We start by distinguishing between flats and houses (defined as all types of property excluding flats). Analysing only ===== PAGE 34 (score 186, numeric-paren 152, table-words 2, var-words 8) ===== 33 which is statistically significant at the 1% level of confidence but not statistically different from the corresponding estimate of 0.195 in column (1) in Table 3. Table 4. Repeat-sales model: Additional results (Area impacted: 6-digit postcode) 1 (1) (2) (3) (4) (5) Variables Reduced Time Window House Flat Urban Rural (A) Type of property (see note 2) Bracket sample Bracket (B) 0.147*** 0.145*** 0.163*** 0.123*** 0.269*** (0.00153) (0.00161) (0.00162) (0.00161) (0.00482) B*sdetached -0.0109*** -0.0163*** 0.00209*** -0.0591*** (0.000638) (0.000652) (0.000713) (0.00141) B*terraced -0.0299*** -0.0365*** -0.0129*** -0.102*** (0.000673) (0.000694) (0.000740) (0.00160) B*flat -0.0316*** - - -0.0186*** -0.111*** (0.00142) (0.00151) (0.00465) B*free 0.0510*** 0.0595*** 0.0418*** 0.0462*** 0.0939*** (0.00126) (0.00132) (0.00472) (0.00133) (0.00414) B*rural 0.0241*** 0.0306*** -0.0186*** (0.000564) (0.000583) (0.00218) B*coastal 0.0173*** 0.0162*** 0.00251*** 0.0124*** 0.0224*** (0.000442) (0.000507) (0.000890) (0.000479) (0.00114) B*quartile -0.0633*** -0.0703*** -0.0535*** -0.0551*** -0.107*** (0.000276) (0.000314) (0.000584) (0.000292) (0.000770) Bracket-defence sample B*Defence (D) 0.265*** 0.200*** 0.00854 0.137*** 0.0748 (0.104) (0.0442) (0.0488) (0.0509) (0.1042) (B) Type of property (see note 2) B*D*sdetached -0.0526 -0.0391 - -0.00132 -0.0298 (0.0407) (0.0255) (0.0277) (0.0311) B*D*terraced -0.0749 -0.0742*** - -0.0301 -0.0520 (0.0455) (0.0239) (0.0307) (0.0434) B*D*flat -0.235*** - - -0.167*** -0.00650 (0.0746) (0.0374) (0.0737) B*D*free -0.0302 -0.0298 0.0666 -0.0410 0.0562 (0.0207) (0.0286) (0.0760) (0.0284) (0.0443) B*D*rural 0.0105 -0.00528 -0.0409 - - (0.0291) (0.0161) (0.0441) B*D*coastal 0.00540 0.00435 -0.0533* 0.0232 -0.0224 (0.0352) (0.0183) (0.0317) (0.0204) (0.0245) B*D*quartile -0.0570*** -0.0398*** -0.0118* -0.0185 -0.0456*** (0.0237) (0.0113) (0.0068) (0.0127) (0.0170) Defence_ desig B*D*sop -7.42e-05 0.000240 2.92e-05 0.000229** -0.000102 (6.33e-05) (0.000157) (8.43e-05) (0.000110) (7.82e-05) B*D*length -4.57e-05 -1.18e-05 2.31e-05 -7.71e-06 -1.86e-05 (4.79e-05) (2.77e-05) (6.39e-05) (3.88e-05) (3.71e-05) B*D*(sop*length) 4.71e-07 -1.74e-07 -4.13e-07 2.68e-08 -3.13e-07 (4.05e-07) (2.58e-07) (1.06e-06) (3.95e-07) (3.42e-07) Type of defence (see note 3) B*D*embankment -0.0419 0.0140 -0.0400** -0.00560 -0.0143 (0.0311) (0.0172) (0.0152) (0.0194) (0.0274) B*D*bridgeabt - - - - - B*D*highground -0.0874 0.00242 -0.0572 -0.0350 -0.0411 (0.0616) (0.0253) (0.0495) (0.0248) (0.0661) B*D*demount -0.0228 0.0473 -0.0350 - -0.0319 (0.0815) (0.0484) (0.0219) (0.0491) B*D*floodgate - 0.0547 - -0.104 0.0836 (0.0749) (0.100) (0.0919) Flood_ perception B*D*months(sqrt) 4 -0.00362*** -0.00253** -0.00295** -0.00379*** -0.00323* (0.00123) (0.00103) (0.00138) (0.00113) (0.00171) B*D*duration 0.000510*** 0.000210* -9.27e-05 8.47e-05 0.000700*** (0.000188) (0.000115) (0.000217) (0.000127) (0.000233) B*D*months(sqrt)4*duration -1.32e-05 4.24e-06 1.09e-05 8.19e-06 -3.40e-05 ===== PAGE 35 (score 24, numeric-paren 10, table-words 3, var-words 5) ===== 34 (1.59e-05) (1.01e-05) (1.57e-05) (9.33e-06) (2.14e-05) (Continued) Table 4.Continued Lyear (s) -0.000425*** -0.00174*** -0.00681*** -0.000231*** -0.00189*** (2.98e-05) (3.25e-05) (7.18e-05) (3.26e-05) (7.45e-05) Year (t) 0.000477*** 0.00180*** 0.00676*** 0.000284*** 0.00195*** (2.97e-05) (3.24e-05) (7.17e-05) (3.25e-05) (7.43e-05) Observations 7,222,401 5,802,937 1,419,464 5,925,246 1,201,622 Treated Obs. 487 1,498 326 1,215 609 COUNTY FE YES YES YES YES YES R-squared 0.112 0.138 0.042 0.113 0.129 Notes: Robust standard errors in parentheses. *, ** and *** means rejection of the null hypothesis at the 10%, 5% and 1% significance level. 1 Properties impacted by the construction of a flood defence are identified using the 6-digit postcode area where the defence was constructed. 2 Omitted categories are dummy variables for detached property, urban location and fluvial flood risk. 3 The omitted category is a dummy variable for floodwall. 4 Square root of the number of months since the previous flood with respect to the second sale. 6.2. Treatment group: 5-digit postcode area The consequences of flood defences, in terms of the reduced risk of flooding and the disamenity impacts, are likely to find expression at different geographical distances. Quite plausibly, flood-relief benefits might be experienced by those in the immediate vicinity of the flood defence construction, whereas disamenity impacts might occur over a somewhat wider area. As previously noted these disamenity impacts are likely to include visual intrusion as well as impacts associated with restricted access to water bodies.16 There may also be adverse ecosystem impacts arising out of the construction of the defences. Determining which of these concerns is uppermost in the mind of households however, is probably better investigated in other ways rather than through hedonic analysis. Furthermore, it also seems likely these impacts differ across contexts. For example, the visual impact of a flood defence will depend upon the vista and this will differ between urban and rural areas. To determine the true benefits of flood defence projects, it is important not to confine any analysis only to 16 Very recently glass floodwalls have been installed at several sites in England. Given that these are much more expensive this is affirmation that visual disamenity impacts are an important issue. ===== PAGE 36 (score 9, numeric-paren 0, table-words 3, var-words 3) ===== 35 those areas enjoying benefits from a reduced risk of flooding. Examining a larger area might be necessary in order to guarantee capture of other sorts of impacts.17 In this sub-section we present the same regressions as in Table 3, but we now define the area impacted by flood defences as 5-digit postcode area in which the defence is located. That is, we now focus on a much larger area, one which is quite likely to include properties not experiencing any significant reduction in the risk of flooding but potentially suffering disamenity impacts.18 The results appear in Table 5. Including properties not in the immediate vicinity of the flood defence will alter the extent to which the impacts of flood defences are capitalised into property prices in the area under scrutiny. It could be that some characteristics of flood defences that enhance the benefits of those living in the immediate vicinity of the construction actually worsen any disamenity impacts felt by households at greater geographical distances from the flood defence. For example, whilst many households in a 5-digit postcode area might experience only disamenity impacts from an increase in the standard of protection i.e. taller flood defences, households living in the much smaller 6-digit postcode area in which the flood defence is situated might experience a mixture of both positive and negative impacts. This might explain changes in the signs of some coefficients as the focus switches from the 6-digit postcode area to the 5-digit postcode area. 17 It is possible that reducing the risk of flooding in some locations might increase the risk in others in a manner revealed only through hydrological and hydraulic modelling. Whether this is addressed simply by expanding the geographical area assumed impacted by the construction of a flood defence seems doubtful. Equally doubtful is whether any households affected by redirected flooding would be perfectly aware of the threat. Even though the Environment Agency takes care to avoid such a situation some households might nevertheless perceive an increase in the risk of flooding. 18 A consequence of this is a sizeable increase in the number of treated observations. ===== PAGE 37 (score 100, numeric-paren 80, table-words 1, var-words 8) ===== 36 Table 5. Repeat-sales model: The effect of flood defences on property prices (Area impacted: 5-digit postcode) 1 (1) (2) Variables All Flood Plain Bracket sample Bracket (B) 0.147*** 0.186*** (0.00154) (0.00431) (A) Type of property (see note 2) B*sdetached -0.0107*** -0.0280*** (0.000641) (0.00169) B*terraced -0.0295*** -0.0483*** (0.000676) (0.00175) B*flat -0.0307*** -0.0642*** (0.00143) (0.00399) B*free 0.0511*** 0.0409*** (0.00127) (0.00362) B*rural 0.0242*** 0.0455*** (0.000568) (0.00135) B*coastal 0.0172*** 0.0189*** (0.000443) (0.00104) B*quartile -0.0632*** -0.0702*** (0.000277) (0.000712) Bracket-defence sample B*Defence (D) 0.0613*** 0.0490* (0.0201) (0.0260) (B) Type of property (see note 2) B*D*sdetached -0.0327*** -0.0233** (0.00808) (0.0118) B*D*terraced -0.0407*** -0.0281** (0.00860) (0.0121) B*D*flat -0.0711*** -0.0613*** (0.0168) (0.0224) B*D*free 0.0257* 0.0362* (0.0154) (0.0203) B*D*rural -0.00636 -0.0332*** (0.00606) (0.00894) B*D*coastal 0.0688*** 0.0767*** (0.00982) (0.0123) B*D*quartile -0.0114*** -0.00856 (0.00376) (0.00523) Defence design B*D*sop -6.28e-05*** -7.36e-05*** (1.54e-05) (2.12e-05) B*D*length -1.77e-05** -3.00e-05*** (7.99e-06) (1.09e-05) B*D*(sop*length) -1.50e-07*** -1.38e-07*** (2.74e-08) (3.31e-08) Type of defence (see note 3) B*D*embankment -0.0130** -0.00368 (0.00587) (0.00822) B*D*bridgeabt -0.0497 -0.0522 (0.0483) (0.0505) B*D*highground -0.00832 -0.00296 (0.0121) (0.0164) B*D*demount -0.0694*** -0.0528* (0.0180) (0.0271) B*D*floodgate -0.0125 0.0200 (0.0137) (0.0219) Flood_ perception B*D*months(sqrt) 4 -0.00124*** -0.00232*** (0.000425) (0.000631) B*D*duration 9.93e-05*** 0.000127** (3.45e-05) (5.37e-05) B*D*months(sqrt)4*duration 7.45e-06*** 9.63e-06** (2.85e-06) (4.34e-06) (Continued) ===== PAGE 38 (score 23, numeric-paren 2, table-words 5, var-words 10) ===== 37 Table 5.Continued Lyear (s) -0.000421*** -8.34e-05 (2.98e-05) (7.69e-05) Year (t) 0.000473*** 0.000116 (2.98e-05) (7.67e-05) Observations 7,217,966 1,135,690 Treated Obs. 14,716 7,417 County FE YES YES R-squared 0.111 0.110 Notes: Robust standard errors in parentheses. *, ** and *** means rejection of the null hypothesis at the 10%, 5% and 1% significance level. 1 Properties impacted by the construction of a flood defence are identified using the 5-digit postcode area where the defence was constructed. 2 Omitted categories are dummy variables for detached property, urban location and fluvial flood risk. 3 The omitted category is a dummy variable for floodwall. 4 Square root of the number of months since the previous flood with respect to the second sale. Turning now to the results for the 5-digit postcode areas, beginning with column (1), it can be observed that detached properties in urban areas impacted by the construction of a floodwall and threatened by fluvial flooding were resold on average, for a price 6.1% higher than those properties where exposure to flood risk remained unchanged. This is before accounting for those factors that modify the impact of the construction of flood defences. This coefficient is moreover significant at the 1% level of confidence. At the same time however, as the sample is restricted such that the comparison is made only with properties located in a floodplain in column (2) the statistical significance of the coefficient diminishes. Once more the results in column 2 are our preferred specification. Considering now the differential effects for other sorts of properties there appear to be significant differences. Semi-detached properties, terraced properties and in particular flats appear to enjoy significantly lower benefits compared to detached properties. Once again however, as the sample is restricted the statistical significance of these differential impacts slips although in the case of flats the benefits of the construction of a floodwall in an urban area continue to be significantly smaller than for detached properties. ===== PAGE 42 (score 148, numeric-paren 119, table-words 1, var-words 7) ===== 41 5 are indeed associated with the construction of flood defences and their design characteristics. Table 6. Repeat-sales model. Robustness test: Placebo regression 6-Digit 1 5-Digit 2 (1) (2) (3) (4) Variables All Flood Plain All Flood plain Bracket sample 𝐵𝑟𝑎𝑐𝑘𝑒𝑡 ̅̅̅̅̅̅̅̅̅̅̅ (𝐵̅) 0.200*** 0.204*** 0.189*** 0.194*** (0.000838) (0.00227) (0.000839) (0.00228) (A) Type of property (see note 3) 𝐵̅*sdetached -0.0376*** -0.0462*** -0.0433*** -0.0519*** (0.000344) (0.000888) (0.000345) (0.000891) 𝐵̅*terraced -0.0528*** -0.0616*** -0.0507*** -0.0595*** (0.000369) (0.000935) (0.000369) (0.000939) 𝐵̅*flat -0.0712*** -0.0831*** -0.0775*** -0.0896*** (0.000639) (0.00175) (0.000640) (0.00176) 𝐵̅*free 0.0369*** 0.0353*** 0.0436*** 0.0419*** (0.000518) (0.00148) (0.000520) (0.00149) 𝐵̅*rural 0.0265*** 0.0419*** 0.0224*** 0.0378*** (0.000305) (0.000702) (0.000306) (0.000707) 𝐵̅*coastal 0.00418*** 0.0131*** 0.00385*** 0.0128*** (0.000482) (0.000672) (0.000485) (0.000676) 𝐵̅*quartile -0.0755*** -0.0775*** -0.0798*** -0.0818*** (0.000166) (0.000430) (0.000166) (0.000431) Bracket-defence sample 𝑩̅*Defence (𝑩̅*D) 0.0163 0.00872 -0.00297 -0.0212 (0.0288) (0.0290) (0.00986) (0.0139) (B) Type of property (see note 3) 𝐵̅*D*sdetached 0.00697 0.0172 -0.00344 -0.00181 (0.0168) (0.0171) (0.00555) (0.00814) 𝐵̅*D*terraced 0.000853 0.0116 -0.00684 0.00102 (0.0170) (0.0172) (0.00554) (0.00802) 𝐵̅*D*flat -0.0121 -0.00126 -0.0120 0.0159 (0.0190) (0.0191) (0.00754) (0.0105) 𝐵̅*D*free -0.0152 -0.0161 -0.00832 0.00193 (0.0143) (0.0146) (0.00563) (0.00804) 𝐵̅*D*rural 0.0132 -0.00876 0.00506 -0.00846* (0.0107) (0.0110) (0.00353) (0.00493) 𝐵̅*D*coastal 0.00681 0.00171 0.00529 -0.00466 (0.0105) (0.0106) (0.00422) (0.00483) 𝐵̅*D*quartile -0.00482 -0.00215 0.000434 0.00437 (0.00631) (0.00633) (0.00208) (0.00279) Defence_ design 𝐵̅*D*sop -3.70e-05 -4.14e-05 9.29e-06 -4.11e-06 (3.08e-05) (3.06e-05) (8.17e-06) (1.12e-05) 𝐵̅*D*length -1.44e-05 -1.18e-05 -7.75e-06 -7.09e-06 (1.73e-05) (1.71e-05) (4.98e-06) (6.53e-06) 𝐵̅*D*(sop*length) 2.07e-07 2.00e-07 4.77e-09 3.14e-08 (1.48e-07) (1.48e-07) (2.29e-08) (2.89e-08) Type of defence (see note 4) 𝐵̅*D*embankment -0.0135 -0.00818 -0.00508 0.00166 (0.00892) (0.00902) (0.00331) (0.00445) 𝐵̅*D*bridgeabt - - -0.0150 -0.0121 (0.0179) (0.0229) 𝐵̅*D*highground -0.0161 -0.00227 -0.00862 0.00669 (0.0247) (0.0272) (0.00588) (0.00959) 𝐵̅*D*demount 0.00526 0.00918 -0.00896 0.00855 (0.0390) (0.0392) (0.0185) (0.0244) 𝐵̅*D*floodgate -0.0336 -0.0292 0.00609 0.0113 (0.0297) (0.0291) (0.00847) (0.0110) Floo d_ perce ti 𝐵̅*D*months(sqrt) 5 0.000439 8.43e-05 0.000307 0.000181 (0.000679) (0.000700) (0.000219) (0.000326) 𝐵̅*D*duration 6.48e-05 5.15e-05 1.52e-05 -3.66e-05 ===== PAGE 43 (score 27, numeric-paren 8, table-words 5, var-words 6) ===== 42 (5.88e-05) (6.03e-05) (2.16e-05) (3.06e-05) 𝐵̅*D*months(sqrt)5*duration -5.06e-06 -3.45e-06 -8.15e-07 1.27e-06 (4.21e-06) (4.24e-06) (1.53e-06) (2.19e-06) (Continued) Table 6.Continued Lyear (s) -0.00159*** -0.00141*** -0.00159*** -0.00141*** (2.94e-05) (7.57e-05) (2.94e-05) (7.57e-05) Year (t) 0.00162*** 0.00144*** 0.00162*** 0.00144*** (2.93e-05) (7.54e-05) (2.92e-05) (7.54e-05) Observations 1,030,068 107,537 1,030,068 107,537 Treated Obs. 2,502 2,459 20,440 10,415 County FE YES YES YES YES R-squared 0.173 0.170 0.149 0.147 Notes: Robust standard errors in parentheses. *, ** and *** means rejection of the null hypothesis at the 10%, 5% and 1% significance level. 1 Properties located impacted by the construction of a flood defence are identified using the 6-digit postcode area where the defence was constructed. 2 Properties impacted by the construction of a flood defence are identified using the 5-digit postcode area where the defence was constructed. 3 Omitted categories are dummy variables for detached property, urban location and fluvial flood risk. 4 The omitted category is a dummy variable for floodwall. 5 Square root of the number of months since the previous flood with respect to the second sale. 7. Interpretation of Results From the results given in column (2) of Tables 3 and 5 it is possible to estimate the typical effect on prices of the construction of flood defences inter alia for fluvial and coastal flooding as well as for urban and rural properties and for different types of properties.20 Results corresponding to sample average values appear in Table 7. For the case of flats we report estimates only for urban areas since few flats are found elsewhere. We present separate estimates for 5-digit and 6-digit postcode areas. The results in Table 7 suggest that the overall effect of the construction of flood defences on property prices varies considerably, ranging from a price increase of 16.7% for houses in 6- digit postcode urban areas protected to a decrease of 5.7% for flats in 5-digit postcode urban areas. Furthermore, providing flood protection appears to result in a price decrease of between 0.8 and 5.0% for houses located in rural 5-digit postcode areas, something we suggest is the result of the disamenity impacts of flood defences outweighing the benefits of 20 For areas exposed to fluvial flood risk, there is an average of 188 months with respect to the previous flood, with and average duration of 58 days. For areas exposed to coastal flood risk, the average months with respect to the previous flood is 177 months, with an average duration of 32 days. ===== PAGE 44 (score 11, numeric-paren 0, table-words 3, var-words 5) ===== 43 flood protection. However, in 6-digit postcodes in rural areas the reverse is true; the benefits of flood protection appear to outweigh any disamenity impacts presumed present. Table 7. Capitalisation of flood defences into property prices in England, 1996-2014 (95% confidence interval and mean capitalisation rate in parentheses) Type of property Fluvial risk Coastal risk 6-digit postcode House Urban £28,829*** [£12,676 – £45,029] £45,420*** [£20,521 – £70,675] (12.6%) (16.7%) Rural £14,804* [-£2,768 – £32,399] £28,784** [£2,736 – £54,723] (6.5%) (10.5%) Flat Urban -£8,290* [-£17,340 – £760] -£576 [-£14,165 – £13,006] (-4.4%) (-0.3%) 5-digit postcode House Urban £3,203 [-£6,635 – £13,042] £15,048** [£1,833 – £28,511] (1.4%) (5.5%) Rural -£11,509** [-£22,125 – -£915] -£2,326 [-£16,937 – £12,230] (-5.0%) (-0.8%) Flat Urban -£10,819*** [-£15,858 – -£5,780] -£2,852 [-£9,127 – £3,460] (-5.7%) (-1.5%) Note: These estimates are based on column (2) in both Table 3 and 5. The monetary values are calculated using the sample average price of a house located in a fluvial or coastal flood risk area: £228,804 and £273,617, respectively. For the case of flats we use the average price of £190,142 for a flat located in a fluvial floor risk area. 95% confidence interval appears in brackets. Mean capitalisation rate appears in parentheses. *, ** and *** means rejection of the null hypothesis at the 10%, 5% and 1% significance level. All prices are in July 2014 GBP. Finally, the construction of flood defences always results in a decrease in the price of flats ranging from 0.3 to 5.7%. We suggest that this is the result of two aforementioned effects: lower or no benefits from flood protection and the disamenity impact of flood defences. Before closing, it is worth reiterating that the rates of capitalisation in Table 7 are based on the perceived change in the level of the risk of flooding. Obviously it would be interesting to investigate the extent to which the change in objective risk brought about by flood protection is perceived by homeowners. Furthermore, if there is to begin with a lack of perception of flood risk in an area then the construction of a flood defence itself conveys information. This might result in a situation in which the construction of a flood defence actually reduces ===== PAGE 53 (score 181, numeric-paren 149, table-words 2, var-words 7) ===== 52 Table A2. Repeat-sales model. Robustness test: Excludes extreme values (Excludes top 1% and bottom 1% of observations) 6-Digit 1 5-Digit 2 (1) (2) (3) (4) Variables All Flood Plain All Flood Plain Bracket sample Bracket (B) 0.166*** 0.199*** 0.165*** 0.199*** (0.00144) (0.00407) (0.00145) (0.00413) (A) Type of property (see note 2) B*sdetached -0.0105*** -0.0278*** -0.0103*** -0.0270*** (0.000619) (0.00163) (0.000621) (0.00164) B*terraced -0.0251*** -0.0466*** -0.0247*** -0.0461*** (0.000653) (0.00169) (0.000656) (0.00171) B*flat -0.0383*** -0.0654*** -0.0377*** -0.0650*** (0.00133) (0.00374) (0.00134) (0.00380) B*free 0.0416*** 0.0355*** 0.0415*** 0.0343*** (0.00117) (0.00336) (0.00117) (0.00342) B*rural 0.0241*** 0.0454*** 0.0242*** 0.0465*** (0.000546) (0.00129) (0.000550) (0.00131) B*coastal 0.0193*** 0.0198*** 0.0193*** 0.0198*** (0.000427) (0.00101) (0.000427) (0.00102) B*quartile -0.0673*** -0.0737*** -0.0672*** -0.0737*** (0.000269) (0.000692) (0.000270) (0.000700) Bracket-defence sample B*Defence (D) 0.207*** 0.194*** 0.0527*** 0.0418* (0.0390) (0.0399) (0.0196) (0.0258) (B) Type of property (see note 2) B*D*sdetached -0.0308 -0.0123 -0.0343*** -0.0253** (0.0204) (0.0209) (0.00782) (0.0114) B*D*terraced -0.0699*** -0.0494** -0.0425*** -0.0281** (0.0218) (0.0224) (0.00828) (0.0116) B*D*flat -0.202*** -0.181*** -0.0638*** -0.0515** (0.0321) (0.0324) (0.0164) (0.0221) B*D*free -0.0241 -0.0264 0.0355** 0.0508** (0.0257) (0.0264) (0.0151) (0.0202) B*D*rural -0.0177 -0.0475*** -0.00669 -0.0364*** (0.0146) (0.0149) (0.00588) (0.00868) B*D*coastal 0.0123 0.0107 0.0659*** 0.0774*** (0.0161) (0.0164) (0.00979) (0.0122) B*D*quartile -0.0363*** -0.0320*** -0.0114*** -0.00944* (0.00990) (0.0101) (0.00364) (0.00509) Defence_ design B*D*sop 7.86e-05 7.39e-05 -6.45e-05*** -8.02e-05*** (6.90e-05) (6.91e-05) (1.53e-05) (2.06e-05) B*D*length -4.07e-05* -3.83e-05 -1.95e-05** -3.75e-05*** (2.34e-05) (2.35e-05) (7.63e-06) (1.01e-05) B*D*(sop*length) 3.93e-08 5.95e-08 -1.37e-07*** -1.22e-07*** (2.25e-07) (2.26e-07) (2.71e-08) (3.19e-08) Type of defence (see note 3) B*D*embankment -0.0107 -0.00651 -0.0120** -0.00155 (0.0143) (0.0146) (0.00567) (0.00793) B*D*bridgeabt - - -0.0524 -0.0580 (0.0482) (0.0501) B*D*highground -0.0189 -0.0184 -0.00327 0.00685 (0.0241) (0.0251) (0.0114) (0.0149) B*D*demount 0.0259 0.0321 -0.0693*** -0.0503* (0.0484) (0.0482) (0.0179) (0.0270) B*D*floodgate 0.0336 0.0353 -0.0147 0.0155 (0.0739) (0.0741) (0.0137) (0.0218) Flood_ perception B*D*months(sqrt) 4 -0.00290*** -0.00322*** -0.00126*** -0.00241*** (0.000900) (0.000914) (0.000411) (0.000608) B*D*duration 0.000152 0.000125 8.52e-05** 8.98e-05* (9.32e-05) (9.42e-05) (3.33e-05) (5.13e-05) B*D*months(sqrt)4*duration 1.10e-05 1.32e-05* 8.93e-06*** 1.28e-05*** (6.98e-06) (7.06e-06) (2.77e-06) (4.13e-06) (Continued) ===== PAGE 54 (score 18, numeric-paren 6, table-words 3, var-words 4) ===== 53 Table A2.Continued Lyear (s) -0.000177*** 0.000136* -0.000173*** 0.000151** (2.90e-05) (7.52e-05) (2.90e-05) (7.53e-05) Year (t) 0.000234*** -0.000102 0.000230*** -0.000117 (2.90e-05) (7.51e-05) (2.90e-05) (7.51e-05) Observations 7,073,555 1,113,592 7,073,555 1,113,592 Treated Obs. 1,804 1,774 14,574 7,357 County FE YES YES YES YES R-squared 0.130 0.125 0.130 0.124 Notes: Robust standard errors in parentheses. *, ** and *** means rejection of the null hypothesis at the 10%, 5% and 1% significance level. 1 Properties located impacted by the construction of a flood defence are identified using the 6-digit postcode area where the defence was constructed. 2 Properties impacted by the construction of a flood defence are identified using the 5-digit postcode area where the defence was constructed. 3 Omitted categories are dummy variables for detached property, urban location and fluvial flood risk. 4 The omitted category is a dummy variable for floodwall. 5 Square root of the number of months since the previous flood with respect to the second sale.