Hedonic pricing approach to estimate flood damage in. Tokyo Metropolitan Area

Transcription

1 Hedonc prcng approach to estmate flood damage n Tokyo Metropoltan Area Abstract There exsts a great deal of lterature on economc losses from flood dsasters. However, examnaton of the lterature rases concerns wth regard to napproprate selecton of flood rsk varables and omtted varable bas n estmatng flood damage. In ths paper, we estmate a Hedonc land prce model by employng two-step procedures to correct the omtted varable bas on flood hazard. The flood rsk s estmated to lower the land prce by 10.24% and the flood damage s estmated to be 1,200,000 yen/m2. Ths estmate s farly large compared wth the estmate by Tokyo Metropoltan Government, whch ndcates that the ndrect damage cost s lkely to be much hgher than the drect damage cost. Keywords: Flood hazard, Perceved externalty, Hedonc approach, Self-selecton bas JEL: Q51, Q54 1

2 1. Introducton Recent studes report that future clmate change s lkely to brng ncreased frequency of natural dsaster ncludng floodng events (Parry et al. (2007)). Thus, much attenton has been pad to adaptaton polcy aganst floodng such as the constructon of levee and dam and the extenson of the capacty of sewage system to reduce flood damage. In order to mplement cost effectve polcy, the cost and beneft of the polcy measures should be examned. Prevous studes analyze Hedonc model to estmate perceved cost of flood damage usng a varable representng flood hazard, such as nformaton on Specal Flood Hazard Areas provded by the Federal Emergency Management Agency (Hallstrom and Smth (2005)), mappng data of floodplans that would flood n a 100-year flood event (Bn and Polasky (2004)) and so on, and fnd that flood hazard reduce land prce. However, concerns about omtted varable bas and/or seemng correlaton on a flood hazard varable exst n prevous studes. For example, flood hazard s affected by not only clmate n the area n queston but also development of nfrastructure such as constructon of levee and dam, network of sewage system, geographc characterstcs, whch are also lkely to nfluence the land prce. If some of these factors are omtted n the analyss, the estmaton on the varable representng flood hazard s based. The purpose of ths study s to estmate Hedonc land prce model correctng the bas due to omtted varable and to explore how flood hazard s valued. Our man fndng s that the estmaton wthout correcton of the bas underestmates the value of flood hazard. Secton 2 descrbes the model and the data and Secton 3 reports estmaton results. Fnely, concludng remarks are summarzed n Secton Estmaton model and data 2.1 Estmaton model (Hedonc Land Prce Model) Followng the standard hedonc prcng approach, we formulate the Hedonc Land Prce model as follows: ln LP = α + X β + γdrsk + u (1) where LP, X,, DRsk, and u denote land prce, a vector of the attrbutes such as the elevaton, the dstance to the nearest staton, value of floor-to-area rato (FAR) regulaton, etc., flood rsk dummy and error term of the ste, respectvely. The varables used n ths 2

3 model are defned n Table 1. [Insert Table 1] DsStaton and Tme are used as the proxes of the transportaton cost. Thus, the more DsStaton and Tme are expected to lower the land prce. FAR, BLR and Fre capture strength of the regulatory varables. The hgher values, weaker regulaton, are expected to rase the land prce due to hgher land productvty. Elevaton s expected to capture the effect of the amenty on the land prce. The hgher elevaton and the shorter dstance are expected to ncrease the resdent s utlty because of good vew and better amenty. The zonng dummes are expected to capture how the land-use regulatons affect the land productvty. Tokyo 23 specal ward dummes (W) are expected to capture regon specfc effect such as the dfference n the polcy and publc servces among 23 local authortes. Ralway lne dummes (R) are expected to capture the characterstc of the area such as the accessblty to the popular shoppng area; the ralway lne whch accesses to popular shoppng area are convenent s lkely to rase the land prce of the area where the lne s used. The flood rsk dummy, DRsk, s constructed based on the flood hazard map made by the local governments as s explaned n the data secton. The dummy s equal to one, f the ste n queston s located n the area desgnated as one exposed to the flood rsk, otherwse zero. If the ste s located n the hazardous area, ts land prce s expected to decrease, whch means the negatve parameter of the flood rsk dummy. We should note that flood rsk dummy varable cannot be treated as exogenous, because some unobservable attrbutes of a ste are lkely to affect not only ts land prce as well as the flood hazard at the ste n queston. For example, whle the constructon of the levee along the nearest rver s lkely to reduce the flood hazard, t s also lkely to lower the land prce because of bad landscape. If ths s the case, the perceved damage of flood could be estmated wth bas, whch s a type of omtted varable bas. In order to correct for a self-selecton bas due to the omtted varables n estmaton of equaton (1), we use the two-step procedure suggested by Lee and Trost (1978) and Barnow et al. (1981). Rsk = δ + Y θ + (2) e ln LP = α + X β + γprsk + u, (3) where PRsk = Prob(DRsk = 1 Y ) = 1 Φ( δ θy ) (4) 3

4 In the frst step, we construct the model to determne the flood hazard, whch s expressed n equaton (2), usng the probt model and calculate the predcted probablty, PRsk, of the ste to be ncluded n the flood hazard area, whch s expressed n equaton (4). In the second step, after replacng DRsk n equaton (1) wth PRsk, we estmate equaton (3) by lnear least squares. (Flood Hazard Model) We apply the probt model to determnant of the flood hazard. Whether the ste s ncluded n the hazardous area depends on the degree of the flood rsk of the ste, Rsk, whch s specfed as follows: ln LP = α + X β + γprsk + u (3) where e s normally dstrbuted wth mean 0 and standard devaton of 1, and we denote Y as a vector of the attrbutes of the ste such as the elevaton of the ste, the straght-lne dstance between the ste and the nearest rver and 23 specal ward (admnstratve dstrct n Tokyo) dummes. The varables used n ths model are defned n Table 1. In ths model, we expect that DsRver, Snk and Tokyo 23 specal ward dummes (W) are determnants of the flood hazard. The flood hazard s lkely to be ncreased, as the ste s closer to the rver. Thus, the coeffcent of DsRver s expected to be negatve. Tokyo 23 specal ward dummes are expected to capture the regonal specfc effect on the flood hazard. Snk s defned as the mnmum value of the elevatons of the representatve ponts n grds surroundng the ste mnus Elevaton. As s shown n Fgure 1, the grd where ste locates are surrounded by eght grds (Grd 1 to Grd 8). SSSS = mn EEEEEEEEE n, EEEEEEEEE, where n = 1, 2, 8 If Snk s postve, whch means mnmum value of the elevatons of the representatve ponts n surroundng grds s larger than the elevaton of ste, ste locates n the bottom of the snk. As the results, the floodng water s lkely to be backed up at ste. On the other hand, f Snk s negatve, ste does not locate at the bottom of the snk and the floodng water s not lkely to be backed up. Therefore, the parameter on Snk s expected to be postve. It should be noted that Snk s expected to affect the flood hazard but not the land prce, though the elevaton s lkely to affect the land prce. Thus, Snk s expected to be an nstrumental varable. 1 Wthout loss of generalty, we normalze =1. 4

5 In practce, Rsk s unobservable but whether ste s n the hazardous area s observable. Thus the followng relaton s consdered to be held: DRsk=1 f Rsk>0 DRsk=0 otherwse Gven ths specfcaton, the probablty that the ste s ncluded n the flood hazard area s gven by: PRsk = Prob(DRsk = 1 Y ) = 1 Φ( δ θy ) (4) where Φ denotes the cumulatve normal functon. It should be noted that we consder the mnmum value among elevatons of the surroundng stes mnus the elevaton of the ste as an nstrumental varable n equaton (3). Because the elevaton of the ste s lke to affect the land prce, snce the ste wth the hgher elevaton s lkely to have a good vew. However, the dfference between the maxmum value among elevatons of the surroundng stes and the elevaton of the ste n queston rather than the elevaton of the ste n queston s lkely to affect the flood hazard. For example, f the maxmum value mnus the elevaton of the ste n queston s negatve, the ste s lkely to face less flood hazard. On the other hand, t s lkely to face more flood hazard, f t s postve. 2.2 Data We construct the flood rsk dummy based on the flood hazard map publshed by the Tokyo Metropoltan Government n (Fg 2) The rsk nformaton on the hazard maps s derved from the result of flood smulaton on the assumpton of the equvalent heavy ran under the Toka Flood Dsaster 2 n September, Durng September 11-12, 2000, a heavy storm by the typhoon attacked the Toka area of Japan, the metropoltan area of Nagoya, and resulted n one of the most severe flood dsasters n Japan. Ths has been wdely studed n varous academc felds as a model megalopoltan flood. (Zha & Sato(2002))- The maxmum amount of ranfall per hour and the maxmum daly ranfall n Nagoya were 97 mm and 428 mm respectvely. The maxmum daly ranfall was two tmes as average monthly ranfall. The Shnkawa and the Shona Rvers burst ther banks and submerged many offce buldngs, subway statons and neghborng houses. (Takao, et al (2002)) Thus, total economc loss was 270 bllon yen (approxmately 32 bllon US dollar) caused by overflow, blackout, transportaton and dstrbuton stops. 5

6 We used offcal land prce 3 reported by Mnstry of Land, Infrastructure, Transport and Toursm (MLIT, hereafter) n The data set of the offcal land prce provded by MLIT ncludes not only land prce but also the attrbutes of the ste address, land shape, land-use zonng, accessblty to the gas, water and sewerage utltes, the name of the nearest staton, the dstance from t, lot percentage regulaton, floor-to area rato regulaton and so on. All the attrbutes we used n our regresson are lsted n Table 1. The objectve area are 19 specal wards of 23, excludng Sumda, Arakawa, Katsushka and Shbuya whose hazard maps are not avalable on the nternet web ste. We plot the locaton of 1516 stes on Fg 3. We collected the nformaton on accessblty to the nearest central busness dstrcts (CBDs), elevaton and dstance to the nearest rver. The CBDs of Tokyo metropoltan area are often defned as some gateway statons of JR Yamanote loop lne 4. The accessblty to CBD s measured as the tme dstance from the nearest staton of the subway, the prvate or the other JR lne to Shnjuku, Ikebukuro, Tokyo, Shnagawa or other termnal statons whch connect to the Yamanote loop lne. We calculated the tme dstance n September 2009 usng the Yahoo! route search ( 5. Table 2 shows the summary statstcs. The mean and maxmum values of Snk ndex are -1.34m and 0.81m, respectvely, whch means that most of stes are not the bottom of snk of surroundng 8 grds. [Insert Table 2] 3. Estmaton results (Flood Hazard Model) Man estmaton results are reported n Table 3. 3 The offcal land prce s land prce apprased by Land Apprasal Commttee under MLIT. Snce t s dffcult to obtan transacton land prce data n Japan, the offcal land prce s wdely used for the analyss of hedonc land prce functon n Japan (for example, Nakagawa et al. (2009)). It s well known that ths prce s correlated wth the transacton land prce. Detaled nformaton on the offcal land prce s avalable at 4 The Yamanote Lne s Tokyo's most mportant tran lne whch connects major gateway statons to CBD, such as Shnjuku, Tokyo, Shbuya and so on. A trp around the whole crcle takes approxmately one hour. (japan-gude.com( 5 When names of a departure staton and an arrval staton are entered, the Yahoo! Route search shows route, fare, travel tme by tran between these statons and so on. 6

7 [Insert Table 3] The parameter on DsRver s sgnfcant wth a negatve sgn, whch means that the hgher flood hazard rsk s, the closer the ste s from a rver. The parameter on Snk s also sgnfcant wth a postve sgn, whch means that the ste locates further from the bottom of the snk wth hgher Snk and face lower rsk of flood hazard. (Hedonc Land Prce Model) In ths study, we estmate Hedonc Land Prce Model not only by two stage procedure, equaton (2) and (3), but also by least square method, equaton (1) for comparson. Man estmaton results are reported n Table 4. The model 1 s the basc model. We added cross terms between the flood rsk and strength of the regulatory varables n the model 2 and 3 as the flood damage mght be larger f the total floor space sze s larger. As the result of a Sargan test, the equaton s completely dentfed. The Hausman test for endogenety rejects the null hypothess that the dummy varable of flood hazard s an exogenous varable. [Insert Table 4] The parameter on flood hazard s sgnfcant wth a postve sgn n the case of two stage procedure, whch means that hgher rsk of flood hazard reduces land prce. On the other hand, the parameter on flood hazard s nsgnfcant n the case of OLS. The parameter on flood hazard s negatve, but s nsgnfcant n the model 2. However, the parameter on cross term between FAR and flood rsk s sgnfcant wth negatve sgn. Ths means that the flood damage s larger where FAR s larger. Usng the mean value of land prce, we derved the flood damage effect on land prce. Table 5 shows the reducton rate of land prce wth flood hazard area s % n the model 2. Both parameters on flood rsk and cross term between flood rsk and heght are sgnfcant wth negatve sgns. Ths means that people who lve or work on upper floor also suffer from flood. As n the case of model 2, we calculated the average reducton of land prce by flood rsk s % (Table 5). The other varables are sgnfcant wth expected sgn. Elevaton, Lot Sze and FAR have a postve sgn. Ths ndcates that the hgher the elevaton of a ste s, the larger the lot sze s or the weaker the regulaton s, the hgher the land prce s. Ths s because the ste wth hgher elevaton has a better vew and the land productvty of the ste wth larger lot sze and weaker regulaton s hgher due to more effcent land use. On the other 7

8 hand, DsStaton and Tme have a negatve sgn. Ths ndcate that the closer the ste s from the nearest staton or the closer the nearest staton s from Yamanote lne, the hgher the land prce s. Ths s because of hgher transportaton costs. Usng the estmated parameter on Flood Hazard and the equaton (1), we estmate how the land market values zero rsk of flood hazard as follows. land prce of the ste wth zero rsk V = land prce of the ste wth flood hazard = (e γ 1) 100 = (e ) 100 = (%) From the above calculaton, the flood rsk s estmated to lower the land prce by 10.24% whch ndcates that the average reducton n land prce s 143,946 yen/m 2 (approxmately 1,137 Euro), usng means of the observatons locatng n flood hazard area. Fnally, from the above dscusson, we would lke to estmate the perceved flood damage cost. The present value of the expected flood damage should be equal to reducton n the land prce, f the land owner s rsk neutral. Therefore, the followng equaton holds by gnorng the present value of the flood damage after hundred years. 100 (Reducton n Land Prce) = d(1 d)99 D (1 + ρ) t t=0 where d, D and ρ denote probablty of occurrence of the Toka Flood Dsaster n each year, perceved flood damage and dscount rate. It should be noted that d s equal to , snce the flood rsk nformaton on the hazard maps are calculated under the assumpton of the Toka Flood Dsaster whch s expected to occur once n every 100 years. D = (Reducton n Land Prce) = 1,196,209 From the above equaton and estmated average land prce reducton, the average flood damage s estmated to be 1,196,209 yen/m 2 (approxmately 9,446 Euro/m 2 ), when we use 3% for the assumpton of the dscount rate, whch s often used the calculaton of present value n the context of long term envronmental ssues such as global warmng. Tokyo Metropoltan government estmated that the flood damage could be 50,000 yen/m 2 n 2007 based on the smulaton of the physcal damage. However, although the 6 d meets the followng equaton; 0.01=100C1*d*(1-d) 99. 8

9 complete recovery tme could be long n the case of the large scale dsaster, t dd not nclude the ndrect damage such as loss of lfe, loss of the proft whch the companes would earn, f the flood dd not occur, long term mental damage, and so on. Thus, the estmates employed by the government could be farly low. We note that we may underestmate the land prce reducton caused by the flood damage f the people expect the government to mplement any balout polcy when severe dsaster. 4. Concluson Future clmate change s lkely to brng ncreased frequency of natural dsaster ncludng floodng events. Thus, much attenton has been pad to adaptaton polcy aganst floodng. From the perspectve of the cost effectve polcy, estmaton of the beneft of reducton n the flood damage s an mportant research topc. However, prevous studes suffer from the concern on the omtted varable bas n estmatng flood damage. We estmate Hedonc land prce model by employng two step procedures to correct ths bas on flood hazard and we fnd that the prevous studes are lkely to underestmate the perceved flood damage. The flood rsk s estmated to lower the land prce by 10.24% whch reduces 143,946 yen/m 2 (approxmately 1,137 Euro) on average and the perceved flood damage s estmated to be 1,196,209 yen/m 2 (approxmately 9,446 Euro/m 2 ). Ths estmate s farly larger than the estmate by Tokyo metropoltan government, whch ndcates that the ndrect damage cost s lkely to be much hgher than the drect damage cost (cost estmaton based on the physcal damage). 9

10 Fg 1 Snk ndex of ste Grd 1 Grd 2 Grd 3 Elevaton 1, Grd 4 Ste Elevaton 3, Grd 5 Observaton pont 50m Grd6 Grd 7 Grd 8 50m 10

11 Fg 2 The Flood Hazard Map publshed by the Tokyo Metropoltan Government Tokyo 23 specal wards Shnjuku staton area Tsukj area 11

12 Fg stes used for estmatons 12

13 Table 1 Defnton of Varables and data source Flood Land Prce Varables Defnton Data Source Hazard Hedonc Model Model Eq. (2) Eq. (3) Land General LP Land prce n January 1 st, 2009 of Informaton System ste (Yen /m 2 ) ( No Yes Flood hazard dummy: Equal to DRsk one, f the ste s located n the area desgnated as one exposed Flood Hazard map Yes No to the flood rsk, otherwse zero. Lot sze Lot sze of ste (m 2 ) Land General Informaton System No Yes DsStaton Dstance from ste to the Land General nearest ralway staton (m) Informaton System No Yes Tme dstance from the nearest Tme staton of ste to the Yahoo! Route search No Yes Yamanote lne (mnute) Fre Fre protecton dstrct dummy: Equal to one, f the ste s Land General located n the fre protecton Informaton System No Yes dstrct, otherwse zero. Floor-to-area rato regulaton FAR appled to ste (%): Upper lmt Land General of total floor space area per unt Informaton System No Yes lot sze BLR Buldng-to-land rato regulaton Land General appled to ste (%) Informaton System No Yes Elevaton Elevaton of ste (m) Geographcal nformaton database No Yes DsRver Dstance from ste to the Geographcal nearest rver (m) nformaton database Y Yes Snk Mnmum value of the elevaton Geographcal of grds surroundng the ste nformaton database Yes No 13

14 mnus Elevaton (m) Insde of JR Yamanote lne Central Tokyo dummy: Center of Tokyo metropoltan area. Equal to one, f the ste s located n the Yes No center area, otherwse zero. Tokyo 23 specal ward dummes: Equal to one, f the ste n queston s located n W ward, W otherwse zero. (W=Chuo, Mnato, Shnjuku, Bunkyo, Tato, Koto, Shnagawa, Meguro, Ota, Land General Informaton System No Yes Setagaya, Nakano, Sugnam, Toshma, Kta, Itabash, Nerma, Adach, Edogawa) Ralway lne dummes: Equal to R one, f the nearest ralway lne from the ste n queston s R ward, otherwse zero. (R=Chuo, Land General Informaton System No Yes Kehn Tohoku,..) Zonng dummes: Equal to one, f the ste locates n Z dstrct, otherwse zero. (Z=Resdental zone, Exclusve Md-hgh Resdental Z zone, Exclusvely Low-story Resdental zone, commercal zone, Land General Informaton System No Yes Industral zone, Lght Industral zone, Exclusvely Industral zone, Neghborhood Commercal zone, and Quas-Resdental zone) 14

15 Table 2 Man Descrptve statstcs Mean Medan Max Mn Std. dev Land prce (yen) Elevaton (m) DsRver (m) Snk ndex (m) Lot sze (m 2 ) DsStaton (m) Tme (mn) BLR (%) Heght (FAR/BLR) (%) FAR (%)

16 Table 3 Man estmaton results : Flood Hazard Model Coef Std error Constant DsRver (m) *** Snk ndex (m) ** Chuo dummy ** Mnato dummy ** Shnjuku dummy Bunkyo dummy ** Tato dummy *** Koto dummy *** Shnagawa dummy ** Meguro dummy Ota dummy Setagaya dummy Nakano dummy Sugnam dummy Toshma dummy ** Kta dummy Itabash dummy Nerma dummy Adach dummy *** Edogawa dummy *** Lklhood rato test Note: *, ** and *** ndcate sgnfcant at the 10% level, the 5% level and the 1% level, respectvely. 16

17 Table 4 Man estmaton results of Land Prce Hedonc Model Model 1 Model 2 Model 3 2-stage OLS 2-stage OLS 2-stage OLS Coef Std Error Coef Std Error Coef Std Error Coef Std Error Coef Std Error Coef Std Error Constant *** *** *** *** *** *** Flood rsk *** ** *** FAR (%) x Flood rsk *** ** BLR (%) x Flood rsk *** *** Heght x Flood rsk *** DsRver (m) *** ** * * Elevaton (m) *** *** *** *** *** *** Lot sze (m 2 ) *** *** *** *** *** *** DsStaton (m) *** *** *** *** *** *** Tme (mn) *** ** *** *** *** *** Centeral Tokyo dummy ** *** ** ** * ** FAR (%) *** *** *** *** *** *** Adjusted R-squared Test for endogenety Accepted Accepted Accepted Note: *, ** and *** ndcate sgnfcant at the 10% level, the 5% level and the 1% level, respectvely. The results of Ralway dummes, 23 specal ward dummes and Zonng dummes are excluded n the table. The results of the full verson are provded upon request. 17

18 Table 4 Land prce reducton and flood damage 2-stage OLS Land prce reducton Flood damage Flood damage Land prce reducton (yen/m 2 ) (yen/m 2 ) Model % -1,196, % -256,008 Model % -1,726, % -269,833 Model % -1,560, % -154,977 18

19 Reference: O. Bn and S. Polasky (2004) Effects of flood hazards on property values: evdence before and after hurrcane Floyd, Land Economcs, vol.80, pp Danel G. Hallstrom and V. Kerry Smth (2005) Market responses to hurrcanes, Journal of Envronmental Economcs and Management, vol.50, pp Masayuk Nakagawa, Makoto Sato and Hsak Yamaga (2009) Earthquake Rsks and Land Prces: Evdence from the Tokyo Metropoltan Area, The Japanese Economc Revew, Vol. 60, pp M.L. Parry, O.F. Canzan, J.P. Palutkof, P.J. van der Lnden and C.E. Hanson (eds) (2007) Contrbuton of Workng Group II to the Fourth Assessment Report of the Intergovernmental Panel on Clmate Change, Cambrdge Unversty Press, Cambrdge, Unted Kngdom and New York, NY, USA. Kenj Takao, Tadahro Motoyosh, Teruko Sato, Kam Seo, Saburo Ikeda and Teruk Fukuzono (2002) Influences of the Flood Dsaster Experence and the Threat of Flood Dsaster on Preparedness. The Toka Flood Dsaster Case, Report of the Natonal Research Insttute for Earth Scence and Dsaster Preventon, No.63, pp (n Japanese wth Englsh abstract) L. Lee and R. P. Trost (1978) Estmaton of Some Lmted Dependent Varable Models wth Applcaton to Housng Demand, Journal of Econometrcs, Vol. 8, pp Guofang Zha and Teruko Sato (2002) Structural Analyss of the Toka-Flood Dsaster of 2000 n Japan, Proceedngs of Second Annual IIASA-DPRI Meetng INTEGRATED DISASTER RISK MANAGEMENT: Megacty Vulnerablty and Reslence Towards Establshng a Framework for Megalopoltan Flood Rsk Analyss ( 19