The Dynamic Evolution of the Structure of an Urban Housing Investment Niche Network and Its Underlying Mechanisms: A Case Study of 35 Large and Medium-Sized Cities in China

Abstract

With the growth of urban agglomerations, the spatial diffusion of housing investment is clear; however, little research has been carried out to address its network characteristics and underlying mechanisms of influence. Using data on 35 large and medium-sized cities, this paper applies niche theory to housing investment, constructing a housing investment niche index that includes resources, the housing market, the social economy, and policy. The purpose is to study the characteristics of the network structure and its mechanisms of influence based on an improved gravity model and a temporal exponential random graph model (TERGM). Supported by the analysis of the network structure, we find that the node degree within the network is low, the network density exhibits an inverted “V” shape, and the network level suggests the existence of a “rich cities club”. According to the local network clustering analysis, the cities are divided into three clusters: the Yangtze River Delta region, the Beijing–Tianjin–Hebei region, and the central and eastern regions. Furthermore, analysis of the endogeneity in the structure reveals that there is a hierarchy of cities with high economic development levels, which makes it difficult to establish an investment network with strong relationships. The effects of the attributes are consistent with the predictions of location theory and new economic geography theory. The external network effects conform to the law of the general gravity model. Our research provides insights into the ways in which the misplaced competition in housing investment between cities in a region and the flow of production factors can be reasonably guided.

1. Introduction

Network externality theory suggests that urban development depends on the ability of a city to establish and maintain linkages with other cities [1], and that the contribution of such linkages to urban development may exceed the impact of local economic agglomeration [2,3]. With the rapid development of the real estate industry, behaviours such as off-site investment and mergers and acquisitions have already broken geographical boundaries. In addition, competition among similar cities and resource misallocations across cities have exacerbated the problem of imbalanced regional development, leading to both competition and cooperation in housing investment between cities, a situation that is becoming more complex [4]. To promote the ethical development of the real estate market, the Chinese government implemented a policy titled “Houses are for living in and not for speculative investment”; however, after years of development, there remains a mismatch between urban housing and the urban population. China’s Housing Stock Report 2021 shows that the ratio of housing units in first tier and second tier cities in 2020 is 0.97 and 1.08, respectively, with regional supply and demand differences. Large and medium-sized cities generally have higher housing prices, faster growth rates, and excess investment. For this reason, determining how to implement policies and provide guidance tailored to the needs of the city, that is, how to guide the misplaced competition in housing investment between different cities, reasonably guide the flow of the factors of production, and prevent overheating and cooling in urban housing prices, has become necessary in order to promote balance in the development of housing investment between cities.

The existing real estate investment literature mainly discusses the relationships among urbanization, urban scale, economic growth and real estate investment, as well as the factors that influence real estate investment [5,6]. As large and medium-sized city centres have developed, increasing the strength of their siphon and diffusion effects, the volume of research on urban real estate investment from regional and spatial perspectives has increased [7,8]. For example, Alkay argued that accessibility, environmental quality, and the size and quality of the housing stock affect the diversity and heterogeneity in the spatial distribution of real estate investment [9]. Zhang et al. used Moran’s I to measure the extent of the spatial aggregation in China’s housing investment [10]. Janoschka et al. studied cross-border investment in Spanish real estate and its social and spatial effects [11]. Bao et al. used a spatial econometric model to study the impact of real estate investment on PM2.5 concentrations in East, Central, and Western China [12]. Yang et al. analysed the ways in which the evolution of China’s high-speed railway network impacted the aggregation of real estate investment [13].

Nevertheless, existing research that adopts a spatial perspective has mainly focused on real estate investment at the national or regional level [14,15], while urban-level studies have mostly focused on the spatial variation in and factors influencing different types of real estate investment [16], ignoring the relationships between cities. With the increase in the flow of production factors between cities, spatial connections tend to become flat and diverse, and the trends in the development of urban networking have become more obvious [17]. The existing literature on urban networks has focused on the impact of the networks affiliated with listed companies, the timing of investment and financing for intercity enterprises, corporate culture, and other factors related to the structural characteristics, spatial patterns, and style of evolution of urban networks [18]; on the other hand, there have been few analyses of urban networks from the perspective of housing investment.

Niche theory refers to the basic survival space occupied by the biological populations in a geographic environment [19]. The most representative form of niche theory is multidimensional hypervolume niche theory, proposed by Hutchinson [20], which refers to the relative position of an organism within a population and its interactions with its population and the environment in the natural ecosystem. Subsequently, Frosch et al. applied this version of niche theory to industrial systems to study the development of a synergistic relationship between industrial activities and the environment [21]. Subsequently, niche theory has been used to analyse spatial patterns and competitive advantages in industrial development [22,23]. Pan proposed a two-fold relationship of competition and cooperation between cities; the status of cities within their network is determined by their resources and production factors [24]. Because resources are limited, a large number of scholars have utilized niche theory to study urban problems, extending concepts such as the urban niche [25]. However, niche theory has seldom been applied in the field of urban housing investment, even though urban housing investment can be regarded as an ecological niche system; its components are analogous to biological units within a natural ecosystem, which can influence and even dominate their environment. This image reflects the development of housing investment within a city and the competition among cities.

This paper studies 35 large and medium-sized cities in China, constructing a comprehensive four-dimensional index of urban housing investment niches. The four dimensions are urban resources, the housing market, the social economy and real estate policy. Furthermore, a modified gravity model and a temporal exponential random graph model (TERGM) are used to study the network structure of the urban housing investment niche and the mechanisms by which it forms. As shown by the above analysis, this article differs from the existing literature in the following ways:

(1)

Unlike earlier studies that analysed investment data, this paper introduces niche theory into the field of urban housing investment research, constructs a comprehensive index of housing investment niches, and provides theoretical guidance for the efficient allocation of housing investment resources.

(2)

Due to the directionality of the spatial correlations within the housing investment network, the impacts of population flows, capital flows, traveling distance, and economic distance on the urban housing investment niche network are fully addressed.

(3)

Furthermore, regarding research methods, a TERGM (based on a static ERGM) containing endogenous structural effects, attribute effects, exogenous spatial effects, and temporal effects was constructed and dynamic network analysis was used to conduct an in-depth study of the mechanisms influencing the urban housing investment niche network.

The remaining sections of this paper are as follows. Section 2 presents the theoretical foundation and research model. Section 3 presents the research design. Section 4 presents the characteristics of the dynamic changes within the network structure. Section 5 presents the results of the mechanism analysis. Finally, Section 6 summarizes the conclusions and provides a discussion.

2. Theoretical Foundation and Research Model

2.1. Complex Urban Housing Investment Niche System

The complex urban housing investment niche system is a mechanism by which linkages are developed, and is composed of a resources–housing–people system with the ultimate goal of satisfying people’s well-being. The complex urban housing investment niche system includes systems of resource niches, housing market niches, social economic niches, and real estate policy niches (Figure 1). Specifically, public resources, human resources and land resources constitute the first ecological niche, that of resources; the rational use of resources can ensure that the scale and structure of the supply of land for housing is optimal, which is a necessary condition for housing investment. Second, market demand and market supply constitute the ecological niche of the housing market; the linkage between housing prices and land prices can promote the beneficial circulation of capital in the real estate market. Third, the economic environment and the social environment together constitute the socioeconomic ecological niche. A good investment environment can ensure that the needs of real estate enterprises for the development of investment and financing can be met. Fourth, housing policy, monetary policy, and land policy constitute the ecological niche for real estate policy. The regulation of housing investment via the relevant policy levers can balance supply and demand within the housing market and improve the real estate system to a certain extent. The flow of production factors between cities forms a network that can be thought of as the niche for housing investment. Overall, the local and individual characteristics of the network are affected by the niche structure and the factors determining housing investment, and these characteristics exhibit spatial variation. Moreover, the urban economy, population size, and travelling distance affect the formation of the network defining the housing investment niche. Thus, this paper focuses on the “structure–factor–network” concept.

As shown in the above analysis, in this paper we consider urban fundamentals such as housing market capacity, market supply and demand, and purchasing power. The urban housing investment niche index considered here includes the urban resource niche, the housing market niche, the socioeconomic niche, and the policy niche (Table 1).

We chose the observation sample time for this paper on the basis of the following two considerations. First, the impact of major events; followings the COVID-19 outbreak in 2020, China’s economy is under additional strain, leading to an even steeper decline in the housing market. The “China Real Estate Statistical Yearbook (2021)” reveals that for urban housing investment, the land acquisition area in 2020 was 255.363 million square meters, which is much lower than in 2019 and equivalent to the data for 2017. In contrast to the era of steady growth from 2011 to 2019, China’s housing investment in 2020 is entering a period of turbulent adjustment due to exogenous uncertainties [26]. If data for years after 2020 were included in this study for calculation, there would be a large deviation in the interpretation of China’s housing market as a whole [27]. Second, the robustness of the model results; for analysis of dynamic changes in network structure, TERGM is the tool of choice, allowing investigation of the impact of the t − 1 period on the network structure of the t period [28]. However, abnormal year data are often affected by exogenous uncertainty, making it difficult to evaluate the endogenous effects of the system and hence jeopardizing the robustness of the TERGM results. For both of the above reasons, this paper establishes the sample observation time as 2011–2019.

Table 1. Index system for the urban housing investment niche.

Type of Urban Housing Investment Niche	Primary Index	Secondary Index (Unit)	Description of Secondary Indicators	Indicator Source
Urban resources niche	Public resources	Per capita area of urban green space (m2/person)	Urban green space/total population at year end	China Urban Statistical Yearbook
		Per capita area of urban roads (m2/person)	Urban road area/population at year end	China Urban Statistical Yearbook
		Number of hospital beds per 10,000 people (bed/104 persons)	(Total number of hospital beds/total population) × 10,000	China Urban Statistical Yearbook
	Human resources	Number of college students per 10,000 people (students/104 persons)	(Total number of college students/population) × 10,000	China Urban Statistical Yearbook
		Number of employed persons in urban areas (103 persons)	Total number of employees	China Urban Statistical Yearbook
	Land resources	Built area (km2)	Urban built area	China Urban Statistical Yearbook
		Residential land supply area (hectare)	Includes land for indemnificatory and commercial housing	Bureau Of Urban Planning and Natural Resources
Housing market niche	Market demand	Housing sales area (103 m2)	Commercial housing sales area	China Urban Statistical Yearbook
		Natural population growth rate (‰)	Birth rate-death rate	China Urban Statistical Yearbook
		Urbanization rate (%)	Urban population/permanent population	China Urban Statistical Yearbook
	Market supply	Expected rate of return from real estate (%)	The average housing price growth rate over the previous three years	CEIC Macroeconomic Database
		Land area purchased by real estate enterprises (104 m2)	Land area purchased by real estate development firms	CEIC Macroeconomic Database
		Number of real estate development firms (firms)	Number of real estate development firms	China Urban Statistical Yearbook
Social economic niche	Economic environment	Per capita GDP (yuan/person)	GDP/Total population at year end	CEIC Macroeconomic Database
		Per capita disposable income of the urban population (yuan/person)	Per capita disposable income of the urban population	CEIC Macroeconomic Database
		Balance of loans from financial institutions (104 yuan)	Balance of loans from financial institutions	China Urban Statistical Yearbook
	Social environment	Infrastructure construction (108 yuan)	Fixed asset investment-real estate investment	China Urban Statistical Yearbook
		Share of the tertiary industry in GDP (%)	Tertiary industry output/GDP	China Urban Statistical Yearbook
Real estate policy niche	Housing policy	Weighted average interest rate on individual housing loans (%)	Converted to real interest rates with city-specific CPIs	CEIC Macroeconomic Database
	Monetary policy	M2 growth rate-GDP growth rate-inflation rate (%)	M2 growth rate-GDP growth rate-inflation rate	CEIC Macroeconomic Database
	Land policy	Land price (yuan)	Average transaction price of land	CEIC Macroeconomic Database

Note: In order to eliminate the influence of time, all prices used in the above indicators are real prices with 2011 as the base year. Certain index data could not be obtained directly from the relevant yearbooks and are therefore indirectly calculated; for details, please see the indicator descriptions in Table 1. Note that the proxy variable for housing policy is the weighted average interest rate on loans for individual housing [29]. In order to capture differences in the housing policies of the 35 large and medium-sized cities, the CPI for each city (with 2011 as the base period) is used to convert nominal interest rates into real interest rates. China has long adopted a combination of both quantitative and price-based monetary policy; because China currently depends mainly on quantitative easing its major monetary policy tool, this article draws upon the existing literature [30] in selecting the M2 growth rate net of the real GDP growth rate and the city-specific CPI growth rate as a proxy variable for monetary policy. For our land policy indicator we refer to the research of Ye Jianping [31] and adopt land prices as our proxy variable. The data in this paper are taken from the China Urban Statistical Yearbook, the Urban Planning and Natural Resource Bureaus of the 35 cities under study, the CEIC Macroeconomic Database, and the China Real Estate Statistical Yearbook. The geographical distances between cities used in the gravity model are taken from the China Transportation Atlas. source: Authors, 2022.

Table 1. Index system for the urban housing investment niche.

Type of Urban Housing Investment Niche	Primary Index	Secondary Index (Unit)	Description of Secondary Indicators	Indicator Source
Urban resources niche	Public resources	Per capita area of urban green space (m2/person)	Urban green space/total population at year end	China Urban Statistical Yearbook
		Per capita area of urban roads (m2/person)	Urban road area/population at year end	China Urban Statistical Yearbook
		Number of hospital beds per 10,000 people (bed/104 persons)	(Total number of hospital beds/total population) × 10,000	China Urban Statistical Yearbook
	Human resources	Number of college students per 10,000 people (students/104 persons)	(Total number of college students/population) × 10,000	China Urban Statistical Yearbook
		Number of employed persons in urban areas (103 persons)	Total number of employees	China Urban Statistical Yearbook
	Land resources	Built area (km2)	Urban built area	China Urban Statistical Yearbook
		Residential land supply area (hectare)	Includes land for indemnificatory and commercial housing	Bureau Of Urban Planning and Natural Resources
Housing market niche	Market demand	Housing sales area (103 m2)	Commercial housing sales area	China Urban Statistical Yearbook
		Natural population growth rate (‰)	Birth rate-death rate	China Urban Statistical Yearbook
		Urbanization rate (%)	Urban population/permanent population	China Urban Statistical Yearbook
	Market supply	Expected rate of return from real estate (%)	The average housing price growth rate over the previous three years	CEIC Macroeconomic Database
		Land area purchased by real estate enterprises (104 m2)	Land area purchased by real estate development firms	CEIC Macroeconomic Database
		Number of real estate development firms (firms)	Number of real estate development firms	China Urban Statistical Yearbook
Social economic niche	Economic environment	Per capita GDP (yuan/person)	GDP/Total population at year end	CEIC Macroeconomic Database
		Per capita disposable income of the urban population (yuan/person)	Per capita disposable income of the urban population	CEIC Macroeconomic Database
		Balance of loans from financial institutions (104 yuan)	Balance of loans from financial institutions	China Urban Statistical Yearbook
	Social environment	Infrastructure construction (108 yuan)	Fixed asset investment-real estate investment	China Urban Statistical Yearbook
		Share of the tertiary industry in GDP (%)	Tertiary industry output/GDP	China Urban Statistical Yearbook
Real estate policy niche	Housing policy	Weighted average interest rate on individual housing loans (%)	Converted to real interest rates with city-specific CPIs	CEIC Macroeconomic Database
	Monetary policy	M2 growth rate-GDP growth rate-inflation rate (%)	M2 growth rate-GDP growth rate-inflation rate	CEIC Macroeconomic Database
	Land policy	Land price (yuan)	Average transaction price of land	CEIC Macroeconomic Database

Note: In order to eliminate the influence of time, all prices used in the above indicators are real prices with 2011 as the base year. Certain index data could not be obtained directly from the relevant yearbooks and are therefore indirectly calculated; for details, please see the indicator descriptions in Table 1. Note that the proxy variable for housing policy is the weighted average interest rate on loans for individual housing [29]. In order to capture differences in the housing policies of the 35 large and medium-sized cities, the CPI for each city (with 2011 as the base period) is used to convert nominal interest rates into real interest rates. China has long adopted a combination of both quantitative and price-based monetary policy; because China currently depends mainly on quantitative easing its major monetary policy tool, this article draws upon the existing literature [30] in selecting the M2 growth rate net of the real GDP growth rate and the city-specific CPI growth rate as a proxy variable for monetary policy. For our land policy indicator we refer to the research of Ye Jianping [31] and adopt land prices as our proxy variable. The data in this paper are taken from the China Urban Statistical Yearbook, the Urban Planning and Natural Resource Bureaus of the 35 cities under study, the CEIC Macroeconomic Database, and the China Real Estate Statistical Yearbook. The geographical distances between cities used in the gravity model are taken from the China Transportation Atlas. source: Authors, 2022.

2.2. Comprehensive Index of Urban Housing Investment Niches: Niche Theory

The subject of niche theory is the status of biological units and their influence on the environment. The “ecostate” of a biological unit refers to the state of the unit, and is formed by its biological growth and development over a period of time and its interactions with the environment. The ecostate includes the number of individual organisms and the amount of available energy, resource occupancy, and adaptability. The influence of biological units on the environment is referred to as the “ecorole” of those units, and includes the biological growth rate, energy-to-matter conversion ratio, population ratio, and ability of units to occupy a new environment [32]. Therefore, in this paper an urban housing investment niche refers to the position and influence of a city’s housing investment relative to the urban housing investments of numerous other cities. Specifically, it includes the “ecostate” and “ecorole”, with “ecostate” including the stock of resources needed to promote investment such as the population, market environment, and urban resources and the “ecorole” including the influence of investment on city characteristics such as its economic level, infrastructure construction, real estate policy, and monetary policy. The organic combination of the ecostate and the ecorole fully expresses the width and spatial influence of the urban housing investment niche. The formula for calculating a single-dimensional niche is shown in Formula (1)

$$N_i=\frac{S_i+A_i{P}_i}{{{\displaystyle \sum }}_{j=1}^n\left(S_j+A_j{P}_j\right)}$$

(1)

where i, j = 1, 2, …, n, N_i is the single-dimensional housing investment niche in city i, S_i and P_i are the ecostate and ecorole of city i, respectively, S_j and P_j are the ecostate and ecorole of city j, respectively, A_i and A_j are dimensional conversion coefficients, and S_i + A_iP_i and S_j + A_jP_j are the absolute niches of city i and city j, respectively.

The formula for calculating the full multidimensional niche is shown in Formula (2)

$$M_{ij}={\displaystyle \sum }_{j=1}^n{N}_{ij}/n$$

(2)

where M_ij is the comprehensive ecological niche for urban housing investment, N_ij is a single-dimensional niche for city i, n is the number of dimensions, and j is the number of cities.

2.3. Establishing Niche Networks for Urban Housing Investment: A Modified Gravity Model

This paper uses a modified gravity model to analyse the relationships in the network of urban housing investment niches. Not only are modified gravity models able to completely account for geographical distance, they can further reveal the trends in the evolution of spatial relationships. The basic equation for the urban gravity model is shown in Formula (3)

$$F_{ij}=k\frac{Q_i{Q}_j}{D_{ij}^b}$$

(3)

where F_ij is the gravitational value for the correlation between city i and city j while Q_i and Q_j measure the “quality” of cities i and j, respectively; D_ij is the geographical distance between city i and city j, k is the gravitational coefficient, and b is the distance attenuation coefficient.

The traditional gravity model has no spatial directionality, and the economic relations between cities are asymmetric. Hence, in order to highlight the directionality of the spatial correlations within the urban housing investment niche, the traditional gravity model is modified in this paper. The share of real estate and housing development investment in one city in the total housing investment in two cities is used for the corrected gravity coefficient, k_ij. However, given the influence of geographical and economic factors on the correlation in housing investment, the total distance between cities i and j is represented by the ratio of the geographical distance (D_ij) and the difference in per capita GDP (g_i − g_j) between cities. In combination with the overall index for the urban housing investment niche, housing development investment, the total population of the city at the end of the year, and gross regional product are used to measure the spatial correlation in the urban housing investment niche. The formula for the modified gravity model is shown in Formula (4)

$$F_{ij}=k_{ij}\frac{\sqrt[3]{M_i{H}_i{P}_i{G}_i} \times \sqrt[3]{M_j{H}_j{P}_j{G}_j}}{d_{ij}^2}, k_{ij}=\frac{H_i}{H_i+H_j}, d_{ij}=\frac{D_{ij}}{g_i-g_j}$$

(4)

where M_i and M_j are the comprehensive indexes for the housing investment niches in cities i and j, respectively, H_i and H_j are the amount of housing investment in cities i and j, respectively, P_i and P_j are the total populations of cities i and j, respectively, at the end of the year, and G_i and G_j are the GDPs of cities i and j, respectively.

2.4. Analysis of the Spatial Correlations in the Network Structure of the Urban Housing Investment Niche: A Temporal Exponential Random Graph Model (TERGM)

The exponential random graph model (ERGM) is a statistical model that uses a probabilistic model of a network to explain the observed characteristics of a real network [33]. It can completely account for the influence of internal and external factors on the formation of the network structure. As a frontier method for revealing the mechanism by which network relations are generated, its basic principle is as follows. Suppose that there is a real network, G = (V, E), where V and E are the set of real nodes and real edges in the network. If M = {(i, j); i ∈ V, i ≠ j} is the set of all possible edges between all nodes in network G, then E is a subset of M. First, a binary random variable Y is chosen to represent the elements of M. When (i, j) ∈ E, y_i,j = 1; otherwise, y_i,j = 0. Second, the random adjacency matrix y = [y_i,j] is constructed based on the binary random variable Y, and all the random adjacency matrices form the network adjacency matrix set Y. Finally, Pr(Y = y|θ), which expresses the probability of y appearing in set Y conditional on θ. The formula for calculating the ERGM is shown in Formula (5)

$$Pr\left(Y=y|\theta \right)=\left(\frac{1}{c}\right)exp\left\{{\sum }_H{\theta }_H{g}_H\left(y\right)\right\}$$

(5)

where c = c(θ) is a standardized constant that ensures that the probability remains within the range 0~1; H represents the structural factors that affect the formation of the network; which can be divided into the endogeneity of the network structure, the node attributes, and external network effects; g_H(y) is the network statistic of the corresponding attribute; and θ_H is the coefficient of this statistic.

This paper uses the TERGM developed by Hanneke et al. [34]. Unlike the ERGM, the TERGM can be used to evaluate the evolution of a multistage investment network by continually simulating and modifying parameters. The specific calculation principle is as follows. Suppose there are a series of real networks, G_t = (V_t, E_t), where V_t and E_t are the nodes and edges in the network at time t, and the specific network at time t is denoted y^t. According to the discrete time Markov chain principle, a TERGM of order k can be defined. The formula for calculating the TERGM is shown in Formula (6).

$$Pr\left(Y^t=y^t|Y^{t-k},\dots ,Y^{t-1},\theta \right)=\frac{exp\left\{{{\displaystyle \sum }}_H{\theta }_{H}g\left(y^t,y^{t-1},\dots ,y^{t-k}\right)\right\}}{c\left(\theta ,y^{t-k},\dots ,y^{t-1}\right)}$$

(6)

As shown in the formula, the formation of the t-phase network depends on the structure of the network in the previous period. However, in addition to the endogeneity of the structure, the node attributes, and the external network effects present in the ERGM network, in a TERGM network time is an influencing factor as well.

TERGM estimation is based on bootstrapping and pseudo-maximum likelihood estimation (PMLE) as well as Markov chain Monte Carlo maximum likelihood estimation (MCMC MLE) [35,36]. Although both methods are suitable for small samples, MCMC MLE is more accurate; therefore, we use bootstrapped PMLE as a robustness check.

3. Urban Housing Investment Niche Network Construction and Impact Mechanism Design

3.1. Construction of the Urban Housing Investment Niche Network

First, according to the urban housing investment niche index system, we calculated the comprehensive index M_ij of the urban housing investment niche in the nine years from 2011 to 2019 (Appendix A). Second, the niche gravity value of urban housing investment was obtained using the modified gravity model. When the gravity value in a cell was higher than the average gravity value for the corresponding row, we recorded a 1 for that cell. A value of 1 indicates that the housing investment niches of the row cities and column cities are related. When the cell value was less than the row average, we recorded a 0, which indicates that the housing investment niches of the row cities and column cities are not related [37]. Thus, a directed binary network G = (V_i, V_j, A) of urban housing investment niches was constructed where V_i = [v_i], (i = 1, 2, 3, …, n), in city i, V_j = [v_j], (j = 1, 2, 3, …, n) in city j, and A is the adjacency matrix, with A = [a_ij], (i = 1, 2, 3, …, n; j = 1, 2, 3, …, n) indicating the relationship in housing investment between city i and city j.

(1)

Spatial characteristics of the network as a whole

A chord diagram can be used to describe the intuition underlying the overall spatial pattern in the urban housing investment niche network. Weighted arcs are used in the connections between the nodes in the graph to reflect the relationships between the nodes. The relative size of each arc reflects the position of that connection in the network. The number of arcs represents the number of cities associated with investment in the focal city. The width of the arc indicates the correlation in investment between the focal city and the target city. As a result, the internal structure of the network as a whole can be described in terms of network density, network reciprocity, and network correlations.

(2)

Spatial characteristics of locally clustered networks

Local clustering is a modular community detection algorithm; we used the Louvain method in R. Specifically, the Louvain method aims to maximize modularity, Q. It assigns each node i in the network to an independent community. Therefore, changes in the modularity (ΔQ) are compared before and after the allocation of nodes to clusters. If the maximum ΔQ > 0, then node I is assigned to the community of the node that generates the largest ΔQ; otherwise, the allocation of node I remains unchanged until all communities of nodes remain unchanged. The simplified form of the corresponding calculation is shown in Formula (7)

$$\mathsf{\Delta }Q=\left[\frac{k_{i,in}}{2m}-\frac{{{\displaystyle \sum }}_{tot}{k}_i}{2m^2}\right]$$

(7)

where m is the sum of the weights of all edges, $k_i$ is the sum of the weights of all edges connected to node i, $k_{i,in}$ is the sum of the edge weights for all nodes in the same community as node i, and ${\displaystyle \sum }_{tot}{k}_i$ is the sum of the weights of the nodes in a community and the edges connected to node i.

(3)

Spatial characteristics of individual networks

Node degree is used to analyse the characteristics of the network nodes. It describes the degree of correlation in housing investment between cities. Cities with a high node degree play an important role in the investment network. Node degree encompasses two indicators, the out degree and in degree, which reflect the capacity of a city’s housing investment niche to send and to receive, respectively.

3.2. Description of the Mechanisms of Influence

(1)

Endogeneity of the network structure

The endogeneity in the network structure refers to the formation of microconfigurations through self-organizing processes within the network. Such configurations occur only inside the network relationships [38]. Specifically, reciprocity is the most important binary structure affecting network relationships, and is used to test the strength of the link between two nodes in the network. A closed network is a three-part structure involving three nodes and three sides, and is regarded as a simple expression of the smallest possible cohesive subgroup. Transitive triads and Cyclic triads indicate whether there are transitive or cyclic tripartite relationships in the network (

Table 2. TERGM: Main variables and descriptions.

Classification	Variable Name	Description	Configuration	Statistic	Definition
Endogenous network structures	Edge	Intercept		${\displaystyle \sum _{i,j}}{y}_{ij}{}^t$	The intercept term is the same as that in a linear regression model
	Mutual	Reciprocity		${\displaystyle \sum _{i,j}}{y}_{ij}{}^t y_{ji}{}^t$	Examines the reciprocity in the relationship between housing investment in two cities in the network
	Ctriple	Cyclic triad		${\displaystyle \sum _{i,j,k}}{y}_{ij}{}^t y_{jk}{}^t y_{ki}{}^t$	Examines whether there is a circular tripartite relationship in urban housing investment
	Ttriple	Transitive triad		${\displaystyle \sum _{i,j,k}}{y}_{ij}{}^t y_{jk}{}^t y_{ik}{}^t$	Examines whether there is a transitive tripartite relationship in urban housing investment
Node attributes	Homophily	Presence of the same attributes		${\displaystyle \sum _{i,j}}{y}_{ij}{}^t \left|{\delta }_i{}^t-{\delta }_j{}^t\right|$	Indicates whether cities with the same attributes tend to have investment relationships
	Sender	Sender effect		${\displaystyle \sum _{i,j}}{y}_{ij}{}^t {\delta }_i{}^t$	Indicates whether cities with certain attributes are more likely to invest in other cities
	Receiver	Receiver effect		${\displaystyle \sum _{i,j}}{y}_{ij}{}^t {\delta }_j{}^t$	Indicates whether cities with certain attributes are more likely to attract investment from other cities
External network effects	Edgecov	Exogenous effect		${\displaystyle \sum _{i,j}}{y}_{ij}{}^t x_{ij}{}^t$	Indicates whether spatial distance affects the tendency of cities to build housing investment relationships
Time effect	Stability	Degree of stability		${\displaystyle \sum _{i\ne j}}{y}_{ij}{}^t y_{ij}{}^{t-1}+\left(1-y_{ij}{}^t\right)\left(1-y_{ij}{}^{t-1}\right)$	Indicates whether the network pattern in period t − 1 affects the network pattern in period t

Source: Authors, 2022.

).

(2)

Node attributes

Node attributes play an important role in the formation of the network (Table 2). Of the various node attributes, intermarriage refers to the possibility of establishing relations between cities with the same attributes. Positive matching means that investment relations can be easily established between cities with the same attributes. The sender effect refers to the possibility of establishing housing investment relationships between the focal city and other cities. A positive coefficient indicates that there is a strong probability of an investment relationship being established. The recipient effect refers to the possibility of a city accepting housing investment relationships with other cities. A positive acceptance effect indicates that the focal city is more likely to accept investment from other cities.

(3)

External network effects

External network effects refer to the links between relationships outside the housing investment network that affect the housing investment network. In the traditional gravity model, an increase in travelling distance has a negative impact on investment between cities [24], with travelling distance being regarded as an exogenous variable that affects the formation of the network (Table 2).

(4)

Temporal effect

The temporal effect is a structural effect that is unique to the TERGM. The temporal effect is used to test whether the network structure is stable over time, and refers to the impact of development in t − 1 on the development of the network in period t, which is used to evaluate the robustness of the network. A significantly positive coefficient indicates that the network structure is stable.

4. Characteristics of the Dynamic Changes in the Network Structure

4.1. Spatial Characteristics of the Network as a Whole

In order to visually present the spatial–temporal disparities in the overall network from 2011 to 2019, we defined four-year intervals, selecting 2011, 2015, and 2019 as the years under investigation (Figure 2). Overall, there are clear differences in node size and connection strength in the network. For example, Beijing, Guangzhou, Shenzhen, Shanghai, Nanjing, and Hefei are the core cities in the urban agglomeration of Beijing–Tianjin–Hebei, the Pearl River Delta, and the Yangtze River Delta, respectively, and these cities are highly connected to other cities. As shown by the arc connections in Figure 2a, in 2011, the relationship in housing investment between Beijing and Shijiazhuang was the largest, followed by those between Guangzhou and Shenzhen and between Tianjin and Shijiazhuang. Figure 2b shows that in 2015, those cities with large intercity housing investment relationships were Beijing and Shijiazhuang, Nanjing and Hefei, and Guangzhou and Shenzhen. Finally, in 2019, as shown by Figure 2c, Beijing, Tianjin, Guangzhou, Shenzhen, Nanjing, and Hefei had large intercity housing investment relationships. Therefore, these results show that the housing investment network is a “rich cities club”.

The density values in 2011, 2015, and 2019 are 0.212, 0.230, and 0.213, respectively, thus exhibiting an inverted “V”-shaped trend. On the one hand, this shape is related to the real estate investment cycle and the unsustainability of an overreliance on investment to drive development. On the other hand, real estate policy is related to macro-level regulations. For example, after 2015, real estate destocking was implemented, real estate speculation was prohibited, and real estate was prohibited from being used as a short-term means to stimulate the economy. Moreover, the reciprocity values in 2011, 2015, and 2019 were 0.755, 0.723, and 0.721, respectively. Although these values are high, reciprocity decreased after 2011 because the hierarchical nature of a rich cities club is not conducive to the formation of a network.

The average value of the network correlations in 2011, 2015, and 2019 was 1, indicating significant levels of spatial correlation and strong spillover effects. The network density levels were 0.245, 0.272, and 0.279, respectively, showing an upwards trend. Meanwhile, network efficiency was 0.811, 0.792, and 0.810, indicating that stability improved with an increase in the number of connections in the network.

4.2. Spatial Characteristics of the Local Clustering Networks and Individual Networks

The local clustering results are shown in Figure 3 and

Table 3. Clusters and cluster density for the 35 large and medium cities studied.

Year	Cluster	City	Density
2011	Cluster 1	Shanghai, Nanjing, Hangzhou, Ningbo, Hefei, Fuzhou, Xiamen, Nanchang, Xi’an, Urumqi	0.533
	Cluster 2	Beijing, Tianjin, Shijiazhuang, Taiyuan, Hohhot, Shenyang, Dalian, Changchun, Harbin, Jinan, Qingdao, Zhengzhou, Chongqing, Yinchuan	0.528
	Cluster 3	Wuhan, Changsha, Guangzhou, Shenzhen, Chengdu, Nanning, Haikou, Guiyang, Kunming, Lanzhou, Xining	0.582
2015	Cluster 1	Beijing, Tianjin, Shijiazhuang, Taiyuan, Hohhot, Shenyang, Dalian, Changchun, Harbin, Jinan, Qingdao, Zhengzhou, Chongqing, Xi’an	0.550
	Cluster 2	Shanghai, Nanjing, Hangzhou, Ningbo, Hefei, Fuzhou, Xiamen, Nanchang, Wuhan, Changsha, Guangzhou, Shenzhen, Chengdu, Nanning, Guiyang, Kunming, Lanzhou, Xining, Yinchuan, Urumqi, Haikou	0.448
2019	Cluster 1	Nanjing, Hangzhou, Ningbo, Hefei, Fuzhou, Xiamen, Nanchang, Nanning	0.607
	Cluster 2	Beijing, Tianjin, Shijiazhuang, Taiyuan, Hohhot, Shenyang, Dalian, Changchun, Harbin, Shanghai, Jinan, Qingdao, Yinchuan	0.474
	Cluster 3	Zhengzhou, Wuhan, Changsha, Guangzhou, Shenzhen, Chongqing, Chengdu, Xi’an, Guiyang, Kunming, Lanzhou, Xining, Urumqi, Haikou	0.550

. Specifically, Beijing, Shanghai, and Chongqing were core cities within the network in 2011. The results divide the network into three clusters: the Yangtze River Delta, Beijing–Tianjin–Hebei, and the central and eastern regions. For each cluster, network density is 0.533, 0.528, and 0.582, respectively, indicating that the intercity housing investment connections in the central and eastern regions were highest. In 2015, Beijing, Shanghai, Guangzhou, Shenzhen, Chongqing, Nanjing, and Hangzhou were at the core of the network. The network is divided into inland and coastal areas, with respective densities of 0.550 and 0.448. The housing investment relationships among inland cities are the strongest. In 2019, Beijing, Shanghai, Shenzhen, Nanjing, Hangzhou, and Wuhan were core cities, with the network again being divided into the Yangtze River Delta cluster, the Beijing–Tianjin–Hebei cluster, and central and eastern region cluster. The densities of these clusters were 0.607, 0.474, and 0.550, respectively, with network density in the Yangtze River Delta being the highest. Overall, although the clusters within the residential investment niche network involving 35 large and medium-sized cities changed slightly, the internal network connections within the clusters changed significantly. Therefore, it is necessary to further analyse the characteristics of their respective spatial–temporal patterns.

The distributions of the 2011, 2015, and 2019 node degrees for the urban housing investment niche network were used to draw histograms of the out-degree distribution and the in-degree distribution (Figure 4). The peak value of the in-degree distribution was between 0 and 5 in 2011, 2015, and 2019, indicating that the distribution is right-tailed and thus that most cities are unable to attract investment. In 2011, the peak value of the out-degree distribution was between 4 and 6, again indicating a right-tailed distribution. The peak value of the out-degree distributions for 2015 and 2019 were between 8 and 10, suggesting that the right-tailed tendency of the distribution increased. This increase implies that foreign investment is increasing and that there remains little mutual investment between cities. This lack suggests that there is a massive amount of room for growth in the future.

5. Mechanism Analysis

5.1. Results

The existing empirical research in the new economic geography field has shown that economic development, population size, and regional transportation are important factors affecting investment in housing. This paper uses city-specific GDP and population size as node attributes. We divided the cities according to GDP following the approach used by Xu Helian [39]. If a city ranked in the top 50% of the 35 cities in terms of GDP, it was classified as GDPHigh; otherwise, it was classified as GDPLow. We used the classification criteria for megacities (>10 million people), super cities (5–10 million people), and large cities (1–5 million people) to classify cities as PeopleHigh, PeopleMid, and PeopleLow. In addition, we measured travelling distances using the intercity traffic accessibility index as the actual measurement to obtain a 35 × 35 spatial distance matrix.

The Statnet package in R along with MCMC MLE were used to explore the mechanisms underlying the evolution of the network. Furthermore, in order to verify the estimated TERGM results, we compared the estimated results obtained using the ERGM and the TERGM (

Table 4. Comparison of the ERGM and TERGM results.

Variable	ERGM	TERGM
Edges	−0.47 ***	−1.55 ***
	(0.29)	(0.19)
Mutual	1.76 ***	1.15 ***
	(0.27)	(0.17)
Ctriple	−0.46 ***	−0.28 ***
	(0.12)	(0.06)
Ttriple	−0.09	−0.00
	(0.05)	(0.01)
Homophily (GDPHigh)	0.06	−0.04
	(0.16)	(0.12)
Homophily (PeopleHigh)	0.17	0.05
	(0.17)	(0.12)
Sender (GDPLow)	−0.34	−0.16
	(0.25)	(0.15)
Sender (PeopleLow)	0.33	0.12
	(0.28)	(0.17)
Receiver (GDPHigh)	0.71 ***	0.43 ***
	(0.20)	(0.12)
Receiver (PeopleHigh)	2.65 ***	1.35 ***
	(0.30)	(0.14)
Stability		2.87 ***
		(0.06)

Notes: Standard errors in parentheses. *** p < 0.001. source: Authors, 2022.

). Specifically, the coefficient of edges increased from −2.47 to −1.55 and the coefficient of Ctriple increased from −0.46 to −0.28, indicating that the influence of the static variables decreased significantly when the TERGM was used. The coefficients of Mutual and Stability from the TERGM are 1.15 and 2.87, respectively, and are significant at the 1% level, indicating that dynamic variables have a significant impact on the formation of network relations.

On the basis of the above analysis, we drew box diagrams by simulating the index values for the various network characteristics. The closer the midpoint is to the actual line of fit, the better the fit of the model is [40]. The results from the ERGM and the TERGM are shown in Figure 5 and Figure 6, respectively. Within these figures, panels (a–e) present the results of a simulation of 100 random network characteristic indexes. Furthermore, by comparing the simulated values presented in the box graphs with the actual characteristic index values, we find that the actual network values from the TERGM fall essentially at the midpoint of each box in the box graph; the TERGM fits the data better than the ERGM. In addition, the red line in (f) is the ROC curve. The larger the area under the curve, the better the model fit is. Therefore, the TERGM fits the data better.

The data are divided into two periods, 2011–2015 and 2015–2019, as shown in

Table 5. Comparison of TERGM estimation results for 2011–2015 and 2015–2019.

Variable	Model 1	Model 2	Model 3
	2011–2019	2011–2015	2015–2019
Edges	−0.79 ***	−0.58 ***	−0.74 *
	(0.06)	(0.15)	(0.36)
Mutual	0.95 ***	1.14 ***	0.70 *
	(0.13)	(0.25)	(0.28)
Ctriple	−0.29 ***	−0.40 ***	−0.26 **
	(0.06)	(0.11)	(0.10)
Ttriple	−0.03	−0.02	−0.06 *
	(0.02)	(0.03)	(0.03)
Homophily (GDPHigh)	−0.03	−0.34	0.04
	(0.11)	(0.26)	(0.18)
Homophily (PeopleHigh)	0.11	−0.14	0.15
	(0.13)	(0.24)	(0.20)
Sender (GDPLow)	−0.22	−0.14	−0.14
	(0.14)	(0.29)	(0.24)
Sender (PeopleLow)	0.27	−0.03	0.26
	(0.17)	(0.34)	(0.27)
Receiver (GDPHigh)	0.43 ***	1.55 ***	0.43 *
	(0.12)	(0.30)	(0.19)
Receiver (PeopleHigh)	1.53 ***	0.91 **	1.57 ***
	(0.15)	(0.28)	(0.23)
Distance	−0.15 ***	−0.17 ***	−0.13 ***
	(0.05)	(0.06)	(0.04)
Stability	2.81 ***	3.12 ***	2.62 ***
	(0.06)	(0.12)	(0.09)

Notes: Standard errors in parentheses. *** p < 0.001; ** p < 0.01; * p < 0.05. source: Authors, 2022.

. In model 1, which uses 2011–2015, the coefficient of Mutual is significantly positive, reflecting the high level of dependence in the investment relations between cities. The coefficient of Cyclicality is significantly negative, indicating that the number of closed cyclic structures in the network decreased over time. However, the coefficient of Transitivity is not significant, indicating that the establishment and maintenance of niche network relationships in the housing investment of some cities has been hindered. Regarding the effects of node attributes, the coefficients of Homophily (GDPHigh) and Homophily (PeopleHigh) are not significant, indicating that the level of connectivity between cities with similar attributes is not high. The coefficients of Sender (GDPLow) and Sender (PeopleLow) are not significant, indicating that there is no strong trend towards foreign investment in cities with low economic development levels and small populations. The coefficients of Receiver (GDPHigh) and Receiver (PeopleHigh) are significantly positive, indicating that cities with high per capita incomes and cities with large populations clearly attract investment and that these two attributes together constitute the core elements needed to attract investment inflows. In addition, in terms of external network effects, there is a negative relation between the transportation network and urban housing investment, which is in conformity with the structure of a general gravity model (the less accessible a region is, the weaker the investment correlations are). Finally, the coefficient of Stability is significantly positive, indicating that the network structure is stable overall.

Models 2 and 3 estimate the dynamic evolution in the model between 2011–2015 and 2015–2019. From the perspective of the endogeneity of the network structure, the coefficient of Mutual passed the significance test, while that of Reciprocity decreased from 1.14 to 0.7. The coefficient of Cyclicality increased significantly, from −0.40 to −0.26. The coefficient of Transitivity failed to pass the significance test, indicating that the negative impact of the circular structures within the network has been alleviated and that obstacles to network formation at the network level have been further reduced. However, the correlations within the housing investment network are in line with those predicted by resource flow theory. With the establishment of cooperative circles in urban agglomerations, the barriers between cities are expected to gradually weaken. However, the reciprocity in the investment network is not stable, which shows that during a mutual investment game between cities there are no network connections without a direction or goal. Instead, we see a trend toward directional competition, reflecting the positive externalities of the network.

In terms of the effect of attributes, the coefficients of Homophily (GDPHigh), Homophily (PeopleHigh), Sender (GDPLow), and Sender (PeopleLow) were not significant in either time period. The coefficient of Receiver (GDPHigh) decreased from 1.55 to 0.43, and this difference passed the significance test, indicating that cities with a high economic level in the network were less likely to become receivers. On the one hand, in the context of national policy regulations those cities with excessively high housing prices have had purchase and sale restrictions introduced in order to curb real estate speculation and strengthen demand-side regulations. On the other hand, given limited resources, cities with high economic levels have placed strong barriers to competition over public and human resources, preventing the establishment of “alliances between giants” relationships within the investment network. The coefficient of Receiver (PeopleHigh) increased from 0.91 to 1.57, passing the significance test and indicating that cities in the network with large populations were more likely to become receivers. However, there is an obvious long tail in China’s population distribution. The population is mainly concentrated within a few large cities, and the first-mover advantages and scale effects enjoyed by those cities will exist for a long time. The population is expected to continue to agglomerate in large cities and metropolitan areas in the future.

In addition, regarding the external network effects the coefficient of Distance increased from −0.17 to −0.13, passing the significance test and indicating that the negative effect of travelling distances on urban housing investment has weakened. In recent years, China’s high-speed railways and other transportation facilities have continually broken down urban boundaries, and the time cost of communicating between cities has been significantly reduced. With improved attractiveness for flows of people and capital as well as logistics and capital flow the status of cities and their development potential will be greatly improved, especially among regional and even national transportation hub cities. In addition, regarding the temporal effects, the coefficient of Stability decreased significantly from 3.12 to 2.62. The stability of the investment network has declined. Because areas undergoing urbanization generally transform from cities to metropolitan areas to urban agglomerations, as the carrying capacity of large cities becomes saturated, new changes in the pattern of housing investment tend to arise in the future.

5.2. Robustness Check

In this paper, the MPLE method was used to test the estimated results (

Table 6. Robustness test.

Variable	2011–2019	2011–2015	2015–2019
Edges	−1.02 *	−0.99 *	−0.76 *
	[−1.41; −0.59]	[−1.34; −0.21]	[−1.16; −0.19]
Mutual	1.27 *	1.76 *	0.66 *
	[0.78; 1.62]	[1.03; 2.18]	[0.56; 0.74]
Ctriple	−0.33 *	−0.55 *	−0.20 *
	[−0.45; −0.17]	[−0.66; −0.18]	[−0.26; −0.12]
Ttriple	−0.01	0.02	−0.06
	[−0.07; 0.03]	[−0.08; 0.09]	[−0.14; 0.01]
Homophily (GDPHigh)	−0.02	−0.30 *	0.04
	[−0.21; 0.13]	[−0.54; −0.15]	[−0.12; 0.26]
Homophily (PeopleHigh)	0.12	−0.19	0.15
	[−0.16; 0.29]	[−1.11; 0.33]	[−0.16; 0.35]
Sender (GDPLow)	−0.17	−0.11	−0.09
	[−0.55; 0.31]	[−0.62; 0.68]	[−0.63; 0.44]
Sender (PeopleLow)	0.31 *	0.03	0.25
	[0.04; 0.55]	[−0.85; 0.65]	[−0.12; 0.74]
Receiver (GDPHigh)	0.44 *	1.65 *	0.42 *
	[0.18; 0.62]	[1.04; 1.93]	[0.11; 0.61]
Receiver (PeopleHigh)	1.54 *	0.89 *	1.57 *
	[1.25; 1.79]	[0.34; 1.55]	[1.22; 1.86]
Distance	−0.13 *	−0.15 *	−0.12 *
	[−0.18; −0.07]	[−0.22; −0.08]	[−0.15; −0.07]
Stability	2.82 *	3.18 *	2.63 *
	[2.61; 3.15]	[2.96; 4.27]	[2.50; 3.03]

Note: 95% confidence intervals are in square brackets, * p < 0.05. source: Authors, 2022.

). The results show that in terms of endogenous structure effect, the coefficient of Mutual is significantly positive, the coefficient of Cyclicity is significantly negative, and the coefficient of Transitivity is not significant. In terms of the attribute effect, the coefficients of Homophily (GDPHigh), Homophily (PeopleHigh), Sender (GDPLow), and Sender (PeopleLow) are not significant in the 99% confidence interval, while the coefficients of Receiver (GDPHigh) and Receiver (PeopleHigh) are significantly positive. In terms of the external network effect, the coefficient of Distance is significantly negative. In terms of the time effect, the coefficient of Stability is significantly positive. Moreover, the change trends of the four effects in the two time periods are consistent with Table 5. Therefore, the fitting result of TERGM is robust.

6. Conclusions and Discussion

In studying 35 large and medium-sized cities from 2011 to 2019, we constructed a comprehensive index to measure urban housing investment niches. Network correlations were established through a modified gravity model, and TERGM was used to explore the formation of the urban housing investment niche network. The main conclusions of this paper are presented below.

First, this paper uses niche theory to construct a niche evaluation model for urban housing investment. From a structural and functional standpoint, it is proposed that sustainable development of the urban housing investment niche requires the systematic integration of a resources–housing–people system. Correspondingly, four aspects were built into the evaluation system, which encompassed a resource niche, housing market niche, socio-economic niche, and real estate policy niche.

Second, the spatial correlations in the housing investment niche network are low. Network density follows an inverted “V”-shaped trend. The results at the network level suggest the existence of a “rich cities club”. However, the spillovers between urban networks are uneven, indicating the presence of a core–edge distributional pattern which may further widen the gap between different cities in terms of their ability to attract housing investment.

Third, the clustering results for the urban housing investment niche network depict a development model with urban agglomerations at the centre. The network can be divided into the Yangtze River Delta region, the Beijing–Tianjin–Hebei region, and the central and eastern regions. In addition, mutual investment between cities is low. Due to the large differences in the network relationship between different cities, it is of great significance to seek differentiated resource positioning, unleash multi-level market demand, and implement active and efficient policy intervention for building a healthy and sustainable urban housing investment niche network.

Fourth, the results on the endogeneity in the structure verify that reciprocal behaviour in housing investment among most cities is substantial. However, there are quality barriers between certain cities which affect the development of the network. This shows that the high level of investment in urban housing is an important factor for attracting investment in urban economic development. The analysis of the external network effects verifies that the relationship between the transportation network and the urban housing investment network conforms to the law of general gravity models. In addition, the temporal results suggest that the network structure is not stable. As the carrying capacity of certain large cities becomes saturated, future housing investment patterns exhibit new characteristics.

Our research can provide insights into the ways in which misplaced competition in housing investment between cities in a region can be corrected and the flow of factors can be guided reasonably. On the one hand, utilizing niche theory we chose resources, housing, population, and institutions as indicators for measuring the housing investment niche. On the other hand, we discuss the mechanism by which the spillover relationships between cities in terms of urban housing investment form and evolve from the perspective of a spatial network. Most existing studies have focused on describing spatial features. In contrast with the existing literature [41], we present a TERGM analysis of the endogenous and exogenous factors affecting dynamic change in the network structure. This paper reveals the internal mechanisms that drive changes in the network structure and the mechanisms that influence those changes.

Nevertheless, our research has limitations. First, the index system explores the indicators of physical factors that affect the housing investment niche. In future research, this index should be expanded to incorporate items such as the livability of housing, quality of life, and other non-physical factors. Multi-dimensional indicators will help to enhance the comprehensiveness of the index system and to make it more systematic, allowing it to keep pace with the actual development of housing investment in the new era. Second, with the continued future growth of digital networks is expected to allow multi-source heterogeneous databases including statistical data, social survey data, POI data, and map information data to be constructed, which can further improve the scientific basis of urban housing investment niche network evaluation and guidance. Third, if the structure of the housing investment niche changes various old and new dynamic factors can be expected to transform, as can the attribute impact variables of network nodes. The properties of network nodes may be affected by the population, economy, transportation, and other influencing factors with “man–land” as the core as well as by virtual flows such as information flow and network flow as the information network develops. The mapping relationship between network virtual space and real space is strengthened, and the influence of “man–land” may be supplanted by the influence of “man–land–virtual network” [42]. Therefore, further research into “invisible” node attribute factors should be explored.

Several policy implications regarding the urban housing investment network can be noted based on our findings. First, the urban housing investment niche can be depicted with a development model in which urban agglomerations are at the centre and development is driven by cities with absolute advantages. Therefore, the government can formulate more perfect policies to direct the upgrading and economic development of the regional industrial structure in order to achieve balanced overall regional development. Second, due to the coexistence of urban competition and cooperation, “alliances between giants” have not been established between cities with high economic development levels within the investment network. The government can act as a “visible hand” and encourage cities with high economic development levels to take the lead in becoming “senders” of network relationships. By reducing obstacles at the network level, the imbalance in the development of network relations can be improved. Third, the size of the urban population is an important factor affecting the development of China’s housing investment network at this stage. The push–pull theory of population migration suggests that the “pull power” of urban investment can be improved. Measures should be taken to improve the alignment between population agglomeration and the provision of public services such as education, medical treatment, and the environment.