Content uploaded by Hua Zheng
Author content
All content in this area was uploaded by Hua Zheng on Mar 28, 2024
Content may be subject to copyright.
国際開発研究フォーラム
FORUM OF INTERNATIONAL DEVELOPMENT STUDIES
研究論文
ARTICLE
名古屋⼤学⼤学院国際開発研究科
GRADUATE SCHOOL OF INTERNATIONAL DEVELOPMENT
NAGOYA UNIVERSITY
54-4
ISSN: 2189-9126
Unveiling the Spatial Heterogeneity of
Macroeconomic Factors for Real Estate
Investment: A Comprehensive Analysis
Based on City-level Data Analysis in China
Hua ZHENG
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
1
doi: 10.18999/forids.54.4
Unveiling the Spatial Heterogeneity of
Macroeconomic Factors for Real Estate
Investment: A Comprehensive Analysis
Based on City-level Data Analysis in China
Hua ZHENG*
Abstract
ɉ This paper investigates the macroeconomic factors driving real estate investments, focusing on
variations across different regions in China. Recognizing the significance of location in real estate and
building upon existing literature that addresses spatial dependence through spatial econometrics, this
study further explores the issue of spatial non-stationarity by employing the Multi-scale Geographical
Weighted Regression model (MGWR). The MGWR approach processes city-level cross-sectional data
through subsampling to yield individual coefficients for each city, thereby uncovering the degree of spatial
non-stationarity present. The findings suggest that income levels, the student-to-population ratio, and
per capita government expenditure have consistently impact on real estate investments, albeit with
minor variations. Conversely, factors such as population size, the presence of real estate agents, and
green coverage have a more variable influence across different regions.
Keywords: Real Estate Investment, Spatial Non-stationarity, MGWR Model
1ᶮIntroduction
ɉ Efficient investment decision-making in the capital market is crucial for both investors and
economic growth as it can accelerate economic development and stimulate increased investment
(Klimczak 2010). In most of the developing country, real estate, being the most prominent physical
asset closely tied to human life, significantly interacts with economic growth and the business cycle.
ɉ China’s real estate market experienced rapid growth in the beginning of the 21st century,
driven both by the market-oriented housing market reform in 1998 and the rapid urbanization and
industrialization processes (J. Zhang et al. 2012). After China’s accession to the WTO, there was
a rapid acceleration in industrialization, fueled by Foreign Direct Investment (FDI) and a surge in
international trade. Affordable land, competitive labor costs, and easy market access in coastal areas
enhanced China’s attractiveness to both domestic and international investors. Recognizing this
*ɉPh. D. student at the Graduate School of International Development (GSID), Nagoya University, Japan. I am deeply grateful to
my supervisors, Professor Someya, Professor Otchia and Professor Mendez, for their invaluable guidance and support. I also
appreciate the opportunity to present my work at both GSID’s 30th-anniversary conference and the TOKAI Young Researcher
Conference. These forums were instrumental in the progression and refinement of my research. Additionally, my sincere
thanks go to the anonymous reviewers of the Forum of International Development Studies for their insightful feedback, which
significantly contributed to the development of this paper.
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
2
potential, a pivotal responsibility of local governments became the enhancement of infrastructure
development, a task that carries significant costs. To boost fiscal revenue, Chinese authorities actively
facilitated the transfer of land to the private sector. The swift rise of the real estate sector catalyzed
the growth of scale economies in urban area, further stimulated by market-oriented reforms. Enhanced
employment prospects in urban regions further spurred rural-to-urban migration, accelerating the pace
of urbanization. According to World Bank data, China’s urbanization rate1 increased from 17.9% in 1978
to 49.7% in 2010 and further to 60.8% in 2020, as a significant number of people migrated from rural
to urban areas. This migration surge presented a lucrative opportunity for real estate developers, who
capitalized on the demand by selling housing. As a result, real estate investment in China flourished,
with the sector witnessing an average growth rate of 24.4% from 2000 to 20102.
ɉ Horizontally, the real estate sector has significantly boosted fiscal revenue, enabling the local
government to enhance social infrastructure construction and promote industrial development, thereby
driving economic growth. Vertically, due to the expansive supply chain associated with the real estate
sector, it is closely interlinked with industrial development. This connection has spurred growth in
both its upstream and downstream sectors, including raw material production, labor intermediation,
construction, and real estate brokerage. As a result, these factors have made notable contributions to
economic advancement. Consequently, China’s economic development has become deeply reliant on
the real estate sector. According to the IMF, real estate, together with the construction sector, directly
constituted 15% of China’s GDP in 2019. Additionally, these industries represented a quarter of fixed-
asset investment, 14% of total urban employment, and nearly 24.1% of bank loans. In 2020, the share
of real estate related loan of total bank loan raise to 42.3%.
ɉ While in the past decade, attention from economists and policymakers has been drawn to the
slowing economic growth in China. Inefficient and spatially misallocated real estate investment has
been doubted as one of the main factors, given its significant influence on resource allocation (Huang &
Du 2017; Wu et al. 2020). To slow down and prevent such misallocations, it is crucial to understand the
factors driving real estate investment in different area. This knowledge enables developers to make
informed decisions on investment allocation, while governments can formulate timely and appropriate
policies to support and regulate the real estate market.
ɉ This study utilizes city-level data from China spanning the years 2016 to 2018 for its analysis.
With some missing values, the final sample consists of 273 cities. By taking the mean values across
the three-year period, this research conducts cross-sectional analysis. To ensure robustness, both
conventional regression models with province-fixed effects and the MGWR analysis are employed.
ɉ The structure of this paper is as follows: Section 2 provides a comprehensive literature review and
highlights the research contributions of this study. Section 3 introduces the models used and discusses
the selection of data for the analysis. Section 4 presents the empirical results obtained from the
analysis and provides a thorough discussion of these findings. Finally, Section 5 offers the conclusion
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
3
of the study, summarizes the key findings, provides insights based on the research outcomes and states
the limits of this paper.
2ᶮLiterature Review
2.1ᶮGeneral Study about Determinants of Real Estate Investment
ɉ The most studied determinant is economic growth. According to DiPasquale and Wheaton’s model
(1996), both the real estate investment (supply side) and housing price (demand side) are highly
correlated with the economic growth, population growth, labor market, interest rate and financial
policy. The closed relationship between real estate investment and economic growth and business
cycle attracted remarkable attention. Green (1997) discussed the taut relationship between real estate
investment and economic growth and found it is real estate investment that cause economic growth,
but non-real estate investment is caused by economic growth. As supplementary to this, Zhang et al.
(2012) empirically studied provincial data of China and found different interactive directions, telling
that both ways of causality in the eastern area, one-way causality in the middle-developed area and no
causality in the poor area. Regarding the effectiveness of the impact, Kim (2000) argued there is no
cointegration relationship between real estate investment and GDP, and later, Wigren & Wilhelmsson
(2007) supplied that real estate investment has a significant effect on short-run GDP growth but has a
weak effect on the long-run.
ɉ Population, along with population mobility, population structure and population density, directly
affect real estate demand, not only from DiPasquale and Wheaton’s model but also from many other
scholars. Wang et al. (2018) found the increase in elderly dependency ratio and inter-regional migration
pull up the demand of house investment. Cui et al. (2022) recognized due to the need for stable life,
both aging population and new generation drive the demand of houses, but it has geographical limits.
ɉ Urbanization is another important factor. Lieser & Groh (2014b) confirmed that rapid urbanization
and compelling demographics attract real estate investments. Dasgupta et al. (2014) concluded that
urbanization raises the demand of housing investment by analyzing data from more than 90 countries.
In China’s case, J. Chen et al. (2011) suggests that the exiting high urbanization levels are still
boosting the development of housing sector in developed province. Zhang et al. (2014) supplied it by
studying the effect from different types of urbanizations and found population urbanization has very
limited contribution for attracting real estate investment, while geographical urbanization and capital
urbanization has greater impact. Liu et al. (2018) learned that urbanization Granger-cause real estate
investment in the northeastern regions but not in eastern and western regions in China.
ɉ Some other factors are also addressed. For example, He et al. (2022) identified the migration
is the key determinants for real estate investment from overseas. Wolny (2016) clarified that the
accessibility to real estate is also an important determinant for selecting location. Saiz (2010)
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
4
introduced satellite-generated data to test the impact of geographic features such as oceans, lakes,
mountains, and wetlands. He found both regulatory policy and physical geographical condition can
induce a relative scarcity of the developable land. Mak et al. (2012) identified that major determinants
causing real estate investment to vary among Chinese regions are demographic, economic and policy
factors.
2.2ᶮSpatial Characteristic of Real Estate Investment
ɉ There’s a widespread acknowledgment that the value or potential worth of real estate is intricately
tied to its location and the value of real estate from its neighboring areas. Historically, scholars
incorporated regressors such as distances to key centers as independent variables (Belsley et al. 1980)
to measure the impact from neighboring regions. However, incorporating such variables into real
estate models was not widespread, primarily due to the heavy computation. Early initiatives to weave
a spatial weight matrix into regression models for urban and real estate analysis were led by Griffith
(1981) and Anselin and Can (1986). It was Anselin (1998) who ushered in a more refined spatial
approach for the real estate market, and Dubin (1992, 1998) subsequently adapted this for hedonic
house price models. In recent years, Chinese scholars, such as Y. Wang et al. (2017) have included the
spatial autocorrelation into real estate investment studies, aiming to bolster estimator accuracy.
ɉ Yet, another facet of spatial data analysis, termed spatial heterogeneity, pivots attention from
spatial autocorrelation to spatial non-stationarity. Fotheringham et al. (2003) introduced the GWR
model to reveal the exitance of spatial non-stationarity. While this model has found applications in
environmental studies, such as haze pollution analysis (Shaojian et al. 2020; S. Wang et al. 2014), its
application in real estate remains limited. Given the objective of this paperŻunearthing determinants
of spatially-varied real estate investmentŻand considering China’s vast geographical disparities and
structural differences, it’s imperative to consider the potential for spatial variations in determinant
effects. J. Zhang et al. (2012) substantiated this by identifying divergent directions in the Granger
relationships between real estate investment and GDP levels across eastern, central, and western
regions.
ɉ Regarding the determinants of real estate investment in China, two primary limitations emerge
from past literature. In general, many papers neglect spatial factors, potentially leading to missing
variable problem and endogenous issues caused by spatial autocorrelations in the residuals. Spatial
studies about real estate investment is limited. Among those limited studies, spatial autocorrelationŻ
often referred to as spatial homogeneityŻhas been addressed. On the other hand, spatial non-
stationarity remains largely unconsidered.
ɉ This paper provides two main contributions. First, in line with the trend of examining real
estate investment from a spatial dependence perspective, this study approaches the topic more
comprehensively by considering spatial non-stationarity by applying MGWR model. Secondly, unlike
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
5
many previous studies using national or province-level data in this domain, our research taps into
subregional data. This granular level of data allows for a more detailed analysis. By doing so, this
paper captures local variations and nuances that are often overlooked in broader studies, offering a
fresh perspective on determinants of real estate investment.
2.3ᶮSpatiotemporal Characteristic of the Data
ɉ The rationale for employing spatial econometrics in this study is mainly attributed to the notable
spatial clustering observed in real estate investment. In order to mitigate the impact of annual
fluctuations, this paper takes the average value from 2016 to 2018 to reduce the cyclical impact, and
then utilize the Geographic Information System software (ArcGIS Pro) to plot the value. Figure 1
below illustrates the distribution of real estate investment across various cities. Notwithstanding
the presence of some missing data points, the figure reveals two distinct spatial patterns. Firstly, it
unequivocally demonstrates a pronounced spatial disparity, characterized by substantial investment in
the Eastern regions that gradually diminishes towards the Western areas. Secondly, the capital city
(in red dots) of each province exhibits higher levels of investment, which subsequently extends to its
surrounding localities.
Figure 1 Spatial Distribution of Real Estate Investment
Note: Mapped by ArcGIS pro software. Currency unit is Chinese RMB.
Source: Chinese City Statistics Yearbook
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
6
3ᶮMethodology, Data Selection and Modeling
3.1ᶮReason for Methodology Selection
ɉ To better understand how the impact of determinant vary across space, there are two main parts of
reasons why this paper applies MGWR model. Theoretically, the presence of spatial dependence and
variation underscores the importance of considering space, aiming to have more accurate estimation
of impact estimators. From a technical standpoint, firstly, pinpointing determinants across different
regions by using non-spatial approach requires extensive time series data. Given that China’s real
estate development has only spanned 25 years, the outcomes from the time series analysis are not
entirely reliable. The GWR and MGWR models, which utilize cross-sectional data from a sample of
276 cities, enhance the degree of freedom. Secondly, in the non-spatial analysis, there is potential
endogenous problem caused by spatial autocorrelation in the residual term. It is efficient to remove or
lower down the correlation in residual term by using GWR or MGWR model (Gu et al. 2021).
3.2ᶮGeographically Weighted Regression (GWR)
ɉ Before delving into the MGWR model, it is useful to begin with an introduction to GWR. The basic
idea behind GWR is that correlation is uneven among space, which is so called spatial non-stationarity.
In practice, by using cross-sessional data, Ordinary least squares (OLS) calculate the overall effect
out of full samples, while GWR aims to compute impact coefficient for each sample point. With the
geographical information of each region embedding in GWR, GWR collects samples by grouping those
sample points within a certain scale, which is so called bandwidth. Akaike Information Criterion
(AICs) and Cross Validation (CV) are used to identify the best bandwidth, either in distance unit using
fixed kernal technique or in number of adjacent points using the adaptive kernel technique. Taking
consideration of the huge size difference of cities in China, in order to reduce the spatial distortion,
this paper uses the latter one. Within the bandwidth, GWR conduct the weighted regression based on
Gaussian function or bi-square weighting function. To have efficient computation, this paper uses bi-
square weighting function as below:
(1)
ɉ Where the dmax is the maximum distance from jth farthest province to the regression city i, dij is the
distance between city i and city j.
The GWR model can be expressed as follows:
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
7
(2)
ɉ Where yi represents the explained variable, (ui, vi) is the location information, ui is the easting,
and vi is the northing. m is the total amount of independent variables; βm (ui, vi) is the coefficient of
independent variable m. xim is the observation value of the variable m at the sample point i; εi is the
error term. The parameters area solved by the following matrix form.
(3)
ɉ Where T is the matrix transpose operation, X is the matrix of independent variables, Y is the matrix
of dependent variable, W (ui, vi) is weighting matrix with element wij.
3.3ᶮMGWR Modeling
ɉ The MGWR model is one of the extensions of GWR that relaxes the assumption that all spatially
varying processes for each independent variable vary at the same spatial scale. The bandwidth for
each independent factor is flexible, as demonstrated in the work of Fotheringham, Yang and Kang (2017).
The fundamental concept behind MGWR is that different explanatory variables have varying degrees of
spatial influence on the explained variable. In practice, different sample sizes are assigned to different
variables based on their spatial influence. The formula is as below:
(4)
ɉ Compared with equation (2), the difference of equation (4) is the bwm in βbwm including the
bandwidth used for calibrating the mth conditional relationship, and variable m has different bandwidth
bwm. M is the number of variables. (ui, vi) is the location information, εi is the error term.
3.4ᶮData Selection
ɉ To ascertain the spatial pattern of determinants influencing real estate investment, this study
employs city-level cross-sectional data obtained from the Chinese City Statistics Yearbook 2016
Ż
2018. To mitigate the influence of spatial price discrepancies, the study adopts provincial Consumer
Price Index (CPI) data to adjust the price variables for each city within the defined operational range.
Furthermore, to address the issue of annual volatility, the cross-sectional data is averaged over
the period spanning from 2016 to 2018. These methodological choices are implemented to ensure
comparability and stability in the analysis of the determinants’ spatial effects on real estate investment.
ɉ Based on previous literature, the following variables are selected for analysis. The summary of
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
8
variable explanation is listed in Table 1.
ɉ Income level (GDPpc): The correlation between income level and real estate investment has
been extensively discussed (Green 1997; Kim 2000; Wigren & Wilhelmsson 2007; J. Zhang et al.
2012). Following DiPasquale and Wheaton’s model (1996), real estate investment strongly depends
on economic performance. This study employs real per capita GDP as a measure of economic
development. The per capita GDP for each city is adjusted using provincial CPI, with 2000 as the base
year.
ɉ Industrial structure (ISO): While not directly examined, Krugman (1991) and Alonso (1964) suggest
that regions with advanced manufacturing tend to experience industrial agglomeration, attracting more
real estate investment and migration. This study measures the industrial structure by introducing
industrial structure optimization (ISO)3.
ɉ Population structure (Pop, Student): Population structure is considered a critical factor influencing
real estate demand. Numerous studies have emphasized its importance (X. Wang et al. 2018; Cui
et al. 2022; Lieser & Groh 2014b; Dasgupta et al. 2014; J. Chen et al. 2011; Liu et al. 2018; Zhang et
al. 2014). This study examines the impact of total population. In term of number of college student,
Johnes & Taylor (1989) argue that cities with higher education institutions are likely to attract
enterprise investment and expand manufacturing sector. Additionally, graduates are stickier to
local place. They tend to stay in their local cities due to geographic convenience and interpersonal
networks, making them an important indicator of real estate demand. Luo (2020) notes that in China,
the distribution of higher education resources greatly influences housing prices. This, in turn, spurs
investment. Based on this understanding, this paper introduces the share of college student to total
population (Student) as a pivotal variable.
Table 1 Variable Explanation and Summary Statistics
Notion Description Obs Mean Std.
Dev. Min Max
REinv The natural logarithm of per capita real estate investment 294 7.287 0.963 3.403 10.098
GDPpc The natural logarithm of per capita GDP 295 18.870 0.537 17.369 20.333
ISO The level of balanced growth among various parts of the industry 293 0.559 0.303 0.030 1.745
Green The natural logarithm of per capita total green land cover ratio 289 3.679 0.145 2.49 4.037
Pop The natural logarithm of population 295 15.053 0.767 12.172 17.34
LTD Year-end loan balance/Year-end deposit balance 294 2.860 0.980 0.056 6.289
Land The natural logarithm of per capita land area 295 0.704 0.183 0.264 1.545
Student The natural logarithm of number of collages student 281 10.605 1.300 7.385 13.871
Gov The natural logarithm of per capita government expenditure 295 9.422 0.878 7.285 12.847
Agent The natural logarithm of population working as real estate agent
over total population
295 8.047 0.486 7.233 10.561
Note: Obs indicates the observations, and Std. Dev. refers standard deviation.
Source: Chinese City Statistics Yearbook
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
9
ɉ Finance development (LTD): Financial security and sustainability significantly influence real estate
investment (Frunza & Minoiu 2015; Moshirian & Pham 2000; D. Zhang et al. 2018). This study uses
the loan-to-deposit (LTD ) ratio of year-end balance in banks as a proxy for finance development.
ɉ Natural environment (Green, Land): While not directly examined, Y. Chen et al. (2020) argue that
real estate investment affects air quality and green coverage, showing there is correlation between
real estate investment and green coverage. Gu & Jie et al. (2021) demonstrate that green coverage
is a key factor in attracting migrants to settle. Referring on the requirement of living, green coverage
contributes to the land price. Additionally, due to land scarcity, land area per capita is included as an
explanatory variable, which can be treated as the inverse of population density.
ɉ Government expenditure (Gov): From a supply-side perspective, the government gains revenue by
leasing land to developers. This revenue allows them to enhance infrastructure and other services.
The more the government spends, the greater their revenue or the higher their financial needs. (F. Liu
et al. 2016). Additionally, the government facilitates improved connectivity within the city, which in
turn, attracts more investment. To measure this, per capita government expenditure is utilized.
ɉ Other variables (Agent): Housing agents play a crucial role in the real estate trading market (Agarwal
et al. 2019; Baryla & Ztanpano 1995). This study utilizes the share of the population working in real
estate leasing and sales as a proxy for real estate agent services.
4ᶮEmpirical Result and Discussion
4.1ᶮGlobal Regression, Regional Regression and Model Comparison
ɉ Table 2 presents the outcomes of global regression analyses by using various sample groups and
province-fixed effect. Model 1 employs an OLS model without province-fixed effects, revealing that all
independent variables are statistically significant. Upon incorporating provincial controls in Model 2,
the ISO variable loses its significance, while the coefficient magnitude for land per capita experiences a
substantial reduction.
ɉ Model 3 and Model 4 control for Area_1 and Area_2 respectively. Despite the varying scales
of regional controls, most coefficients remain consistent. However, the impact coefficient for ISO
experiences a notable decline compared to other variables.
ɉ Models 5 through 7 present regression results for samples within each region following the Area_1
classification. The variation in coefficient magnitude and significance across regions highlights spatial
heterogeneity. Key variables such as per capita GDP and population significantly influence real estate
investment only in the eastern region. This finding aligns with J. Zhang et al. (2012), who confirmed
the bidirectional causality between real estate investment and economic growth but found no causal
link from economic growth to real estate investment in either the middle or western regions.
ɉ Although population is expected to be a crucial factor affecting real estate investment, it is only
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
10
marginally significant in the eastern region but not western region. The population coefficient’s
magnitude exhibits considerable variation. Green coverage is significant solely in the western region,
which is consistent with the observed pattern of green coverage’s impact on researcher distribution (Gu
& Yu et al. 2021). Lastly, ISO is not significant in any region.
ɉ Models 8 and 9 examine sub-economic size samples, conducting separate regressions for first-tier
cities and the remaining cities. Comparing the two sets of results, the impact of economic growth,
agents, financial development, and land per capita does not exhibit noticeable differences. The
ISO variable is significant in both groups; however, the magnitude varies considerably. Population,
green coverage, and government expenditure are significant only in the non-first-tier cities group,
highlighting the importance of these factors for real estate investment in less developed cities.
4.2ᶮComparison of Result from Different Models
ɉ Table 3 presented below offers a comparative analysis of the outcomes derived from OLS, GWR, and
Table 2 Result of Regression of OLS for Full Sample and Regional Samples
REinv
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Full Sub-area sample group Sub-economic size
sample
Model 1 Model 2 Model 3 Model 4 East Middle West Big city Non-Big
city
GDPpc 0.283*** 0.238** 0.219** 0.230*** 0.380*** 0.121 0.173 0.355** 0.261**
(0.0893) (0.0954) (0.0905) (0.0865) (0.138) (0.189) (0.174) (0.164) (0.110)
ISO 0.431*** 0.0104 0.272** 0.232** 0.294 0.00946 0.0644 0.750*** 0.358***
(0.113) (0.109) (0.115) (0.113) (0.189) (0.229) (0.232) (0.247) (0.133)
Green 0.531** 0.418** 0.648*** 0.421** 0.341 0.273 1.070*** 0.144 0.625**
(0.213) (0.191) (0.210) (0.205) (0.429) (0.404) (0.302) (0.457) (0.244)
Pop 0.304*** 0.205*** 0.265*** 0.249*** 0.154* 0.532*** 0.173 0.0417 0.483***
(0.0619) (0.0689) (0.0628) (0.0615) (0.0915) (0.144) (0.124) (0.110) (0.0836)
Agent 0.322*** 0.282*** 0.313*** 0.314*** 0.358*** 0.253*** 0.283*** 0.403*** 0.285***
(0.0468) (0.0484) (0.0473) (0.0476) (0.0755) (0.0872) (0.0969) (0.0941) (0.0549)
LTD 1.010*** 0.913*** 0.922*** 0.937*** 0.902*** 0.598* 1.021*** 1.069*** 0.871***
(0.164) (0.168) (0.160) (0.159) (0.282) (0.326) (0.256) (0.268) (0.206)
Student 0.0946*** 0.104*** 0.109*** 0.113*** 0.0319 0.128** 0.108 0.147* 0.0281
(0.0357) (0.0317) (0.0348) (0.0336) (0.0547) (0.0627) (0.0692) (0.0757) (0.0443)
Land ᶭ0.305*** ᶭ0.121*** ᶭ0.284*** ᶭ0.268*** ᶭ0.185** ᶭ0.360*** ᶭ0.237** ᶭ0.337*** ᶭ0.317***
(0.0402) (0.0414) (0.0399) (0.0396) (0.0752) (0.0722) (0.0903) (0.0814) (0.0479)
Gov 0.453*** 0.505*** 0.407*** 0.404*** 0.398*** 0.631** 0.283 0.183 0.708***
(0.0994) (0.101) (0.0976) (0.0967) (0.134) (0.281) (0.238) (0.172) (0.144)
Constant ᶭ8.236*** ᶭ6.863*** ᶭ6.565*** ᶭ6.373*** ᶭ6.934*** ᶭ8.382* ᶭ5.213 ᶭ2.619 ᶭ11.87***
(1.657) (2.123) (1.826) (1.763) (2.589) (4.284) (3.583) (3.078) (2.034)
Observations 276 276 276 276 110 105 61 68 208
R-squared 0.796 0.890 0.811 0.830 0.844 0.720 0.819 0.874 0.688
FE No Province Area_1 Area_2 No No No No No
Note: Standard errors in parentheses; ***p<0.01, **p<0.05, *p<0.1; FE refers to Fixed effect. Area_1 and Area_2 refer to
different division of China, Area_1 divides China into three regions, east, middle, and west regions, Area_2 divides China
into 7 regions, East, Northeast, North, Central, South, Northwest, Southwest
Source: Indicated in Table 1
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
11
MGWR models. For the GWR and MGWR models, the mean and standard deviation of the coefficient
estimates are displayed, as opposed to the complete set of local coefficient estimates.
ɉ The bandwidth selection for both GWR and MGWR models employs adaptive bandwidths, utilizing
a bi-square kernel function as the weighting mechanism. The higher R2 values and lower AIC values
suggest that the GWR and MGWR models exhibit superior goodness of fit. Furthermore, Moran’s I
value are calculated for the residuals of each model to assess spatial autocorrelation. As noted by Gu,
Yu, et al. (2021), a significant Moran’s I value for residuals may potentially lead to endogeneity issues
and biased result. Examining the Moran’s I value in two different weight matrixes for residuals reveals
that the OLS model demonstrates a lower fit compared to the GWR and MGWR models. Given that
the MGWR model accounts for varying scales of impact and displays relatively lower AIC values and
higher R2 values, it is concluded that the MGWR model provides the optimal fit. Additionally, GWR
identifies clustering in variables with similar spatial patterns, potentially increasing multicollinearity
risks, MGWR’s variable bandwidths can mitigate these spatial similarities between independent
variables.
Table 3 Comparison of Regression Result of OLS, GWR and MGWR Model
Variable OLS GWR MGWR
AIC 322.438 295.078 231.932
RMSE 0.426 0.381 0.338
R20.796 0.881 0.907
Adj R20.789 0.855 0.886
Moran’s I
in residual
Contiguity matrix 0.213*** 0.048 ᶭ0.058
Inverse-distance matrix 0.126*** 0.004 ᶭ0.041
Note: Analysis by author. *** p<0.01, ** p<0.05, * p<0.1. RMSE refers to Root-mean-square
deviation statistics.
Source: Indicated in Table 1
4.3ᶮResult of MGWR Model and Discussion
ɉ As explained in the methodology part, MGWR allows each explanatory variable has its own
bandwidths, which indicates the impact scales for the spatial process. Having real estate investment
per capita as dependent variable, the bandwidth for each independent variable is showing in Table 4.
To be noted, the full samples size for regression is 276. Having bandwidth as 275 indicate a global
influence of this variable to real estate investment. There are economic growth (GDPpc), industrial
structure (ISO), number of college student (Student) and Government expenditure (Gov) as the global
parameter. Population (Pop) has the lowest bandwidth indicate a small influencing scale and a higher
level of spatial heterogeneity. The coefficient variance result from MGWR model is shown in the Table
5.
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
12
ɉ The parameter estimates for significant variables and noteworthy cities are illustrated using GIS
software. Cities with missing values are represented in white, while dark grey indicates cities with
parameter estimates not statistically significant at the 5% level. GDP per capita, Student, and Gov are
global parameters, causing their estimates to vary minimally across space, and resulting in a simplistic
spatial pattern of their influence. The maps in Figure 2 provide further explorations.
Table 4 Bandwidth of Each Explanatory Variable from MGWR Model
Variable Bandwidth ENP_j Adj t-val (95%) DoD_ j
Intercept 43 14.958 2.960 0.519
GDPpc 275 1.179 2.039 0.971
ISO 275 1.255 2.065 0.960
Green 98 6.204 2.669 0.675
Pop 55 13.176 2.920 0.541
Agent 160 4.003 2.515 0.753
LTD 119 5.913 2.652 0.684
Student 275 1.199 2.046 0.968
Land 214 2.770 2.379 0.819
Gov 275 1.173 2.037 0.972
Note: Analysis by author. ENP_j represents the effective quantity of parameters, while the adjusted
t-value at a 95% confidence level is denoted by Adj t-val (95%); DoD_j signifies the extent of
dependency.
Source: Indicated in Table 1.
Table 5 Summary Statistics for MGWR Parameter Estimates
Mean STD Min Median Max
GDPpc ɉ0.097 0.003 0.08 ɉ0.097 ɉ0.102
ISO ɉ0.013 0.002 ɉ0.010 ɉ0.013 ɉ0.018
Green ɉ0.011 0.120 ᶭ0.203 ɉ0.009 ɉ0.199
Pop ɉ0.172 0.127 ᶭ0.111 ɉ0.163 ɉ0.463
Agent ɉ0.263 0.051 ɉ0.189 ɉ0.244 ɉ0.377
LTD ɉ0.192 0.072 ɉ0.013 ɉ0.211 ɉ0.310
Student ɉ0.142 0.003 ɉ0.138 ɉ0.141 ɉ0.149
Land ᶭ0.152 0.035 ᶭ0.215 ᶭ0.152 ᶭ0.098
Gov ɉ0.224 0.002 ɉ0.221 ɉ0.223 ɉ0.229
Intercept ɉ0.088 0.240 ᶭ0.473 ɉ0.079 ɉ0.549
Note: Analysis by author. STD refers to standard deviation.
Source: Indicated in Table 1.
ɉ From the demand side, household income level, indicated by GDP per capita, correlates with real
estate investment globally. Despite minor differences, the effects of GDP per capita are slightly more
pronounced in southern and eastern cities, where developed cities such as Shenzhen, Guangzhou,
and Shanghai are located. This finding aligns with the consensus among researchers that a stronger
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
13
economic situation attracts more investment (D. Zhang et al. 2018; J. Zhang et al. 2012). From the
supply side, government still plays an important role for real estate market overall. Government
spending significantly influences middle and western cities, which are relatively less developed
areas. Two possible explanations for this phenomenon are: first, the implementation of the Belt and
Road policy has led to the central government placing greater emphasis on western regions; second,
considering the scarcity of land, real estate development in the eastern and southern regions started
earlier than in the central region, making the central area has more abundant land available for leasing.
However, in general, in terms of the supply side, the impact of government expenditure on real estate
investment is still comprehensive.
ɉ Regarding the impact of share of student to total population, the influence is somewhat more
substantial in northeastern China, an area grappling with severe population and talent loss issues.
The effect gradually diminishes towards the west. This study posits that university graduates are
more likely to remain in the area, and their anticipated higher income enables them to afford housing.
Consequently, this stimulates investment in the housing market. Based on the findings, the marginal
effect of college students in the northeastern region is more pronounced.
ɉ Apart from the global parameters, land emerges as a significant factor with a relatively higher
influence scale, ranking 215 out of 276 cities in total. The land per capita demonstrates a negative
marginal effect on real estate investment across the board. The lower land per capita in the developed
eastern area hampers investment to some extent, which can be attributed to industrial agglomeration.
ɉ The influence of remaining four factors, namely Pop, Green, Agent, and LTD, exhibits a relatively
smaller handwidth indicating a higher level of spatial heterogeneity. Notably, a significant negative
marginal effect of green coverage is observed in the central region of China. This negative effect
implies that rapid real estate investment is associated with a reduction in green land. Moreover, the
impact of green coverage on real estate investment in the eastern region is not found to be statistically
significant. This phenomenon may be attributed to the fact that, compared to the quality of living
conditions, higher economic growth and other incentives hold greater appeal for real estate investors.
Conversely, the western region exhibits a higher attraction for real estate investment with respect to
green coverage, which aligns with the findings of previous studies conducted by other scholars (Gu &
Yu et al. 2021).
ɉ Population, considered to be a key explanatory variable for real estate, was found to be statistically
insignificant in the eastern, southern, and middle-western regions of China. This implies that
population is not the primary driving factor behind real estate investment in these areas. This lack of
significance should raise concerns from the government. Additionally, the smallest bandwidth of 55
indicates a significant variation in the impact across different spatial areas compared to other variables.
ɉ In contrast, Agent exhibits a higher level of influence in the south regions, particularly in areas near
Guangdong Province. From there, the influence gradually decreases as we move from the southern to
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
14
Figure 2 Spatial Patterns of Estimators of Significant Explanatory Variables
Note: Analysis by author. Plotted by ArcGIS Pro software.
Source: Indicated in Table 1.
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
15
northern regions and towards the middle. This may be attributed to the rapid economic development
of Guangdong Province in southern China, particularly after 2015, which led to a surge in migration
and created a greater space for Agent to play a role in the real estate market.
ɉ In terms of the LTD, the eastern coastal area and certain cities in the central region exhibit the
highest coefficients, gradually diminishing towards the northeastern and southwestern areas. This
pattern differs from the findings of the regional OLS analysis presented in Table 2. The insignificance
of some cities in the central and northeastern areas may contribute to this disparity. The relationship
between the LTD and bank profitability is notable. During periods of economic stability or growth,
an increase in LTD signifies market confidence, leading banks to be more inclined to extend credit.
Conversely, in times of economic recession, a higher LTD indicates heightened financial risk.
ɉ The bandwidth for the intercept is notably small, indicating significant spatial variation in how the
intercept affects real estate investment. This might result from pronounced spatial discrepancies in
initial real estate investments and unaccounted-for factors like housing prices. As seen in the previous
figure, developed eastern and southern regions experience a greater impact from the intercept,
whereas the central region remains largely unaffected. Surprisingly, the northeastern area appears to
have a negative influence from the intercept.
5ᶮConclusion
ɉ The real estate sector plays a pivotal role in China’s economic development. While numerous
studies have explored the determinants of real estate investment, there is growing recognition of the
importance of spatial factors for enhancing estimation accuracy. The advancement of spatial statistics
has also brought attention to issues of spatial non-stationarity. Given China’s vast geographical
landscape, this study addresses this oversight. Utilizing the MGWR model, this paper aims to
investigate the spatial variation of determinant of real estate investment.
ɉ Compared to OLS, both GWR and MGWR account for spatial non-stationarity, resulting in more
accurate estimators. They allow for sub-sampling to halt when a new observation significantly alters
the estimator. Instead of producing a generalized estimator for each determinant, GWR and MGWR
offer specific estimators for each determinant at every observation. Upon a comparative analysis with
the GWR model, it becomes evident that the MGWR offers the flexibility for independent variables to
function within different bandwidths. This provision infers vary spatial impact originating from diverse
variables, thereby amplifying the fidelity and granularity of the resultant findings.
ɉ Through OLS regression, this study selects key factors for GWR and MGWR analysis. From the
demand side, income level, education, and from supply side, government expenditure are found to have
a global influence on real estate investment, albeit with some slight variations.
ɉ Population, real estate agents, loan-to-deposit ratio and green coverage exhibit significant spatial
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
16
variation on their impact on real estate investment.
ɉ Based on empirical findings, several key insights emerge relevant to developer’s decision making.
Beyond the already economically prosperous regions, particularly the eastern and south-eastern areas,
indicators like the scale of governmental expenditure and the loan-to-deposit ratio highlight promising
investment opportunities in central regions such as Shaanxi and Inner Mongolia. Moreover, with
higher educational institutions in northeastern China intensifying efforts to boost student enrollment,
there’s a potential uptick in investment attractiveness in this area. Simultaneously, the rising presence
of agencies in the southwest suggests a likely increase in real estate demand in the near term.
ɉ Based on our findings, there are several insights that the government may take into account when
formulating regulations for the real estate market. Different regions may require distinct approaches.
For instance, the northeastern area, which was previously an industrial hub, later faced a significant
recession resulting in numerous abandoned houses. If there’s an intention to stimulate real estate
investment in this region, the government might consider enhancing higher education institutions and
promoting the housing leasing market. Conversely, if the aim is to regulate real estate investment in
the southwestern area, a potential strategy could be to manage the housing leasing market.
ɉ Several regulation recommendations for government sector are provided as well:
ɉ 1. For fostering economic growth in the northeastern region through real estate investment, the
government might consider bolstering higher education initiatives to attract and retain talent.
ɉ 2. In the eastern and southern regions, it’s crucial for governments to first, regulate housing price
and construction requirements; second, to enhance rental policies, addressing housing concerns for
residents, and institute safeguards against housing speculation and potential bubbles.
ɉ 3. Emphasizing the preservation of green spaces in the eastern and central regions is paramount for
maintaining sustainable land use practices.
ɉ 4. To diminish dependency on the banking sector, diversifying the sources of real estate investment
funds is advisable. The adoption of Real Estate Investment Trusts (REITs) could be explored as an
alternative investment avenue.
ɉ Three key limitations of this study must be highlighted: Firstly, the findings rely on average data
spanning from 2016 to 2018; more in-depth cross-sectional or panel analyses might be necessary.
Secondly, missing values could influence bandwidth selection, potentially distorting spatial
representation. Lastly, considering the constraints of this model, the methodology for computing
t-tests and p-values warrants reevaluation.
Endnotes
1 According to the World Bank, urbanization refers to “the share of the population living in urban areas in the given
year”.
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
17
2 Source: China Statistical Yearbook
3 Drawing on Yu’s work (2015), this paper introduces modifications to certain components of the formula and
subsequently computes the Theil index to assess the degree of balanced growth across different industries. The
Theil index is used as a measure, where a larger value indicates greater imbalances among the various industries.
The referenced formula is presented below:
Where i represents the i-th industry, and n is the total number of industries. Y is the total industrial output value;
L denotes the total employment in industry. There are many ways to decompose the whole industrial system. Here
in this paper, I utilize three industries: the primary industry, secondary industry and tertiary industry.
References
Agarwal, S., He, J., Sing, T. F., & Song, C. 2019. Do real estate agents have information advantages in housing
markets? Journal of Financial Economics. 134(3): 715
Ż
735.
Alonso, W. 1964. The Historic and the Structural Theories of Urban Form: Their Implications for Urban Renewal.
Land Economics. 40(2): 227
Ż
231.
Anselin, L. 1998. GIS research infrastructure for spatial analysis of real estate markets. Journal of Housing Research.
9(1): 113
Ż
133.
Anselin, L., & Can, A. 1986. Model Comparison and Model Validation Issues in Empirical Work on Urban Density
Functions. Geographical Analysis. 18(3): 179
Ż
197.
Baryla, E., & Ztanpano, L. 1995. Buyer search duration in the residential real estate market: The role of the real
estate agent. Journal of Real Estate Research. 10(1): 1
Ż
13.
Belsley, D. A., Kuh, E., & Welsch, R. E. 1980. Regression Diagnostics Identifying influential Data and Source of
Collinearity. New York: NY Wiley.
Chen, J., Guo, F., & Wu, Y. 2011. Chinese urbanization and urban housing growth since the mid
Ż
1990s. Journal of
Housing and The Built Environment. 26(1): 219
Ż
232.
Chen, T., Liu, L., Xiong, W., Zhou, L.-A., & others. 2017. Real estate boom and misallocation of capital in China. China
Economic Review. 60(1): 101400.
Chen, Y., & Lee, C.-C. 2020. The impact of real estate investment on air quality: Evidence from China. Environmental
Science and Pollution Research. 27(1): 22989
Ż
23001.
Cui, Z., Zhang, Z., & Li, C. 2022. Research on the Impact of Aging and New Generation in the Population Structure on
China’s Real Estate Price Volatility. Journal of Finance Research. 6(1): 5
Ż
14.
Dasgupta, B., Lall, S. V., & Lozano-Gracia, N. 2014. Urbanization and housing investment. World Bank Policy Research
Working Paper. 115004.
DiPasquale, D., Wheaton, W. 1994. Housing Market Dynamics and the Future of Housing Prices. Journal of Urban
Economics. 35(1): 1
Ż
27.
Dubin, R. A. 1992. Spatial autocorrelation and neighborhood quality. Regional Science and Urban Economics. 22(3):
433
Ż
452.
Dubin, R. A. 1998. Predicting House Prices Using Multiple Listings Data. Journal of Real Estate Finance and
Economics. 17(1): 35
Ż
59.
Fotheringham, A. S., Brunsdon, C., & Charlton, M. 2003. Geographically weighted regression: The analysis of spatially
varying relationships. John Wiley & Sons.
Fotheringham, A. S., Yang, W., & Kang, W. 2017. Multiscale Geographically Weighted Regression (MGWR). Annals of
the American Association of Geographers. 107(6): 1247
Ż
1265.
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
18
Frunza, M.-C., & Minoiu, C. 2015. The determinants of real estate investment in Europe: Evidence from panel data.
Procedia Economics and Finance. 32(1): 164
Ż
171.
Green, R. K. 1997. Follow the leader: How changes in residential and non-residential investment predict changes in
GDP. Real Estate Economics. 25(2): 253
Ż
270.
Griffith, D. A. 1981. Towards a Theory of Spatial Statistics: A Rejoinder. Geographical Analysis. 13(1): 91
Ż
93.
Gu, H., Jie, Y., Li, Z., & Shen, T. 2021. What Drives Migrants to Settle in Chinese Cities: A Panel Data Analysis.
Applied Spatial Analysis and Policy. 14(2): 297
Ż
314.
Gu, H., Yu, H., Sachdeva, M., & Liu, Y. 2021. Analyzing the distribution of researchers in China: An approach using
multiscale geographically weighted regression. Growth and Change. 52(1): 443
Ż
459.
He, L., Bennett, M. M., & Jiang, R. 2022. The uneven geography of real estate investment by Mainland Chinese state-
owned and private enterprises in the US: Local market conditions, migration, and ethnic networks. Environment
and Planning A: Economy and Space. 54(4): 1
Ż
22.
Huang, Z., & Du, X. 2017. Government intervention and land misallocation: Evidence from China. Cities. 60(1): 323
Ż
332.
Johnes, J., & Taylor, J. 1989. The first destination of new graduates: Comparisons between universities. Applied
Economics. 21(3): 357
Ż
373.
Klimczak, K. 2010. Determinants of Real Estate Investment. Economics & Sociology. 3(2): 58
Ż
66.
Koichi, M., & Bertrand, R. 2016. Asia’s Financial Crisis and the Role of Real Estate. Routledge.
Krugman, P. 1991. Increasing Returns and Economic Geography. Journal of Political Economy. 99(3): 483
Ż
499.
Lieser, K., & Groh, A. P. 2014. The Determinants of International Commercial Real Estate Investment. Journal of Real
Estate Finance and Economics. 48: 611
Ż
659.
Liu, T.-Y., Su, C.-W., Chang, H.-L., & Chu, C.-C. 2018. Is urbanization improving real estate investment? A cross-
regional study of China. Review of Development Economics. 22(2): 862
Ż
878.
Luo, H. 2020. Research on the Interaction between Higher Education Resource Allocation and Real Estate Price.
Open Journal of Social Sciences. 08(04): 58
Ż
68.
Mak, S., Choy, L. H. T., & Ho, W. K. O. 2012. Region-specific Estimates of the Determinants of Real Estate
Investment in China. Urban Studies. 49(4): 741
Ż
755.
Moshirian, F., & Pham, T. 2000. Determinants of US investment in real estate abroad. Journal of Multinational
Financial Management. 10(1): 63
Ż
72.
Saiz, A. 2010. The Geographic Determinants of Housing Supply. Quarterly Journal of Economics. 125(3): 1253
Ż
1296.
Shaojian, W., Shuang, G. A. O., & Jing, C. 2020. Spatial heterogeneity of driving factors of urban haze pollution in
China based on GWR model. Geographical Research. 39(3): 03000651.
Wang, S., Fang, C., Ma, H., Wang, Y., & Qin, J. 2014. Spatial differences and multi-mechanism of carbon footprint
based on GWR model in provincial China. Journal of Geographical Sciences. 24(4): 612
Ż
630.
Wang, X., Hui, E. C.-M., & Sun, J. 2018. Population aging, mobility, and real estate price: Evidence from cities in
China. Sustainability. 10(1): 3140.
Wang, Y., Wang, S., Li, G., Zhang, H., Jin, L., Su, Y., & Wu, K. 2017. Identifying the determinants of housing prices in
China using spatial regression and the geographical detector technique. Applied Geography. 79: 26
Ż
36.
Wigren, R., & Wilhelmsson, M. 2007. Construction investments and economic growth in Western Europe. Journal of
Policy Modeling. 29(3): 439
Ż
451.
Wolny, A. 2016. Accessibility of Real Estate by Transportation as a Determinant of the Development of Suburban Real
Estate Markets - Case Study. Real Estate Management and Valuation. 24(1): 5
Ż
18.
Wu, Y., Heerink, N., & Yu, L. 2020. Real estate boom and resource misallocation in manufacturing industries:
Evidence from China. China Economic Review. 60: 101400.
Zhang, D., Cai, J., Liu, J., & Kutan, A. M. 2018. Real estate investments and financial stability: Evidence from regional
commercial banks in China. The European Journal of Finance. 24(16): 1388
Ż
1408.
Zhang, J., Wang, J., & Zhu, A. 2012. The relationship between real estate investment and economic growth in China:
A threshold effect. Annals of Regional Science. 48(1): 123
Ż
134.
Forum of International Development Studies. 54
Ż
4ᶨMar. 2024ᶩ
19
Appendix
ɉ Appendix 1 displays the results of the Variance Inflation Factor (VIF) test for collinearity, conducted
for each model presented in Table 2. A VIF value below 10 indicates an absence of significant
collinearity problem or that the collinearity issue can be disregarded. As demonstrated in the table, all
models pass the collinearity test.
Appendix 1 Collinearity Inspection of The Explanatory Variables Based on Different Sample Groups
VIF (1) (2) (3) (4) (5) (6)
Full East Middle West Big city Non-Big city
Model 1
Ż
4 Model 5 Model 6 Model 7 Model 8 Model 9
GDPpc 3.31 3.87 5.01 3.03 3.15 3.16
Student 3.28 3.23 3.46 4.08 3.82 2.07
Gov 3.18 4.12 5.01 4.28 5.14 3.52
Pop 2.95 2.33 5.04 4.96 2.95 3.59
Agent 2.92 5.06 2.32 3.29 5.08 1.87
Land 1.53 2.01 2.12 1.93 2.02 1.66
ISO 1.49 1.74 1.94 2.34 2.01 1.39
LTD 1.36 1.33 1.73 1.59 1.66 1.28
Green 1.32 1.41 1.15 1.48 1.16 1.41
Note: Analysis by author.
Source: Indicated in Table 1
