The Effect of Localized Density on Housing Prices in Singapore

Transcription

1 The Effect of Localized Density on Housing Prices in Singapore June 13, 2017 Eric Fesselmeyer a,, Kiat Ying Seah b, Jonathan Ci Yi Kwok a Department of Economics and IRES, National University of Singapore b Department of Real Estate and IRES, National University of Singapore Abstract This paper measures how localized residential density impacts housing prices in Singapore. Using exogenous variation in residential density, we find that an increase in density causes non-trivial decreases in property values: a 10% increase in density decreases price per square foot by between 1.3% and 2%. To the best of our knowledge, ours is the first paper to measure the causal effect of localized density. The finding that residents prefer living in less dense environments carries important policy implications. For example, our findings suggest that land use restrictions such as anti-sprawl measures that increase urban density as a by-product may have unattended negative effects on welfare. Keywords: Density, Willingness-to-pay, Land-use policy JEL Classifications: R20, R21, R38 Corresponding author. ecsef@nus.edu.sg, Tel: (65)

2 1. Introduction As urban populations grow, households live in denser buildings and housing developments. Higher density means that shared facilities and common spaces are more crowded, there are longer waits for elevators, and there is more interaction with neighbors and less privacy. 1 Surprisingly, while there is a vast literature on a diverse range of topics related to density, 2 to the best of our knowledge, there are no papers that attempt to estimate the causal effect of this type of localized density on price. 3 In this paper, we estimate the effect of project density on price in the secondary-market for private apartments in Singapore. A project is a collection of adjacent apartment buildings that share a land parcel, a name (e.g., The Anchorage ), and facilities. Our measure of project density is the number of apartments in a project per acre of land. We find that an increase in project density causes price to decrease. Our conclusion that localized density negatively impacts welfare carries important policy implications as almost all cities regulate density in one way 1 It has been argued that overcrowding and lack of privacy promotes alienation and pathological behavior (Durkheim, 1951). The well-cited Calhoun (1962) report finds that rats kept in a high population density environment exhibit increased aggressive behavior, disruption of mating and nesting behavior. Experimental crowding studies on people have found that crowding acts as a stressor, increasing blood pressure and heart rate and lowering tolerance for frustration (Evans, 1979). 2 For example, Ciccone and Hall (1996), Glaeser and Mare (2001), and Ahlfeldt et al. (2015) consider the relationship between density and wages and productivity. Glaeser et al. (2001) considers the relationship between density and urban amenities. Brueckner and Largey (2008) considers how density affects social interaction. 3 Lee (2016) and Tang (2010) study a similar question but do not address the endogeneity of density. 1

3 or another. Consider, for example, land use restrictions such as anti-sprawl measures that are intended to minimize urban expansion but also tend to increase urban density. Our findings suggest that the positive effects of any such policy measures are at least partially offset by the negative effects on welfare of increase density. In a spatial equilibrium, the price of otherwise identical housing at a given location that has more of a negative amenity, such as density, will be lower to just compensate buyers and render them indifferent between properties. This difference in price due to different amounts of the attribute, or, in other words, the implicit price of the attribute, can be estimated using a hedonic regression. The challenge in estimating the implicit price of project density consistently is that unobserved demand factors render project density endogenous: developers find it profitable to provide more apartments in locations that are in high demand and thus have a high price per square foot. 4 We address this endogeneity by instrumenting for project density using Singapore s plot ratio allowance. Equivalent to Floor Area Ratio (FAR) restrictions, the plot ratio allowance restricts the amount of floor area per land area. Plot ratio allowance is clearly correlated with project density as a greater plot ratio allowance allows developers to build more apartments. 5 4 Cai et al. (2016), for example, find that developers in China exceed land-use restrictions in more attractive locations. Some other papers that study land-use regulations include Brueckner et al. (2017), Fu and Somerville (2001), and Glaeser et al. (2005). 5 A profit-maximizing developer trades off the negative effect of increased density on 2

4 The same unobserved demand factors that render OLS inconsistent may also affect plot ratio allowances. We minimize this threat to identification by utilizing neighborhood fixed effects, which capture time-constant neighborhood-level unobservables. Our identification scheme relies, then, on the exogenous variation in plot ratio allowance within neighborhoods. Given the size of each neighborhood we consider, on average about 20 acres, it seems unlikely that there is enough variation in demand to cause planners to vary plot ratio allowance for this reason. In fact, at least part of this variation of plot ratio allowance in nearby locations is unrelated to demand and instead stems from Singapore s use of checkerboard planning in which high-density developments are interspersed with low-rise developments to create structural heterogeneity and a more interesting and varied cityscape (Centre for Liveable Cities, 2013; Cheong-Chua Koon Hean, 2012). One important and widespread implementation of this approach occurred under the Central Area Structure Plan of the early 1980s, which specifically focused on interspersing developments of varied density that would contribute to the visual diversity of the overall physical form and skyline of the city (Chor and Heng, 2017; Urban Redevelopment Authority, 1993, 2011). 6 price per square foot versus the gain of selling more apartments. With no plot ratio restrictions, a developer chooses the number of apartments such that the profit from an additional apartment is just off-set by the negative effect of the apartment on price per square foot. A restrictive plot ratio allowance means that a developer never reaches the optimal density and is constrained at a density such that adding apartments is profit-increasing. Variation in plot ratio allowances thus results in variation in density (DiPasquale and Wheaton, 1996). 6 Another example is a scheme in the early 1990s in which Urban Redevelopment Au- 3

5 Even though we argue that variation in plot ratio allowance is plausibly exogenous given our identification strategy, there could always be unobserved demand factors that we failed to recognize that influenced planners choice of plot ratio allowance at the project level. Consider, for example, a location that has a particularly attractive unobserved attribute compared to other locations within the same neighborhood. Planners may have assigned a greater plot ratio allowance for this location, which ultimately increases density. This would introduce a positive bias in our estimates. Given that we find a negative density effect, the direction of the bias suggests that, if anything, our estimates underestimate the effect of project density on prices and can be interpreted as a conservative, upper bound of the negative impact of density. The rest of the paper is organized as follows. Section 2 presents a brief overview of Singapore s planning and institutional background. Section 3 describes the data and our definition and choice of neighborhoods. Section 4 contains the model, section 5 contains regression estimates and results, and section 6 concludes. 2. Institutional Background Singapore is a square kilometer city-state (roughly half the size of Los Angeles or London) with an urbanized population of 5.96 million, of thority (URA) subdivided small land plots so that there would be more variety in housing design (Urban Redevelopment Authority, 2000). 4

6 which 3.9 million are citizens or residents. 7 More than 90% of households headed by a citizen or permanent resident own their own homes (Singapore Department of Statistics, 2015). Most of these homes are apartments in high-rise buildings developed by Singapore s public housing authority, the Housing and Development Board (HDB), established in 1960 shortly before independence. The remaining population mostly live in high-rise apartment buildings built by private developers. Privately developed apartments are sold to both Singaporeans and foreigners, at market prices. We examine transactions of private apartments as their purchases are not subsidized or restricted. Given its small size, planning in Singapore has always been important and has a long history dating back to its colonial roots. A land use plan was implemented in the 1820s, shortly after the founding of Singapore. However, planning was mostly piecemeal and haphazard. The ad-hoc approach was deemed a failure by the early 1920s, and the downtown area was overcrowded with slums. This was then rectified in 1958, whereby a comprehensive and systematic approach to planning was approved in the form of the Master Plan. The Master Plan is a statutory plan which details various land uses (residential, commercial, industrial, etc) and provides the framework for development. Each land parcel is subject to an intensity control, which initially 7 In contrast, Los Angeles has a population of 3.88 million and London, 8.7 million. 5

7 differed by whether the development was a residential or commercial development. Commercial developments were regulated by plot ratio allowance, the ratio between gross floor area and land area, much akin to the Floor Area Ratio (FAR). Residential developments, instead, were subject to a maximum density, which was defined as the number of residents per hectare of land. In subsequent revisions of the Master Plan after 1989, residential density controls were converted to plot ratio allowances by a fixed conversion factor so that now all land is regulated by the same measure. 8 The more visionary Concept Plan, first adopted in 1971, is meant to be a long-range blueprint of Singapore s development that aims to coordinate the conflicts that may arise from competing land uses and makes room for major infrastructure development such as tunnels, major expressways, and train networks. While the Master Plan is a statutory document with the provisions contained in it legally binding, the Concept Plan is non-statutory, though many of its suggestions are later implemented through subsequent revisions to the Master Plan. In essence, the Master Plan makes the visions of the Concept Plan statutory. 9 8 All new developments and redevelopments are subject to the Master Plan through a planning approval application to the Urban Redevelopment Authority (URA). Besides the Master Plan, all new and redevelopment plans are subject to other existing design guidelines that include (a) special height control plans that are dictated by flight paths, telecommunication installations, sensitive governmental and military buildings, and military airports, (b) building height restrictions, (c) conservation areas and monuments to preserve historical enclaves or structures, and (d) street block plans with building setback requirements. 9 A forum for public consultation is provided to receive feedback from professional bodies (e.g., architects, and builders) and the public prior to the introduction of each Concept 6

8 In the early 1990s, after a major review of the Concept Plan, the Urban Redevelopment Authority (URA) divided Singapore into 55 planning areas with accompanying Development Guide Plans (DGPs) that included detailed policies and control guidelines on land use, building restrictions, urban design, conservation and redevelopment guidelines of heritage buildings and monuments, parks and water bodies. These 55 DGPs formed the basis of the Master Plan of 1998, the last major revision to the Master Plan. We include only projects in our sample that were subject to and built after the 1998 Master Plan restrictions. Further there were no revisions of plot ratio allowances of the land lots in our neighborhoods in the subsequent reviews of the Master Plan, ruling out any change in plot ratio allowance in response to changes in demand over the sample period. 3. Data and Neighborhood Definition We use data from the Real Estate Information System (REALIS) database of the Singapore Land Authority (SLA), which contains information on private residential property transactions in Singapore since Data in RE- ALIS comes from caveats, a legal document lodged with the SLA by a buyer or mortgage provider to register the buyer s legal rights to a property. 10 The caveat contains the transaction date, address, project name, price, area, Plan. Subsequent ad hoc changes and lobbying for deviations from the Master Plan are uncommon in Singapore. 10 The vast majority of buyers lodge caveats. For example, Fesselmeyer, Liu, and Salvo (2016) estimate that REALIS contains 93% of all new sale transactions. 7

9 floor, type of sale (new or resale), and property type (apartment, landed, etc.). We restrict our sample to apartments sold in the secondary market to ensure that the buyer observes the density of the project at the time of purchase (in contrast to new purchases, which typically take place years before construction is completed). Further, to eliminate any spillover effect of new project construction on already built apartments in the same neighborhood, we only use transactions that occurred after a neighborhood was completely developed. Apartments are located in a project, a collection of adjacent buildings, each containing housing units, sharing a land parcel, a name (e.g., Cavendish Park ), and facilities. Projects form neighborhoods, a set of projects in close vicinity that is delineated by major roads, rivers, canals, etc. We include neighborhoods in our sample that contain variation in plot ratio allowance. As we discuss in detail in the introduction, by choosing small neighborhoods, including neighborhood fixed effects in the regression, and Singapore s desire for projects that are structurally heterogeneous and vary over the skyline, variation in plot ratio allowance within a neighborhood is arguably exogenous. Figure 1 contains examples of two neighborhoods in our sample. The neighborhood in panel (a) is 17.2 acres in size and contains three projects, Yishun Sapphire, Yishun Emerald, and Eight Courtyards, which are delineated by a flood canal to the north, a medium sized road to the west, and larger roads to the east and south. Note the variation in plot ratio allowance 8

10 within this small area: 2.1 and 2.5. The neighborhood in panel (b) is 19 acres in size and also contains three projects, West Bay, Blue Horizon, and West Cove, which are delineated by large roads to the west, north, and east, and a flood canal to the south. (West Coast Crescent that runs between the projects is a small, local access road with little traffic.) Again, note the wide variation in plot ratio allowance in this small area: 1.6, 1.8, and 2.8. The 337 projects in our sample are plotted on the map of Singapore in Figure 2. The map shows that the neighborhoods are found throughout Singapore. Like many cities, the plot ratio allowance is more generous near the Central Business District (CBD), where demand is greatest, and decreases as distance to the CBD increases. A simple regression of the natural log of plot ratio allowance of each project on distance to the CBD in km renders (with standard errors in parentheses): ln (plot ratio allowance) = distance to CBD (0.004) (0.004) On average, plot ratio allowance decreases by 2% in distance to CBD. Our final sample consists of 11,913 transactions from 2002 to 2016 in 337 projects in 44 neighborhoods. The top panel of Table 1 describes the data. For example, the mean unit size and purchase price are, respectively, 1300 square feet and S$959,209, in January 2000 dollars, or roughly US$685,000. Average apartment age was 9.4 years. Most apartments are in high-rise buildings; the average floor was 7.5, with a maximum of 38. About 44% of 9

11 transactions involved apartments with leases greater than 99 years. These leases were mostly either 999 years or freehold. 11 We define project density as the number of apartments in a project per acre of land. Number of apartments was collected and cross-checked on three Singapore real estate websites, propertyguru.sg, iproperty.sg, and keylocation.sg. Land area and plot ratio allowance is taken from the Urban Redevelopment Authority s OneMap webpage, From the lower panel of Table 1, the average project contains 168 apartments. There is a wide variation in project size: the smallest project contains only 6 units and the largest contains over 1,700. Average project land area is about 2.4 acres. The smallest land lot is only 0.04 acres and the largest is 18.5 acres. Project density averages 85 units per acre of land, with a minimum density of 17.9 and a maximum of 645. Our instrument, plot ratio allowance, averaged over 2. The smallest allowance was 1.4 and the largest was 4. Neighborhoods are small, the average area is acres. The variation in plot ratio allowance should be exogenous with respect to local residential demand as long as planners do not expect demand to vary by location within small areas. Though we think this is generally true in Singapore, one possibly important exception are locations with a sea view, which may be deemed to be of high demand and allocated a higher plot ratio 11 As a former British colony, Singapore follows the British leasehold system in which land is either freehold, i.e., owned in perpetuity, or leased from a freeholder for a certain number of years. For further details on the leasehold system in Singapore, readers may refer to Fesselmeyer, Liu, and Salvo (2016). 10

12 allowance compared to other locations in the same neighborhood that would not have a sea view in their line of sight. We control for this particular amenity with an indicator variable for such projects, which make up 4% of our sample. One threat to our identification strategy stems from the possibility that developers alter the quality of construction in response to the plot ratio allowance. For example, a developer may respond to a low plot ratio allowance by increasing construction quality in order to market the apartments as upscale or respond to a high plot ratio with higher quality construction to compensate for the negative price effect of high density. To control for these possibilities, we collected quality data from the Construction Quality Assessment System (CONQUAS) of the Building and Construction Authority. Under the CONQUAS program introduced in 1989, developers can submit their buildings for assessment of their workmanship standards. The CONQUAS score of a building is the sum of points awarded for three main components: Structural Works, which focuses on the structural integrity and safety of the building, Architectural Works, which focuses on the aesthetics and quality of finishes and components, and M&E Works, which focuses on mechanical and electrical services. In the lower panel of Table 1, one sees that about half the buildings in the sample were assessed, and the average score of assessed buildings was 83.8, with a minimum score of 68.4 and a maximum score of 97.9 (out of a possible 100). In the regressions below, we standardize CON- QUAS score, subtracting off the sample mean and dividing by the sample 11

13 standard deviation. Finally, we collected project facility data from the three aforementioned real estate websites. Facilities include a tennis court, a gym, a function room, a jacuzzi, and a sauna. Summary statistics of facilities are included in the lower panel of Table Model and Estimation Approach We model the relationship between price per square foot and project density using a hedonic regression as follows. Let log deflated price per square foot of unit i in project j in neighborhood c transacted in period t be determined as: ln (p ijct ) = α c + δ ln D j + βx i + λ t + ɛ ijct, (1) where α c is a neighborhood fixed effect, D j is project density, X i are observable apartment and project characteristics, λ t includes year and month time effects, and ɛ ijct is an idiosyncratic error. The parameter of primary interest is δ, the effect of project density on price per square foot. As discussed in the introduction, the biggest threat to identification is the endogeneity of density. We address this by including neighborhood fixed effects to control for the variation in unobserved demand factors across neighborhoods, and we instrument for density with plot ratio allowance, which we argue is credibly exogenous within neighborhood. 12

14 5. Results We estimate equation (1) by OLS and by two-stage least squares. We include neighborhood fixed effects and a full set of apartment and project characteristics as well as a flexible time trend. Standard errors are clustered by project. Since CONQUAS score is not available for all buildings, we estimate three variations of the model. The first specification excludes building quality. We estimate this model with the full sample and with a subsample of apartments from CONQUAS-assessed buildings. The second specification includes an indicator variable for whether the building that contains the apartment is high quality, defined as having a CONQUAS score in the upper 25 th percentile of scores. The omitted category includes apartments in buildings below the 75 th percentile and in buildings that were not submitted for assessment. The third specification contains an indicator for whether the building was assessed and an interaction of this indicator with the CONQUAS score. OLS results are contained in Table 2. In each of the specifications the effect of density on price is negative, large, and statistically significant. In column (1), with no control for quality, the effect of a 10% increase in density is a.8% decrease in price per square foot. The effect is very similar in column (2), when we only include apartments in CONQUAS-assessed buildings. Adding CONQUAS building quality reduces the estimate slightly to.55% and.6% in columns (3) and (4). This suggests that buildings built under a larger plot ratio allowance are of lower quality. 13

15 The other estimates are all reasonable and statistically significant. A sea view is estimated to increase price between 17 to 20%. An apartment in a high quality building fetches a premium of 6% (column (3)). Column (4) shows that a one standard deviation increase in CONQUAS score increases price by 4.7% for assessed buildings. Age negatively affects price, as does unit size. Because of the tropical climate, to avoid bugs and to enjoy the breeze Singaporeans prefer living on high floors, and these units cost more as shown by the floor estimate. Dummy variables for first floor and top floor indicate a negative effect. Apartments on these floors tend to have priced area that is outside, which is less desirable than inside space, and thus price per square foot is lower, conditional on unit size. As expected, an apartment with a very long lease earns a premium, from 11 to 16% across columns. The OLS estimates are expected to be biased upwards as density is likely endogenous. We turn now to two stage least squares results than instrument for project density with plot ratio allowance. First-stage results are contained in Table 3. We see that plot ratio allowance has a significant, positive effect on project density, as expected. The smallest F -statistic in the four regressions is 61, easily surpassing the rule of thumb of 10 suggested in Stock and Yogo (2005), indicating that the instrument is not weak. Second-stage results are contained in Table 4. Again we see that the effect of project density is negative and statistically significant at the 1% level. Moreover, the estimated effect is more negative than the OLS estimates, as expected given the discussion in the introduction on the direction of any bias. 14

16 In column (1), excluding CONQUAS score, the effect of a 10% increase in density is to lower price per square foot by nearly 1.6%. The effect increases to 2% for the subsample from CONQUAS-assessed buildings. Adding quality diminishes the effect slightly to 1.3% and 1.4% in columns (3) and (4), but the effect is still large. In other words, using a careful identification approach, we find that project density has a non-trivial, negative effect on price. The other explanatory variables estimates, similar to the OLS estimates interpreted above, are all significant, save the indicator for CONQUAS assessment in column (4), and all have the expected signs. 6. Conclusion In this paper, we use plausibly exogenous variation in project density to measure how very localized density affects prices in the secondary-market for apartments in Singapore. To the best of our knowledge ours is the first paper to measure the causal effect of very localized density on welfare. We find that a 10% increase in density decreases price per square foot by between 1.3% and 2%. Our findings carry important policy implications. Consider for example land use restrictions such as anti-sprawl measures that are intended to minimize urban expansion but also tend to increase density. Our findings suggest that any positive effects of such policy measures are at least partially offset by the negative effects of increased localized density on utility. 15

17 DATA: Cluster Dummies (cont d) (a) Three projects delineated by roads and a flood canal West Bay Blue Horizon West Cove (b) Three projects delineated by roads and a flood canal Eligible cluster (planning region Clementi) 0.09 km 2 (< 9.49 km 2 ) Figure 1: Two neighborhoods in the sample, with project names and plot area allowance Not an elig Overview Methodology Data Collection Reg. Estimates Validity Checks 16

18 which may not have government-imposed plot ratios. Also excluded were private developments directly situated above retail outlets and transportation nodes (e.g. bus interchanges) as the allocation of these plot ratios may take into account the density allocated for commercial and transportation purposes. Developments that are not near other developments are also dropped as they are not part of a cluster. This resulted in a total of 44 cluster dummies, which comprise 336 developments. Notwithstanding, there remains a wide spatial distribution of developments across the state, and more importantly, across many planning districts. Figure 4.2 depicts the developments that fall within these cluster dummies. FIGURE 4.2 SPATIAL LAYOUT OF PRIVATE DEVELOPMENTS Finally, by checking the plot ratios Figuredepicted 2: Map of in projects the Masterplans over the years 8, plot ratios of all the private developments within the above clusters were found to be unchanged over time, providing a cleaner identification. 8 Masterplans were updated in 2003, 2008 and 2014 for the time period concerning the dataset. 17

19 Table 1: Summary Statistics Apartment-level variables VARIABLES Mean StDev Min Max Price (2000S$) 959, , ,235 16,834,112 Price per square foot (2000S$) , Unit size (square feet in 1000s) Age (in years) Floor First floor of building Top floor of building Lease > 99 years Transaction Year Observations 11,913 Project-level variables VARIABLES Mean StDev Min Max Plot ratio allowance Project units ,715 Project land area (acres) Project density (units/acre) Sea view Tennis court Gym Function room Jacuzzi Sauna CONQUAS assessed CONQUAS score Projects per neighborhood Neighborhood area (acres) Projects 337 Neighborhoods 44 18

20 Table 2: OLS regressions: transaction price per square foot, in log (1) (2) (3) (4) Project density, in log *** *** ** *** (0.025) (0.030) (0.024) (0.021) 5 < Age 10 years *** *** *** *** (0.013) (0.015) (0.013) (0.012) 10 < Age 15 years *** *** *** *** (0.012) (0.014) (0.015) (0.014) Age > 15 years *** *** *** *** (0.016) (0.019) (0.020) (0.019) Unit size, square feet in 1000s *** *** *** *** (0.015) (0.017) (0.015) (0.015) Floor 0.005*** 0.005*** 0.005*** 0.005*** (0.001) (0.001) (0.001) (0.001) First floor of building *** *** *** *** (0.009) (0.009) (0.009) (0.009) Top floor of building *** *** *** *** (0.013) (0.015) (0.013) (0.013) Lease > 99 years 0.130*** 0.112*** 0.147*** 0.162*** (0.019) (0.019) (0.023) (0.022) Sea view 0.166*** 0.180*** 0.188*** 0.195*** (0.029) (0.026) (0.036) (0.034) CONQUAS high quality 0.061*** (0.015) CONQUAS assessed 0.036* (0.019) CONQUAS score, standardized 0.047*** CONQUAS assessed (0.008) Constant 6.523*** 6.631*** 6.408*** 6.428*** (0.120) (0.140) (0.117) (0.107) Project facility effects Yes Yes Yes Yes Neighborhood effects Yes Yes Yes Yes Year effects Yes Yes Yes Yes Month effects Yes Yes Yes Yes R Observations 11,913 10,765 11,913 11,913 Number of regressors Notes: *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is price per square foot in 2000 S$, in log. Project density is number of housing units per acre of land. Project facilities include dummy variables for: tennis court, gym, function room, jacuzzi, and sauna. Columns (1), (3), and (4) use all observations. Column (2) uses observations from CONQUAS-assessed buildings only. Standard errors clustered by project are in parentheses. 19

21 Table 3: First-stage regressions: project density, in log (1) (2) (3) (4) Plot ratio, in log 0.820*** 0.826*** 0.799*** 0.809*** (0.095) (0.106) (0.091) (0.089) 5 < Age 10 years *** *** *** *** (0.023) (0.023) (0.025) (0.022) 10 < Age 15 years *** *** *** *** (0.033) (0.034) (0.037) (0.032) Age > 15 years *** *** *** *** (0.047) (0.052) (0.052) (0.048) Unit size, square feet in 1000s *** *** *** *** (0.019) (0.016) (0.018) (0.018) Floor 0.003** 0.004*** 0.003*** 0.003*** (0.001) (0.001) (0.001) (0.001) First floor of building 0.025*** 0.025*** 0.025*** 0.026*** (0.008) (0.007) (0.007) (0.007) Top floor of building 0.043** ** 0.039** (0.017) (0.013) (0.016) (0.016) Lease > 99 years ** * ** *** (0.057) (0.064) (0.058) (0.055) Sea view ** *** ** ** (0.083) (0.082) (0.086) (0.085) CONQUAS high quality (0.031) CONQUAS assessed (0.050) CONQUAS score, standardized CONQUAS assessed (0.030) Constant 3.746*** 3.567*** 3.765*** 3.788*** (0.127) (0.149) (0.123) (0.124) Project facility effects Yes Yes Yes Yes Neighborhood effects Yes Yes Yes Yes Year effects Yes Yes Yes Yes Month effects Yes Yes Yes Yes R Observations 11,913 10,765 11,913 11,913 Number of regressors Notes: *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is project density, the number of housing units per acre of land, in log. Project facilities include dummy variables for: tennis court, gym, function room, jacuzzi, and sauna. Columns (1), (3), and (4) use all observations. Column (2) uses observations from CONQUAS-assessed buildings only. Standard errors clustered by project are in parentheses. 20

22 Table 4: Second-stage regressions: transaction price per square foot, in log (1) (2) (3) (4) Project density, in log *** *** *** *** (0.039) (0.055) (0.034) (0.036) 5 < Age 10 years *** *** *** *** (0.012) (0.014) (0.012) (0.011) 10 < Age 15 years *** *** *** *** (0.013) (0.016) (0.014) (0.013) Age > 15 years *** *** *** *** (0.019) (0.025) (0.020) (0.018) Unit size, square feet in 1000s *** *** *** *** (0.015) (0.017) (0.015) (0.015) Floor 0.006*** 0.006*** 0.006*** 0.006*** (0.001) (0.001) (0.001) (0.001) First floor of building *** *** *** *** (0.009) (0.009) (0.009) (0.008) Top floor of building *** *** *** *** (0.013) (0.014) (0.013) (0.013) Lease > 99 years 0.111*** 0.093*** 0.130*** 0.141*** (0.019) (0.023) (0.020) (0.020) Sea view 0.166*** 0.176*** 0.185*** 0.192*** (0.024) (0.021) (0.030) (0.028) CONQUAS high quality 0.051*** (0.014) CONQUAS assessed (0.019) CONQUAS score, standardized 0.043*** CONQUAS assessed (0.008) Constant 6.894*** 7.102*** 6.725*** 6.796*** (0.184) (0.252) (0.162) (0.172) Project facility effects Yes Yes Yes Yes Neighborhood effects Yes Yes Yes Yes Year effects Yes Yes Yes Yes Month effects Yes Yes Yes Yes Observations 11,913 10,765 11,913 11,913 Number of regressors Notes: *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is price per square foot in 2000 S$, in log. Project density is number of housing units per acre of land. Project facilities include dummy variables for: tennis court, gym, function room, jacuzzi, and sauna. Columns (1), (3), and (4) use all observations. Column (2) uses observations from CONQUAS-assessed buildings only. Standard errors clustered by project are in parentheses. 21

23 References Ahlfeldt, G.M., Redding, S.J., Sturm, D.M. and Wolf, N. (2015): The economics of density: Evidence from the Berlin Wall, Econometrica, 83(6), Bertaud, A. and Brueckner, J.K. (2005): Analyzing building-height restrictions: predicted impacts and welfare costs, Regional Science and Urban Economics, 35, Brueckner, J.K. and Largey, A.G. (2008): Social interaction and urban sprawl, Journal of Urban Economics 64, Brueckner, J.K., Fu, S., Gu, Y. and Zhang, J. (2017): Measuring the stringency of land-use regulation: The case of China s building-height limits, Review of Economics and Statistics, forthcoming. Cai, H., Wang, Z., and Zhang, Q. (2016): To build above the limit? Implementation of land use regulations in urban China, Journal of Urban Economics, 98, Calhoun, J.B. (1962): Population density and social pathology, Scientific American, 306, Centre for Liveable Cities (2013): 10 principles for liveable highdensity cities: Lessons from Singapore. Singapore: Urban Land Institute. Cheong-Chua, K. H. (2012): Singapore: Housing a nation, Urban Solutions, July. Chor, G.H. and Heng, C.K. (2017): Shaping Singapore s cityscape through urban design, in 50 Years of Urban Planning in Singapore, edited by C.K. Heng. Singapore: World Scientific Publishing. 22

24 Ciccone, A. and Hall, R.E. (1996): Productivity and the density of economic activity, American Economic Review, 86(1), DiPasquale, D. and Wheaton, W.C. (1996): Urban Economics and Real Estate Markets. Prentice Hall: Upper Saddle River, New Jersey. Durkheim, E. (1951): Suicide. Glencoe: Free Press. Evans, G.W. (1979): Behavioral and physiological consequences of crowding in humans, Journal of Applied Social Psychology, 9(1), Fesselmeyer, E., Liu, H., and Salvo, A. (2016): How do households discount over centuries? Evidence from Singapore s private housing market, working paper. Fu, Y. and Somerville, C.T. (2001): Site density restrictions: Measurement and empirical analysis, Journal of Urban Economics 49, Glaeser, E.L., Gyourko, J., and Saks, R. (2005): Why is Manhattan so expensive? Regulation and the rise in housing prices, Journal of Law & Economics, 48, Glaeser, E.L., Kolko, J., and Saiz, A. (2001): Consumer city, Journal of Economic Geography, 1, Glaeser, E.L. and Mare, D.C. (2001): Cities and skills, Journal of Labor Economics, 19(2), Lee, J.S. (2016): Measuring the value of apartment density? The effect of residential density on housing prices in Seoul, International Journal of Housing Markets and Analysis, 9(4), Singapore Department of Statistics (2015): Singapore in Figures. 23

25 Stock, J.H. and Yogo, M. (2005): Testing for weak instruments in linear IV regression, Chapter 5 in Identification and Inference in Econometric Models: Essays in Honor of Thomas J. Rothenberg, edited by D.W.K. Andrews and J.H. Stock. Cambridge: Cambridge University Press. Tang, B. and Yiu, C.Y. (2010): Space and scale: A study of development intensity and housing price in Hong Kong, Landscape and Urban Planning 96, Urban Redevelopment Authority (1993): Towards a tropical city of excellence. Singapore: URA. Urban Redevelopment Authority (2000): housing designs, Skyline, Sep./Oct. Encouraging variety in Urban Redevelopment Authority (2011): Conserving the past, Skyline, Supplement. 24