1# RECORD 933204aac69df35d44733b61b90a1210852e7b682# TITLE: A new Mass Rapid Transit (MRT) line construction and housing wealth: Evidence from the Circle Line3# YEAR: 2017.0 VENUE: nan DOI: 10.24294/jipd.v1i1.224# AUTHORS: M. Diao; Yi Fan; Tien Foo Sing5# PAGES: 26 CANDIDATE TABLE PAGES (1-based): [9, 12, 17, 19, 21, 22]6# ABSTRACT: This study uses the opening of a new Mass Rapid Transit (MRT) line, known as the Circle Line (CL), between 2010 and 2012 in Singapore as the exogenous event to empirically test the impact of the new CL on housing wealth of households. Using a differences-in-differences (DID) approach, and with the non-landed private housing transaction data covering the period from 2009 to 2013, we find that the average housing prices in our sample increase by 1.6% in the post-opening of the CL. We also find significant capitalization of the new CL into housing prices, especially households living within a 400 meter radius (the treatment zone) from the closest MRT stations on the CL. The treatment effects that are measured by the marginal willingness to pay for houses located within the treatment zone is 13.2% relative to houses located outside the treatment zone. The new CL opening creates an estimated of S$1.23 billion housing effects for households living in the close proximity to the CL MRT stations. However, we do not find significant “anticipative” effects on house prices in the 6-month window prior to the opening of CL. The strongest treatment effect is found after the opening of the phase 1 of CL, and the intensity of treatment effects declines in phases 2 and 3 of the CL opening. The results show strong evidence of the causality from the accessibility to MRT stations on housing prices.789===== PAGE 1 (score 9, numeric-paren 0, table-words 0, var-words 9) =====10116412Journal of Infrastructure, Policy and Development (2017) Volume 1 Issue 1, pp.64-89.13DOI: 10.24294/jipd.v1i1.2214Original Article15A new Mass Rapid Transit (MRT) line construction and 16housing wealth: Evidence from the Circle Line17Mi Diao181, Yi Fan191 and Tien Foo Sing202*211 Department of Real Estate, National University of Singapore222 Department of Real Estate/Institute of Real Estate Studies (IRES), National University of Singapore23ABSTRACT24This study uses the opening of the new Mass Rapid Transit (MRT) in stages between 2010 and 2012 25in Singapore as the exogenous event to empirically test the impact of the new Circle Line (CL) on 26transaction data covering the period from 2009 to 2013, we find that the average housing prices 27into housing prices, especially households living within a 400-meter radius (the treatment zone) 28willingness to pay” for houses located within the treatment zone is 13.2% relative to houses located 29outside the treatment zone. The new CL opening creates an estimated S$1.23 billion housing wealth 30intensity declines in phases 2 and 3 of the CL opening. 31Keywords32marginal willingness-to-pay; housing wealth331.34Introduction35Investments in public infrastructure projects, such as new MRT lines, 36highways etc., are highly risky and have a long-gestation period. It is 37difficult to attract sufficient private capitals into public infrastructure 38projects, based solely on economic considerations. Governments 39invariably take on the key roles and responsibilities in developing 40infrastructure projects. Public infrastructure investments constitute a 41large portion of government’s fiscal expenditures in many countries. 42and policymakers exercise due restraints to avoid over-committing on 43public infrastructure project spending. 44Public infrastructure investments boost economic growth and create 45new employment in productive sectors, and they also generate positive 46spillovers that improve social welfare and environmental quality. At the 47urban level, new MRT line investments generate significant economic 48activities, which increase prices of lands surrounding MRT stations. 49New MRT lines radiating out of the central business district (CBD) 50ARTICLE INFO51Received: January 17, 2017 52Accepted: March 8, 2017 53Available online: March 14, 201754*CORRESPONDING AUTHOR55Tien Foo Sing, Department of 56Real Estate/Institute of Real Estate 57Studies (IRES), National University 58of Singapore, 4 Architecture Drive, 59Singapore 117566; 60rststf@nus.edu.sg61CITATION62 63 64 65 Diao M, Fan Y and Sing TF (2017). “A 66New Mass Rapid Transit (MRT) Line 67Construction and Housing Wealth: 68Evidence from the Circle Line”. 69Journal of Infrastructure, Policy 70and Development, 1(1): 64-89.71http://doi.org/10.24294/jipd.v1i1.2272COPYRIGHT73Copyright © 2017 by author(s) and 74EnPress Publisher LLC. This work is 75licensed under the Creative Commons 76Attribution-NonCommercial 4.0 77International License (CC BY-NC 4.0). 78http://creativecommons.org/licenses/79by/4.0/80, Singapore81, Singapore828384===== PAGE 9 (score 84, numeric-paren 60, table-words 3, var-words 5) =====8586A new Mass Rapid Transit (MRT) line construction and housing wealth: Evidence from the Circle Line877288Table 1. Descriptive statistics89Variable90Symbol91distmrt<=400092after==093after ==194dumMRT ==095dumMRT==196Observation9781,316.009833,739.009947,577.0010074,752.001016,564.00102Sale price (in SGD)103price104$1,546,549.00105$1,629,331.00106$1,487,845.00107$1,557,934.00108$1,416,896.00109($1,259,147.00)110($1,278,837.00)111($1,241,668.00)112($1,289,634.00)113($826,063.00)114ln-sale price115lnprice11614.06911714.12011814.03411914.07112014.052121(0.557)122(0.568)123(0.547)124(0.566)125(0.446)126Floor level127floor1289.51412910.6891308.6801319.5431329.182133(8.445)134(9.461)135(7.533)136(8.543)137(7.229)138Unit size (sqm)139area140110.715141117.371142105.995143111.298144104.078145(57.682)146(59.804)147(55.649)148(58.569)149(45.881)150Building age151age1524.3181534.2601544.3591554.3391564.079157(7.552)158(7.556)159(7.549)160(7.589)161(7.114)162Freehold tenure163dumfh1640.6161650.6611660.5841670.6401680.349169(0.486)170(0.473)171(0.493)172(0.480)173(0.477)174Condominium dummy175dumcond1760.5881770.6251780.5611790.5871800.599181(0.492)182(0.484)183(0.496)184(0.492)185(0.490)186Apartment dummy187dumapt1880.4121890.3751900.4391910.4131920.401193(0.492)194(0.484)195(0.496)196(0.492)197(0.490)198Resale dummy (secondary199market)200dumresale2010.4102020.4002030.4172040.4162050.338206(0.492)207(0.490)208(0.493)209(0.493)210(0.473)211Newsale dummy (sales by212developers)213dumnews2140.4722150.4462160.4912170.4672180.531219(0.499)220(0.497)221(0.500)222(0.499)223(0.499)224Sub-sale dummy (sale by225individuals)226dumsubs2270.1182280.1542290.0922300.1162310.131232(0.322)233(0.361)234(0.289)235(0.321)236(0.337)237HDB buyer dummy238hdbbuyer2390.3432400.3002410.3742420.3442430.335244(0.475)245(0.458)246(0.484)247(0.475)248(0.472)249Distance to the nearest MRT250station (in meter)251distmrt2521536.3572531509.9332541555.0962551646.854256278.000257(903.328)258(836.038)259(947.718)260(857.723)261(89.980)262Note: The table shows the means (first row for each variable) and standard deviations (in parentheses) for the dependent variable, which is the natural logarithm of housing prices (lnprice), and other control 263variables (the descriptions and symbols are given in the first and the second columns). The full sample includes 81,360 non-landed private housing sales in a 4000 meter boundary from the MRT stations on the 264CL. Based on the time of the transactions, the sample is further sorted into a sub-sample of sales before the opening of CL (after =0) (Column 4) and a sub-sample of sales after the opening of the CL (after=1) 265(Column5). Based on the cutoff distance of 400 meters from the closest MRT station, the sample is also sorted into the treatment group ( 400 meters) (the treatment group) (Column 6) and the control group 266(>400 meters) (Column 7).267268269===== PAGE 10 (score 9, numeric-paren 0, table-words 1, var-words 7) =====270271Mi Diao, Yi Fan and Tien Foo Sing27273273apartment units in buildings that are below six-storey in height to foreigners. The housing sample is 274made up of 58.8% of condominiums and 41.2% of apartments. More condominiums (62.5%) were 275sold in the pre-MRT opening period compared to 56.1% of condominiums sold in the post-MRT 276opening period. 277By sale type, we divide the sale samples into the primary market (new and uncompleted units) 278(“dumresale” =0) and the secondary market (“dumresale” =1), which consist mainly of completed 279units sold by individual owners. The primary market consists mainly of housing units sold by 280developers in new developments (“dumnews” 47.2%); some of the new units in projects under 281construction are sold by individual buyers to other buyers before temporary occupation permits 282(TOPs)2839 are granted, and these sales are classified as “subsales” (“dumsubs” 11.8%). The resales 284make up about 41.0% of the sample. The resale units take up 40.0% and 41.7% of the full samples 285in the “before” and “after” periods, respectively. 286By the linear distance (in meter) to the closest CL MRT stations, the average distance is 287estimated at 1,536.357 meters. For the “before” and “after” the CL opening subsamples, the average 288distances to the closest CL MRT stations are estimated at 1,509.933 meters and 1,555.096 meters, 289respectively. Figure 2 shows a scatter plot of log-sale price (“lnprice”) against distance to the closest 290MRT station (“distmrt”). The sale distributions appear random within the 4-km bounded area, 291though we observe a marginal downward trend in housing price over the MRT distance range. We 292conduct further tests in the next section to see if the price gradient is monotonically decreasing.2933.2. 294Cut-off boundary (discontinuity) to the closest MRT stations295In our empirical tests for MRT proximity effects on housing prices, the cut-off distance—296that shows discontinuity in housing price changes with respects to distance (accessibility) to the 297nearest MRT station—is important. As an identification strategy, we need to households’ “marginal 298willingness to pay” (MWTP) for being close to a MRT station; or in other words, we need to 299identify how far (maximum distance) a typical buyer is willing to “walk” to the nearest MRT station 300so as to take a MRT ride to his/her destination. Instead of having an arbitrarily selected distance,30110 302we use the pre-CL opening sales of non-landed housing that are located within 2-km boundary of 303the six MRT stations on the existing lines (operational stations) (“MRTold” =1) to test the MWTP 304for MRT accessibility effects. We create discrete distance dummies starting with the first dummy: 305“dist2 200 meters”, and an increment of 100-meter in length for subsequent dummies (i.e., “dist3 = 306201 to 300 meters”, “dist4 = 301 to 400 meters” ... to “dist15 = 1,401 to 1,500 meters”). 307We run OLS regressions with two sets of distance “treatment” dummies: the first set includes 308houses located within the first 1-km boundary from the closest MRT stations, and the second set 309extends the boundary area to 1.5 km. The models control for heterogeneity in housing and spatial 310attributes, and sale type. The results are summarized in Table 2. Models 1 and 2 do not have the 311year of sale fixed effects (“saleyear”); whereas Models 3 and 4 include the year of sale fixed effects. 312The estimated coefficients on the control variables are consistent across the models, and have the 3139. When a building is completed, a developer can apply to the Commissioner of Building Control for a Temporary Occupation Permit 314(TOP) before giving keys to flats in a project and allowing buyers / owners to move into the building.31510. Diao, Leonard and Sing (2016) propose a local-polynomial-regression-based approach to identify the housing price discontinuity 316boundary to the MRT stations in Singapore.317318319===== PAGE 12 (score 166, numeric-paren 141, table-words 3, var-words 4) =====320321Mi Diao, Yi Fan and Tien Foo Sing32275323Table 2. Boundary discontinuities in housing prices 324 325(1)326(2)327(3)328(4)329Distance<200m330dist23310.090*3320.090*3330.172***3340.173***335(0.052)336(0.049)337(0.051)338(0.048)339201 to 300m340dist33410.118***3420.125***3430.191***3440.193***345(0.024)346(0.023)347(0.023)348(0.022)349301 to 400m350dist4351-0.0293520.0093530.054**3540.085***355(0.022)356(0.021)357(0.022)358(0.021)359401 to 500m360dist5361-0.063**362-0.040*3630.0143640.034365(0.025)366(0.023)367(0.024)368(0.023)369501 to 600m370dist63710.069***3720.088***3730.140***3740.158***375(0.026)376(0.025)377(0.026)378(0.025)379601 to 700m380dist7381-0.035*382-0.053***3830.0303840.008385(0.019)386(0.018)387(0.019)388(0.018)389701 to 800m390dist8391-0.156***392-0.161***393-0.087***394-0.096***395(0.019)396(0.018)397(0.019)398(0.018)399801 to 900m400dist9401-0.026402-0.052***4030.062***4040.033**405(0.017)406(0.016)407(0.017)408(0.016)409901 to 1000m410dist104110.030*4120.087***4130.113***4140.162***415(0.017)416(0.016)417(0.017)418(0.016)4191001 to 1100m420dist114210.197***4220.166***423(0.013)424(0.012)4251101 to 1200m426dist124270.279***4280.229***429(0.018)430(0.018)4311201 to 1300m432dist134330.208***4340.236***435(0.022)436(0.021)4371301 to 1400m438dist144390.150***4400.181***441(0.018)442(0.017)4431401 to 1500m444dist154450.055**4460.097***447(0.023)448(0.022)449Floor level450floor4510.013***4520.013***4530.013***4540.013***455(0.001)456(0.001)457(0.001)458(0.001)459Unit size460area4610.006***4620.006***4630.006***4640.006***465(0.000)466(0.000)467(0.000)468(0.000)469Condominium 470dummy471dumcond4720.226***4730.219***4740.231***4750.235***476(0.009)477(0.008)478(0.009)479(0.008)480Freehold dummy481dumfh4820.251***4830.259***4840.223***4850.235***486(0.011)487(0.011)488(0.011)489(0.011)490Age491age492-0.011***493-0.012***494-0.010***495-0.011***496(0.001)497(0.001)498(0.001)499(0.001)500Resale dummy501dumresale502-0.069***503-0.091***504-0.072***505-0.093***506(0.011)507(0.011)508(0.011)509(0.011)510Constant511Constant51213.038***51312.969***51412.968***51512.905***516(0.016)517(0.015)518(0.016)519(0.015)520R-squared5210.6745220.7065230.6925240.722525Note: This table summarizes the OLS regression results with the ln-housing price (lnprice) as the dependent variables. The models 526are estimated using only 7,679 housing transactions located within 2,000 meters from the 6 existing MRT stations (Buona Vista, 527Paya Lebar (East-West line), Bishan, Dhoby Ghaut inter-change (North-South line), Serangoon and Harbour Front (North-East 528line) before the opening of the CL. The “dist2” and “dist_k”, where k=(3…15) are distance dummies with a 100-meter interval. 529Column (4) includes the planning sector fixed effects (FE), which is defined by the first 2 digits of the postal code. Standard errors in 530parentheses. *** p<0.01, ** p<0.05, * p<0.1531correct signs. The models show that housing attributes, such as floor level, unit size, freehold tenure, 532and condominium dummy all have positive effects on housing prices; whereas, the age variable has 533a negative effect on housing prices. Housing prices in the resale market are lower than prices for 534comparable houses in the developers’ sale market.535In Models 1 and 2, where housing sales in the 1-km to 2-km boundary are used as the control 536sample, we find that houses located in the first 200 meters and between 201 and 300 meters have 537positive price premiums relative to the control sample of houses located outside the 1-km boundary 538from the nearest CL MRT stations. The proximity premiums ranging between 9.0% and 12.5% are 539540541===== PAGE 13 (score 11, numeric-paren 0, table-words 3, var-words 5) =====542543A new Mass Rapid Transit (MRT) line construction and housing wealth: Evidence from the Circle Line54476545statistically and economically significant. However, the premiums disappear for houses in the 301-546m to 400-m region, and then become negative for houses located in the 401-m to 500-m region. 547The premium is positive again in the 501-m to 600-m region, and then becomes negative beyond 548the 601-m radius range. The results are consistent in the two models, and we observe a reasonable 549clear discontinuity in housing price at the 301-m radius, though we believe the 601-m radius could 550be caused by the trade-off against other negative externalities associated with being close to MRT 551stations, such as noise, traffic congestion, etc. (Bollinger and Ihlanfeldt, 1997; and Bowes and 552Ihlanfeldt 2001)553In Models 3 and 4, where we include 5 more discrete distance dummies ranging from 1,001 to 5541,500 meters into the treatment group, and use housing sales between 1,600 meters and 2,000 meters 555as the control sample. The two models are consistent and the coefficients do not vary significantly 556in Model 4, where the year of sale fixed effects are added. We observe a similar pattern of price 557increases for houses located near the MRT stations, and the positive premiums are economically 558and statistically significant and positive in the first 400 m range. The coefficient on “dist5” (401- 559to 500-meter range) is positive but insignificant, and the coefficient on “dist6” (501- to 600-meter 560range) is again positive and significant. Figure 3 plots the coefficients on the “dist_i”, where [i= (2, 5613, ….., 15]; and we could see the clear discontinuity in the price premiums for houses located up to 562400 meters from the closest MRT stations on the CL. The positive premiums for houses located in 563the range from 1,000-m to 1,500-m range relative to the control sample (1,501-m to 2,000-m range) 564could be due to factors other than being in close proximity to the CL MRT stations. 565Figure 3. Estimated coefficients of the distance to the closest CL MRT stations 566Note: The graph shows the estimated ln-housing price regression against different dummies on the distance to the 567closest MRT stations (“dist2” and “dist_k”, where k=(3, 4, …15), where the sample includes houses that are located 568within 2,000 meters from the nearest CL MRT stations. We control for other variables in the models as shown in Table 2 569(Column 4). 570Therefore, based on the above results, we set the treatment boundary at 400 meters to capture 571the proximity (accessibility) effects; we use a dummy variable, “dumMRT”, to sort houses into 572the treatment group if they are located within 400 meters from the nearest CL MRT station, and 573574575===== PAGE 17 (score 81, numeric-paren 60, table-words 5, var-words 5) =====576577A new Mass Rapid Transit (MRT) line construction and housing wealth: Evidence from the Circle Line57880579Table 3. Treatment effects of the opening of new CL MRT stations580Sample criteria: Distance to the MRT stations 581(“distMRT”) 582(1)583(2)584(3)585<=4000m586<=4000m 587<=2000m588Floor level589floor5900.015***5910.007***5920.007***593(0.000)594(0.000)595(0.000)596Unit size597area5980.007***5990.007***6000.007***601(0.000)602(0.000)603(0.000)604Condominium dummy605dumcond6060.133***6070.165***6080.154***609(0.003)610(0.002)611(0.002)612Freehold dummy613dumfh6140.149***6150.099***6160.086***617(0.003)618(0.002)619(0.003)620Age621age622-0.011***623-0.009***624-0.008***625(0.000)626(0.000)627(0.000)628Resale dummy629dumresale6300.017***631-0.052***632-0.051***633(0.003)634(0.002)635(0.003)636After the opening of MRT line637after6380.051***6390.016***6400.015***641(0.002)642(0.002)643(0.003)644Proximity to MRT (≤ 400m)645dumMRT6460.126***6470.042***6480.049***649(0.007)650(0.005)651(0.005)652After × Proximity to MRT653aftdumMRT654-0.078***6550.132***6560.122***657(0.009)658(0.006)659(0.006)660Constant661Constant66212.959***66313.532***66413.539***665(0.004)666(0.008)667(0.009)668Observations66981,31667081,31667157,417672R-squared6730.6696740.8656750.859676Planning Sector FE677No678Yes679Yes680Year of Sale FE681No682Yes683Yes684Note: This table summarizes the OLS regression results with the ln-housing price (lnprice) as the dependent variables. Models in 685Columns (1) and (2) are estimated using a sample of 81,316 transactions found within a 4,000-meter boundary from the closest 686CL MRT stations, and Column (3) is estimated using a sub-sample that includes only 57,417 transactions within a 2000 meter 687boundary. The models include a vector of control variables that include floor level (“floor”), unit size (“area”), condominium 688dummy (“dumcond”), freehold dummy (“dumfh”), age, and resale dummy (“dumresale”). “After” is the treatment dummy that 689have a value of 1, if a sale occurs after the CL opening; and otherwise 0 for a sale occurs before the CL opening. “dumMRT” is the 690proximity dummy that has a value of 1, if a house is located within 400 meters from the closest CL MRT station; and 0 otherwise; and 691“aftdumMRT” is the interactive term of the two treatment variables. The planning sector fixed effects (FE) is defined by the first 2 692digits of the postal code, and the year of sale fixed effect is also included in the model. Standard errors in parentheses. *** p<0.01, 693** p<0.05, * p<0.16945. Empirical Analysis and Results6955.1. 696Housing wealth effects and MRT line opening 697We estimate the baseline OLS model as in Equation 1 without the year of sale and the planning 698sector (spatial) fixed effects, and use only the housing sales that fall within 4,000 meters from the 699closest MRT station so as to minimize boundary discontinuity effects. The results are summarized in 700Column 1 of Table 3. The adjusted R7012 of the model is estimated at 0.669. The estimated coefficients 702on the vector of hedonic attributes are all statistically significant and have the correct signs. 703Variables such as floor level, unit size, condominium dummy, freehold dummy and resale dummy 704are positively correlated with housing prices; whereas, housing age is negatively correlated with 705housing prices. 706The results on the key treatment variables in the model shows that the CL opening has positive 707impact on housing prices, and houses located within 400-meter radius from the CL MRT stations 708are sold for higher prices than other houses located outside the vicinity. The results seem to be 709consistent with the expected treatment outcomes. However, when we look at the interactive term, it 710711712===== PAGE 18 (score 17, numeric-paren 0, table-words 6, var-words 5) =====713714Mi Diao, Yi Fan and Tien Foo Sing71581716has an incorrect sign, which implies that housing prices in the treatment area decrease more than the 717house prices in the control area after the CL opening. The results could be influenced by unobserved 718spatial heterogeneity and time variations that are not controlled for in the model.719In Column 2, we include the planning sector and the year of sales fixed effects to the model, 720and the goodness of fit of the model increases significantly to 0.865. While most of the estimated 721coefficients on the hedonic variables are significant and have the correct signs, but the values of 722the coefficients are different. For the resale dummy, we find that the coefficient becomes negative 723after controlling for the spatial and temporal variations in the housing price model. We show that 724the treatment effects are more consistent with the expected outcomes. The results show that housing 725prices in the post-CL opening period increase at a more moderate rate of 1.6% compared to 5.1% 726estimated in the baseline model. We find that houses located within 400 meters from the closest CL 727MRT station sell for a price of 4.2% higher than comparable houses located outside the 400-m MRT 728station region. The interactive term (“aftdumMRT”) is also statistically significant and has a positive 729sign, which implies that houses with the accessibility convenience to the new CL (“dumMRT” = 7301) sell for 13.2% more expensive than other houses without the accessibility convenience to the 731new CL. The results confirm that the opening of CL generates positive housing wealth effects to 732households who live and own houses in close proximity to MRT stations along the CL. 733We conduct a robustness test by reducing the sample coverage area from 4,000 meters to 2,000 734meters, which cut the sample size from 81,316 observations to 57,417 observations. The adjusted 735R7362 of the model as shown in Column 3 of Table 3 decreases only marginally to 0.859. The estimated 737coefficients for the Model (3) are consistent with the estimated coefficients of the early Model (2). 738We find significant treatment effects, which show that the opening of the new CL increases prices 739for houses located within 400 meters from the closest CL MRT station by 12.2% in the post-opening 740period relative to price changes of other houses that are located outside the 400-m range. The results 741again support the story that the new CL opening generates positive economic effects on houses that 742are within an accessibility distance to the MRT stations along the line.7435.2. 744Effects on existing MRT stations 745The CL is connected to other major MRT lines via six existing MRT stations, which are denoted 746by the dummy “MRTold”. Except for the two terminating stations on the CL (Harbour Front in the 747west and Dhoby Ghaut in the centre), four of the existing stations serve as the interchanges that 748connect the CL to other major MRT lines. These four interchanges are Buona Vista station and Paya 749Lebar station on the East-West line, Bishan station on the North-South line, and Serangoon station 750on the North-East line. Commuters living in the city fringe areas could travel to the outskirts of the 751island and the CBD with MRT via these four interchanges. We first use the six existing stations to 752establish the cut-off distance for the MWTP in Table 2; and as additional robustness tests, we next 753test how the opening of new stations on the CL affects housing prices around these existing stations 754relative to other stations. 755The results of the panel regressions with the spatial and time fixed effects are summarized in 756Table 4. Column 1 shows the results when all the six existing stations as identified by the dummy 757“MRTold” are included in the triple-interactive term, (“after×dumMRT×MRTold”); whereas 758Column 2 replaces the “MRTold” with the “interchg” dummy to capture the incremental effects of 759760761===== PAGE 19 (score 60, numeric-paren 40, table-words 5, var-words 6) =====762763A new Mass Rapid Transit (MRT) line construction and housing wealth: Evidence from the Circle Line76482765the four interchanges on the CL line (Buona Vista, Bishan, Serangoon and Paya Lebar stations). The 766estimated coefficients on the control variables are significant, and the overall results are consistent 767with those reported for Model 2 in Table 3. The results show an average growth in housing prices 768of 2.6% after the CL opening, and houses located within the 400-m radius from the CL stations 769enjoy significant price premiums of about 13.0% relative to other houses located outside the 400-m 770range. We find that the coefficients on the two triple-interactive terms—“after×dumMRT×MRTold” 771(Model 1) and “after×dumMRT×interchg” (Model 2)—are significant, but have negative values 772of -0.093 and -0.096, respectively. The results imply that while housing prices near the CL MRT 773stations in general increase following the opening of CL, housing prices increase at a faster rate 774in other CL stations than housing prices in the existing stations (both “MRTold” and “interchg”) 775(“dumMRT|MRTold=0” and “dumMRT|interchg=0”). The new CL generates incrementally higher 776housing wealth effects to households, who were previously not served by the CL line, relative to 777households who have already had the access to the existing lines via the six stations. 778Table 4. Treatment effects in existing MRT stations and inter-changes779(1)780(2)781Floor level782floor7830.007***7840.007***785(0.000)786(0.000)787Unit size788area7890.007***7900.007***791(0.000)792(0.000)793Condominium dummy794dumcond7950.163***7960.163***797(0.002)798(0.002)799Freehold dummy800dumfh8010.094***8020.094***803(0.002)804(0.002)805Age806age807-0.009***808-0.009***809(0.000)810(0.000)811Resale dummy812dumresale813-0.050***814-0.050***815(0.002)816(0.002)817After the opening of MRT line818after8190.026***8200.026***821(0.002)822(0.002)823Proximity to MRT (≤ 400m)824dumMRT8250.129***8260.130***827(0.003)828(0.003)829After × Proximity × Existing MRT station830aftdumMRTold831-0.093***832(0.009)833After × Proximity ×Interchange834aftdumMRTinterchg835-0.096***836(0.009)837Constant838Constant83913.509***84013.511***841(0.008)842(0.008)843Observations84481,31684581,316846R-squared8470.8648480.864849Planning Sector FE850Yes851Yes852Year of Sale FE853Yes854Yes855Note: This table summarizes the OLS regression results with the ln-housing price (lnprice) as the dependent variables. The Models 856are estimated using a sample of 81,316 transactions found within a 4,000-meter boundary from the closest CL MRT stations. The 857models include a vector of control variables that include floor level (“floor”), unit size (“area”), condominium dummy (“dumcond”), 858freehold dummy (“dumfh”), age, and resale dummy (“dumresale”). “After” is the treatment dummy that have a value of 1, if a sale 859occurs after the CL opening; and otherwise 0 for a sale occurs before the CL opening. “dumMRT” is the proximity dummy that has 860a value of 1, if a house is located within 400 meters from the closest CL MRT station; and 0 otherwise; and “aftdumMRT” is the 861interactive term of the two treatment variables. The triple-interactive terms “aftdumMRTold” and “aftdumMRTinterchg” interact 862“aftdumMRT” with the two station-related variables, where “MRTold” is used to identify six of the existing MRT stations ((Buona 863Vista, Paya Lebar (East-West line), Bishan, Dhoby Ghaut interchange (North-South line), Serangoon and Harbour Front (North-East 864line)), and “interchg” identifies four of them that are used as the inter-changes on the CL line (Buona Vista, Paya Lebar (East-West 865line), Bishan (North-South line), Serangoon (North-East line). The planning sector fixed effects (FE) is defined by the first 2 digits 866of the postal code, and the year of sale fixed effect is also included in the model. Standard errors in parentheses. *** p<0.01, ** 867p<0.05, * p<0.1.868869870===== PAGE 20 (score 9, numeric-paren 0, table-words 2, var-words 5) =====871872Mi Diao, Yi Fan and Tien Foo Sing873838745.3. 875Heterogeneity Tests876We run some heterogeneity tests to further identify if the differential economic spillovers effects 877of the new CL between new sales (consist mainly of pre-completion sales by developers) and resales 878(completed), and private buyers and public housing upgraders. The results are summarized in Table 8795. Column 1 shows the incremental effects of the new CL line on the resale housing market vis-à-880vis the new (pre-completion) sale markets. We find that the new CL have larger incremental on the 881treatment group of houses in the new sale markets than those in the resale market. The differential 882price effects associated with the accessibility benefits of the CL are about 6.4% with the developers’ 883new houses enjoy the greater price increases relative to houses sold in the secondary market. There 884is anecdotal evidence showing that private developers capitalize on the opening of the CL in the 885launch of new projects, such as Sky Habitat88611 that was labelled as the most expensive suburban 887residential project when it was launched in April 2012 by CapitaLand, one of the largest listed 888developers in Singapore.88912890By buyer type, we separately identify new private housing buyers, who previously live in HDB 891flats identified with public housing addresses, and other housing buyers with private housing 892addresses. We denote the “public housing upgraders” group by a new dummy variable, “HDBbuyer”; 893and they make up about 34.3% of the sample buyers. We observe a significant increase in the 894number of upgraders from 30.0% to 37.4% in the two periods “before” and “after” opening of the 895CL. In Column 2, the triple-interactive term shows that HDB upgraders pay 4.3% more than other 896private housing buyers for houses located within 400 meters from the closest MRT stations after the 897CL opening, ceteris paribus. Public housing buyers, who are more likely to use MRT to commute on 898daily basic, place higher MWTP premiums for accessibility benefits to the CL MRT stations.8995.4. 900Temporal variations of the MRT line opening 901As the CL opened in stages over a 2.5-year period from May 2009 to Jan 2012, we test the 902temporal variations in the treatments associated with the CL opening events. Based on the gradual 903information diffusion argument, we expect the earliest phase of CL opening (the stretch from 904Bartley station to Marymount station) to have larger incremental price impact than the last phase of 905the CL line extension from Promenade station to Marina Bay station. We represent each of the four 906phases of CL opening with a dummy “Ph_k”, where k = (1, 2, 3, 4), and replace the treatment term 907“after” in Equation (1) with the four new interactive terms: “aftPh_k” and “aftPh_k×dumMRT”. 908The regression results are reported in Column 1 of Table 6. 909The results show significant negative price changes after the phase 1 (“aftPh_1”: May 2009), 910phase 3 (“aftPh_3”: October 2011) and phase 4 (“aftPh_4”: January 2012); and the only period that 911show positive price growth is after the phase 2 of the CL opening (“aftPh_2”: April 2010). The first 912period decline could be caused by the negative sentiment from the aftermath of the Subprime Crisis 913in 2007–2008; whereas the negative growth in October 2011 and January 2012 could be caused by 914the impact of cooling measures introduced by the government during the periods. We next examine 91511. “CapitaLand sells 125 units at Sky Habitat”, by Kalpana Rashiwala, The Business Times, 16th Apr 2012.91612. “Sky Habitat effect’ lifts sales in other estates”, by Esther Teo, The Straits Times, 16th Jun 2012.917918919===== PAGE 21 (score 57, numeric-paren 40, table-words 4, var-words 5) =====920921A new Mass Rapid Transit (MRT) line construction and housing wealth: Evidence from the Circle Line92284923the treatment effects shown by the triple “diff-in-diff” variables in Model 1, and find that stronger 924price increases when the phase 1 CL opened in May 2009, which creates positive price effects of 92519.2% for houses located within 400 meters from the stations along the Bartley and Marymount 926stretch. The incremental price effects are still positive in the phases 2 and 3 of the CL opening, but 927with a weaker magnitude of 12.5% and 3.4%, respectively. There is no significant price effect shown 928in the phase 4 extension of the CL to the Marina Bay in January 2012. The results are consistent 929with the information diffusion hypothesis, which predicts strong price effects when the CL opening 930event was first revealed in the market. 931Table 5. Effects on developers’ sales and public housing upgraders932(1)933(2)934Floor level935floor9360.007***9370.007***938(0.000)939(0.000)940Unit size941area9420.007***9430.007***944(0.000)945(0.000)946Condominium dummy947dumcond9480.164***9490.163***950(0.002)951(0.002)952Freehold dummy953dumfh9540.096***9550.095***956(0.002)957(0.002)958Age959age960-0.009***961-0.009***962(0.000)963(0.000)964Resale dummy965dumresale966-0.048***967-0.052***968(0.002)969(0.002)970After the opening of MRT line971after9720.026***9730.024***974(0.002)975(0.002)976Proximity to MRT (≤ 400m)977dumMRT9780.134***9790.111***980(0.003)981(0.003)982After × Proximity × Resale983aftdumMRTresale984-0.064***985(0.006)986After × Proximity × HDB buyer987aftdumMRThdbbuyer9880.043***989(0.006)990Constant99113.507***99213.514***993(0.008)994(0.008)995Observations99681,31699781,316998R-squared9990.86410000.8641001Planning Sector FE1002Yes1003Yes1004Year of Sale FE1005Yes1006Yes1007Note: This table summarizes the OLS regression results with the ln-housing price (lnprice) as the dependent variables. The Models 1008are estimated using a sample of 81,316 transactions found within a 4000 meters boundary from the closest CL MRT stations. The 1009models include a vector of control variables that include floor level (“floor”), unit size (“area”), condominium dummy (“dumcond”), 1010freehold dummy (“dumfh”), age, and resale dummy (“dumresale”). “After” is the treatment dummy that have a value of 1, if a sale 1011occurs after the CL opening; and otherwise 0 for a sale occurs before the CL opening. “dumMRT” is the proximity dummy that 1012has a value of 1, if a house is located within a 400meter from the closest CL MRT station; and 0 otherwise; and “aftdumMRT” is 1013the interactive term of the two treatment variables. The triple-interactive terms “aftdumMRTresale” and “aftdumMRThdbbuyer” 1014interact “aftdumMRT” with the two new dummies, which include “dumresale” that indicates non-developers’ sales, and “hdbbuyer” 1015identifies public housing upgraders. The planning sector fixed effects (FE) is defined by the first 2 digits of the postal code, and the 1016year of sale fixed effect is also included in the model. . Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.11017Like McDonald and Osuji (1995) and McMillen and McDonald (2004), we test if “anticipative” 1018effects occur around the CL opening event. We use a 6-month pre-opening dummy, “bef6mth”, 1019which has a value of 1 if a house were sold in less than 180 days (6 months) before the opening 1020of the closest CL MRT stations, to capture the “anticipative” effects, alongside the post-opening 1021effects in the Model 2. Our results show no “anticipative” effects associated with the CL opening, 1022but prices of houses located close to the MRT stations were 4.4% lower than prices of other houses 1023located outside the 400m cut-off boundary in the pre-opening of CL. However, houses located in 102410251026===== PAGE 22 (score 114, numeric-paren 89, table-words 6, var-words 4) =====10271028Mi Diao, Yi Fan and Tien Foo Sing1029851030the treatment zone increase by 14.3% more than prices of comparable houses outside the treatment 1031zone in the post-CL opening period. We next replace the “after” dummy by “Ph_k”, which captures 1032different opening event time effects, and estimate the regression model as shown in Column 3 1033of Table 6. The results are consistent with the earlier findings, where we observe negative price 1034changes in the pre-opening period, and the information impact is positive in the next three phases 1035of the CL opening but at a declining rate. The news of the extension of the CL (phase 4) has no 1036incremental impact on housing prices in the treatment zone.1037Table 6. Temporal effects of the opening of CL 1038(1)1039(2)1040(3)1041Floor level1042floor10430.007***10440.007***10450.007***1046(0.000)1047(0.000)1048(0.000)1049Unit size1050area10510.007***10520.007***10530.007***1054(0.000)1055(0.000)1056(0.000)1057Condominium dummy1058dumcond10590.168***10600.165***10610.168***1062(0.002)1063(0.002)1064(0.002)1065Freehold dummy1066dumfh10670.099***10680.098***10690.098***1070(0.002)1071(0.002)1072(0.002)1073Age1074age1075-0.009***1076-0.009***1077-0.009***1078(0.000)1079(0.000)1080(0.000)1081Resale dummy1082dumresale1083-0.052***1084-0.053***1085-0.053***1086(0.002)1087(0.002)1088(0.002)1089Proximity to MRT (≤ 400m)1090dumMRT10910.038***10920.031***10930.028***1094(0.005)1095(0.005)1096(0.005)10976 months before MRT opening1098bef6mth1099-0.007**1100-0.017***1101(0.003)1102(0.003)11036 months before MRT opening x 1104Proximity to MRT1105bef6mthdumMRT1106-0.044***1107-0.037***1108(0.010)1109(0.010)1110After the opening of MRT line1111after11120.021***1113(0.003)1114After×Proximity to MRT1115aftdumMRT11160.143***1117(0.007)1118AfterxPhase 11119aftph11120-0.041***1121-0.028***1122(0.005)1123(0.006)1124AfterxPhase 21125aftph211260.046***11270.058***1128(0.003)1129(0.003)1130AfterxPhase 31131aftph31132-0.007*1133-0.0011134(0.004)1135(0.004)1136AfterxPhase41137aftph41138-0.022***1139-0.020**1140(0.008)1141(0.008)1142After×Phase1×Proximity to MRT1143aftph1dumMRT11440.192***11450.202***1146(0.008)1147(0.008)1148After×Phase2×Proximity to MRT1149aftph2dumMRT11500.125***11510.134***1152(0.008)1153(0.008)1154After×Phase3×Proximity to MRT1155aftph3dumMRT11560.034***11570.044***1158(0.011)1159(0.011)1160After×Phase3×Proximity to MRT1161aftph4dumMRT1162-0.0801163-0.0701164(0.054)1165(0.054)1166Constant116713.545***116813.535***116913.549***1170(0.008)1171(0.008)1172(0.008)1173Observations117481,316117581,316117681,3161177R-squared11780.86611790.86511800.8661181Planning Sector FE1182Yes1183Yes1184Yes1185Year of Sale FE1186Yes1187Yes1188Yes1189Note: The table summarizes the regression with the ln-housing price (lnprice) as the dependent variable. The Models are estimated 1190using a sample of 81,316 transactions found within a 4,000-meter boundary from the closest CL MRT stations. “Phase 1” = 28 May, 11912009: (Bartley–Marymount); “Phase 2” = 17 April, 2010: (Dhoby Ghaut–Bartley) (eastern stretch); “Phase 3” = 8 October, 2011: 1192(Marymount–Harbour Front) (western stretch); “Phase 4”= 14 January, 2012: CL Extension (CLe) (Promenade–Marina Bay). 1193Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1119411951196===== PAGE 23 (score 9, numeric-paren 0, table-words 0, var-words 9) =====11971198A new Mass Rapid Transit (MRT) line construction and housing wealth: Evidence from the Circle Line11998612006. Conclusion1201The bid-rent theory of Alonso (1964), Muth (1969), and Mills (1972) has shaped urban growth 1202of monocentric cities for decades. Household preference for bigger space and tranquillity of living 1203environment in suburban areas induces them to live further away from central areas. The trade-1204off between commuting costs and housing consumption results in a negative bid-rent curve, which 1205predicts that housing price declines with respect to distance from the CBD. However, with large 1206public investments in transit and MRT networks that improve connectivity and accessibility to 1207suburban areas, the Alonso-Muth-Mill’s bid-rent theory has become less relevant in the urban 1208structure in today’s modern cities. There are two competing hypotheses that explain possible 1209positive rent gradient from the CBD: First, Richardson (1977) argues that less congested and low-1210density living environment in suburbs creates externality rents for high income households, who 1211pay higher housing prices in areas located further away from the CBD. Second, the decentralisation 1212trend also reduces the CBD effect. The emergence of multiple regional centres changes a 1213monocentric city structure to a poly-centric one. Richardson et al. (1990) show that the distance-1214to-CBD effects on bid rent is still statistically significant but weak in 1970, and the distance effects 1215disappeared in 1980 in some US cities. 1216This study empirically tests housing wealth effects associated with the opening of new CL MRT 1217Line in Singapore. Using non-landed private housing transactions for the period from January 2009 1218to May 2013, we find that the opening of the new CL MRT line increases housing prices by 1.6%, 1219on average. Houses in the treatment zone, which is within 400 meters from the CL MRT stations, 1220sell for 4.2% higher in price than comparable houses located outside the 400-m MRT zone. Housing 1221price increases in the MRT treatment region are 13.2% more than housing price increases of houses 1222located outside the treatment regions after the CL opening. Based on 6,564 houses sold for an 1223average price of S$1,416,896 in the post-CL opening period, which is translated into an aggregate 1224value of S$9.30 billion, we estimate the housing wealth of S$1.23 billion accrued to households 1225living in 400-m radius from the closest MRT stations after the opening of the CL. We also find that 1226the accessibility premiums are larger in developers’ market, and for public housing buyers (HDB 1227upgraders) who pay relatively higher accessibility premiums relative to private housing buyers. 1228When we test temporal variations in the treatment effects, we find no significant “anticipative” 1229effects in the six-month window prior to the opening of the new CL. Housing price in the treatment 1230zone increases by more than 14.3% relative to housing prices outside the treatment zone after the 1231CL opening. We find significant positive treatment effects in the first three phases of the CL opening, 1232but the treatment intensity diminishes after phase 1. The results are consistent with the information 1233diffusion hypothesis, and the results find no sentiment-driven momentum effects. 1234Governments of rapidly expanding cities invest heavily in public transit projects to meet 1235increasing transportation demand. In countries including Singapore, new public transit lines are seen 1236as a key driver for economic and employment growth.13 There are also other non-pecuniary benefits 1237associated with public transit lines, which include reductions in road congestion and greenhouse 123813. Mr Ngiam Tong Dow, the then permanent secretary of the Ministry of Communications, used the economic motivations to 1239persuade the government to commit on the MRT project in Singapore. Sources: “Success and Failure of Public Policies: The 1240Singapore Experience, 1960-2000”. In A Mandarin and the Making of Public Policy: Reflections by Ngiam Tong Dow. Singapore: 1241NUS Press, 2006, pp 150.1242