1# RECORD 2344ebd24eeebb31c33011f3e72a12383c3c1e882# TITLE: Hedonic Price Modeling of New Residential Property Values in Xi’an City, China3# YEAR: 2015.0 VENUE: nan DOI: 10.11114/ijsss.v5i9.25104# AUTHORS: Kai Liu; T. Ichinose5# PAGES: 15 CANDIDATE TABLE PAGES (1-based): [4, 8, 9]6# ABSTRACT: This study analyzed new residential property values of Xi’an City in March, 2014. Results show that accessibility indices, such as distance to CBD, have been well capitalized into the residential property values. Particularly, a within-zone housing unit is sold 9.4% more than if it was outside the attendance zone of a Key Primary School, i.e. home buyers have how much willingness-to-pay for the high-quality educational resource. Another corollary is got that the accessibility to subway stations has not significantly been capitalized, only with a low premium reflecting in the real estate market of Xi’an City. Considered that spatial local singularities caused by unobserved variables or estimation bias can be associated with multi-regression errors, this study herein has explored an unconventional viewpoint to residual problem, which combines the regional differences (coming from real world) and the spatial distributions of singularities (feedback from data). Furthermore, whole samples are classified into 5 agglomerations for revealing the underlying reasons about the future trend and variation of real estate market within each region. It is effective to provide scientific basis of decision making for the real estate investors and planners.789===== PAGE 1 (score 14, numeric-paren 0, table-words 0, var-words 14) =====1011International Journal of Social Science Studies 12Vol. 5, No. 9; September 2017 13ISSN 2324-8033 E-ISSN 2324-8041 14Published by Redfame Publishing 15URL: http://ijsss.redfame.com 1642 17Hedonic Price Modeling of New Residential Property Values in Xi’an City, 18China 19Kai Liu1 & Toshiaki Ichinose2, 3 201Doctoral Program in Geoenvironmental Science, University of Tsukuba, Tsukuba City, Japan 212Center for Social and Environmental Systems Research, National Institute for Environmental Studies, Tsukuba City, 22Japan 233Graduate School of Environmental Studies, Nagoya University, Nagoya City, Japan 24Correspondence: Kai Liu, Doctoral Program in Geoenvironmental Science, University of Tsukuba, 1-1-1 Tennodai, 25Tsukuba City, 305-8577 Ibaraki, Japan. 26 27Received: June 23, 2017 Accepted: July 20, 2017 Available online: September 1, 2017 28doi:10.11114/ijsss.v5i9.2510 29 30URL: https://doi.org/10.11114/ijsss.v5i9.2510 31 32Abstract 33This study analyzed new residential property values of Xi’an City in March, 2014. Results show that accessibility 34indices, such as distance to CBD, have been well capitalized into the residential property values. Particularly, a 35within-zone housing unit is sold 9.4% more than if it was outside the attendance zone of a Key Primary School, i.e. 36home buyers have how much willingness-to-pay for the high-quality educational resource. Another corollary is got that 37the accessibility to subway stations has not significantly been capitalized, only with a low premium reflecting in the real 38estate market of Xi’an City. Considered that spatial local singularities caused by unobserved variables or estimation bias 39can be associated with multi-regression errors, this study herein has explored an unconventional viewpoint to residual 40problem, which combines the regional differences (coming from real world) and the spatial distributions of singularities 41(feedback from data). Furthermore, whole samples are classified into 5 agglomerations for revealing the underlying 42reasons about the future trend and variation of real estate market within each region. It is effective to provide scientific 43basis of decision making for the real estate investors and planners. 44Keywords: Hedonic price modeling, new residential property values, residuals 451. Introduction 46Residential property value is one of the most important indicators of urban economic development, and it concerns 47everybody in a city. Especially in recent years, residential problem in major Chinese cities has generated global interest 48among investors, policy makers, scholars and journalists (Wu, Deng, & Liu, 2014). In Chinese cities, variations of 49housing price are directly affected by macroeconomic and policy, apparently showing characteristic of spatial 50disproportion. Cities with location and resource superiority have a comparatively higher and more rapidly growing 51housing price, and this brings a bigger pressure on the housing market in the area near the quality public resource and 52local public goods such as Key Primary Schools (see Note 1), main stations, etc. (Hu, Zheng & Wang, 2014). Zheng and 53Kahn (2007) proved that subway stations, schools, green spaces and other public resources can be able to explain a 54quarter of the spatial differences about residential property values in Beijing. Chinese scholars also found that property 55values of residences inside a Key school's attendance zone had a higher and faster growth rate in Shanghai (Wang, Ge, 56& Zhang, 2010; Shi & Wang, 2014; Liu & Sun, 2015). In addition, transportation infrastructures, high schools, parks 57and other amenities have also been proved to be the important determinants of housing prices (Kim & Zhang, 2005; 58Zheng & Kahn, 2007). Certainly, extensive researches about the impact of public rail transit on housing price and land 59use pattern have been done in Beijing, Shanghai and various other cities, where the appreciation effect is observed and 60premiums caused by local public goods are quantitatively estimated (Wang, Zhu, & Zhang, 2004; Pan, Ren, & Yang, 612007). Wei, Wei, & Jiang (2014) exposed that development of urban rail transport has a marked impact on real estate 62industry. In short, properties bought and sold on real estate market are a kind of compound commodities, whose values 63are usually influenced by the following two key factors: commuter cost and education environment. It is because that 64these factors can contain the main part of implicit price used to embody property buyers' convenience will and 65willing-to-pay for these location accessibilities. Moreover, lower commuter cost and quality education environment are 666768===== PAGE 4 (score 18, numeric-paren 0, table-words 3, var-words 8) =====6970International Journal of Social Science Studies Vol. 5, No. 9; 2017 7145 72a typical case for making quantitative analyses to estimate the capitalization effect of local public goods such as CBD, 73subway stations and Key Primary Schools. 742.2 Structure of the Hedonic Pricing Model 75Before using hedonic pricing model, the form for the hedonic price function must be chosen. Housing is not usually 76considered as a homogeneous commodity (McLeod, 1984). A house buyer purchases both a dwelling and a set of site 77characteristics. The price at which a residence is traded is depending on lots of factors: housing structure, location, 78neighborhood, and environmental characteristics. The household must find the dwelling with the best combination of 79features at the most apposite price (O’Sullivan, 2002). The implicit prices of spillover benefits (premium) from the 80factors can be inferred based on the hedonic hypothesis (Rosen, 1974). Consequently, the traditional hedonic pricing 81model takes the following form: 82 𝑃= 𝑓(𝐿,𝑆,𝑁) (1) 83where P is market price of the housing. Here we defined total price of a new residence as the dependent variable P. L, S, 84N denotes independent variables, which represents location characteristics, structure characteristics, and neighborhood 85characteristics, respectively (Ridker & Henning, 1967). In this study, 11 variables were chosen as hedonic price indices. 86Data with detailed descriptions and descriptive statistics, can be found in Table 1. 87Table 1. Selection of variables used to be added into the hedonic pricing mode and results of descriptive statistics* 88Dependent 89variable 90Unit 91Description 92Mean 93SD 94Min. 95Max. 96Expected 97sign** 98P 9910,000RMB 100The total price of a new residence 10192.25 102109.92 10317.70 1041230.00 105Independent 106variable 107 108DC 109km 110A 111measure 112of 113the 114ease 115of 116accessibility 117to 118CBD, 119i.e. 120straight-line 121distance 122to 123central 124business district 1258.58 1264.88 1271.06 12832.89 129- 130DSX 131km 132A 133measure 134of 135the 136ease 137of 138accessibility to railway station, i.e. 139straight-line distance to Xi'an Station 1408.75 1414.73 1421.30 14330.58 144- 145DS 146km 147A measure of the impact of the 148nearest subway station which is a 149straight-line 150distance 151from 152the 153residence 1541.37 1551.60 1560.10 15715.73 158- 159DR 160km 161A measure of the integrated impact 162of the nearest river or lake from the 163residence to the boundary of the 164nearest river or lake 1651.63 1661.15 1670.01 1686.63 169- 170S 171m2 172Gross Floor Area 173110.86 17448.66 17524.02 176456.00 177+ 178BR 179individual 180Number of bedroom 1812.47 1820.79 1830.00 1845.00 185+ 186LR 187individual 188Number of living room 1891.93 1900.36 1911.00 1925.00 193+ 194WR 195individual 196Number of washroom 1971.43 1980.64 1991.00 2004.00 201+ 202HB 203Dummy variable 204Whether the residence is in a 205high-rise building above 12 floors or 206not 2070.67 2080.47 2090.00 2101.00 211- 212SN 213Dummy variable 214Whether the residence is a south 215facing house or not 2160.86 2170.35 2180.00 2191.00 220+ 221E 222Dummy variable 223A 224measure 225of 226education 227environment. Ed=1, where d is the 228radius of buffer made by a Key 229Primary School when the residence 230is located into this area; or Ed=0, 231when the residence is located outside 232this buffer. 2330.61 2340.49 2350.00 2361.00 237+ 238* Data source from biggest Chinese homepage of real estate website named An Ju Ke. (https://xa.anjuke.com) 239** + and - represent increasing and decreasing effects on the housing price, respectively. 240The hedonic pricing model employed in the evaluation of resources typically uses multiple regression techniques to relate 241property price details to diverse characteristics of differing properties, and sort out the different contributions. Many 242economists have stressed that economic theory does not suggest an appropriate functional form for hedonic price 243equations (Rosen, 1974; Freeman, 1979; Halvorsen & Pollakowaski 1981; Cassel & Mendelsohn 1985). Consequently, it 244is necessary to try reasonable functional forms to each variable and utilize a multiple regression equation (Cassel & 245Mendelsohn 1985). Halvorsen and Pollakowaski (1981) recommend the use of the Box-Cox flexible functional form in 246247248===== PAGE 5 (score 7, numeric-paren 0, table-words 2, var-words 3) =====249250International Journal of Social Science Studies Vol. 5, No. 9; 2017 25146 252hedonic analysis, which would measure best performance with goodness-of-fit. This form is commonly applied in the 253literatures using the hedonic pricing model (Shultz & King 2001, Kong et al., 2007). However, Cassel and Mendelsohn 254(1985) had pointed out a disadvantage of this methodology: that using a best-fit criterion to choose functional forms does 255not necessarily lead to more accurate estimates of characteristic prices. Maurer et al. (2004) thought that simplicity of the 256model and explanatory in economics are more important comparing to the accurate Box-Cox conversion. In general, this 257study just considers coefficient of determination (R2) as the indictor to judge whether the function form of hedonic pricing 258model has measured better performance with goodness-of-fit. 259Equation 1 is a very classical defined formula which is frequently referenced. However, as Rosen (1974) pointed out, 260there is no reason to expect the relationship between the price and factors to be liner. Another situation is that different 261factors based on different units and orders of magnitude. In most cases, non-liner is to be expected, because purchasers 262cannot treat individual housing attributes as discrete items from which they can pick and mix until the desired 263combination of characteristics ID is found (Morancho, 2003). Thus following previous studies (Harrison & Rubinfeld 2641978; Can, 1992; Zheng & Kahn, 2013), this study employs a function form combining with logarithmic and 265semi-logarithmic form, which can be respectively defined as follows: 266 𝐿𝑛(𝑃) = 𝑎0 + ∑267𝑎𝑖𝐿𝑛(𝑍𝑖) + 𝜀268𝑖269𝑛=1270 (2) 271 𝐿𝑛(𝑃) = 𝑎0 + ∑272𝑎𝑖𝑍𝑖+ 𝜀273𝑖274𝑛=1275 (3) 276where Zi denotes explanatory variables, a0 denotes a constant, ai denotes regression coefficients, and ε is the residual 277error term. 278This study applies a form which is combined with Equation 2 &3 to deal with a variety of situations which happened 279correspondingly for the selected variables as the case may be. In other words, some variables must be regarded as 280dummy variables (0 or 1). This is because logical variables or category variables cannot make logarithmic 281transformation (Li, Silva, & Bo, 2011). This detail and regression process will be brought to light in the next Section. In 282this study, real estate transaction prices were collected for Mar. 2014 in Xi’an City. Based on the hedonic pricing model, 283this study attempts to estimate the capitalization effect of location characteristic such as CBD, subway stations and Key 284Primary Schools. At the same time, implicit prices (premium) for characteristics specific to the housing itself and 285neighborhood characteristics are also estimated. 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305Figure 2. Study area 306 307308309===== PAGE 8 (score 25, numeric-paren 0, table-words 7, var-words 6) =====310311International Journal of Social Science Studies Vol. 5, No. 9; 2017 31249 313Table 2. Collinearity diagnostics 314Start 315End 316Variable 317Collinearity Statistics 318Variable 319Collinearity Statistics 320Tol 321VIF 322Tol 323VIF 324Ln_DC 3250.212 3264.720 327Ln_DC 3280.638 3291.567 330Ln_DSX 3310.234 3324.278 333Ln_DS 3340.697 3351.434 336Ln_DS 3370.721 3381.386 339DR 3400.951 3411.052 342DR 3430.957 3441.045 345Ln_S 3460.234 3474.281 348Ln_S 3490.829 3501.206 351BR 3520.311 3533.215 354LR 3550.694 3561.441 357WR 3580.422 3592.367 360HB 3610.888 3621.126 363HB 3640.894 3651.118 366SN 3670.824 3681.213 369SN 3700.846 3711.182 372E 3730.822 3741.217 375E 3760.853 3771.172 378Table 3. Judgment process of variable selection 379Variable selection in the model 380Coefficient of determination (R2) 381Remarks 382Every variable is added. 3830.81 384Start 385Every variable Except Ln_DC 3860.77 387Hold Ln_DC 388Every variable Except Ln_DSX 3890.80 390Drop Ln_DSX 391Hold Ln_S and remove BR,LR & WR 3920.78 393Hold Ln_S 394Hold BR and remove Ln_S, LR & WR 3950.57 396Drop BR 397Hold LR and remove Ln_S, BR & WR 3980.38 399Drop LR 400Hold WR and remove Ln_S, BR & LR 4010.63 402Drop WR and end 4033.4 Focusing on a Key Primary School 404As Section 3.2 announced, there is a remaining problem about variable E. Owing to the value of dummy variable E is 405determined by the size of d-meters buffer drawn in GIS software, the radius d needs to be done some adjustments 406through changing the threshold value at several times, for the purpose of quantifying the capitalization of premium from 407a Key Primary School in the hedonic pricing model, also must meet the condition passing in T-test at statistically 408significance (Table 4). Finally when setting an 825-meters buffer, we can get the best regression result with statistical 409significance at 5% level. This study had tried 8 times and finally found a certain threshold value, 825 meters. Results 410prove that, within a scope of 825-meters, Key Primary Schools, as one of the most representative local public goods, 411have the strongest impact on new residential property values in Xi’an City. As noted in Sections 3.3 and 3.4, this study 412had selected 7 variables and simplified the model successfully. 413Table 4. Adjustment for the variable E 414E 415Coefficients 416t-Ratio 417Degree of confidence in T-text 418Passed or not 419E_1000 4200.018 4210.692 42251.1% (<<95%, try to set d<1000) 423Resolutely not passed 424E_800 4250.040 4261.620 42789.4% (become better, try to set d<800) 428Not passed 429E_600 4300.020 431-0.169 432Regression failed 433Resolutely not passed 434E_700 4350.027 4361.079 43771.9% (worse than d=800) 438Not passed 439E_750 4400.025 4411.024 44269.4% (even worse than d=700, must set d>800) 443Not passed 444E_900 4450.042 4461.605 44789.1% (become better, but worse than d=800, try to set 448800<d<900) 449Not passed 450E_850 4510.048 4521.865 45393.7% (become much better, try to set d=825) 454Could be passed but 455statistically significant only 456at the 10% level 457E_825 4580.090 4590.043 46095.7%>95% (It means when d=825, variable E works 461best in hedonic pricing model) 462Passed T-test at the 5% level 463and can be added into the 464regression 4654. Results of Hedonic Pricing Model 466Results of parameter estimates are presented in Table 5. The estimated coefficients generated using hedonic pricing 467model exhibits signs in accordance with expectations. They provide a clear indication that factors of housing area, CBD; 468floors; subway station; Key Primary School and south facing do exert a significant impact on new residential property 469values. Final regression equation denoting the whole results can also be created (Equation 4). 470 471 472 473 474475476===== PAGE 9 (score 29, numeric-paren 1, table-words 8, var-words 9) =====477478International Journal of Social Science Studies Vol. 5, No. 9; 2017 47950 480Table 5. Regression results after eliminating collinear variables, dependent variable: Ln (P), where P is total price 481Independent variables 482Regression parameters 483(R2 =0.782, adjusted R2 =0.779, F-value=267***, No. of Obs.=378) 484Coefficient 485t-Ratio 486P-value 487Constant 4881.413 4895.139 4900.000 491Ln_S*** 4921.245 49332.311 4940.000 495Ln_DC*** 496-0.268 497-0.924 4980.000 499HB*** 500-0.142 501-4.538 5020.000 503Ln_DS*** 504-0.061 505-3.174 5060.002 507E** 5080.090 5092.097 5100.037 511SN* 5120.044 5131.690 5140.092 515*, ** and *** indicate statistical significance at 10%, 5% and 1% level, respectively 516 𝐿𝑛(𝑃) = 1.413 + 1.245𝐿𝑛(𝑆) −0.268𝐿𝑛(𝐷𝐶) −0.142𝐻𝐵−0.061𝐿𝑛(𝐷𝑆) + 0.90𝐸+ 0.044𝑆𝑁 (4) 517From the value of every coefficient, the calculation based on the regression equation can be completed, in order to 518quantify the estimation value of capitalization corresponding every variable, respectively. Premiums can be calculated 519according to series of processes of anti-logarithmic transformation. 520Main findings are summarized as follows: 5211. P (total price in 10,000 RMB) increases by 1.245% along with S (Area in m2) increases by 1%. According to the 522average level of hosing prices got from the 378-samples (922,500 RMB), this is equivalent to about 11,500 RMB more 523expensive. 5242. P decreases by about 0.268% along with DC (distance to CBD in km) becomes 1% farther, which is about 2500 RMB 525cheaper. Generally, between a residence and CBD there exist strong distance decay effects. 5263. A residence at the floors which is above 12th (≧12) is 14.2% cheaper than one at a lower floor below 12th (<12). 527According to PCDI (per capita disposable income) of Xi’an in 2013 (see Note 6), two ratiocinations can be generalized 528that: 1) a high premium caused by the number of the floor; 2) home-buyers should pay more about 121,800 RMB to buy a 529low-rise residence, which is equivalent to nearly 3.7 times the amount of PCDI in Xi’an in 2013. 5304. When distance to the nearest subway station (in km) becomes 1% farther, housing price will become 0.061% cheaper, 531correspondingly. It means that about 560 RMB cheaper at the average level. Comparing with other metropolis cities that 532have well-developed rail transportation, Xi’an is during a period of the initial stage (until 2014 at least) as a developing 533city because of a low-level premium (see Note 7). In contrast, Metropolis cities like Beijing, Shanghai or Guangzhou, 534there are much higher premium about the factor of rail transportation (Gu & Zheng, 2010). Li et al. (2011) found that 535inside CBD within 8km of Beijing, premiums caused by subway station is not significant and cannot be passed in 536regression processing, that is to say the factor of subway station failed to join the hedonic pricing model because the result 537of T-test is substandard. Only outside the area which is 8km far from the CBD, premiums on the housing price caused by 538subway station seemed to be statistically significant. This contradiction can be considered relating to the commuting cost. 539The metro rail transit is more inclined to give scope to its advantages about commuting cost in the suburban districts far 540from the CBD (Dong, Ding, & Zhao, 2009). Conversely, taking other modes of transportation like bus is cheaper and 541more efficient than taking subway rails for commuting activities happening around the CBD. 5425. With regard to the premium from a Key Primary School, three arguments can be supported: 1) In the case that when 543setting an 825-meter buffer of a Key Primary School, the best regression results can be got with statistical significance at 5445% level. For the future study, another approach named matching regression can be used to specifically investigate the 545so-called Xue Qu Fang problem (Hu et al., 2014); 2) a within-zone housing unit is sold 9.4% more than if it was outside 546the attendance zone of a Key Primary School; 3) At average, premiums are nearly 86,900 RMB from the factor, a Key 547Primary School, which is about 2.6 times the amount of 2013 year’s PCDI of Xi’an. It also indicates that educational 548resources have been well capitalized into the residential property value. 5496. For the factor about the situation of North and South transparent, there is a slight premium, which is only 41,500 RMB. 550Moreover, the regression result is roughly statistically significance at only 10% level (strictly speaking it is normal at least 5515% in statistics). This is because the vast majority of new residences are south facing, which can be clearly read from the 552raw data. Therefore, it seems that data set are lack of comparison samples (not N-S facing houses) when just discussing 553about new residential property values. 5547. After T-text, all of these chosen variables in the hedonic pricing model yield good explanatory power and can also 555reflect the preferences of homeowners in Xi’an City. However, during the regression process, there is a meaningless 556variable, DR, which was dropped in the regression process. In general, housing price trends to become higher when the 557residence is closer to a river or lake according to the expectation. But the regression coefficient of DR got a positive 558number, which is out of expected results and the actual situation. Perhaps, this situation is related to the limitation of the 559560561===== PAGE 10 (score 4, numeric-paren 0, table-words 1, var-words 2) =====562563International Journal of Social Science Studies Vol. 5, No. 9; 2017 56451 565samples, or else unobserved variables or errors. Thence, this study has proved a mathematical corollary that omitted 566variables may result in regression failure and estimated bias. Besides, water system in Xi’an City is very complicated, 567which may cause geographical recognition errors, such as the Xi’an Moat surrounding the City Wall and the Weihe 568Reservoir. So nearest neighborhood analysis done by GIS may also cause accidental and systematic errors. Additionally, 569for these new residential properties which located in the undeveloped suburban areas in Xi’an, the impacts of water on 570housing price are relatively weak, even though they are near a river or lake with a very low price but a closed distance to 571some suburban rivers (see Note 8). Furthermore, real estate development is faster but development of urban infrastructure 572lags behind. So this factor failed to join and estimate the regression equation. All of these above situations could not be 573controlled; consequently, it is necessary to employ the approach of residual analysis, which will be brought to light in 574Chapter 5. 5755. Residual Analysis 576According to the above work, firstly the reasons why residual analysis is necessary need to be discussed: 5771. Variable DR failed to join the hedonic pricing model and was dropped by using stepwise regression method, caused 578by spatial local singularities or data limitations. 5792. Unobserved variables are existent. Quantity of residual can reflect the effectiveness of regression fitting, as well as 580residuals are related to the existence of unobserved variables in a sense. 5813. Positive or negative numbers of residuals indicate smaller or larger biased estimate of regression equation, 582respectively. 5834. It provides a reference for future studies using residual analysis. Specifically, through the process of residual analysis, 584we can try to explain what places have more caused residuals and the reason in consideration of city layout of Xi’an. 5855. According to the results running in SPSS software, we can probe the local singularity from a table of residuals. 586Distribution of residuals was approximately following the Normal distribution, and this indicated that the chosen 587variables in the hedonic pricing model yielded good explanatory power and residuals were under a reasonable situation. 588Herein, map of the location of residuals can be illustrated in Figure 4a. Moreover, sample cases with a high residual are 589picked out and visualized by deviation multiples (Predictive values / observed values), can used for reflecting and 590quantifying the local singularities (Figure 4b). 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605Figure 4. Spatial distribution of regression residuals (left: a) and local singularities (right: b) 606On Figure 4a, we can find that most of residuals are located in the developing area between 2nd Ring Road and 3rd 607Ring Road, which have more real estate development and appreciation space. Even though these areas are about 6084km-8km far from the CBD, more and more new residences have been built in recent years. In other words, in these 609places effects of unobserved variables on the housing price are more significant. So, the results of hedonic pricing 610model show a situation of smaller biased estimates. Eight of the local singularity points that have the highest residual is 611illustrated in Fig. 4b. They are located in three areas in Xi’an City. First area (A) in the southwest of the 2nd Ring Road, 612High-tech New Development Zone is located. It seems that homeowners in High-tech New Development Zone can 613