1# RECORD 11ddac3108d5fcc806866635ea2276160f2ca53d2# TITLE: Coastal armoring and sinking property values: the case of seawalls in California3# YEAR: 2020.0 VENUE: Environmental Economics and Policy Studies DOI: 10.1007/s10018-020-00278-34# AUTHORS: Arlan Brucal; John M. Lynham5# PAGES: 23 CANDIDATE TABLE PAGES (1-based): [7, 9, 10, 15, 17, 18, 19]6# ABSTRACT: Rising sea levels necessitate careful consideration of different forms of coastal protection but cost-benefit analysis is limited when important non-market social costs have not been measured. Seawalls protect individual properties but can potentially impose negative externalities on neighboring properties via accelerated beach loss. We conduct a hedonic valuation of seawalls in two coastal California counties: San Diego and Santa Cruz. We find no strong evidence to suggest that the presence of a seawall is positively correlated with the value of the home protected. However, we find that seawalls are strongly negatively correlated with the value of neighboring properties in Santa Cruz but not in San Diego county, suggesting that the effect of seawalls depend on certain geographical attributes. Our results are robust to accounting for the public-good nature of locational attributes and the potential spatial dependence of housing prices. Simulation reveals that doubling the extent of seawalls in San Diego and Santa Cruz could reduce property tax revenues by $7 million and $54 million, respectively.789===== PAGE 1 (score 7, numeric-paren 0, table-words 0, var-words 7) =====1011Vol.:(0123456789)12Environmental Economics and Policy Studies (2021) 23:55–7713https://doi.org/10.1007/s10018-020-00278-3141 315RESEARCH ARTICLE16Coastal armoring and sinking property values: the case 17of seawalls in California18Arlan Brucal1 · John Lynham219Received: 10 September 2019 / Accepted: 4 May 2020 / Published online: 4 June 2020 20© The Author(s) 202021Abstract22Rising sea levels necessitate careful consideration of different forms of coastal protec-23tion but cost-benefit analysis is limited when important non-market social costs have 24not been measured. Seawalls protect individual properties but can potentially impose 25negative externalities on neighboring properties via accelerated beach loss. We con-26duct a hedonic valuation of seawalls in two coastal California counties: San Diego 27and Santa Cruz. We find no strong evidence to suggest that the presence of a seawall 28is positively correlated with the value of the home protected. However, we find that 29seawalls are strongly negatively correlated with the value of neighboring properties in 30Santa Cruz but not in San Diego county, suggesting that the effect of seawalls depend 31on certain geographical attributes. Our results are robust to accounting for the public-32good nature of locational attributes and the potential spatial dependence of housing 33prices. Simulation reveals that doubling the extent of seawalls in San Diego and Santa 34Cruz could reduce property tax revenues by $7 million and $54 million, respectively.35Keywords Seawalls · Hedonic valuation · Coastal erosion · Property values36JEL Classification H23 · Q26 · Q51 · Q5437Arlan Brucal: Brucal acknowledges support from the Grantham Foundation and the Economic and 38Social Research Council (ESRC) through the Centre for Climate Change Economics and Policy. 39John Lynham: Lynham acknowledges support from the Center for Ocean Solutions and valuable 40research assistance from Qingran Li.41Electronic supplementary material The online version of this article (https://doi.org/10.1007/s1001428-020-00278-3) contains supplementary material, which is available to authorized users.43 * Arlan Brucal 44 45a.z.brucal@lse.ac.uk46 47John Lynham 48 49lynham@hawaii.edu501 51Grantham Research Institute on Climate Change and the Environment, London School 52of Economics and Political Science, Houghton Street, London WC2A 2AE, UK532 54Department of Economics, University of Hawaii at Manoa, 2424 Maile Way, Saunders Hall 542, 55Honolulu, HI 96822, USA565758===== PAGE 7 (score 30, numeric-paren 0, table-words 4, var-words 9) =====596061611 362Environmental Economics and Policy Studies (2021) 23:55–77 63(ERF) is reduced to zero. To compute this variable, the erosion rate at each prop-64erty location is calculated first. Since in some areas erosion rates can be determined 65from historical records, while in others they are affected by protection devices such 66as seawalls, we applied the same method used in the Heinz Center’s erosion haz-67ard report. A detailed description of this formulation is presented in Appendix B. A 68squared logarithm of Geotime is added to the model to test for a non-linear effect. 69Built Post-FIRM is an indicator variable for whether or not the home was built post 70the publication of a Flood Insurance Rate Map (FIRM) for the surrounding area. 71Post-FIRM homes typically have lower insurance premiums if they are built above 72base flood elevation levels, therefore we expect the coefficient on this variable to be 73positive.74Table 1 presents summary statistics for all of our main variables. On the average, 75properties in Santa Cruz are generally more expensive compared to those in San 76Diego. This is probably due to bigger houses and lot sizes. Houses in San Diego are 77also situated in higher places and closer to the water, which implies that these houses 78are potentially in high cliffs above water (Griggs and Savoy 1985). The difference in 79Table 1 Summary statistics80Source of basic data: The Heinz III Center for Science, Economics, and Management (Center 2000)81San Diego82Santa Cruz83Mean84SD85Min86Max87Mean88SD89Min90Max91Price (’000, in 2014 92US$)93588.5694412.209566.15963120.0097699.0898600.859971.251003120.00101Seawall (0–1)1020.181030.381040.001051.001060.191070.391080.001091.00110Percent seawalled (%)11139.5911217.791130.0011461.5411542.4911621.001170.00118100.00119House size (sq. ft.)1201784.05 811.17121700.001225000.001232571.56 2377.86 450.0012422000.00125Lot size (sq. ft.)1265843.64 5019.31 863.0012728239.00 8032.56 4873.24 399.0012828106.00129Years since last trans-130action13114.4213210.521331.0013442.0013518.6213612.841370.0013855.00139Years since built14040.0914119.861422.00143109.0014444.7814520.561461.00147104.00148Fireplace (0–1)1490.811500.391510.001521.001530.721540.451550.001561.00157Waterfront (0–1)1580.471590.501600.001611.001620.441630.501640.001651.00166Waterfront nourish-167ment (1–0)1680.061690.241700.001711.001720.021730.121740.001751.00176Beach nourishment 177(1–0)1780.131790.341800.001811.001820.051830.211840.001851.00186Bedroom1872.931881.091891.001906.001913.191921.471931.0019412.00195Geotime1962289.42 6609.13 0.4719745772.00 4895.74 8636.80 55.9619846059.00199Elevation (ft.)2007801.26 4220.72 −12.30 10158.70 1327.22 3360.50 − 18.00 10086.40201Distance to water (ft.)202186.75203162.452041.00205803.35206249.77207161.5720829.05209670.38210Beach width21181.21212286.962130.002143000.0021538.39216179.092170.002182000.00219Built Post-FIRM2200.052210.212220.002231.002240.072250.252260.002271.00228Year of purchase229198523010.5223119572321998233198023412.8423519442361999237Observations238132239151240241242===== PAGE 8 (score 19, numeric-paren 0, table-words 5, var-words 9) =====24324462245 246Environmental Economics and Policy Studies (2021) 23:55–772471 3248this locational attribute should provide more insights on the effect of seawalls on 249inland property. In particular, seawalls in Santa Cruz should affect inland properties 250much more than they could in San Diego. The extent of seawalled properties are 251about the same for the two counties, which makes these two cases comparable.2523 Results2533.1 OLS results254We start our analysis with an ordinary least squares (OLS) method using all the vari-255ables listed in Table A.6. We did separate analysis for San Diego and Santa Cruz 256to account for the potential difference in hedonic price functions between these 257two market segments. We also run combined observations from the two counties 258to formally test this difference by adding a San Diego indicator variable that turns 259on when the property is situated in San Diego County. Results are summarized in 260Table 2.261For San Diego, we reject the null hypothesis of no correlation for only 4 var-262iables. In particular, we find that price increases by about 0.50% as the size of a 263house increases by one percent.9 This translates to a marginal implicit price of house 264size by about $164/sq.ft.10 Lot size has smaller marginal implicit price, with houses 265being valued more by about $37 as lot size increases by sq.ft. Waterfront is surpris-266ingly not significant. This is possibly due to fact that most houses in San Diego are 267situated in high cliffs above water and beach is not easily accessible. Surprisingly, 268homeowners value having a fireplace in their homes, considering that most areas of 269San Diego rarely require a fireplace for heating. This is possibly because we did not 270explicitly know whether it is outdoor or indoor fireplaces. Outdoor fireplaces may be 271associated with homeowners’ valuation of outdoor living, which makes this struc-272tural attribute a worthy investment.11273In Santa Cruz, house size is positively correlated with housing prices. However, 274since houses in Santa Cruz are larger on the average, the implicit marginal price of 275house size is smaller at $135/sq.ft. The same observation holds true for lot size, with 276Santa Cruz houses being valued by $32 more per additional sq.ft. Waterfront, in 277contrast to San Diego, is significantly associated with higher housing prices in Santa 278Cruz. In particular, waterfront houses in Santa Cruz are valued more by about 40%, 2799 We recognize that all of our regression coefficients should be interpreted as elasticities, although we 280remain silent on the direction of causality. See Gopalakrishnan et al. (2011) for a discussion of the endo-281geneity problem in hedonic beach value models and for a potential solution.28210 We calculate marginal implicit price ($ per unit) for the logged variables as:283where 훽k is the estimated coefficient of the regressor xk , and ̄xk and ̄p are the unconditional means of the 284regressor xk and the dependent variable p, respectively.285Marginal Implicit Price = 𝛽k286̄xk287× ̄p28811 The National Association of Realtors 2013 Home Features Survey reveals that 57% chose homes with 289at least one fireplace.290291292===== PAGE 9 (score 165, numeric-paren 138, table-words 1, var-words 7) =====293294632951 3296Environmental Economics and Policy Studies (2021) 23:55–77 297Table 2 Results from simple OLS298Regression299All variables300Reduced model301Reduced model302Combined303San Diego304Santa Cruz305Combined306San Diego307Santa Cruz308Combined309San Diego310Santa Cruz311Seawall (0−1)3120.0993130.2163140.1003150.1143160.2233170.0683180.0793190.216320−0.007321(0.130)322(0.182)323(0.206)324(0.110)325(0.159)326(0.157)327(0.107)328(0.157)329(0.146)330Percent seawalled (%)331−0.080***332−0.038333−0.111**334−0.091***335−0.061336−0.100***337−0.088***338−0.065*339−0.094***340(0.030)341(0.045)342(0.050)343(0.024)344(0.038)345(0.028)346(0.024)347(0.038)348(0.028)349House size (sq. ft.)3500.516***3510.498***3520.523***3530.496***3540.455***3550.480***3560.494***3570.460***3580.504***359(0.113)360(0.125)361(0.163)362(0.089)363(0.105)364(0.131)365(0.088)366(0.104)367(0.129)368Lot size (sq. ft.)3690.150**3700.370***3710.0573720.133**3730.373***3740.0793750.137***3760.375***3770.076378(0.061)379(0.097)380(0.073)381(0.053)382(0.089)383(0.065)384(0.053)385(0.089)386(0.063)387Years since last transaction388−0.387***389−0.392***390−0.375***391−0.374***392−0.379***393−0.355***394−0.372***395−0.380***396−0.359***397(0.040)398(0.058)399(0.057)400(0.036)401(0.055)402(0.048)403(0.036)404(0.055)405(0.047)406Years since built407−0.174*408−0.134409−0.218410−0.064411−0.050412−0.106413−0.070414−0.052415−0.102416(0.095)417(0.125)418(0.174)419(0.065)420(0.099)421(0.085)422(0.065)423(0.097)424(0.087)425Fireplace (0−1)4260.232***4270.267***4280.1814290.213***4300.255**4310.187*4320.211***4330.241**4340.191*435(0.075)436(0.096)437(0.115)438(0.074)439(0.105)440(0.104)441(0.074)442(0.103)443(0.104)444Waterfront (0−1)4450.290***4460.1654470.333**4480.375***4490.1964500.463***4510.365***4520.2074530.428***454(0.106)455(0.119)456(0.159)457(0.088)458(0.134)459(0.117)460(0.086)461(0.133)462(0.112)463Waterfront nourishment (1−0)464−0.1474650.147466−0.216467−0.2704680.118469−0.447**470(0.232)471(0.620)472(0.234)473(0.205)474(0.582)475(0.204)476Beach nourishment (1−0)4770.056478−0.0014790.1244800.087481−0.2054820.209**483(0.180)484(0.456)485(0.168)486(0.137)487(0.408)488(0.104)489Bedrooms490−0.036491−0.0664920.032493(0.088)494(0.116)495(0.137)496Geotime4970.008498−0.034499−0.001500(0.066)501(0.322)502(0.076)503504505===== PAGE 10 (score 51, numeric-paren 37, table-words 3, var-words 1) =====50650764508 509Environmental Economics and Policy Studies (2021) 23:55–775101 3511Robust standard errors are in parentheses. *, **, ***Denote statistical significance at 10, 5 and 1% level.512Source of basic data: The Heinz III Center for Science, Economics, and Management513Table 2 (continued)514Regression515All variables516Reduced model517Reduced model518Combined519San Diego520Santa Cruz521Combined522San Diego523Santa Cruz524Combined525San Diego526Santa Cruz527Geotime squared528−0.005529−0.005530−0.000531(0.005)532(0.021)533(0.007)534Elevation (ft.)5350.013536−0.0045370.020538(0.016)539(0.027)540(0.023)541Distance to water (ft.)542−0.0095430.027544−0.023545(0.043)546(0.109)547(0.049)548Beach width5490.0155500.0195510.013552(0.021)553(0.031)554(0.029)555Built Post-FIRM556−0.540**557−0.580558−0.476559(0.239)560(0.356)561(0.401)562San Diego (1−0)5630.132564−0.015565−0.013566(0.104)567(0.076)568(0.075)569Constant5709.587***5717.889***57210.385***5739.380***5747.445***57510.067***5769.374***5777.414***5789.913***579(0.767)580(1.402)581(1.132)582(0.652)583(0.863)584(0.961)585(0.653)586(0.866)587(0.974)588Observations589247590130591117592281593130594151595281596130597151598R-sq.5990.5576000.6236010.5506020.5166030.5746040.4976050.5136060.5726070.486608R-sq. (adj.)6090.5226100.5666110.4736120.4966130.5386140.4616150.4976160.5446170.457618619620===== PAGE 11 (score 16, numeric-paren 0, table-words 4, var-words 8) =====621622656231 3624Environmental Economics and Policy Studies (2021) 23:55–77 625posting a marginal implicit price of more than $232,000.12 The existence of fire-626place, while positive, is not significantly correlated with housing prices.627Overall, we find no strong evidence to suggest that there are systematic differ-628ences in housing prices between the two counties or market segmentation (Taylor 6292003). Our OLS regression finds that the San Diego dummy, while positive, is statis-630tically insignificant. We also performed Wald test for San Diego dummy and found 631the F-statistic insignificant, with the value of 1.60 and p value equivalent to 0.207.632The variables indicating flooding and erosion risks (e.g.Geotime, Elevation) do 633not appear to be significantly correlated with property prices. This is true when we 634analyze housing prices in each county separately and after combining observations 635from the two counties. This is different from the findings in Kriesel and Friedman 636(2003). This could potentially be explained by the different nature of the shorelines 637between the two regions. Unlike the Atlantic and Gulf coast, much of the Pacific 638coastline consists of narrow beaches backed by steep cliffs. Cliff erosion is site spe-639cific and episodic, which is often hidden by the low long-term average erosion rate. 640Only 45 observations in our entire sample have a Geotime less than 100 years. It is 641likely that erosion risk is not reflected in property prices till decades later.642Although the low long-term erosion rate might reduce homeowners’ perceptions 643of the erosion threat to their properties, shoreline management policies can still 644affect house buyers’ awareness of erosion risk levels. Regarding the seawall indica-645tors, we obtain remarkably similar results to Kriesel and Friedman (2003) and Krie-646sel et al. (2005). The coefficient on Seawall is positive but not statistically significant 647in either county. The imprecision of this estimate relative to Kriesel and Friedman 648(2003) could be due to our smaller sample size. We evaluate the effect of seawalls 649on housing prices using combined samples to increase the sample size and observe 650same result.651Meanwhile, we notice a significant negative effect of Percent Seawalls on prop-652erty prices in Santa Cruz. This suggests that property owners in the county are worse 653off as a result of extensive coastal armoring. Furthermore, our coefficient of −0.11 is 654more than three times larger than the −0.03 found by Kriesel and Friedman (2002). 655This suggests that, as the percentage of waterfront coastal armoring in a community 656increases by 100%, the associated value of a house located on the Santa Cruz coast 657drops by $77,597. In contrast, we do not find any strong evidence to suggest that the 658extent of armored coastline in San Diego has an effect on property values. If there 659is an effect, house values in San Diego declines by about $22,365 if the share of 660waterfront coastal armoring in a community increases by 100%. Interestingly, our 661estimate falls within the ballpark of Kriesel and Friedman (2003) who found that for 662the same percentage increase in coastal armoring, house prices on the Atlantic and 663Gulf coasts drop by $41,091 in 2014 dollars.66412 For dummy variables, the calculated marginal implicit price is expressed as:665where 훽k and p are as previously defined.666(5)667Marginal Implicit Price = (e𝛽k −1) × ̄p668669670===== PAGE 15 (score 16, numeric-paren 10, table-words 2, var-words 1) =====671672696731 3674Environmental Economics and Policy Studies (2021) 23:55–77 675horizontal axis and the average prices of neighboring houses weighted by the spatial 676lag vector 퐖 in the vertical axis. We assume that the degree of spatial dependence 677of a particular house is inversely related to the geographic distance between the two 678properties. That is,679We find evidence of positive spatial dependence for houses located in Santa Cruz. 680The Moran statistic is 0.0867 and is statistically significant at 5% level. In contrast, 681there is no strong evidence to suggest that housing prices in San Diego are spatially 682dependent.683We formally test for the presence of global spatial autocorrelation by calculating 684Moran’s I (Moran 1950) and Geary c (Cliff and Ord 1970). We then determine the 685significance of these statistics following the procedure suggested by Pisati (2001). 686Table 3 summarizes the results. The tests confirm our graphical approach suggest-687ing no strong spatial dependence of house prices in San Diego. In contrast, there is 688strong evidence of spatial autocorrelation of house prices in Santa Cruz. The spatial 689autocorrelation persists even if we combined the two datasets.6903.2.2 Hedonic prices with spatial dependence691With an indication of potential spatial dependence, we re-estimate Eq. (2) using data 692on Santa Cruz county house prices. This time, we relax the assumption that 휆 and 휌 693are both zero. We estimate parameters based on models with spatial dependence in 694a maximum likelihood framework using the procedure suggested by Drukker et al. 695(7)696퐖=697⎛698⎜699⎜700⎜701⎜⎝7020703w1,2 w1,3 … w1,n704w2,17050706w2,3 … w2,n707w3,1 w3,27080709… w3,n710⋮711⋮712⋮713⋱714⋮715wn,1 wn,2 wn,3 …7160717⎞718⎟719⎟720⎟721⎟⎠722wi,j = 1723dij724.725Table 3 Tests for global spatial 726autocorrelation727The estimated standard deviation of each statistic are in parentheses. 728*,**,***Represents significance at 10, 5 and 1%, respectively729San Diego730Santa Cruz731Combined732Distance of nearest 733neighbor (in kms)734Mean7351.2797360.042737Min7380.9317390.006740Max7413.3367420.239743Moran’s I7440.0237450.087**7460.054*747(0.049)748(0.037)749(0.022)750Geary c7510.9437520.900**7530.924**754(0.051)755(0.040)756(0.024)757758759===== PAGE 16 (score 18, numeric-paren 0, table-words 6, var-words 6) =====76076170762 763Environmental Economics and Policy Studies (2021) 23:55–777641 3765(2015). We employ different specifications to potentially account for different forms 766of spatial autocorrelation, namely spatial lag model, spatial error model and the spa-767tial ARAR(1,1,) model.17 Then, for each model, we run three different specifications 768similar to the simple OLS regression in which we start with all the variables and 769drop some of them to slightly increase our sample. We do this to test the stability of 770our estimates. Results are presented in Table 4.771We start the analysis by looking at the estimated spatial autocorrelation param-772eters. We find no strong evidence to suggest that there exists spatial autocorrela-773tion. In particular, we fail to reject the null hypotheses that (1) 휆= 0, 휌≠0 ; (2) 774휆≠0, 휌= 0 ; and (3) 휆≠0, 휌≠0 in Eq. (6).18775The estimated effect of changes in the variables of interest, even after account-776ing for potential spatial dependence, remains robust and consistent with our base-777line results. The coefficient remains insignificant for the indicator variable Seawall. 778Nonetheless, the coefficient for the extent of coastline armoring (Percent Seawalls), 779remains negative albeit slightly smaller and statistically significant. This confirms 780our earlier result that as the percentage of waterfront coastal armoring in a beach 781community increases, the associated value of houses located in San Diego and 782Santa Cruz drops significantly. Moreover, our estimates remain stable across differ-783ent forms of assumed spatial dependence and under different sets of controls, thus 784increasing our confidence on the robustness of the results.7854 Discussion786We find that the sales price of a house located within 1,000 feet of the coast in San 787Diego and Santa Cruz counties is correlated with its structural characteristics and 788the shoreline management policies taken in the community. Although seawalls pro-789tect coastal structures from erosion damages, the negative coefficient on Percent 790Seawalls calls for attention to the welfare impacts of coastal armoring on waterfront 791and non-waterfront households.792Using our baseline estimates, we predicted the potential change in total hous-793ing values for the two counties when the share of those that are protected by sea-794walls increases by a certain percentage (see Table 5). Results suggest that dou-795bling the current level of properties protected by seawalls may be associated with 796a $20,954.16 drop in the resale value of an average coastal home in San Diego that 797is protected by seawalls. For those that are not protected by seawalls, the magnitude 79817 Thanks to an anonymous referee, we recognized that the estimation of a SARAR model may intro-799duce more problems than solving them. Nonetheless, the employment of different specifications, includ-800ing the SARAR, is meant to show that our estimates, particularly on seawall protection, are consistent 801and robust over different specifications of spatial dependence. We refrain to interpret the actual estimates 802due to uncertainties in the type of spatial dependence.80318 We also perform diagnostic tests for spatial dependence in an OLS framework after estimating 804Eq. (2), following the procedure as suggested by Pisati (2001). Results are summarized in Table A.8. We 805find no strong evidence to suggest that there is spatial dependence amongst observed housing prices in 806each county.807808809===== PAGE 17 (score 161, numeric-paren 134, table-words 1, var-words 7) =====810811718121 3813Environmental Economics and Policy Studies (2021) 23:55–77 814Table 4 Results for models with spatial dependence815Spatial lag model816Spatial error model817Spatial ARAR(1,1) model818(1)819(2)820(3)821(1)822(2)823(3)824(1)825(2)826(3)827Seawall (0−1)8280.0948290.068830−0.0088310.1038320.060833−0.0118340.0848350.046836−0.026837(0.161)838(0.138)839(0.128)840(0.163)841(0.136)842(0.126)843(0.157)844(0.132)845(0.123)846Percent seawalled (%)847−0.098*848−0.100**849−0.093**850−0.112**851−0.100***852−0.094**853−0.081*854−0.090**855−0.084**856(0.053)857(0.041)858(0.041)859(0.054)860(0.037)861(0.038)862(0.047)863(0.035)864(0.035)865House size (sq. ft.)8660.496***8670.480***8680.502***8690.519***8700.484***8710.508***8720.472***8730.468***8740.489***875(0.155)876(0.117)877(0.118)878(0.155)879(0.114)880(0.115)881(0.149)882(0.112)883(0.113)884Lot size (sq. ft.)8850.0568860.0798870.0768880.0568890.0738900.0718910.0648920.0648930.062894(0.075)895(0.060)896(0.059)897(0.077)898(0.058)899(0.058)900(0.070)901(0.055)902(0.055)903Years since last transaction904−0.379***905−0.355***906−0.360***907−0.374***908−0.358***909−0.363***910−0.382***911−0.363***912−0.367***913(0.057)914(0.053)915(0.053)916(0.058)917(0.053)918(0.053)919(0.057)920(0.053)921(0.052)922Years since built923−0.196924−0.106925−0.100926−0.210*927−0.113928−0.107929−0.212*930−0.102931−0.097932(0.124)933(0.080)934(0.081)935(0.127)936(0.079)937(0.080)938(0.125)939(0.078)940(0.079)941Fireplace (0−1)9420.1769430.187*9440.191*9450.1839460.1779470.184*9480.1429490.1659500.171951(0.116)952(0.108)953(0.108)954(0.117)955(0.108)956(0.108)957(0.119)958(0.108)959(0.108)960Waterfront (0−1)9610.333**9620.463***9630.426***9640.326**9650.480***9660.441***9670.371***9680.470***9690.434***970(0.131)971(0.101)972(0.099)973(0.136)974(0.100)975(0.098)976(0.133)977(0.096)978(0.094)979Waterfront nourishment (1−0)980−0.208981−0.446*982−0.216983−0.458*984−0.200985−0.453*986(0.263)987(0.251)988(0.264)989(0.252)990(0.264)991(0.251)992Beach nourishment (1−0)9930.1329940.2099950.1279960.2209970.1289980.196999(0.207)1000(0.166)1001(0.208)1002(0.163)1003(0.203)1004(0.163)1005Bedrooms10060.03310070.02810080.0541009(0.134)1010(0.136)1011(0.132)1012Geotime10130.02110140.0101015−0.0071016(0.080)1017(0.093)1018(0.076)101910201021===== PAGE 18 (score 61, numeric-paren 47, table-words 2, var-words 1) =====10221023721024 1025Environmental Economics and Policy Studies (2021) 23:55–7710261 31027Table 4 (continued)1028Spatial lag model1029Spatial error model1030Spatial ARAR(1,1) model1031(1)1032(2)1033(3)1034(1)1035(2)1036(3)1037(1)1038(2)1039(3)1040Geotime squared1041−0.0021042−0.0011043−0.0001044(0.006)1045(0.007)1046(0.006)1047Elevation (ft.)10480.02710490.02010500.036*1051(0.021)1052(0.020)1053(0.020)1054Distance to water (ft.)1055−0.0331056−0.0301057−0.0091058(0.049)1059(0.059)1060(0.049)1061Beach width10620.01010630.01310640.0051065(0.022)1066(0.022)1067(0.023)1068Built post-FIRM1069−0.4261070−0.4491071−0.512*1072(0.290)1073(0.312)1074(0.302)1075Constant10768.372***107710.034***10789.640***107910.398***108010.114***10819.959***10826.603***10838.362***10848.102***1085(2.221)1086(1.904)1087(1.895)1088(1.043)1089(0.803)1090(0.809)1091(2.470)1092(2.122)1093(2.077)1094Lambda10950.15610960.00310970.02210980.29410990.14611000.1551101(0.152)1102(0.136)1103(0.135)1104(0.181)1105(0.166)1106(0.162)1107Rho11080.0561109−0.1341110−0.1071111−0.3411112−0.3121113−0.2941114(0.250)1115(0.204)1116(0.201)1117(0.339)1118(0.287)1119(0.280)1120Observations1121117112215111231511124117112515111261511127117112815111291511130Degrees of Freedom (model)1131171132101133811341711351011368113717113810113981140Log-likelihood1141−71.1031142−106.3901143−107.9661144−71.5861145−106.1671146−107.8331147−70.6691148−105.8461149−107.4581150Chi-sq.1151125.7541152138.5321153132.6541154134.3381155155.2621156146.4711157123.6931158128.2711159123.7481160Robust standard errors are in parentheses. *, **, ***Denote statistical significance at 10, 5 and 1% level116111621163===== PAGE 19 (score 26, numeric-paren 0, table-words 2, var-words 8) =====116411657311661 31167Environmental Economics and Policy Studies (2021) 23:55–77 1168is larger at $22,365. Overall, the said increase in the share of protected properties 1169translates to more than $718 million reduction in the value of affected houses in 1170the county. In terms of property tax revenues, a very crude back-of-the-envelope 1171Table 5 Potential effect of increasing the share of waterfront households protected by seawalls1172I. San Diego County has 70 miles of coastline. Based on a very conservative estimate of housing density 1173(4 homes per acre), this means that there are approximately 33,939 homes within 1,000 feet of the shore-1174line. 42% of these homes are currently protected by seawalls. Property tax in San Diego County is a little 1175more than 1%.1176II. Santa Cruz County has 29 miles of coastline. Based on an even more conservative estimate of housing 1177density (2 homes per acre), this means that there are approximately 7,030 homes within 1,000 feet of the 1178shoreline. 40% of these homes are currently protected by seawalls. Property tax in Santa Cruz County is 1179a little more than 1%.1180III. The change in price for seawalled properties is calculated using the estimated coefficients of the Sea-1181wall and Percent Seawall in our baseline model (i.e., OLS including all variables). The unconditional 1182mean property price for each county is used as the base price ($588,000 for San Diego and $699,000 for 1183Santa Cruz.). For non-seawalled properties, we just used the coefficient for Percent Seawall1184% Change in 1185seawalls (from 1186current mean)1187Change in price1188No. of affected houses1189Total effect1190Revenue 1191loss1192(in $)1193(in $)1194Seawalled1195Non-seawalled1196(in ’000$)1197(in ’000$)1198San Diego11991.00120010.471201−223.65120214,397120319,5421204−4,219.921205−42.2012065.0012077.901208−1,118.26120914,967121018,9721211−21,097.371212−210.97121310.001214−95.321215−2,236.53121615,680121718,2591218−42,331.751219−423.32122020.001221−635.101222−4,473.06122317,105122416,8341225−86,161.881226−861.62122730.001228−1,619.351229−6,709.58123018,531123115,4081232−133,391.071233−1,333.91123440.001235−3,048.071236−8,946.11123719,956123813,9831239−185,919.981240−1,859.20124150.001242−4,921.251243−11,182.64124421,382124512,5571246−245,649.311247−2,456.49124860.001249−7,238.901250−13,419.17125122,807125211,1321253−314,479.741254−3,144.80125570.001256−10,001.021257−15,655.70125824,23212599,7071260−394,311.941261−3,943.12126280.001263−13,207.601264−17,892.22126525,65812668,2811267−487,046.531268−4,870.47126990.001270−16,858.641271−20,128.75127227,08312736,8561274−594,584.261275−5,945.841276100.001277−20,954.161278−22,365.28127928,50912805,4301281−718,825.861282−7,188.261283Santa Cruz12841.001285−0.741286−775.9812872,840.1212884,189.881289−32,533.711290−325.3412915.001292−158.541293−3,879.8912942,952.6012954,077.401296−162,879.831297−1,628.80129810.001299−704.131300−7,759.7913013,093.2013023,936.801303−327,267.401304−3,272.67130520.001306−2,956.451307−15,519.5813083,374.4013093,655.601310−667,096.041311−6,670.96131230.001313−6,756.961314−23,279.3713153,655.6013163,374.401317−1,032,546.401318−10,325.46131940.001320−12,105.671321−31,039.1513223,936.8013233,093.201324−1,436,679.041325−14,366.79132650.001327−19,002.571328−38,798.9413294,218.0013302,812.001331−1,892,554.721332−18,925.55133360.001334−27,447.661335−46,558.7313364,499.2013372,530.801338−2,413,233.601339−24,132.34134070.001341−37,440.951342−54,318.5213434,780.4013442,249.601345−3,011,776.641346−30,117.77134780.001348−48,982.431349−62,078.3013505,061.6013511,968.401352−3,701,244.161353−37,012.44135490.001355−62,072.111356−69,838.0913575,342.8013581,687.201359−4,494,696.641360−44,946.971361100.001362−76,709.971363−77,597.8813645,624.0013651,406.001366−5,405,194.881367−54,051.95136813691370===== PAGE 20 (score 7, numeric-paren 0, table-words 0, var-words 7) =====13711372741373 1374Environmental Economics and Policy Studies (2021) 23:55–7713751 31376calculation suggests that this increase would cost the government $7 million in 1377terms of revenue loss and the discounted present value of this annual loss is $143 1378million.191379For Santa Cruz County, the potential effect of increasing the current share of 1380seawalled properties by 100% is significantly greater, reaching to about a $77,000 1381decline in home resale values for protected properties. For non-seawalled proper-1382ties, the effect is about similar at $78,000. This implies that there is no positive net 1383benefit of putting up an additional seawall from the current level. The increase in the 1384seawalled-properties is associated with the loss in the total value of coastal homes in 1385the county by more than $5.4 billion. This translates to a government revenue loss of 1386about $54 million, which is equivalent to a discounted present value of about $1.08 1387billion.1388Although our estimates are large, the associated welfare impacts are consist-1389ent with the existing literature. Our findings are in line with Kriesel and Friedman 1390(2003) and Kriesel et al. (2005) who find negative, albeit smaller, effects on prop-1391erty prices. Landry et al. (2003) also found a negative effect, but their estimated 1392effect was statistically insignificant. However, their contingent valuation study 1393strongly suggests that beachgoers do not like beaches with seawalls. The existing 1394literature suggests that our estimates might even be underestimated. Gopalakrishnan 1395et al. (2011) correct for endogeneity bias and find that the beach width effect on 1396property prices is almost five times larger. Furthermore, they find that the long- term 1397net value of coastal residential property can fall by as much as 52% when the erosion 1398rate triples. To the extent that seawalls can potentially double or triple erosion rates 1399on neighboring non-protected properties, our estimates are very much in line with 1400their results.1401Our paper does not come without limitations. Aside from potential measurement 1402problems associated with the survey-based price variable and the inability to control 1403for unobserved systematic patterns across beach communities, our study may also 1404be confounded by the potential immunity of the property market to climate risks, 1405which may be coupled with households’ potential downward bias against these risks. 1406To be more precise, if property buyers underestimate the risk of floods/erosion, then 1407the benefits of seawalls will be vastly undervalued while the negative consequences 1408would remain. Moreover, we find that variables indicating flooding and erosion risks 1409are not significantly correlated with property prices, which could indicate that home 1410buyers are myopic. Another possibility is that home buyers assess the risk accurately 1411but that it is quite low or even non-existent. Both possibilities could be because 1412there were no flood episodes in the study areas in the years prior to the survey. Most 1413of the prior flood episodes were in Northern California. There has been no major 1414impacts in Central California until 2017 when Hurricane Marie caused flooding in 1415the region.1416Even in the absence of flooding and erosion, more recent information of pro-1417gressing climate change, climate change reporting, and extreme events might have 141819 Property tax in San Diego County is a little more than 1%. Using the annual discount rate of 5%, an 1419infinite stream of annual losses of $7.188 million would be $7.188 m/0.05 = $143.765 m.1420