FORWARD-LOOKING BEHAVIOUR IN SINGAPORE'S PRIVATE HOUSING MARKET: THE IMPACT OF THE NORTHEAST TRANSIT LINE ON THE HOUSING PRICE GRADIENT CHEW KUO TING B.Soc.Sci. (Hons.), NUS A THESIS SUBMITTED FOR THE DEGREE OF MASTER OF SOCIAL SCIENCE DEPARTMENT OF ECONOMICS NATIONAL UNIVERSITY OF SINGAPORE 2011 ACKNOWLEDGEMENTS To my parents and family members, for their support in everything I do. To my supervisor, Associate Professor Anthony Chin Theng Heng, for the academic support and freedom he has given me. To Dr Eric Fesselmeyer, Mun Lai Yoke and Liu Zhenning, for their helpful comments to this thesis. To Zeng Ting, Gao Xin Wei, Cao Qian, Chen Yan Hong, Shao lei, Sun Yifei, Yap Wei Ming, Chan Ying Jie, Leong Chi Hoong and Mok Wen Jie, for the discussions and feedbacks. To the Department of Geography at National University of Singapore for providing me with access to their old Singapore maps. Any existing errors are mine. i Table of Contents Summary .................................................................................................................. iv List of Tables............................................................................................................. v List of Figures ........................................................................................................... v Introduction ............................................................................................................... 1 1. Literature Review .................................................................................................. 2 1.1 Urban Spatial Structures ................................................................................... 2 1.2. Hedonic Price Models and Transportation Developments ................................ 3 1.3. Non-OLS Hedonic Price Models ..................................................................... 6 1.4. Spatial Econometrics ....................................................................................... 7 1.4.1. Spatial Heterogeneity ................................................................................ 7 1.4.2. Spatial Autocorrelation ............................................................................. 8 2. NEL Project Overview ........................................................................................... 9 3. Empirical Analysis .............................................................................................. 12 3.1. Research Hypotheses ..................................................................................... 12 3.2. Empirical Data .............................................................................................. 12 3.2.1. OLS Estimations ..................................................................................... 19 3.2.2 OLS Estimations with Interaction Terms .................................................. 23 3.3. Can’s Spatial Expansion Model ..................................................................... 28 3.3.1. Spatial Expansion Model Estimations ..................................................... 29 4. Conclusion........................................................................................................... 32 Reference ................................................................................................................ 34 Appendix 1: Estimation results ................................................................................ 41 ii Appendix 2: Temporal Effects of the NEL ............................................................... 43 Appendix 3: Construction of Housing Price Index ................................................... 44 iii SUMMARY This thesis seeks to investigate the effects of the North-East Line (NEL) Mass Rapid Transit (MRT) extension on neighbouring private housing resale prices. A hedonic price analysis on the Singapore private housing market is conducted using 99-year non-landed resale private houses and Executive Condominiums (EC) transactions located near the NEL from 1st January 1995 to 31st December 2008. The Ordinary Least Squares (OLS) and the spatial expansion method were used to estimate the hedonic price model. After controlling other variables, the estimations show the presence of positive announcement effects and negative construction effects on the non-landed private resale prices located within 800 metres of the NEL development. In particular, the announcement effects from the NEL were so strong that housing prices were higher than the prices levels when the NEL became operational. This suggests that the private resale housing market was over-reactive to market news on the NEL developments. Keywords: Hedonic price. Spatial econometrics. Transport infrastructure. Housing iv LIST OF TABLES Table 1: Descriptive Statistics .................................................................................. 16 Table 2: Percentage of Housing Transactions in each Proximity Zone ..................... 16 Table 3: Private Resale Estates ................................................................................ 17 Table 4: Breusch-Pagan Heteroskedasticity Test ...................................................... 21 Table 5: Price Indices from OLS Estimations .......................................................... 25 Table 6: Price Indices from the Spatial Model ......................................................... 32 LIST OF FIGURES Figure 1: Map of Singapore’s Transit Network ........................................................ 11 Figure 2: URA Property Price Index of Non-Landed Residential Properties ............. 24 Figure 3: OLS Year-Proximity Price Index .............................................................. 24 Figure 4: Spatial Model Price Index ......................................................................... 31 v INTRODUCTION Singapore’s small land mass of 712.4 square kilometres (Department of Singapore Statistics, 2011) has cultivated a country that is extremely prudent with its urban development and land management practices. One key focus in Singapore’s area of urban management is the mobility of its residents within the city-state. The construction of the North-East Line (NEL) transit system in 2003 is part of the development of a Rapid Transit System based on a comprehensive rail network. The NEL was targeted to serve residents in the North-East, and designed to connect the existing East-West and North-South MRT transit lines. The NEL’s construction phase spanned from November 1997 to June 2003. To ensure the financial feasibility of the NEL project, the catchment area of its alignment consists of high population density areas such as Chinatown, Serangoon and Clarke Quay. Land had to be acquired and some developed areas had to make way for its construction. Tunnelling works had to be carefully managed so as not to damage the structural foundations of nearby residential houses and commercial buildings. Furthermore, the close proximity of several construction sites to residential estates meant that managing negative externalities from the construction of the NEL are of concern to residents along the alignment. This thesis attempts to identify the effects of the NEL on neighbouring private resale housing prices. A hedonic price analysis is chosen to analyse the transaction prices of the private resale houses located near the NEL project from 1995 to 2008. The selection of this period allows the study to investigate how the announcement, construction and operational phases of the NEL project affected neighbouring housing prices. In this thesis, OLS estimations and spatial expansion model are used to 1 estimate the hedonic price model. To my knowledge, no hedonic study of the Singapore housing market has adopted the spatial expansion model, which is an estimation method that can help account for spatial heterogeneity and spatial autocorrelation. The rest of the thesis is organised as follows. Chapter 1 reviews the literature. Chapter 2 provides a brief overview of the NEL project. Chapter 3 introduces the estimation models, results and interpretation. Chapter 4 concludes the main findings and discusses possible extensions to this study. 1. LITERATURE REVIEW 1.1 Urban Spatial Structures Early monocentric urban models suggest that rising land costs in a city’s main central business district (CBD) area will allocate land use according to some left over principle. Monocentric urban models confound that accessibility to central urban areas made these areas highly sought after, and this was reflected through the higher land and rent prices that lead to the negative bid-rent functions (Alonso, 1964; Muth, 1969; Von Thunen, 1826). However, as technological improvements reduced transportation time and costs, cities began to sprawl and they increasingly had spatial layouts that were not predicted by the monocentric spatial models. Polycentric models were developed to focus on how urban spatial developments could be drawn to other parts of a city. Romanos (1977) constructed a two-workplace model, while Madden (1980) modelled and empirically investigated housing choice of dual-income households. Yinger (1992) on the other hand, modelled the effects of segregating suburban and urban workers and their employment locations on urban spatial 2 developments. Several polycentric models have also differentiated employment areas with consumption nodes (Brueckner, 1979; Landsberger and Lidgi, 1978). Developments in the hedonic price model led to its adoption in the areas of empirical housing market research. Lancaster (1966) posits that goods be valued based on the attributes, allowing a more accurate comparison between different goods and their corresponding utilities to consumers. The hedonic price model approach allows housing to be treated as a composite of locational and structural attributes that contributes to its overall price. Rosen (1974) showed that given a housing market with heterogeneous consumer preferences and income, the hedonic price function still reflects a market clearing condition. The applications of hedonic price models have allowed the theoretical spatial models to be tested empirically, and both monocentric (Bailey, Muth and Nourse, 1963; Chung and Chan, 2003; McMillen, 2003) and polycentric housing price gradients (Bender and Hwang, 1985; Heikkila, Gordon, Kim, Peiser and Richardson, 1989; Waddell, Berry and Hoch, 1993) have been observed from various hedonic housing market studies. 1.2. Hedonic Price Models and Transportation Developments Hedonic price models have been commonly used to isolate the marginal price of transportation infrastructure developments on neighbouring housing prices. Access to transportation services increases the accessibility of neighbouring households and also commonly linked to increases in residential home prices. In addition, positive announcement and anticipatory effects on neighbouring housing prices have been commonly linked to transportation infrastructure developments (Bae, Jun and Park, 2003; Damm, Lerman, Lerner-Lam and Young, 1980; McDonald and Osuji, 1995; McMillen and McDonald, 2004; Wang 2010). Even though the announcement and 3 anticipatory effects both occur before the full commencement of the transportation system, these two effects are inherently different. Housing price changes caused by the announcement of the project will constitute the announcement effect, while anticipatory effects are the changes in housing prices when the date of the project commencement draws nearer. As such, if the announcement date and commencement date of the project are very far apart, housing prices surrounding the project may have two distinct periods experiencing announcement and anticipatory effects. For the Southwest Rapid Transit Line in Chicago, McDonald and Osuji (1995) found positive and statistically significant anticipatory effect on residential land prices in 1990, while the anticipatory effects found by McMillen and McDonald (2004) occurred as early as 1987, even though the line only opened in 1993. However, there are also transportation developments had non-positive effects on neighbouring housing prices (Forrest, Glen and Ward, 1996; Gatzlaff and Smith, 1993). Gatzlaff and Smith (1993) found that the announcement and operations of the Miami Metrorail system had no positive effects on neighbouring housing prices, while the Metrolink in Greater Manchester had a negative impact on neighbouring housing prices (Forrest, Glen and Ward, 1996). The announcement of the Supertram construction in Sheffield also caused neighbouring housing prices to decline, although subsequent increases in housing prices after tram’s completion was suggested as evidence of the gained accessibility by neighbouring households (Heneberry, 1998). In Singapore, hedonic price models have been used to construct a constantquality price index for Housing Development Board (HDB) resale flats in Singapore (Ong, Ho and Lim, 2003), and to estimate how political boundaries (Wei and Wong, 2010), ethnicity (Wong, 2008) and proximity of primary schools (Wong, 2008) effect 4 housing prices. In particular, hedonic price models were also used to study the effects of transit developments on housing prices in Singapore. Using four MRT stations, namely Chinese Garden, Chua Chu Kang, Serangoon and Bishan, Lee (2009/2010) found that even after controlling for presence of a shopping centre, proximity to an underground train station provided an average price premium of 2.08% and 2.78% when compared to condominiums located near an above ground station. His findings support the hypothesis that above ground train stations create additional operational noise that adversely affect neighbouring housing prices. A study on the Pioneer MRT station found that announcements of the station’s construction caused neighbouring HDB resale housing prices to rise 22.6% above pre-announcement housing prices (Tan, 2009/2010). HDB resale housing prices were even 6.61% higher than preannouncement levels during its construction phase, suggesting that the anticipation for the benefits of the future station were stronger than the negative construction disturbances experienced by neighbouring households. However, housing prices after the new station became operational were only 7.14% higher than pre-announcement levels, indicating that the announcement of the Pioneer MRT station’s construction had a greater positive effect on HDB resale prices than when the station became operational. With the rapid development of transit lines in Singapore, hedonic housing studies have also focused on how specific transit developments in Singapore affect housing prices within their close vicinity. Ong (2001) found that HDB resale flats nearer to the East-West transit line stations experience a price premium, with those further from the CBD enjoying an even greater premium for locating near an EastWest transit line station. Studies have also found that proximity to NEL train stations 5 had a positive effect on the prices on public and private housing prices (Chan 2004/2005; Quek 2004/2005). In particular, Quek (2004/2005) found positive announcement effects from the NEL project that were as high as 34.9% during the 3rd quarter of 1998. However, other studies on the Circle Line development found both positive and negative announcement effects for houses located near different train stations along the Circle Line (Chia, 2008/2009; Wu, 2007/2008). As such, positive announcement effects could not be said for the entire Circle Line development, but depended on the locational characteristics of each station. 1.3. Non-OLS Hedonic Price Models Despite the common usage of the OLS model in many hedonic price studies of housing markets in both Singapore and other regions, other estimation methods have been created to improve on the OLS framework. Unfortunately, few of these estimations have been used in the Singapore context. Meese and Wallace (1991) adapted the non-parametric locally weighted regression (LWR) by Cleveland and Devlin (1988) and Cleveland, Devlin and Grosse (1988) to the construct of the housing price indices for Alameda and San Francisco Counties. Knight, Dombrow and Sirmans (1995) constructed a seeming unrelated regression (SUR) hedonic price model that allows the marginal effects of housing attributes to change over time. Brunsdon et al.’s (1996) geographically weighted regression (GWR) allows spatial variables to be estimated at each observation point, and weight observations by their distance to this point, eliminating the need for a prior function form in the estimation. Spatial econometrics was developed to address specific issues in the estimation of spatial data (Cliff and Ord, 1981; Upton and Fingleton, 1985; Anselin, 1988, 2001; Can, 1990, 1992), and have been readily applied to hedonic housing price 6 models. Generally, spatial econometrics models seek to address two issues with estimating spatial data: spatial heterogeneity and spatial autocorrelation. This is because presence of these issues in the spatial data violates the OLS assumptions. Such spatial data will then have to be estimated differently from the OLS framework. Thériaut, Des, Villeneuve and Kestens (2003) adopted Can’s (1990, 1992) spatial expansion method with the principal factor analysis to investigate the housing market of the Quebec Urban Community, and showed that the spatial expansion method can address some issues of spatial autocorrelation. Bitter, Mulligan and Dall’erba (2007) showed that although the GWR provides greater explanatory powers and predictive accuracy than the spatial expansion method, the latter’s ability to handle a “large number of variables and interactions” (Bitter et al., 2007, p. 23) makes it a better model to investigate the underlying determinants of housing. 1.4. Spatial Econometrics 1.4.1. Spatial Heterogeneity Spatial heterogeneity is present when locational differences exist between observations across a geographical space, and this is common in studies with observations that are spread over an extensive geographical area. Failure to consider for spatial heterogeneity is a specification problem that will affect the accuracy of the model’s estimation. Regional dummies can be used to account for location-specific differences, and this was done for the Equation 1 and 2 of this study. However, the use of regional dummies ignores the presence of continuous spatial heterogeneity that is natural to the spatial structure of urban housing markets and developments. In addition, regional dummies cannot account for changes in marginal utilities of housing attributes that arise from spatial heterogeneity, such as the marginal utilities 7 for attributes of houses located in urban and rural areas. Bearing in mind these issues, if continuous spatial heterogeneity is deemed to be present, the use of other specifications in the estimations will then be necessary to account for them. 1.4.2. Spatial Autocorrelation On the other hand, spatial autocorrelation is where interactions between neighbouring observations across the geographical area of study exist. Such spatial interactions violate the assumption of independence across variables in standard OLS estimations. Hence, spatial autocorrelation will require regression techniques that can ‘incorporate the spatial dependence into the covariance structure either explicitly or implicitly by means of an autoregressive and/or moving-average structure’ (Cliff and Ord, 1982; p.142). When determining residential housing prices, in addition to establishing it from the locational and physical attributes of the house, realtors commonly factor in the prices of neighbouring houses that were sold recently. This practice leads to the interactions of housing prices across geographically close locations, and leads to the issue of spatial autocorrelation. Anselin (2001) discussed the estimations of different spatial dependence models under different spatial autocorrelation conditions: 1) Pure space-recursive, where dependence is based on neighbouring locations from a previous time period 2) Time-space recursive, where dependence is based on the same location and neighbouring locations from a previous time period 3) Time-space simultaneous, where dependence is based on neighbouring locations of previous and current time periods 4) Time-space dynamic, which is a combination of all spatial dependences 8 With reference to the private housing market investigated in this study, although no test for spatial autocorrelation was conducted in this study, Han’s (2005) Moran’s I spatial test found evidence of spatial clustering and autocorrelation in the Singapore condominium market during the 1990s. This strongly suggests that spatial autocorrelation may exist for the housing market investigated in this study, and accounting for it may improve the estimation results. Following the definitions of Anselin (2001), the pure space-recursive approach is used to account for the effects of the spatial autocorrelation. 2. NEL PROJECT OVERVIEW To meet the commuting needs of the Singapore population, considerations to construct a transit line to serve the North-East residents of Singapore were made by its government in as early as 1984 (Leong, 2003). In January 1996, the Singapore government announced its plans to construct the NEL (Leong, 1996). The NEL was planned to have 16 stations, cover a total length of 20 kilometres and estimated to cost about SGD 5 billion to build. To cover the initial high costs of the NEL construction, and to ensure that initial train fares on the NEL was not set too high, the Singapore government “paid for the first set of operating asserts that included trains and signalling systems” (Leong, 2003, p. 31). Due to the 1997 Asian financial crisis, the NEL constructions were delayed till November 1997. The complexity of the NEL project constructions also led to many other issues. As the NEL was to connect the North-East regions of Singapore to the central locations (Figure 1), constructions were done through several densely populated areas, including Chinatown, Little India, Clarke Quay and Dhoby Ghaut. Due to the scale and proximity of the NEL constructions to populated areas, many housing estates 9 were affected by its construction phase. Residents from “Boon Keng, Sennett, Kandang Kerbau and Farrer Park estates had to face road diversions to facilitate the building of the stations” (Karamjit, 1998) that led to much traffic diversions. The NEL constructions also relocated the Hougang bus interchange, and led to huge changes in several bus routes and new bus stops that greatly affected the lives of many Hougang residents. Several housing estates also had to put up with the noise and pollution that came from the heavy NEL constructions, with some occurring late into the night (Leong, 2003). For example, at one construction site at Clarke Quay, “the soft nature of the ground and the time taken for the installation of each panel – some 40 hours – dictated that the perimeter wall works had to be carried out around the clock” (Leong, 2003, p. 150). On the other hand, despite these woes, many saw the benefits of being near a future train station. Dr Richard Hu, Adviser to Kreta Ayer Grassroots Organisations, advised Clarke Quay residents at a meeting that “If we can all stay the course, we will realise the benefits. This is because having an MRT station nearby is like buying into prime land” (Leong 2003, p. 79). After 5 years of construction, 14 of the 16 NEL train stations began their operations in June 2003. 10 Figure 1: Map of Singapore’s Transit Network 1(Source: Google Maps, 2011) 1 The NEL is represented by the purple line. 11 3. EMPIRICAL ANALYSIS 3.1. Research Hypotheses Past literature investigating the relationship between transportation development projects and housing prices have shown key findings that include announcement, anticipatory, construction and operational effects. However, different projects had observed different effects, and one has to be careful in generalising the effects of a particular transportation development project on neighbouring housing prices. For the NEL project, its announcement is expected to affect neighbouring housing prices, although direction of this effect is ambiguous. Having a future train station in close proximity increases the accessibility of neighbouring households, which in turn increases their housing prices before the line’s actual commencement. However, if households worry about the noise and pollution from the project’s construction phase in the near future, this announcement effect could be negative. Secondly, as the NEL project is an intensive large-scale infrastructure development, the noise and pollution created during its construction phase is expected to adversely affect neighbouring households, and cause neighbouring housing prices to drop. Lastly, the commencement of the NEL project should increase neighbouring housing prices due to the gains in accessibility for the train line. 3.2. Empirical Data The majority of hedonic price analyses done on the Singapore housing market focuses on its public housing. Firstly, the large public housing market provides a huge amount of resale transaction data for empirical analysis. Secondly, Singapore public 12 houses are fairly homogeneous across different regions, which make the model specification less cumbersome. In addition, public flat owners are allowed to sell their flats in the resale market at market prices, allowing hedonic price models to consider for the market forces that affect public housing prices in Singapore. However, despite its relatively small size, the private housing market in Singapore plays an important role in contributing to the overall vitality of the housing market. Studies have found a strong positive correlation between public and private housing markets in Singapore (Phang and Wong, 1997; Lum, 2002; Ong and Sing, 2002; Bardhan, Datta, Edelstein and Lum, 2003; Sing, Tsai and Chen, 2006). Phang and Wong (1997) showed that from 1975 to 1994, several government policies had a significant impact on private housing prices, and that private and public housing prices in Singapore during that period were highly correlated. This strong correlation between the two housing sub-markets suggests that the provision of affordable public housing in Singapore and the general state of its housing market will require the Singapore government to closely monitor, and even intervene in its private housing market when necessary. Landed private residential houses located near the NEL train stations are excluded, because such houses are occupied by high income earner that most likely have their own private vehicles. In addition, landed houses normally go through extensive renovations when they are sold to a new owner, and this will impact the price of the housing. As such, this thesis focuses on 99-ear non-landed private housing and Executive Condominium (EC) resale transactions along the NEL development. ECs were introduced into the Singapore housing market by its government in 1995, and are strata-titled apartments built by private developers. ECs 13 have facilities comparable to private condominiums, but face certain government restrictions, such as a 5-year minimum occupancy period before the unit can be resold (HDB InfoWEB, 2011). In general, 99-year non-landed private houses and ECs in Singapore are cheaper than freehold and landed residential properties. Hence, such non-landed residential estates are commonly occupied by middle-high income households, who are more likely to have family members that will use and benefit from the NEL development. The resale housing transaction prices from 1st January 1996 to 31st December 2008 were downloaded from the Real Estate Information System (REALIS), a database system managed by the Urban Redevelopment Authority of Singapore (URA) (Real Estate Information System, 2011). Transactions around Woodleigh and Buangkok stations were omitted, as these stations did not open with the rest of the stations in June 2003. There were also no 99-year non-landed private or EC resale transactions near the Potong Pasir and Punggol stations during this period. The initial selection of the resale transaction prices were based on the zoning from the REALIS website. In total, 4,706 non-landed private resale housing transactions surrounding 12 NEL train stations were downloaded from the REALIS website. These include the stations Harbourfront, Outram Park, Chinatown, Clarke Quay, Dhoby Ghaut, Little India, Farrer Park, Boon Keng, Serangoon, Kovan, Hougang and Sengkang. A visual check was done through Google Maps along the entire NEL development to ensure that all the private estates selected were not near other existing MRT train stations, except the connecting stations linking the East-West line to the future NEL. As the housing prices from the REALIS website are recorded in nominal terms, the prices were converted to the real terms using the ‘Property Price Index of Non14 landed Private Residential Properties’, which was also downloaded from the REALIS website. All linear distances were measured from the centre of the amenities and housing estates using Google Maps’ distance measure tool. However, as Google Maps only provides up-to-date maps, physical changes along the NEL from 1995 to 2008 will not be captured. A map of Singapore in 2005, loaned from the Department of Geography in the National University of Singapore, and Singapore street directories from 1995 to 2005 were used to check for urban development changes during those periods (Singapore Street Directory, 1995 to 2005). Suspected physical changes were cross-referenced with newspapers references to establish the official opening dates and availability of the different amenities. The periods where these amenities are accessible to neighbouring households are then adjusted accordingly. Table 1 provides a brief descriptive statistics of the data collected for the analysis. Table 2 displays the percentage of estates in each proximity zone, while Table 3 provides a breakdown of each of the estate, their location and their linear proximity to the nearest NEL train station. 15 Variable Nominal Price Storey Area Age Condo Apartment Executive Condo Boon Keng Chinatown Clarke Quay Dhoby Ghaut Farrer Park Harbourfront Hougang Kovan Little India Outram Seng Kang Serangoon Linear proximity to MRT Obs 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 4706 Mean 752177 9.575011 121.2684 153.5686 .5301742 .4396515 .0301742 0.008925 0.063323 0.081386 0.024437 0.060136 0.162558 0.150446 0.044836 0.060349 0.056311 0.069273 0.21802 675.2754 Std. Dev. 446755.3 7.1542 43.30461 106.1864 .4991417 .4963974 .1710847 0.094059 0.24357 0.273455 0.154418 0.237764 0.369002 0.357546 0.206967 0.238157 0.230546 0.253945 0.412945 386.5194 Min 155000 1 38 17 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 62 Max 8482840 37 1016 488 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1700 Counts 2450 1856 142 42 298 383 115 283 722 499 205 284 265 326 1026 - Table 1: Descriptive Statistics Entire Data Set Within 400m Between 400 to 800m Beyond 800m Count Percentage 4706 100 1369 29.09 1889 40.14 1488 30.77 Table 2: Percentage of Housing Transactions in each Proximity Zone 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 Estate Name St Francis Court Fook Hai Building Landmark Towers People's Park Centre People's Park Complex River Place Riverwalk Apartment The Riverside Piazza Gambier Court The Quayside Peace Centre/Mansions Sunshine Plaza Kentish Court Kentish Green Kentish Lodge Kerrisdale Caribbean At Keppel Bay Harbour View Towers The Pearl @ Mount Faber Evergreen Park Regentville Rio Vista The Florida Kovan Melody Toho Green Burlington Square The Bencoolen Pearl Bank Apartment Pearls Centre Spottiswoode Park Craig Place Compass Heights Rivervale Crest Cherry Gardens Sunglade Casa Rosa Central View The Sunnydale Cardiff Court Chiltern Park Chuan Park Minton Rise Condominium The Springbloom Nearest Station Bong Keng Chinatown Chinatown Chinatown Chinatown Clark Quay Clark Quay Clark Quay Clark Quay Clark Quay Dhoby Ghaut Dhoby Ghaut Farrer Park Farrer Park Farrer Park Farrer Park Harbourfront Harbourfront Harbourfront Hougang Hougang Hougang Hougang Kovan Kovan Little India Little India Outram Outram Outram Outram Sengkang Sengkang Serangoon Serangoon Serangoon Serangoon Serangoon Serangoon Serangoon Serangoon Serangoon Serangoon Linear MRT (metres) 580 430 350 120 90 600 210 310 890 780 490 600 380 480 400 400 560 1570 720 800 1700 920 1030 280 1270 630 740 190 200 880 530 62 960 590 310 760 660 670 1040 990 1100 910 970 Table 3: Private Resale Estates 17 Following the literature in hedonic housing studies, linear distances between housing units and the nearest NEL train stations are used to determine the housing accessibility to the transit service. In addition, the level of operational noise and commuter traffic disturbances experienced by households from the train stations is also a proxy of this linear distance. These linear distances are separated into three proximity zones: within 400 metres, between 400 to 800 metres and beyond 800 metres from the nearest NEL train station. These distance zones are based on Wibowo and Olszewski (2005), whom found that 790 metres is the upper limit commuters are willing to walk to an MRT station. The proximity zones are set as dummy variables, with the beyond 800 metres zone omitted to prevent perfect multi-collinearity. The MRT proximity dummy variables in the estimation model start from the announcement of the NEL project construction to allow for the consideration of possible announcement and construction effects. DiPasquale and Wheaton (1996) also suggested that macroeconomic factors like inflation rates and GDP growth can affect housing prices. However, Phang and Wong (1997) found that GDP growth rates were not statistically significant in affecting private resale housing prices in Singapore from the period of 1975 to 1994. However, Phang and Wong’s (1997) conclusion may not be transferrable to the study here due to several reasons. Firstly, as the studies are investigating the housing markets in two different periods, changing socio-economic conditions may also have changed the effects macroeconomic factors have on the Singapore housing market. Secondly, this thesis focuses on non-landed private resale housing, while Phang and Wong (1997) studied the entire Singapore private housing market. High-income earners may not be affected by macroeconomic conditions such as inflation, as they 18 may not require housing loans for their home purchases. Medium or above average income households, on the other hand, may be affected by such macroeconomic changes due to their smaller household wealth holdings. Hence, Singapore’s nominal quarterly Gross Domestic Product (GDP) growth rates are included to account for influence of macroeconomic conditions on the housing prices. 3.2.1. OLS Estimations The first empirical model in this thesis is the semi-logarithm OLS model: N 13 11 i=0 t=1 r=1 lnPi = β0 + ∑ β1Z i + ∑ β2T + ∑ β3 R + β4GDPi + e ------------------------------- (1) Pi = Resale transaction price of housing i Zi = Locational and physical attributes of housing i T = 1 if unit sold in period t, where t = {1996... 2008}, 0 otherwise R = 1 if unit sold is located in region r, 0 otherwise GDPi = Singapore nominal quarterly GDP growth when housing i is sold The semi-logarithm specification follows research by Linneman (1980) and Edmonds (1985). This specification also allows for the construction of the housing price index for the comparisons of housing prices across different periods and regions. Several housing studies adopted the repeated-sales hedonic price estimations (Bailey et al., 1963; Case and Shiller, 1989; McMillen, 2003), but this estimation model is not used in this thesis due to the nature of the private housing market in Singapore. In May 1996, to reduce housing market speculation, the Singapore government imposed a capital gain tax on private housing transactions conducted within 3 years since its last transaction. In addition, the 5-year minimum occupancy restriction imposed on ECs will limit the number of repeated private housing transactions available. Both 19 government restrictions indirectly limit the number of repeated-sale transactions available. As the repeated-sales estimation method has already inherent sample biasness, such as poorer quality houses being sold more frequently, these restrictions will only further worsen the possible sample selection bias problem. A detailed discussion on the biasness of repeated-sales estimations in estimating housing price indices can be found in Dombrow, Knight and Sirmans (1997). The locational variables used in the OLS regression include proximity to shopping malls, NEL stations, bus terminals, primary schools and industrial estates, nearest primary school and a within 1 kilometre proximity to a good performance or good progress primary school2. To prevent perfect multi-collinearity, only within 400 metres and between 400 to 800 metres from the nearest NEL stations dummy variables are used in the estimations. In addition to the housing type dummies 3, ‘Condo’ and ‘Apartment’, facility dummies are included to account for betweenestate differences, with considerations made to exclude highly correlated facilities to prevent multi-collinearity 4. Temporal and regional dummies are also included to account for temporal and locational differences between stations 5. The Breusch-Pagan heteroskedasticity test conducted on the log-transformed prices rejects the null hypothesis that the errors terms of housing resale prices are homogeneous. Thus, robust-standard errors are used in all estimations in this study. 2 3 4 5 Definition of good progress and good performance primary schools are taken from Wong (2008). Condominium estates must occupy at least 4,000 sqm, while smaller private estates are classified as apartment estates. Information on estate facilities were taken from the website, Singapore Condo Directory. ‘1995’, ‘Q1’ and ‘Sengkang’ dummies are omitted to avoid perfect multi-collinearity. 20 Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Ho: Constant variance Variables: fitted values of lrprice F(1 , 4662) Prob > F = = 38.60 0.0000 Table 4: Breusch-Pagan Heteroskedasticity Test All estimation results are shown in Appendix 1. An initial model with only the housing attributes was estimated. This basic OLS estimation is able to show that price is negatively correlated with housing prices while height of the apartment and its floor area are positive correlated. All coefficients from this basic estimation are also at least 10% statistical significant. Building from this, locational variables are included to account for how housing accessibility affects its prices. Most variables obtained coefficients with their expected signs. The estimation results show that controlling for other variables, both private condominiums and apartments are generally more expensive than ECs. This could be attributed to the government restrictions on EC purchases that are not applied to private condominium and apartment buyers. For example, the income ceiling restriction on EC purchases excludes high-income households from bidding up EC prices, while such restrictions do not exist for private condominium and apartment purchases in Singapore. The estimations also show that after controlling for other variables, age of the housing reduces its price, while the size of the housing increases it, and that both effects occur at a decreasing rate. Proximity to malls, houses on higher floors, proximity to a bus interchange and having more bus stops within 400 metres increases the private resale housing prices as well. Intuitively, proximity to malls provides access to retail shops, supermarkets and other amenities. Housing units on higher flats also tend to have a better view, which usually leads to a price premium. Access to a bus interchange or more bus stops 21 should also increase housing price, as these public transportations increase the accessibility of the neighbourhood. Proximity to industrial estates had a negative effect on housing prices, and this is commonly linked to the noise and air pollution associated with industrial estates. With Sengkang omitted, the positive coefficients of all regional dummies indicates that after controlling for other attributes, non-landed private resale houses are cheapest at Sengkang. However, the coefficient for Kovan is statistically insignificant. On the other hand, proximity to good performance primary schools did not provide statistically significant results, although their coefficients were positive, while proximity to good progress primary schools provided statistically significant but negative coefficients. The estimation results also show that proximity to NEL stations reduced non-landed resale prices, while the within 400 metres dummy variable is also statistically insignificant. These estimation results suggest that operational and commuter traffic disturbances had a greater negative effect on neighbouring housing prices than the gains in accessibility from the train stations. As the NEL is an entirely underground transit line, any negative effect on neighbouring houses during the operational phase of the NEL would most probably be due to commuter traffic disturbances. The negative coefficients could also be due to the disturbances faced by residents during the NEL’s construction phase. On the other hand, once the constructions are complete and the neighbouring households gain access to the new transit line, the negative construction effects should disappear and neighbouring housing prices may rise to reflect the increased accessibility. As different stages of the NEL development can affect neighbouring housing prices differently, it is postulated that accounting for the different phases can improve the estimation results of the study. 22 3.2.2 OLS Estimations with Interaction Terms To account for the changing phases, year and proximity to station interactions dummies are created and included into Equation 1 to form Equation 2, and it has the following specification: N 13 11 i=0 t=1 r=1 lnPi = β0 + ∑ β1 Z i + ∑ β 2T + ∑ β3 R + β 4 GDPi 13 13 t=1 t=1 + ∑ β5 (T * mrt 4) + ∑ β 6 (T * mrt 8) + e ------------ (2) However, including the interaction terms did not improve the estimation results (Appendix 1). Most variables retain their coefficients similar to those from Equation 1. Proximity to the NEL train stations remain negative, and the between 400 to 800 metres proximity variable now becomes statistically insignificant. The coefficients signs from proximity to primary schools remain unchanged, while the statistically insignificant coefficient for Kovan also remains. A housing price index for the region along the NEL can be constructed to identify the effects of the NEL on neighbouring private housing prices. This nearNEL housing price index can be constructed by using the coefficients of the year and year-proximity dummies in Equation 2. How the housing price indices are calculated from the year and year-proximity dummy coefficients are shown in the Appendix 5. Figure 2 shows the URA non-landed private resale housing price index from 1995 to 2008 for Singapore. The URA property index shows that the movement of the general private resale market matches with the major events that affected the Singapore economy. This includes the housing market decline during the 1997 Asian Financial Crisis and the global economic slowdown in 1999. 23 Figure 2: URA Property Price Index of Non-Landed Residential Properties (Source: REALIS, 2011) 140 120 100 80 60 40 20 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 URA Index ≤ 400m 400-800m ≥ 800m Figure 3: OLS Year-Proximity Price Index 24 Table 5: Price Indices from OLS Estimations (Source: Author’s calculations) The OLS estimations show that private resale prices located within 400 metres, between 400 to 800 metres, and beyond 800 metres from the NEL stations experience similar resale price trends. 1995 is set as the base year of price 100 to allow the comparison of housing price trends with to 1995 pre-announcement levels. As the areas are in close proximity, their estates share many similar public amenities and are affected by the same physical changes to their neighbourhoods. Hence, their similar price movements are intuitive. The housing price index (Table 5) constructed from the second estimation model shows how the different phases of the NEL development project affected neighbouring private housing prices. Firstly, there is evidence of announcement effects from the NEL project on neighbouring housing prices. For 1997 and 1998, near-NEL housing price index were higher than their 1995 pre-announcement levels, reaching 112.53 and 138.69 for the within 400 metres, 101.65 and 115.76 for the between 400 to 800 metres, and 108.80 and 140.64 for beyond 800 metres. This price 25 increase goes against the general private resale housing market trend in Singapore. Since the price increase after the 1996 announcements, the price increase is highly possible to be due to the NEL announcement. From 1999 to 2000, within 400 metres index dropped by 2% to 5% from pre-announcement levels, between 400 to 800 metres resale prices dropped by about 36% in 1999 and about 20% in 2000 against 1995 prices. This matches the NEL construction phase, where heavy tunnelling works for the NEL project only started in late September 1998 (Straits Times, 1998). In addition, beyond 800 metres housing price index did not go below the 1995 preannouncement levels, suggesting that noise and pollution from the NEL construction affected on neighbouring households. Finally, the booming Singapore private housing market in 1999 and 2000 further strengthens the hypothesis that the decline in the near-NEL price indices was not due to macroeconomic conditions, but probably due to the NEL construction disturbances. From 2001 onwards, within 400 metres resale prices were generally higher than the 1995 price levels by 6-7%, while between 400 to 800 metre resale prices were still lower than their 1995 resale prices. This supports the argument that proximity to train stations provides a positive housing price premium. This however, goes against the maximum walking distance estimated by Wibowo and Olszewski (2005). In addition, in 2001, 400 metre resale prices were already higher than preannouncement levels, even though the NEL was scheduled to open in 2003. Two reasons from the NEL could be attributed to this effect: decrease in construction disturbances and anticipation for the future train station. Unfortunately, it is impossible to separate these two effects, and both effects could exist together. The estimation results also show that within 400 metre resale prices after the NEL became 26 operational were lower than post-announcement levels, but higher than the construction phase price levels. In other words, this suggests that the NEL announcement had a greater impact on neighbouring housing prices than when it became operational. This suggests that households over reacted to the announcement NEL. However, looking at the general Singapore housing market from 2003, the low near-NEL housing prices could also be due to the weak housing market conditions, as seen from the flat URA price index in Figure 3. One issue observed from the price index constructed in this thesis is that houses beyond 800 metres from the NEL stations experienced greater price increases than those within 400 metres after the NEL became operational. This goes against intuition that being closer to a train station increases the household’s accessibility, and in turn, increases its price. One explanation could be that private houses are occupied by middle or high income households who are likely to own cars. Hence, being near a train station may not increase the accessibility for these households who may choose to retain their automobile lifestyle and not utilise the new train service. Another possible explanation is that the OLS estimation does not fully capture the nature of the effects of the NEL project on the housing market. As mentioned in Chapter 2, the inherent geographical nature of such housing market studies may exhibit spatial heterogeneity and spatial autocorrelation that will lead to a model misspecification if estimated by an OLS model. Hence, there is a need to adopt a spatial estimation model to account for these spatial issues in the estimation model. In this thesis, Can’s (1990, 1992) spatial expansion method will be used to tackle the issues of spatial heterogeneity and spatial autocorrelation. 27 3.3. Can’s Spatial Expansion Model Can’s (1990, 1992) spatial expansion model is actually a direct application of Casetti’s (1972) expansion method. The advantage of this spatial model is that it requires no intensive computations of the spatial weight matrices necessary for other spatial regression models. The expansion method allows marginal effects within variables to vary across the geographical space in a continuous manner. To apply the spatial expansion, an ‘initial’ model is first established and expanded by ‘expansion equations’ to form a ‘terminal’ model that can be estimated. In this study, Equation 2 forms the ‘initial’ model that relates the locational and housing attributes to the housing price. The ‘expansion equations’ are the interaction terms of several dependent variables and the spatial variables of the housing units. Spatial coordinates of the private housing estates, taken from Google Earth, are used to measure spatial heterogeneity. For convenience of the analysis, housing units from the same private estates are allocated the same spatial coordinates. Although higher polynomial interaction terms can be included to account for non-linear effects of spatial heterogeneity, extreme multi-collinearity limited the terminal model to firstorder polynomial expansions. Since the effects of the NEL on resale housing prices is this study’s primary focus, year and year-proximity dummies are not spatially expanded to allow easier interpretations. This however, implicitly assumes that these variables exhibit no spatial heterogeneity. As both regional dummies and spatial coordinates are highly correlated, regional dummies are excluded from the spatial expansion model to prevent the problem of multi-collinearity. As the NEL covers more northern than eastern parts of Singapore (Figure 1), only longitude coordinates are used to account for the spatial heterogeneity to reduce the number of interaction 28 terms and simplify the estimation of the spatial expansion model. To account for spatial autocorrelation, an autoregressive spatial lag variable is also included into the spatial expansion model. However, the presence of the autoregressive variable will make its estimations biased and inconsistent if OLS estimations are used. Instead, maximum likelihood estimations (MLE) are used to estimate the spatial expansion model (Cliff and Ord, 1982; Anselin, 1988; Can, 1992). The spatial expansion model estimated in this study then has the following specifications: N J  13 lnPit = β0 + ∑ ( β k0 + β k1 A) Z k + β3 ln ∑ Pj ( t −1)  + ∑ β3T i=0  j=0  t=1 13 13 t=1 t=1 + ∑ β 4 (T * mrt 4) + ∑ β5 (T * mrt 8) + e ---------- (3) A represents the spatial component that accounts for spatial heterogeneity. In this model, this spatial component is the longitude coordinates of the different housing J  units sold. ln ∑ Pj ( t −1)  is the sum of all private resale housing transaction prices  j=0  within 1.3 kilometres radius from resale housing transaction i in the past 6 months, and this variable accounts for the spatial autocorrelation of the spatial data. This distance and time period of spatial autocorrelation was chosen to allow the private residential estates to be easily matched to their own zones for the convenience of data manipulation. In total, 85 resale housing transactions had no such lagged observations, and these transactions were dropped from the spatial expansion model, leaving a total of 4,621 valid housing transaction observations in the spatial expansion model. 3.3.1. Spatial Expansion Model Estimations The spatial expansion estimations show that within 400 metres resale prices have a smaller discount over between 400 to 800 metres resale prices (Appendix 1), 29 suggesting that closer proximity to train stations lead to higher private residential houses prices. However, unlike the OLS estimations, the spatial expansion estimations show that the NEL had a positive impact on the proximity zone of between 400 to 800 metres, which fits more closely to the empirical findings by Wibowo and Olszewski (2005). The general price movement trends of the near-NEL resale prices of both estimation models are also similar, with announcement, construction and operational phase effects observed from the spatial expansion model as well. However, the spatial expansion model showed that near-NEL resale prices were generally higher than the 1995 pre-announcement levels, which is a different result from the OLS estimations. Resale housing prices in 1998 for within 400 metres, between 400 to 800 metres and beyond 800 metres were 49.74%, 80.06% and 37.03% greater the 1995 preannouncement levels respectively. The lowest resale prices estimated from the spatial expansion model, which was observed during the NEL’s heavy construction phase in 1999, were still 6-7% above the 1995 pre-announcement prices. Hence, unlike the OLS estimations, although construction disturbances did reduce neighbouring nonlanded housing prices in comparison to the post-announcement levels, but not below pre-announcement levels. The increase in near-NEL resale housing prices in the periods just before the NEL’s opening was also observed from the spatial expansion model, and its effects are larger than those from the OLS estimations. From 2001 to 2004, within 400 metres resale prices remained slightly over 20% above pre-announcement levels. Between 400 to 800 metres resale prices were about 50-40% above pre-announcement levels. Beyond 800 metres resale prices were also more than 10% above pre30 announcement levels. Post-announcement near-NEL resale prices were also higher than the resale prices when the line became operational (Figure 4). This suggests that consumers were overly optimistic about the NEL during its announcement phase, causing private home buyers at near-NEL regions to accept higher resale prices that were unmatched when the NEL became operational in 2003. However, as mentioned earlier, the stagnating growth of near-NEL housing prices from 2003 to 2006 could be due to the poor housing market conditions in Singapore during the same period. 180 160 140 120 100 80 60 40 20 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 URA Index ≤ 400m 400-800m ≥ 800m Figure 4: Spatial Model Price Index 31 Table 6: Price Indices from the Spatial Model (Source: Author’s calculations) 4. CONCLUSION The positive effects on near-NEL non-landed private resale housing prices during 1997 and 1998 from the announcement of the NEL in January 1996 suggests that non-landed Singapore private home-buyers are forward-looking and appreciative of the gains in accessibility from the future transit line. However, during the NEL’s construction phase, construction disturbances led to a huge decline in neighbouring housing prices. These price fluctuations for the private housing market along the NEL development suggests that private home buyers overly focused on the benefits of the future transit line when its development was announced, and failed to consider the construction disturbances. In addition, the announcement of the NEL development had a greater positive effect on neighbouring housing prices than when the line officially opened, which reinforces the notion that private home buyers over-reacted to the news about the future NEL. This hypothesis of an irrational housing market has also been brought up by studies that apply capital asset pricing models (CAPM) to the 32 study of housing markets (Clayton, 1996, 1997; Meese and Wallace, 1994), although these studies have not ruled out that a more robust model with no assumptions of a frictionless asset-based model and perfect information may prove otherwise. The presence of significant positive announcement effects and negative construction effects on neighbouring housing prices from the NEL project observed in this thesis are consistent with other hedonic housing studies. However, one must be careful in assuming the transferability of these results to other housing markets, because differences in economic and social conditions will affect how households react to transit developments, thereby influence the price movements in the affected housing markets. For example, contrary to the findings in this thesis, no significant positive announcement effects were observed from the Singapore Circle Line transit development (Chia, 2008/2009; Wu, 2007/2008). One reason was that since the Circle Line was constructed after NEL, private home buyers became more aware about the disturbances from a transit line construction from the NEL, and adapted their pricing behaviour accordingly for the Circle Line. Several extensions can be made to this thesis. Spatial tests could be used to identify the proper spatial model necessary for the data, thereby providing a more scientific justification of the choice of spatial model. The use of other spatial models can also help test the validity of this study’s results. Other spatial weights can also be considered to test the robustness of this study’s findings under different spatial interaction assumptions. Future research can consider how the strong land use planning practices by the Singapore government and its rapid transit developments affect Singapore’s overall housing price gradient. 33 REFERENCE Alonso, W. (1964) Location and Land Use. Harvard University Press. Cambridge Anselin, L. (1988) Spatial Econometrics: Methods and Models. Boston, Kluwer Academic Publishers Anselin, L. (2001) Spatial Econometrics. In: B. Baltagi, Editor. A Companion to Theoretical Econometrics. Blackwell. Oxford Bae, C. H. C., Jun, M. and Park, H. (2003) The Impact of Seoul's subway Line 5 on residential property values. Transport Policy, 10, 85 – 94 Bailey, M. J, Muth, R. F. and Nourse, H. O. (1963) A Regression Method for Real Estate Price Index Construction. Journal of the American Statistical Association, 58, 933 – 942 Bardhan, A. D., Datta, R., Edelstein, R. H. and Lum, S. K. (2003) A tale of two sectors: upward mobility and the private housing market in Singapore. Journal of Housing Economics, 12, 83 – 105 Bender, B. and Hwang, H. (1985) Hedonic Housing Price Indices and Secondary Employment Centers. Journal of Urban Economics, 17, 90 – 107 Bitter, C., Mulligan, G. F. and Dall’erba, S. (2007) Incorporating spatial variation in housing attribute prices: a comparison of geographically weighted regression and the spatial expansion method. Journal of Geographical System, 9, 7 – 27 Brueckner, J. K. (1979) A Model of Non-Central Production in a Monocentric City. Journal of Urban Economics, 6, 444 – 463 Brunsdon, C., Fotheringham, S. and Charlton, M. (1996) Geographically Weighted Regression: A method for exploring spatial nonstationarity. Geographical Analysis, 28(4), 281 – 298 34 Can, A. (1990) The Measurement of Neighborhood Dynamics in Urban House Prices. Economic Geography, 66(3), 254 – 272 Can, A. (1992) Specification and Estimation of Hedonic Housing Price Model. Regional Science and Urban Economics, 22, 453 – 474 Casetti, E. (1972) Generating Models by the Expansion Method: Applications to Geographical Research. Geographical Analysis, 4, 81 – 91 Chan, L. Y. (2004/2005) Impact of North East MRT Line on Resale Price of HDB Flats. (Unpublished undergraduate dissertation) Department of Real Estate, National University of Singapore, Singapore Chia, L. S. (2008/2009) Impact of Existing and Improved MRT Systems on Public Residential Property Values. (Unpublished undergraduate dissertation) Department of Real Estate, National University of Singapore, Singapore Chung, Y. T. and Chan, W. H. (2003) Estimating the commuting cost and commuting time property price gradients. Regional Science and Urban Economic, 33, 745 – 767 Clayton, J. (1996) Rational Expectations, Market Fundamentals and Housing Price Volatility. Real Estate Economics, 24(4), 441 – 470 Clayton, J. (1997) Are Housing Price Cycles Driven by Irrational Expectations? Journal of Real Estate Finance and Economics, 14(3), 341 – 363 Cleveland, W. S. and Devlin, S. J. (1988) Locally Weighted Regression: An Approach to Regression Analysis by Local Fitting. Journal of the American Statistical Association, 83, 596 – 610 Cleveland, W. S., Devlin, S. J. and Grosse, I. (1988) Regression by Local Fitting: Methods Properties and Computational Algorithms. Journal of Econometrics, 37, 87 – 144 35 Cliff, A. D. and Ord, J. K. (1981) Spatial Processes: Models & Applications. London: Pion Limited Damm, D., Lerman, S. R., Lerner-Lam, E. and Young, J. (1980) Response of Urban Real Estate Values in Anticipation of the Washington Metro. Journal of Transport Economics and Policy, 14(3), September, 315-336 Department of Singapore Statistics (2011) Key Annual Indicators. Retrieved November 30, 2010, from http://www.singstat.gov.sg/stats/keyind.html#popnarea Dipasquale, D. and Wheaton, W. C. (1996) Urban Economics and Real Estate Markets. Englewoods Cliffs, NJ: Prentice-Hall. Dombrow, J., Knight, J. R. and Sirmans, C. F. (1997) Aggregation Bias in RepeatedSales Indices. Journal of Real Estate Finance and Economics, 14, 75 – 88 Edmonds, R. G. Jr (1985) “Some evidence on the intertemporal stability of hedonic price functions. Land Economics, 61(4), 445 – 449 Forrest, D., Glen, J. and Ward, R. (1996) The Impact of a Light Rail System on the Structure of House Prices: A Hedonic Longitudinal Study. Journal of Transport Economics and Policy, 30(1), January, 15 – 29 Gatzlaff, D. H., and Smith, M. T. (1993) The impact of the Miami Metrorail on the value of residences near station locations. Land Economics, 69(1), 54 – 66 Google Earth (2011) Retrieved November 30, 2010, from http://www.google.com/earth/index.html Google Maps (2011) Retrieved November 30, 2010, from http://maps.google.com/ Han, S. S. (2005) Polycentric urban development and spatial clustering of condominium property values: Singapore in the 1990s. Environment and Planning A, 37, 463 – 481 36 HDB InfoWEB (2011) Housing & Development Board. Retrieved November 30, 2010, from http://www.hdb.gov.sg/ Heikkila, E., Gordon, P., Kim, J. I., Peiser, R. B. and Richardson, H. W. (1989) What Happened to the CBD - Distance Gradient?: Land Values in a Policentric City. Environment and Planning A, 21, 221 – 232 Heneberry, J. (1998) Transport Investments and house prices. Journal of Property Valuation and Investments, 16(2), 144 – 158 Karamjit, K. (1998, 8th February) Building of N-E MRT line - Serangoon to see detours, more parking. Straits Times. Knight, J. R., Dombrow, J. and Sirmans, C. F. (1995) A Varying Parameters Approach to Constructing House Price Indexes. Real Estate Economics, 23(2), 187 – 205 Lancaster, K. J. (1966) A New Approach to Consumer Theory. The Journal of Political Economy, 74(2), 132 – 157 Landsberger, M. and Lidgi, S. (1978) Equilibrium rent differentials in a multi-center city. Regional Science and Urban Economics, 8, 105 – 116 Lee, Y. J. (2009/2010) A Comparison of the Impacts of MRT Attributes on Property Values. (Unpublished undergraduate dissertation) Department of Real Estate, National University of Singapore, Singapore Leong, C. T. (1996, 3rd February) N-E MRT Line - What Will Happen Before The Trains Roll. Straits Times Leong, C. T. (2003) Getting there: The story of the North East line Singapore. Land Transport Authority 37 Linneman, P. (1980) Some empirical results on the nature of hedonic price function for the urban housing market. Journal of Urban Economics, 8, 47– 68 Lum, S. K. (2002) Market Fundamentals, Public Policy and Private Gain: House Price Dynamics in Singapore. Journal of Property Research, 19(2), 121 – 143 Madden J. F. (1980) Urban Land Use and the Growth in Two-Earner Households. Papers and Proceedings of the Ninety-Second Annual Meeting of the American Economic Association. The American Economic Review, 70(2), 191 – 197 McMillen, D. P. (2003) The return of centralization to Chicago: using repeat sales to identify changes in house price distance gradients. Regional Science and Urban Economics, 33, 287 – 304 McMillen, D. P., and McDonald, J. (2004) Reaction of house prices to a new rapid transit line: Chicago’s Midway Line, 1983–1999. Real Estate Economics, 32, 463 – 486 Meese, R. and Wallace, N. (1991) Nonparametric Estimations of Dynamic Hedonic Price Models and the Construction of Residential Housing Price Indices. Journal of the American Real Estate & Urban Economics Association, 19(3), 308 – 332 Muth, R. (1969) Cities and Housing. University of Chicago Press, Chicago Ong, S. E. (2001) Price Effect of MRT Proximity on HDB Resale Flats. CRES Times, 1(1), Centre for Real Estate Studies, Department of Real Estate, School of Design and Environment, National University of Singapore, 9 Ong, S. E. and Sing, T. F. (2002) Price Discovery between Private and Public Housing Markets. Urban Studies, 39(1), 57 – 67 Ong, S.E., Ho, K. H. and Lim, C. H. (2003) A Constant-quality Price Index for Resale Public Housing Flats in Singapore. Urban Studies, 40, 2705 – 2729 38 Phang, Y. S and Wong, W. K. (1997) Government Policies and Private Housing Prices in Singapore. Urban Studies, 34(11), 1819 - 1829 Quek, K. W. (2004/2005) Price Movement of Residential Properties in the Proximity of Northeast MRT Line. (Unpublished undergraduate dissertation) Department of Real Estate, National University of Singapore, Singapore Real Estate Information System (2011) Urban Redevelopment Authority of Singapore. Retrieved November 24, 2010, from https://spring.ura.gov.sg/lad/ore/login/index.cfm Romanos, M. C. (1977) Household Location in a linear multi-center metropolitan area. Regional Science and Urban Economics, 7, 233 – 250 Rosen, S. (1974) Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition. Journal of Political Economy, 82, 34 – 55 Sing, T. F., Tsai, I.C. and Chen, M.C. (2006) Price dynamics in public and private housing markets in Singapore. Journal of Housing Economics 15, 305 – 320 Singapore Condo Directory (2011) Retrieved December 1, 2010, from http://condo.singaporeexpats.com/apartment-condo.htm Singapore Street Directory, 1996 – 2005, Land Transport Authority Straits Times (1998, 24 September) Tunnel works for N-E line begin. Tan, P. J. (2009/2010) The Influence of an MRT Station on Residential Property Value: A “Timeline” Perspective. (Unpublished undergraduate dissertation) Department of Real Estate, National University of Singapore, Singapore Thériaut, M., Des, R.F., Villeneuve, P. and Kestens, Y. (2003) Modelling interactions of location with specific value of housing attributes. Property Management, 21( 1), 25 – 62 39 Upton, G. J. and Fingleton, B. (1985) Spatial Data Analysis by Example. Volume 1: Point Pattern and Quantitative Data. Wiley. New York. Waddell, P., Berry, J. L. and Hoch, I. (1993) Residential property values in a multimodal urban area: new evidence on the implicit price of location. Journal of Real Estate Finance and Economics, 7, 17 – 141 Wei, S. D. and Wong, W. K. (2010) The Political Economy of Housing Prices: Hedonic Pricing with Regression Discontinuity. Journal of Housing Economics, 19(2), 133 – 144 Wibowo, S. S. and Olszewski, P. (2005) Modelling walking accessibility to public transport terminals: Case study of Singapore Mass Rapid Transit. Journal of the Eastern Asia Society for Transportation Studies, 6, 147 - 156 Wong, M. (2008) Estimating Ethic Preferences Using Ethnic Housing Quotas in Singapore. University of Pennsylvania (Unpublished Manuscript) Wong, W. K. (2008) Parental Valuation of Priority Admission in Elementary Schools: The Effects of Academic Reputation and Choices. National University of Singapore, Department of Economics, Working Paper. Wu, L. (2007/2008) Impact of Circle Line on HDB Resale Flat Prices. (Unpublished undergraduate dissertation) Department of Real Estate, National University of Singapore, Singapore Yinger, J. (1992) “City and Suburb: Urban Models with More than One Employment Center. Journal of Urban Economics, 31, 181 – 205 40 APPENDIX 1: ESTIMATION RESULTS Variable Age of flat Age of flat2 Floor area Floor area2 Between levels 6 to 10 Between levels 11 to 15 Between levels 16 to 20 Level 21 and above Condo Apartment Industrial Estate (≤400m) Bus Terminal (≤400m) -0.0043*** (0.00024) 6.36 × 10-7*** (4.89 x× 10-7) 0.0074*** (0.00034) -5.31 × 10-6*** (1.0 × 10-6) 0.0203*** (0.00548) 0.0320*** (0.00654) 0.0441*** (0.00683) 0.0501*** (0.0138) 0.0856 (0.073) 0.280*** (0.071) -0.624*** (0.0567) 0.395*** (0.104) 0.182*** (0.0264) 0.225*** (0.0264) -0.204*** (0.0232) -0.231*** (0.0274) 0.0245*** (0.00452) -0.0192 (0.0221) Year Proximity OLS (3) -0.0041*** (0.00028) 5.8 × 10-6*** (5.93 × 10-7) 0.0074*** (0.00035) -5.3 × 10-6*** (1.0 × 10-6) 0.0190*** (0.00534) 0.0310*** (0.00639) 0.0426*** (0.00684) 0.0516*** (0.0314) 0.0562 (0.763) 0.304*** (0.0763) -0.662*** (0.0582) 0.329*** (0.11) 0.160*** (0.0285) 0.222*** (0.0277) -0.245*** (0.0335) -0.0937 0.066 0.0241*** (0.00454) -0.0083 (0.023) 0.106*** (0.00939) -2.51 × 10-5*** (2.32 × 10-5) -0.0574*** (0.0109) 2.8 × 10-4*** (3.55 × 10-5) 0.0112** (0.0053) 0.0223*** (0.0062) 0.0340*** (0.0068) 0.0250* (0.0316) 13.3535*** (4.088) 5.8694 (4.285) -35.7565*** (8.07) -0.0458 (0.321) 1.9539 (1.576) 28.6811*** (6.407) -1.1714*** (0.155) -1.3500*** (0.179) -0.2946 (0.506) -0.1826*** (0.0502) - 0.0173 (0.0342) 0.0132 (0.0336) -0.7229*** (0.164) - -0.308*** (0.0443) -0.339*** (0.0436) 0.3410*** (0.0931) Basic OLS (1) -0.00345*** (0.00021) 3.9× 10-6*** (4.4× 10-7) 0.0078*** (0.00035) 5.45 × 10-6 (1.02× 10-6) 0.0209*** (0.00585) 0.0348*** (0.0071) 0.0461*** (0.00963) 0.0511*** (0.0144) 0.0821* (0.0388) 0.0874* (0.0254) - Mall (≤400m) - Mall (400800m) - MRT (≤400m) - MRT (400800m) Number of bus stops (≤400m) Primary school (≤400m) Good Performance School (≤1km) Good Progress School (≤1km) - Locational OLS (2) Spatial model (4) 41 Singapore’s Quarterly GDP (%) - 0.0102*** (0.00112) 0.0104*** (0.00114) 0.0095*** (0.0014) Constant 12.39*** (0.0516) 11.6*** (0.17) 11.6*** (0.174) 13.5*** (0.344) Yes Yes Yes Yes No Yes Yes Yes No No Yes Yes No Yes Yes No No No No Yes No No No Yes Estate Amenities 6 Year Fixed Effects (Year*MRT) Effects Regional Fixed Effects Spatial Expansion Effects Spatial lag N F R2 Adjusted R2 RMSE Chi2 6 4706 4706 4621 530.6589 410.2091 0.8754 0.8814 0.8739 0.8792 0.1511 0.1479 4.24 × 104 Legend: * p< 0.1; **p< 0.05; *** p< 0.01 Estate facilities include MPH, gym, tennis court, clubhouse, sauna, jacuzzi, exercise facilities, entertainment room, garden, playground and function room. 42 APPENDIX 2: TEMPORAL EFFECTS OF THE NEL Year 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Basic OLS Model -0.0562*** (0.0154) 0.0953*** (0.0189) 0.3106*** (0.0294) -0.0191 (0.0175) 0.0072 (0.0224) 0.1102*** (0.0235) 0.0792*** (0.0207) 0.0857*** (0.0219) -0.0124 (0.0217) -0.0547** (0.0217) -0.0656*** (0.0217) -0.0103 (0.0235) 0.2751*** (0.0235) ≤400m 0.0138 (0.0308) 0.0338 (0.0378) -0.0136 (0.0536) -0.0766*** (0.0285) -0.1463*** (0.0446) -0.1587*** (0.0421) -0.0165 (0.0414) -0.0416 (0.0362) 0.0137 (0.039) -0.0043 (0.0365) 0.0537* (0.0326) 0.0877** (0.038) 0.0993*** (0.0357) Year-Proximity OLS Model 400-800m -0.1832*** (0.0704) -0.0679 (0.0955) -0.1944* (0.113) -0.5017*** (0.0824) -0.3213*** (0.0911) -0.2949*** (0.0903) -0.1494 (0.0909) -0.1359 (0.0926) -0.1273 (0.0194) -0.2448*** (0.0878) -0.2163** (0.0878) -0.1224 (0.0871) -0.1318 (0.0874) ≥800m -0.0486*** (0.0171) 0.0843*** (0.0277) 0.3407*** (0.0303) 0.0628*** (0.0164) 0.0992*** (0.0201) 0.2352*** (0.0286) 0.0908*** (0.0241) 0.1059*** (0.0245) 0.0003 (0.0242) 0.0024 (0.0246) -0.0384 (0.024) -0.0349 (0.0258) 0.2576*** (0.0258) ≤400m 0.0458 (0.0333) 0.0749* (0.0423) 0.0889* (0.0535) 0.0507 (0.0361) -0.0105 (0.0464) -0.0254 (0.0569) 0.1101** (0.0486) 0.0853* (0.0491) 0.2490*** (0.0498) 0.2959*** (0.0501) 0.3660*** (0.0494) 0.3808*** (0.0523) 0.3979*** (0.0519) Spatial expansion model 400-800m ≥800m 0.1442** -0.0658*** (0.0622) (0.0198) 0.2029*** 0.0875*** (0.0721) (0.0322) 0.2733* 0.3149*** (0.149) (0.0342) 0.0522 0.026 (0.0771) (0.0218) 0.1351* 0.0706*** (0.0742) (0.249) 0.2229*** 0.2306*** (0.073) (0.038) 0.3031*** 0.1056*** (0.0697) (0.0297) 0.2892*** 0.1226*** (0.0715) (0.0335) 0.3630*** -0.0166 (0.0701) (0.0332) 0.2838*** -0.0217 (0.0684) (0.0338) 0.2782*** -0.0677* (0.0669) (0.0347) 0.3517*** -0.0623* (0.0674) (0.0368) 0.3342*** 0.2446*** (0.0687) (0.0382) 43 APPENDIX 3: CONSTRUCTION OF HOUSING PRICE INDEX The construction of the price index for the semi-log models with year-proximity variables uses the following formula: 100* exp [βi + βt + βti], where βi, βt, βti are the coefficients for the year, proximity, and year-proximity respectively. Example of Price Index Construction (within 400 metres, year-proximity) Year 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Coefficient -0.048576 0.084255 0.340719 0.062797 0.099239 0.235192 0.090789 0.105854 0.000299 0.002387 -0.038437 -0.034915 0.257608 mrt4 -0.244663 -0.244663 -0.244663 -0.244663 -0.244663 -0.244663 -0.244663 -0.244663 -0.244663 -0.244663 -0.244663 -0.244663 -0.244663 -0.244663 Year-Proximity Dummy 0.013817 0.033786 -0.013627 -0.076621 -0.146256 -0.158722 -0.016505 -0.041552 0.013728 -0.004318 0.053700 0.087687 0.099331 Exponential 0.78297 0.75622 0.88107 1.08592 0.77222 0.74701 0.84519 0.84334 0.83497 0.79403 0.78146 0.79501 0.82540 1.11882 Normalised 100.00 96.58 112.53 138.69 98.63 95.41 107.95 107.71 106.64 101.41 99.81 101.54 105.42 142.89 44 [...]... announcement of the Pioneer MRT station’s construction had a greater positive effect on HDB resale prices than when the station became operational With the rapid development of transit lines in Singapore, hedonic housing studies have also focused on how specific transit developments in Singapore affect housing prices within their close vicinity Ong (2001) found that HDB resale flats nearer to the East- West transit. .. station 1.3 Non-OLS Hedonic Price Models Despite the common usage of the OLS model in many hedonic price studies of housing markets in both Singapore and other regions, other estimation methods have been created to improve on the OLS framework Unfortunately, few of these estimations have been used in the Singapore context Meese and Wallace (1991) adapted the non-parametric locally weighted regression (LWR)... improve the estimation results Following the definitions of Anselin (2001), the pure space-recursive approach is used to account for the effects of the spatial autocorrelation 2 NEL PROJECT OVERVIEW To meet the commuting needs of the Singapore population, considerations to construct a transit line to serve the North- East residents of Singapore were made by its government in as early as 1984 (Leong, 2003)... 2003 There were also no 99-year non-landed private or EC resale transactions near the Potong Pasir and Punggol stations during this period The initial selection of the resale transaction prices were based on the zoning from the REALIS website In total, 4,706 non-landed private resale housing transactions surrounding 12 NEL train stations were downloaded from the REALIS website These include the stations... in region r, 0 otherwise GDPi = Singapore nominal quarterly GDP growth when housing i is sold The semi-logarithm specification follows research by Linneman (1980) and Edmonds (1985) This specification also allows for the construction of the housing price index for the comparisons of housing prices across different periods and regions Several housing studies adopted the repeated-sales hedonic price estimations... its construction phase is expected to adversely affect neighbouring households, and cause neighbouring housing prices to drop Lastly, the commencement of the NEL project should increase neighbouring housing prices due to the gains in accessibility for the train line 3.2 Empirical Data The majority of hedonic price analyses done on the Singapore housing market focuses on its public housing Firstly, the. .. residential housing prices, in addition to establishing it from the locational and physical attributes of the house, realtors commonly factor in the prices of neighbouring houses that were sold recently This practice leads to the interactions of housing prices across geographically close locations, and leads to the issue of spatial autocorrelation Anselin (2001) discussed the estimations of different... that included trains and signalling systems” (Leong, 2003, p 31) Due to the 1997 Asian financial crisis, the NEL constructions were delayed till November 1997 The complexity of the NEL project constructions also led to many other issues As the NEL was to connect the North- East regions of Singapore to the central locations (Figure 1), constructions were done through several densely populated areas, including... operational phase of the NEL would most probably be due to commuter traffic disturbances The negative coefficients could also be due to the disturbances faced by residents during the NEL’s construction phase On the other hand, once the constructions are complete and the neighbouring households gain access to the new transit line, the negative construction effects should disappear and neighbouring housing. .. and expanded by ‘expansion equations’ to form a ‘terminal’ model that can be estimated In this study, Equation 2 forms the ‘initial’ model that relates the locational and housing attributes to the housing price The ‘expansion equations’ are the interaction terms of several dependent variables and the spatial variables of the housing units Spatial coordinates of the private housing estates, taken from ... from the construction of the NEL are of concern to residents along the alignment This thesis attempts to identify the effects of the NEL on neighbouring private resale housing prices A hedonic price. .. area of urban management is the mobility of its residents within the city-state The construction of the North- East Line (NEL) transit system in 2003 is part of the development of a Rapid Transit. .. Construction of Housing Price Index 44 iii SUMMARY This thesis seeks to investigate the effects of the North- East Line (NEL) Mass Rapid Transit (MRT) extension on neighbouring private housing