Research on the impact and heterogeneity of housing on the allocation of financial assets of households: Evidence from China

Abstract

This paper investigates the impact of housing with both consumption and investment attributes on the risky financial asset allocation of households, constructs Probit and Tobit models using 2019 China Household Finance Survey (CHFS) data, and proceeds to the mediation effect test and heterogeneity analysis. Results indicate that owning only one house exhibits a crowding-out effect on the risky financial asset allocation of urban households, with the degree of risk preference as the mediating effect mechanism, while owning multiple houses exhibits an asset allocation effect. Housing borrowing other than bank loans inhibits urban households from making risky financial asset allocations. The effect of housing on risky financial asset allocation is heterogeneous by income, age, and region.

1. Introduction

As the basic unit of China's economy, the average household asset size of households has been growing year by year. According to a report released by the China Household Finance Survey (CHFS) and Research Center in 2016, the average total household assets in China were 663,000 yuan in 2011, 761,000 yuan in 2013, and reached 929,000 yuan in 2015, with a compound annual growth rate of 8.8 %. In the allocation of household assets, real estate is the main part, and financial assets are auxiliary [1]. Housing largely influences the savings and consumption decisions and investment portfolios of households [2]. Despite the fact that real estate occupies the majority of household asset allocation, the proportion of China's households in financial assets increased in 2013 and 2015 compared to 2011. Coupled with the gradual improvement of China's financial market in recent years, financial management channels have become increasingly diversified, and a variety of financial products have been rapidly developed. Resident families actively participate in Internet financial investment, which has led to the gradual expansion of the Internet financial management market [3,4]. As of December 2018, the scale of Internet wealth management users reached 151.38 million people, an increase of 17.5 % compared with the end of 2017. As of March 2020, the scale of Internet wealth management users grew to 163.56 million people, an increase of 8.1 % compared with the end of 2018. Affected by the COVID-19, the scale of Internet wealth management users as of June 2020 fell to 149.38 million.

On the other hand, due to the influence of traditional views on house ownership, real estate has become the most important part of family assets. Real estate accounts for a significant proportion of family wealth, as a result, housing is one of the most important factors affecting the financial asset allocation of households [1,5]. Additionally, coupled with the rising real estate prices in recent years, it has generated the Real Estate Speculator, resulting in a large amount of money pouring into the real estate market [6]. Moreover, it is also worth noting that risk aversion, which refers to an investor's aversion to investment risk, plays a significant role in the household financial asset allocation [[7], [8], [9]]. For risk-averse investors, the demand for returns is higher than the risk they face. For different investment projects with the same rate of return, investors choose the investment project with low risk first. Barasinska et al. [10] observe that the risk tolerance of households has a direct impact on the allocation of risky assets, with a tendency for risk averse households to choose incomplete portfolios composed primarily of a few risk-free assets. China's household financial investment portfolios suffer from polarized risk, as Lu et al. [11] point out that the risk distribution of China's household financial investment portfolios is U-shaped compared with those in European and the United States, i.e., there are more risk-averse and risk-preferring households. And besides meeting the rigid housing demand, real estate has also been given an investment function. Consequently, owing to its special attributes, housing forms an important part of the household investment portfolios. It is worth investigating whether the consumption and investment of real estate by families will affect investment in the capital market, or whether the wealth effect brought about by the rise in property prices promotes investment in the capital market by families.

With the continuous deepening of China's financial reform and the rapid development of the financial market, the investment varieties of residents show a diversified development trend. Instead of relying solely on savings deposits, their various property income has also increased significantly, and the demand for investment and financial management of households is also rising. For this reason, this paper focuses on the functioning mechanism of the housing with both consumption and investment attributes on the financial asset allocation of households to fill the current research gap. By constructing Probit and Tobit models, the influence direction and degree of housing ownership and housing debt on the allocation probability and depth of risky financial assets of urban households in China were analyzed. Afterwards, risk preference was selected as the mediating variable for the mediation effect test to investigate the influence mechanism of housing ownership on the allocation of risky financial assets among urban households. Finally, the heterogeneous differences in the role of housing on the risky financial asset allocation of urban households across household income, age, and region were further analyzed by constructing Probit and Tobit models in groups. Generally, the findings of this paper could be of great practical reference significance to the investment choices of household assets and the formulation of government policies for the real estate market.

The remainder of this paper presents the following. Section 2 briefly reviews the relevant literature. Section 3 describes the data, variables, and models. Section 4 demonstrates the empirical analysis. Section 5 performs the heterogeneity analysis. Section 6 summarizes this study and provides recommendations.

2. Literature review

In 2006, Campbell coined the term Household Finance, arguing that household finance and corporate finance have similar characteristics [12]. They both aim to optimize the allocation of resources across time and maximize utility through the selection and use of financial instruments such as bonds, funds, and stocks. Therefore, household finance should be treated as an independent research field. In recent years, scholars have conducted progressively richer studies on household finance, which can be categorized into three aspects: household asset allocation, household consumption and household liabilities. In this paper, we will start from the perspective of housing to study the behavior of household financial asset allocation.

Flavin et al. [13] argue that the presence of house price risk causes households to reduce their choice and participation in risky investments due to liquidity constraints. Kullman et al. [14] assume that housing assets have a crowding-out effect on investment in other assets, not only decreasing the probability of households' stock market participation, but also decreasing their holdings of risky financial assets such as equities of the household's stock market participation, but also reduces the proportion of risky financial assets such as stocks. Zhou [15] argues that housing mortgage loans have a crowding-out effect on the total amount and proportion of household financial assets, and empirically confirms that the proportion of bank deposits in households with mortgage loans is lower than the proportion of deposits in households without mortgage loans. Ma et al. [16], based on data from the 2011 CHFS, find that the net value of housing assets of Chinese households crowds out investment in risky financial assets such as stocks due to meeting their own consumption and investment needs. Zhou et al. [17] theoretically explain the crowding-out effect of housing assets on financial assets in households based on the intertemporal asset decision model, and empirically verify the existence of this effect using domestic sample data.

But some scholars hold a different view. Cardak et al. [18] believe that Australian households with housing property rights are more likely to apply for mortgage loans for investment or other purposes than those without such property rights, which increases the proportion of risky financial assets held by households with housing property rights. Wu et al. [19] analyzed and concluded that the number of real estate owned by urban households in China will have an impact on the specific performance of households entering the stock market to some extent, that is, with the increase of the number of real estate owned, the asset allocation effect will gradually take effect, and the behavior of buying a first house will basically crowd out the amount of risky assets held by urban households. Jiang et al. [20] empirically study that the rising housing price in China would prompt the one-room sample households and the overall sample households to allocate risky financial assets and reduce risk-free investment, and encourage the non-owning households to choose risk-free financial assets and the multi-room sample households to invest risk-oriented.

In addition to these, scholars have conducted numerous empirical studies on the impact of other factors on the household's financial asset allocation, mainly including education [12,21], gender and marital Status [22,23], age [24], financial literacy skills [[25], [26], [27]], health status [28], political background [29], and natural hazards [30]. Examination of the domestic and international literature reveals that scholars disagree on whether housing facilitates or inhibits household risky financial asset participation, and that there is currently insufficient research analyzing the heterogeneity of household financial asset allocation. Since real estate accounts for the highest proportion of household assets, it is necessary to investigate in depth the specific and heterogeneous effects of housing on household financial asset allocation. Therefore, according to the 2019 CHFS data, this paper selects the main explanatory variables and a series of control variables related to housing to further explore the impact and heterogeneity of housing on household financial asset allocation in China. Finally, corresponding countermeasures are proposed based on the research findings.

3. Data, variables and models

3.1. Data source

The data used in this study are derived from the 2019 CHFS data from the China Household Finance Survey and Research Center of Southwestern University of Finance and Economics. In the questionnaire of the 2019 CHFS, it is divided into five main parts: demographic characteristics, assets and liabilities, insurance and protection, expenditure and income and financial knowledge, and grassroots governance and subjective evaluation. In this paper, for urban households, a total of 20,230 observations were selected after excluding missing, abnormal or invalid data for the concerned variables, and the data were analyzed using Stata15 software.

3.2. Variable selection

In the structure of financial asset allocation in households in our country, the allocation of risk financial assets is relatively low compared to the relatively high proportion of risk-free financial assets, so for the allocation of risky financial assets to conduct empirical research. Risky financial asset allocation of households is measured in two dimensions: probability and depth, and the selected explanatory variables are: whether to make risky financial asset allocation, and the proportion of risky financial asset allocation. Whether risky financial asset allocation is carried out refers to whether urban resident households hold stocks, funds, financial products, derivatives, non-RMB assets, gold, other financial assets or lending money. The probability of risky financial asset allocation in urban households is measured using a dummy variable, which is assigned a value of 1 if the urban household owns a certain account and allocates assets accordingly, and 0 otherwise. The proportion of risky financial asset allocation in the variable refers to the proportion of the market value of the relevant assets allocated in the urban household to the total market value of the household's financial assets.

Four main explanatory variables for housing were selected, one is whether the urban household owns only one housing unit, which is assigned a value of 1 if it owns only one housing unit, and 0 if it has no housing unit. Two is whether the urban household owns more than one housing unit, which is assigned a value of 1 if the urban household owns more than one housing unit and 0 if it owns only one housing unit among the sample that owns a housing unit. Three is whether the urban household owns a bank loan for housing, which is assigned a value of 1 if there is a bank loan otherwise. bank loan for housing, in the sample of households owning housing, if there is a bank loan, it is assigned a value of 1, otherwise it is 0. Four is whether there is any other housing borrowing other than bank loan in the urban household, if there is any other housing borrowing, it is assigned a value of 1, otherwise it is 0. The control variables refer to the results of the previous scholars' research. In CHFS, the head of the household is the main source and principal of the economy of the resident household, therefore, the age, gender, literacy, marital, and physical status of the head of the household, the risk preference of the household, the total household income, and the total household assets are selected. Considering the non-linear effects of some of the selected variables, the total income and total assets of urban households are logarithmically treated. The descriptions of the variables are shown in Table 1.

Table 1.

Description of variables.

Variable attribute	Variable name	Variable meaning	Variable explanation
Explained variables	Risk	Whether to allocate risky financial assets	Holding risky financial assets is assigned a value of 1, otherwise 0.
	Risk percentage	Allocation ratio of risky financial assets	Ratio of market value of risky financial assets to financial assets.
Primary explained variables	House	Whether to own only one house	1 for owning only one house, 0 for not owning a house.
	Houses	Whether to own more than one house	1 for owning more than one house, 0 otherwise.
	Bankloan	Whether there is a housing bank loan	1 for having a housing bank loan, 0 otherwise.
	Housedebt	Whether there are other loans for housing	1 for other loans with housing, 0 otherwise.
Control variables	Age	Age	Age of head of household.
	Gender	Gender	The age of the head of household is assigned 1 for males and 0 for females.
	Edu	Educational level	The head of the household who has not been to school is assigned 1, the primary school is assigned 2, the junior high school is assigned 3, the high school is assigned 4, the secondary school/vocational high school is assigned 5, the junior college/vocational high school is assigned 6, the university is assigned 7, the master's degree is assigned 8, and the doctoral degree is assigned 9.
	Marriage	Marital status	1 for married head of household, 0 for unmarried and other statuses.
	Health	Physical condition	High risk and high return or slightly high risk and slightly high return is considered risk preference and assigned 1, otherwise 0.
	Riskpre	Risk preference	Head of household answered high risk and high return or slightly high risk and slightly high return, considered risk preferred and assigned 1, otherwise 0.
	Lnincome	ln (Total household income)	Logarithm of total household income
	Lnasset	ln (Total household assets)	Logarithm of total household assets

3.3. Modeling

This paper aims to investigate the influence of variables such as homeownership, housing indebtedness and family characteristics on the allocation of risky financial assets of urban households in China. The explained variable for whether to allocate risky financial assets is a binary discrete variable that takes the value of 1 or 0. Hence, with reference to the model design method of Lu [31], this paper constructs the Probit model to study the probability of risky financial asset allocation of urban households. Additionally, the explained variable for allocation ratio of risky financial assets is a truncated variable, so the Tobit model is constructed to explore the depth of risky financial assets allocation of urban households.

The constructed Probit model is as follows:

$$\begin{array}{cc}Risk=I\left({\alpha }_{1}CoreX+{\alpha }_{2}X+{\mu }_i\right)\text{,}& {\mu }_i\sim N\left(0,{\sigma }^2\right)\end{array}$$ (1)

where Risk is an explained variable in this paper, Risk = 1 indicates that urban households have made allocations to risky financial assets, and Risk = 0 indicates that urban households have not made allocations to risky financial assets, CoreX denotes the core explained variable related to the number of house and debts of urban households, and X is a control variable in terms of household characteristics and other aspects.

The Tobit model is constructed as follows:

$$\begin{gathered} \begin{array}{cc}Riskpercentage={\alpha }_{1}CoreX+{\alpha }_{2}X+{\mu }_i \\ =\max \left(0,Riskpercentage\right)\{\begin{array}{c}Riskpercentage,Riskpercentage>0\\ Riskpercentage,Riskpercentage=0\end{array},& {\mu }_i\sim N\left(0,{\sigma }^2\right)\end{array} \end{gathered}$$ (2)

where Risk percentage is another explained variable indicating the market value of risky financial asset allocations in urban households as a proportion of the total market value of financial assets, similarly, CoreX is the core explanatory variable related to housing, and X is the control variable for the empirical evidence.

4. Empirical analysis

4.1. Full-sample regression analysis

4.1.1. Regression analysis of the impact of home ownership on the financial asset allocation of urban households

Regressions (1) and (2) in Table 2 are based on the criterion of whether urban households own only one house, and the sample data are selected to give the regression results of the Probit and Tobit models. It is evident that whether to own only one house is significantly negatively correlated with both the allocation of risky financial assets and the proportion of allocation at the 1 % significance level, and the marginal effect of owning only one house is 0.2003 and −0.0883, respectively. In contrast, it has a deeper impact on whether urban households participate in risky financial asset allocation. This implies that urban households owning only one home are more cautious in participating in risky financial asset allocation and less willing to invest than urban households without a home. The possible reason for this is that urban households owning only one house is more of a rigid demand of the urban households themselves, based on the attribute of consumption, and the funds for housing purchases crowd out the funds for risky financial asset allocation, and the wealth effect brought by housing does not directly make the household profitable, which is manifested in the crowding-out effect of housing.

Table 2.

Regression results on the effect of the number of houses owned on the risky financial asset allocation.

Variables	(1)	(2)	(3)	(4)
	Whether risky financial assets are allocated	Risky financial asset allocation ratio	Whether risky financial assets are allocated	Risky financial asset allocation ratio
	Probit	Tobit	Probit	Tobit
house	−0.2003***	−0.0883***
	(0.0142)	(0.0068)
houses			0.0469***	0.0310***
			(0.0095)	(0.0053)
age	−0.0044***	−0.0011***	−0.0042***	−0.0010***
	(0.0003)	(0.0002)	(0.0003)	(0.0002)
gender	−0.0044	−0.0156***	−0.0040	−0.0201***
	(0.0083)	(0.0047)	(0.0086)	(0.0048)
edu	0.0209***	0.0150***	0.0217***	0.0160***
	(0.0025)	(0.0014)	(0.0025)	(0.0014)
marriage	−0.0192*	−0.0088	−0.0064	−0.0024
	(0.0108)	(0.0059)	(0.0119)	(0.0064)
health	0.0254**	0.0019	0.0266**	−0.0002
	(0.0106)	(0.0057)	(0.0113)	(0.0060)
riskpre	0.1491***	0.0931***	0.1551***	0.1044***
	(0.0176)	(0.0090)	(0.0167)	(0.0086)
lnincome	0.0307***	0.0101***	0.0411***	0.0115***
	(0.0031)	(0.0016)	(0.0034)	(0.0017)
lnasset	0.0971***	0.0387***	0.1062***	0.0446***
	(0.0033)	(0.0017)	(0.0038)	(0.0019)
Constant		−0.3862***		−0.5758***
		(0.0227)		(0.0272)
Observations	16,300	16,300	17,716	17,716
Pseudo R-squared	0.141	0.393	0.150	0.348

Note: Probit model estimates show marginal effects, ***, **, and * indicate significant at the 1 %, 5 %, and 10 % significance levels, respectively, standard errors of the estimates are in parentheses. Same below.

Regressions (3) and (4) in Table 2 exclude data from the no-housing sample and analyze how owning multiple houses will affect household risky financial asset allocation in urban households with houses. It is observed that there is a positive correlation between the dummy variable of whether to own more than one house and the risky financial asset allocation and the allocation ratio, with the results of the marginal impacts being 0.0469 and 0.0310, respectively, which both pass the significance test at the 1 % level of significance. This implies that households owning multiple houses increase the probability and depth of risky financial asset allocation compared to urban households owning only one house. The main reason for this is that, compared with the decision to purchase the first housing for immediate needs, the purchase of the second and third housing in urban households is more based on the attribute of investment, and the household can rent out the unused housing to obtain stable financial returns. At this point, as the number of houses owned increases, the asset allocation effect of housing will be greater than the crowding-out effect on risky financial assets.

For the remaining explanatory variables, model (1) (2) shows that age can significantly affect the likelihood of urban households to make risky financial asset allocations at the 99 % confidence level, with a marginal effect of −0.0044, and the marginal effect on the depth of urban households' allocations of risky financial assets is −0.0011, which also passes the 1 % significance level test. The specific performance is that as the age of the head of the household increases, the smaller the willingness to carry out the allocation of risky financial assets, and more cautious in its selection. The regression data of age in models (3) and (4) also show the same impact effect, with the results of −0.0042 and −0.0010 on the allocation of risky financial assets, respectively. Models (2) and (4) show that compared to female groups, male heads of urban households lead to a 1.56 % and 2.01 % decrease in the depth of household risky financial asset allocation, respectively. All four regression models show a positive correlation between literacy and risky financial asset allocation. The more educated the head of the urban household is, the higher his corresponding financial literacy will be, and his knowledge of risky financial products and tools will be more in-depth, so that the probability and the depth of the entire household's allocation of risky financial assets will be increased accordingly. Marital status will also affect the allocation of household risky financial assets of residents, and the regression results show a negative correlation, with an impact coefficient of −0.0192 in model (1), which is significant at the 10 % significance level. In order to guarantee the basic living conditions of the family, the allocation of risky financial assets of married urban households will be constrained by the family, while the unmarried households will be more free in the allocation of risky financial assets, and the corresponding possibility will increase. Similarly, the health status of the household head positively influences whether urban households allocate risky financial assets. The healthier and better the head of the household perceives himself, the less likely he is to suffer from physical diseases. As a result, the capital demand for preventive medical treatment in the household will be reduced, which will enable the household to have spare funds for the financial assets allocation. The marginal effect of health on the probability of risky financial asset allocation of urban households is 0.0254 and 0.0266, but the effect on the depth of risky financial asset allocation is not significant. The degree of risk preference of urban households also positively affects the probability of risky financial asset allocation and the size of the market value share of risky financial assets allocated, with regression results all showing significance at the 1 % significance level. The economic status of urban households is the basis for various asset allocations, and the empirical results also significantly show that the better the total income and total assets of urban households, the higher the probability and depth of allocation to risky financial assets such as stocks and funds.

4.1.2. Regression analysis of the impact of housing liabilities on the financial asset allocation of urban households

Table 3 examines the impact of housing liabilities on the allocation of risky financial assets using a data sample of urban households that own homes. To analyze the impact of housing liabilities more clearly, it is divided into two explanatory variables: bank loans and other borrowings for housing. In the regression (1) of the marginal impact on the allocation of risky financial assets, the impact of bank loans for housing on the probability of the allocation of risky financial assets of urban households is negative, but it does not pass the 10 % significance level test, so the original hypothesis is rejected. Housing borrowing other than bank loans is negatively associated with the probability of household participation in risky financial asset allocation such as equity funds, with a marginal effect size of −0.0502, which passes the 1 % significance level test, implying that informal housing other borrowing other than bank loans dampens the possibility of risky financial asset allocation. In terms of the proportion of risky financial asset allocation, bank loans have a facilitating effect and housing other borrowing still has a dampening effect, but the regression results are not significant enough. Possible explanations are that formal bank housing loans have stable repayment terms, cash flows can be rationalized, and repayment pressures on loan borrowers are relatively low. In this context, households can make rational allocations of funds, and bank loans indirectly provide financing for some risky financial market products and instruments. On the other hand, housing loans can be utilized to leverage profits in real estate, and idle funds can then enter the risky financial market for asset allocation. In addition, housing bank loans are also part of housing fund loans. Among the households that contribute to the housing fund, they will choose to take out a provident fund loan because of the poor liquidity of the housing fund and the lower interest rate used for bank loans, etc. The advantages of the housing fund loan also make most of the households more favorable. Housing loans other than bank loans, on the other hand, generally come from informal financial channels, and the repayment period is not fixed; when families have disposable funds, they are forced to make repayments under the pressure of repayment, which inhibits the possibility and depth of urban households' participation in the allocation of risky financial assets.

Table 3.

Regression results on the impact of housing debts on risky financial asset allocation.

Variables	(1)	(2)
	Whether risky financial assets are allocated	Risky financial asset allocation ratio
	Probit	Tobit
bankloan	−0.0030	0.0108
	(0.0114)	(0.0066)
housedebt	−0.0502***	−0.0088
	(0.0129)	(0.0076)
age	−0.0045***	−0.0010***
	(0.0003)	(0.0002)
gender	−0.0022	−0.0194***
	(0.0086)	(0.0048)
edu	0.0206***	0.0154***
	(0.0025)	(0.0014)
marriage	−0.0045	−0.0014
	(0.0119)	(0.0064)
health	0.0247**	−0.0006
	(0.0114)	(0.0060)
riskpre	0.1555***	0.1048***
	(0.0167)	(0.0086)
lnincome	0.0424***	0.0121***
	(0.0034)	(0.0017)
lnasset	0.1117***	0.0473***
	(0.0037)	(0.0019)
Constant		−0.6117***
		(0.0265)
Observations	17,716	17,716
Pseudo R-squared	0.149	0.343

4.2. Robustness tests

It is also possible that the housing asset allocation of urban households may be affected by the allocation of risky financial assets, i.e., reverse causality, which may be endogenous in the model developed. In order to overcome the endogeneity in the model, this paper draws on the research design of Lv et al. [32], and selects the average sales price of residential commercial housing in the province where the urban households are located in the 2019 China Statistical Yearbook as the instrumental variables of whether they own only one house and whether they own multiple houses to conduct IVProbit and IVTobit regressions. Theoretically, the average house price in a certain region affects the allocation of households to real estate and risky financial assets. However, the average house price is for the whole region where the households are located, and it is exogenous and not affected by the allocation of risky financial assets in the households. Thus, the average house price within a provincial area is suitable as an instrumental variable.

Table 4 presents the results of the IVProbit and IVTobit model regressions of the allocation to risky financial assets for the number of houses owned after selecting the average regional house price as the instrumental variable. In this case, models (1) and (2) are conducted with or without the sample of owning only one home, while models (3) and (4) are conducted with or without the sample of owning multiple homes. It can be seen that the F-values estimated in the first stage are both large, 1004.08 and 351.02, respectively, which are both greater than 10, so there is no weak instrumental variable problem, which suggests that the use of average regional house price as an instrumental variable is appropriate. In models (1) and (2), the p-values in the Wald test are 0.9424 and 0.4927, respectively, which are seen to be insignificant, which indicates the appropriateness of the model regression results without using instrumental variables. The p-values in Wald test for models (3) and (4) are 0.0231 and 0.0014, respectively, which are significant at 5 % and 1 % significance. Therefore, it is appropriate to use the instrumental variable method to overcome endogeneity. However, after overcoming the endogeneity problem, the direction of the effect of owning multiple houses on the risky financial asset allocation does not change. Therefore, this verifies that there is mainly a crowding-out effect on financial asset allocation when households acquire their first house, but as the number of house holdings increases, housing exhibits more of an asset allocation effect on financial asset allocation, which is consistent with the results of the referenced study by Lv et al. [32].The reason why there is no endogeneity in the model of whether an urban household owns one house but there is endogeneity in the model of whether a household owns multiple houses may lie in the fact that households owning multiple houses are more based on investment demand. If households can make profits in risky financial assets such as stocks, its corresponding allocation of funds will increase, which will also affect the allocation of housing assets in the opposite direction. However, in the case of households owning one house, which are based on their own needs, the allocation of risky financial assets is not sufficient to affect the allocation of owning one house.

Table 4.

Regression results of the instrumental variable approach to the effect of the number of homes owned on the allocation of risky financial assets.

Variables	(1)	(2)	(3)	(4)
	Whether risky financial assets are allocated IVProbit	Risky financial asset allocation ratio IVTobit	Whether risky financial assets are allocated IVProbit	Risky financial asset allocation ratio IVTobit
house	−0.5483***	−0.2720***
	(0.1372)	(0.0769)
houses			0.5789***	0.3788***
			(0.2011)	(0.1047)
age	−0.0136***	−0.0064***	−0.0110***	−0.0045***
	(0.0008)	(0.0005)	(0.0009)	(0.0005)
gender	−0.0138	−0.0274*	−0.0205	−0.0351***
	(0.0258)	(0.0147)	(0.0246)	(0.0130)
edu	0.0640***	0.0387***	0.0646***	0.0362***
	(0.0088)	(0.0050)	(0.0073)	(0.0038)
marriage	−0.0585*	−0.0308	−0.0335	−0.0130
	(0.0333)	(0.0189)	(0.0341)	(0.0181)
health	0.0793**	0.0424**	0.0769**	0.0367**
	(0.0339)	(0.0197)	(0.0329)	(0.0179)
riskpre	0.4154***	0.2126***	0.3973***	0.1890***
	(0.0459)	(0.0245)	(0.0431)	(0.0210)
lnincome	0.0939***	0.0467***	0.1039***	0.0404***
	(0.0100)	(0.0057)	(0.0110)	(0.0058)
lnasset	0.2955***	0.1628***	0.2547***	0.1307***
	(0.0197)	(0.0111)	(0.0221)	(0.0115)
Constant	−4.5275***	−2.5595***	−4.8900***	−2.4684***
	(0.1529)	(0.0894)	(0.2976)	(0.1557)
Wald test of exogeneity	Prob > chi2 = 0.9424	Prob > chi2 = 0.4927	Prob > chi2 = 0.0231**	Prob > chi2 = 0.0014***
The F-statistics of the first stage	1004.08	1004.08	351.02	351.02
Observations	16,300	16,300	17,716	17,716

4.3. Mediation effect test

According to the empirical analysis, the factors affecting the risky financial asset allocation of urban households are complex and diverse, and risk preference is an important influence variable among the control variables. Therefore, the variable of risk preference is further studied. Risk preference affects the probability and proportion of risky financial asset allocation of urban households, and it is reasonable to assume that housing and other household characteristic factors in the household in turn affect risk preference, which in turn acts on risky financial asset allocations. Referring to the research methods of Lu [31] who takes income as a mediating variable to study the impact of labor experience on the choice of household financial assets and Xiao et al. [33] who test whether risk attitude is a mediating variable of wealth level affecting consumption, this paper employs a stepwise regression method by selecting risk preference as a mediating variable to test whether risk preference is a mediating variable for housing affecting risky financial asset allocation.

Table 5 provides the test effects of using risk preference as a mediating variable of owning only one house on the probability and proportion of risky financial asset allocation of urban households. Models (1) and (4) illustrate that the negative effect of owning only one house on the probability and proportion of risky financial asset allocation of urban households without the risk preference factor is significant at the 1 % significance level, with impact coefficients of −0.2037 and −0.0905, respectively. Models (3) and (6) give the magnitude of the effect of owning only one house on the probability and proportion of risky financial asset allocation of urban households with the inclusion of risk preference as a mediator variable of −0.1990, respectively. The effect sizes of the probability and proportion of risky financial asset allocation are −0.2003 and −0.0883, respectively, and both pass the 1 % significance level test. In models (2) and (5), risk preference is used as an explanatory variable which explains the relationship between owning only one house and other household characteristics and household risk preference. It is observed that owning only one house reduces the risk preference capacity of urban households with an impact coefficient of −0.0207, which is significant at the 99 % confidence level. Evidently, when urban households own only one house, the level of risk aversion will increase by 2.07 %. The significance of the stepwise regression coefficients indicates that the mediating effect from risk preference is significant, and thus the mechanism by which urban households owning only one house affects the allocation of risky financial assets through risk preference exists. Therefore, when an urban household owns only one house, the pressure on the property makes the household's risk tolerance decrease, which reduces the probability and depth of the allocation of risky financial assets among urban households.

Table 5.

A test of the risk preference effect of owning one house on the risky financial asset allocation.

The probabilistic intermediation effect of risky financial asset allocations					Intermediation effects of risky financial asset allocation ratios
Variables	(1)	(2)	(3)	(4)	(5)	(6)
	risk	riskpre	risk	riskpercentage	riskpre	riskpercentage
house	−0.2037***	−0.0207***	−0.2003***	−0.0905***	−0.0207***	−0.0883***
	(0.0141)	(0.0063)	(0.0142)	(0.0069)	(0.0063)	(0.0068)
age	−0.0047***	−0.0018***	−0.0044***	−0.0012***	−0.0018***	−0.0011***
	(0.0003)	(0.0001)	(0.0003)	(0.0002)	(0.0001)	(0.0002)
gender	−0.0015	0.0152***	−0.0044	−0.0138***	0.0152***	−0.0156***
	(0.0083)	(0.0034)	(0.0083)	(0.0047)	(0.0034)	(0.0047)
edu	0.0218***	0.0051***	0.0209***	0.0158***	0.0051***	0.0150***
	(0.0025)	(0.0011)	(0.0025)	(0.0015)	(0.0011)	(0.0014)
marriage	−0.0238**	−0.0197***	−0.0192*	−0.0116**	−0.0197***	−0.0088
	(0.0108)	(0.0054)	(0.0108)	(0.0059)	(0.0054)	(0.0059)
health	0.0263**	0.0047	0.0254**	0.0022	0.0047	0.0019
	(0.0106)	(0.0049)	(0.0106)	(0.0058)	(0.0049)	(0.0057)
riskpre			0.1491***			0.0931***
			(0.0176)			(0.0090)
lnincome	0.0313***	0.0033**	0.0307***	0.0105***	0.0033**	0.0101***
	(0.0031)	(0.0013)	(0.0031)	(0.0016)	(0.0013)	(0.0016)
lnasset	0.0980***	0.0077***	0.0971***	0.0393***	0.0077***	0.0387***
	(0.0033)	(0.0014)	(0.0033)	(0.0017)	(0.0014)	(0.0017)
Constant				−0.3831***		−0.3862***
				(0.0228)		(0.0227)
Observations	16,300	16,300	16,300	16,300	16,300	16,300
Pseudo R-squared	0.137	0.0807	0.141	0.368	0.0807	0.393

Table 6 gives the test effects of using risk preference as a mediating variable of owning multiple homes on the probability and proportion of risky financial asset allocation of urban households. Models (1) and (4) indicate that in the absence of the variable of risk preference, owning multiple houses has a significant positive impact on the probability and proportion of household risky financial asset allocation, with specific degrees of influence of 0.0475 and 0.0319, respectively. Models (3) and (6) are the results of the regression after the addition of the variable of risk preference, which show that the marginal effect on the probability of risky financial asset allocation decreases to 0.0469, and the influence coefficient on the proportion of household risky financial asset allocation decreases to 0.0310. Model (2) and (5) illustrate that owning multiple houses increases the risk tolerance of households and makes them more risk preferred, but the impact is not significant. Therefore, the mechanism by which urban households owning multiple houses affects risky financial asset allocation through risk preference is not significant. The probable reason is that households with multiple houses are more likely to purchase multiple houses based on investment attributes, and their risk preference ability is inherently higher, which will not be significantly enhanced after owning multiple houses.

Table 6.

A test of the risk preference effect of owning multiple houses on the risky financial asset allocation.

The probabilistic intermediation effect of risky financial asset allocations					Intermediation effects of risky financial asset allocation ratios
Variables	(1)	(2)	(3)	(4)	(5)	(6)
	risk	riskpre	risk	riskpercentage	riskpre	riskpercentage
houses	0.0475***	0.0050	0.0469***	0.0319***	0.0050	0.0310***
	(0.0095)	(0.0040)	(0.0095)	(0.0053)	(0.0040)	(0.0053)
age	−0.0045***	−0.0017***	−0.0042***	−0.0012***	−0.0017***	−0.0010***
	(0.0003)	(0.0001)	(0.0003)	(0.0002)	(0.0001)	(0.0002)
gender	−0.0016	0.0128***	−0.0040	−0.0185***	0.0128***	−0.0201***
	(0.0086)	(0.0035)	(0.0086)	(0.0048)	(0.0035)	(0.0048)
edu	0.0227***	0.0054***	0.0217***	0.0169***	0.0054***	0.0160***
	(0.0025)	(0.0011)	(0.0025)	(0.0014)	(0.0011)	(0.0014)
marriage	−0.0114	−0.0230***	−0.0064	−0.0056	−0.0230***	−0.0024
	(0.0119)	(0.0062)	(0.0119)	(0.0064)	(0.0062)	(0.0064)
health	0.0269**	0.0020	0.0266**	−0.0004	0.0020	−0.0002
	(0.0113)	(0.0054)	(0.0113)	(0.0060)	(0.0054)	(0.0060)
riskpre			0.1551***			0.1044***
			(0.0167)			(0.0086)
lnincome	0.0421***	0.0064***	0.0411***	0.0122***	0.0064***	0.0115***
	(0.0034)	(0.0015)	(0.0034)	(0.0017)	(0.0015)	(0.0017)
lnasset	0.1077***	0.0113***	0.1062***	0.0457***	0.0113***	0.0446***
	(0.0038)	(0.0017)	(0.0038)	(0.0019)	(0.0017)	(0.0019)
Constant				−0.5829***		−0.5758***
				(0.0273)		(0.0272)
Observations	17,716	17,716	17,716	17,716	17,716	17,716
Pseudo R-squared	0.146	0.0774	0.150	0.325	0.0774	0.348

5. Heterogeneity analysis

The crowding-out effect and asset allocation effect of housing on risky financial asset allocation and allocation ratio of urban households were analyzed in the previous section, and due to the large differences between urban households of different incomes, ages, and regions, the two effects are significantly heterogeneous among them. Hence, further consideration is devoted to the heterogeneity of these two effects across urban households with different household incomes, ages, and regions. In the study, the sample data will be grouped by per capita annual household income, age, and region. After each grouping, differences in the crowding-out effect of the impact of housing on risky financial asset allocation are examined first among urban households owning only one house, and then differences in the asset allocation effect of the impact of housing on risky financial allocation are analyzed among urban households owning multiple houses.

5.1. Income heterogeneity analysis of the impact of housing on household risk financial asset allocation

Economic status is an important factor influencing the allocation of financial assets and participation in financial markets by urban households. Before examining the heterogeneity of household incomes, it is necessary to classify urban households into income groups. According to the official caliber of the National Bureau of Statistics on the division of middle-income groups in the 2018 National Time Use Survey Bulletin, those with an annual per capita household income of less than 24,000 yuan are classified as low-income households, those with an average annual per capita household income of 24,000–60,000 yuan are classified as middle-income households, those with an average annual per capita household income of between 60,000–120,000 yuan are classified as higher-income families, and those with an annual per capita household income of more than 120,000 yuan are classified as high-income families.

Table 7, Table 8 show the model regression results for the risky financial asset allocation and allocation ratio of urban households owning only one house under different income groups. Consistent with the overall sample regression results above, the impact of owning only one house in each income group on the probability of risky financial asset allocation and allocation proportion of urban households is still significantly negative, i.e., the housing crowding-out effect of urban households owning one house is greater than its asset allocation effect. Specifically, as shown in Table 7, the crowding-out effect of housing assets for low-income urban households is −0.1629, and that of middle-income and higher-income urban households is −0.1710 and −0.1444, respectively, and the regression results all pass the test at the 1 % significance level. The crowding-out effect of owning one house on the probability of allocating risky financial products to middle-income urban households is greater than that of low-income and higher-income urban households. Table 8 shows that the negative effect of owning onehouse on the allocation ratio of risky financial products and instruments for low-income households is −0.0660, and the inhibitory effect on middle-income and higher-income urban households amounts to −0.0864 and −0.0774. The inhibitory effect on the depth of risky financial asset allocation of middle-income households is also higher than that of low-income and higher-income urban households. It is evident that the crowding-out effect of housing on risky financial assets of middle income urban households is greater than that of low-income and higher-income households.

Table 7.

Regression results of Probit model for the sample with owning one house under household income group.

Variables	(1)	(2)	(3)	(4)
	Low income	Middle income	Higher income	High income
house	−0.1629***	−0.1710***	−0.1444***	−0.2319***
	(0.0194)	(0.0226)	(0.0428)	(0.0507)
age	−0.0036***	−0.0052***	−0.0032***	−0.0047***
	(0.0004)	(0.0005)	(0.0009)	(0.0016)
gender	0.0070	−0.0102	−0.0290	−0.0690
	(0.0104)	(0.0136)	(0.0285)	(0.0472)
edu	0.0196***	0.0116***	0.0138*	−0.0007
	(0.0035)	(0.0040)	(0.0082)	(0.0147)
marriage	−0.0137	−0.0391**	−0.0672*	0.0425
	(0.0143)	(0.0199)	(0.0407)	(0.0616)
health	0.0185	0.0269	−0.0027	−0.0434
	(0.0112)	(0.0189)	(0.0505)	(0.0973)
riskpre	0.0885***	0.1754***	0.1701***	0.1317**
	(0.0236)	(0.0296)	(0.0411)	(0.0539)
lnincome	0.0070**	0.1136***	0.1004***	−0.0120
	(0.0029)	(0.0155)	(0.0362)	(0.0375)
lnasset	0.0730***	0.0855***	0.1008***	0.1146***
	(0.0038)	(0.0060)	(0.0138)	(0.0221)
Observations	7933	6400	1505	462
Pseudo R-squared	0.109	0.0983	0.0689	0.114

Table 8.

Regression results of Tobit model for the sample with owning one house under household income group.

Variables	(1)	(2)	(3)	(4)
	Low income	Middle income	Higher income	High income
house	−0.0660***	−0.0864***	−0.0774***	−0.1896***
	(0.0085)	(0.0119)	(0.0288)	(0.0481)
age	−0.0011***	−0.0016***	−0.0006	−0.0005
	(0.0002)	(0.0003)	(0.0006)	(0.0011)
gender	−0.0054	−0.0200**	−0.0372**	−0.0102
	(0.0061)	(0.0078)	(0.0181)	(0.0335)
edu	0.0125***	0.0080***	0.0093*	0.0055
	(0.0021)	(0.0024)	(0.0052)	(0.0103)
marriage	−0.0077	−0.0077	−0.0069	0.0238
	(0.0077)	(0.0110)	(0.0261)	(0.0423)
health	0.0008	0.0053	−0.0113	0.0121
	(0.0063)	(0.0107)	(0.0321)	(0.0745)
riskpre	0.0600***	0.1121***	0.0831***	0.0906**
	(0.0122)	(0.0160)	(0.0277)	(0.0410)
lnincome	−0.0013	0.0316***	0.0113	−0.0206
	(0.0017)	(0.0091)	(0.0228)	(0.0264)
lnasset	0.0273***	0.0369***	0.0533***	0.0887***
	(0.0020)	(0.0032)	(0.0084)	(0.0146)
Constant	−0.1520***	−0.5579***	−0.5416**	−0.5508
	(0.0296)	(0.0996)	(0.2628)	(0.3542)
Observations	7933	6400	1505	462
Pseudo R-squared	−0.542	0.186	0.0777	0.166

As can be seen in the sample subgroups, there is a larger proportion of low-income and middle-income households. Compared with middle-income urban households, the total household income of low-income households is at a low level, and the capital allocation itself is limited to the expenditure on the protection of necessities such as family life and education, and the probability of risky asset allocation will be significantly reduced. When it comes to the allocation of real estate, the crowding-out effect from housing is relatively weak. Higher-income urban households are in a better economic status, compared to middle-income households, with relatively free cash flow and free distribution, so the crowding-out effect from one house is lower.

Table 9, Table 10 display the model regression results of risky financial asset allocation and allocation proportion of urban households with owning multiple houses under different income groups. Among low-income and middle-income urban households, the marginal effects of owning multiple houses on risky financial asset allocation and allocation ratio are 0.0423 and 0.0428, respectively, and the regression results have passed the 95 % significance level test. That is, the asset allocation effect of multiple housing is significant, and owning multiple housing will increase the probability of risky financial asset allocation of low-income and middle-income urban households. However, the effect of owning multiple houses on urban households of other income levels is not significant. The possible reason is that for low-income families, owning multiple houses can obtain funds other than income through rentals, bed and breakfast operations, appreciation realization, etc., and the asset allocation effect of owning multiple houses allows households to have additional funds to invest in risky financial asset allocations. While higher-income households have more funds available to allocate risky financial assets, so the asset allocation effect demonstrated is not significant enough.

Table 9.

Regression results of Probit model for the sample with owning multiple houses under household income group.

Variables	(1)	(2)	(3)	(4)
	Low income	Middle income	Higher income	High income
houses	0.0423***	0.0428***	0.0211	0.0269
	(0.0134)	(0.0148)	(0.0255)	(0.0366)
age	−0.0029***	−0.0055***	−0.0024***	−0.0017
	(0.0004)	(0.0005)	(0.0008)	(0.0013)
gender	0.0101	−0.0109	−0.0252	−0.0575
	(0.0111)	(0.0135)	(0.0247)	(0.0361)
edu	0.0205***	0.0117***	0.0166**	0.0141
	(0.0037)	(0.0039)	(0.0070)	(0.0110)
marriage	−0.0056	−0.0337*	−0.0180	0.0431
	(0.0161)	(0.0202)	(0.0373)	(0.0498)
health	0.0197	0.0211	0.0509	−0.0410
	(0.0121)	(0.0195)	(0.0472)	(0.0845)
riskpre	0.0995***	0.1738***	0.1481***	0.1086***
	(0.0251)	(0.0270)	(0.0341)	(0.0399)
lnincome	0.0126***	0.1159***	0.1161***	0.0021
	(0.0033)	(0.0149)	(0.0313)	(0.0276)
lnasset	0.0795***	0.0903***	0.1019***	0.1072***
	(0.0045)	(0.0066)	(0.0137)	(0.0192)
Observations	7928	7064	2001	723
Pseudo R-squared	0.100	0.0981	0.0763	0.0990

Table 10.

Regression results of Tobit model for the sample with owning multiple houses under household income group.

Variables	(1)	(2)	(3)	(4)
	Low income	Middle income	Higher income	High income
houses	0.0369***	0.0136	0.0245	−0.0012
	(0.0076)	(0.0085)	(0.0167)	(0.0272)
age	−0.0010***	−0.0016***	−0.0004	0.0003
	(0.0002)	(0.0003)	(0.0005)	(0.0010)
gender	−0.0079	−0.0281***	−0.0416**	0.0077
	(0.0065)	(0.0076)	(0.0163)	(0.0275)
edu	0.0122***	0.0097***	0.0101**	0.0187**
	(0.0022)	(0.0023)	(0.0046)	(0.0082)
marriage	−0.0075	0.0025	0.0073	0.0199
	(0.0086)	(0.0111)	(0.0247)	(0.0353)
health	0.0009	0.0049	−0.0035	−0.0209
	(0.0067)	(0.0108)	(0.0307)	(0.0688)
riskpre	0.0695***	0.1151***	0.1020***	0.0821**
	(0.0129)	(0.0147)	(0.0239)	(0.0323)
lnincome	−0.0013	0.0256***	0.0286	−0.0093
	(0.0019)	(0.0085)	(0.0205)	(0.0197)
lnasset	0.0303***	0.0434***	0.0567***	0.0922***
	(0.0023)	(0.0036)	(0.0088)	(0.0137)
Constant	−0.2636***	−0.6682***	−0.9089***	−1.0274***
	(0.0368)	(0.0971)	(0.2481)	(0.2917)
Observations	7928	7064	2001	723
Pseudo R-squared	−3.780	0.188	0.0911	0.163

5.2. Age heterogeneity analysis of the impact of housing on household risky financial asset allocation

To examine the differences in the crowding-out and asset allocation effects of housing across urban households of different ages, the sample is divided into four age groups, with 18–35 as young, 36–50 as young and middle-aged, 51–65 as the middle-aged and elderly, and 65 and above as elderly.

Table 11, Table 12 present the results of the model regressions of risky financial asset allocations and allocation ratios for urban households owning only one house under different age groups. It suggests that owning only one house under each age group has an inhibitory effect on risky financial asset allocation and allocation ratio, and is significant at the 1 % significance level. Same as the full sample regression results, but with significant variability across age groups. The marginal impact effects on the probability of risky financial asset allocation in young, young and middle-aged, middle-aged and elderly, and elderly urban households are −0.2546, −0.1930, −0.1820, and −0.1441, respectively. It is evident that the crowding-out effect of one set of housing asset allocation will gradually diminish as the age group of the head of the household rises in urban households. The coefficients of the impact on the proportion of risky financial assets allocation under different age stages are −0.1045, −0.0941, −0.0864, and −0.0755, respectively. The crowding-out effect on the depth of risky financial assets allocation of urban households will likewise be reduced by the rise of age stages. Therefore, the crowding-out effect of asset allocation for owning only one house is strongest among young urban households and weakest among elderly urban households. Young urban households, which are in the early stage of wealth accumulation, will have significantly reduced funds to allocate to risky financial products or other financial instruments once they have made an allocation to real estate. And with the growth of age and the accumulation of wealth, the financial constraints of one house are gradually eased. In elderly age, the accumulation of wealth will be at its highest, and housing loans, if any, will have been paid off in urban households, so that the crowding-out effect of owning only one house will be significantly reduced.

Table 11.

Regression results of Probit model for the sample with owning one house under different age groups.

Variables	(1)	(2)	(3)	(4)
	young	young and middle-aged	middle-aged and elderly	elderly
house	−0.2546***	−0.1930***	−0.1820***	−0.1441***
	(0.0345)	(0.0278)	(0.0251)	(0.0261)
age	−0.0103***	−0.0040**	−0.0045***	−0.0048***
	(0.0039)	(0.0018)	(0.0013)	(0.0009)
gender	−0.0441	0.0038	0.0000	−0.0082
	(0.0276)	(0.0178)	(0.0135)	(0.0127)
edu	0.0109	0.0179***	0.0235***	0.0219***
	(0.0081)	(0.0052)	(0.0047)	(0.0036)
marriage	−0.0678**	−0.0243	−0.0223	0.0055
	(0.0334)	(0.0278)	(0.0194)	(0.0149)
health	−0.0714	0.0014	0.0188	0.0331***
	(0.0757)	(0.0277)	(0.0157)	(0.0127)
riskpre	0.1422***	0.1472***	0.1360***	0.2453***
	(0.0346)	(0.0294)	(0.0332)	(0.0549)
lnincome	0.0381***	0.0292***	0.0273***	0.0334***
	(0.0097)	(0.0059)	(0.0052)	(0.0056)
lnasset	0.1079***	0.1130***	0.1017***	0.0648***
	(0.0124)	(0.0076)	(0.0055)	(0.0045)
Observations	1720	4371	5610	4599
Pseudo R-squared	0.0875	0.0959	0.121	0.146

Table 12.

Regression results of Tobit model for owning one house samples under different age groups.

Variables	(1)	(2)	(3)	(4)
	young	young and middle-aged	middle-aged and elderly	elderly
house	−0.1045***	−0.0941***	−0.0864***	−0.0755***
	(0.0210)	(0.0143)	(0.0120)	(0.0120)
age	−0.0017	−0.0022**	−0.0006	−0.0018***
	(0.0022)	(0.0010)	(0.0008)	(0.0006)
gender	−0.0090	−0.0153	−0.0212***	−0.0202**
	(0.0161)	(0.0098)	(0.0079)	(0.0082)
edu	0.0062	0.0081***	0.0176***	0.0214***
	(0.0047)	(0.0029)	(0.0028)	(0.0024)
marriage	−0.0521***	−0.0072	−0.0097	0.0066
	(0.0194)	(0.0148)	(0.0107)	(0.0093)
health	−0.0470	−0.0181	−0.0014	0.0151*
	(0.0445)	(0.0146)	(0.0089)	(0.0082)
riskpre	0.0817***	0.0918***	0.0843***	0.1346***
	(0.0207)	(0.0155)	(0.0171)	(0.0247)
lnincome	0.0202***	0.0104***	0.0074***	0.0082***
	(0.0053)	(0.0031)	(0.0028)	(0.0030)
lnasset	0.0390***	0.0527***	0.0428***	0.0288***
	(0.0068)	(0.0039)	(0.0029)	(0.0026)
Constant	−0.3530***	−0.4722***	−0.4391***	−0.2364***
	(0.1024)	(0.0654)	(0.0580)	(0.0534)
Observations	1720	4371	5610	4599
Pseudo R-squared	0.141	0.275	0.403	2.365

Table 13, Table 14 provide the model regression results for the allocation and proportion of risky financial assets allocated to urban households owning multiple houses under the four age groups. Evidently, owning multiple houses has a notable contribution to the risky financial asset allocation and allocation ratio of urban households in the young and middle-aged and above age groups. Among them, the marginal effect of owning multiple houses in elderly households on the probability of risky financial asset allocation is 0.0508, and the impact results in young and middle-aged and middle-aged and elderly households are 0.0477 and 0.0394, respectively. In terms of the probability of risky financial asset allocation, the asset allocation effect of owning multiple houses in urban households in the elderly age group is the strongest. In terms of the depth of risky financial asset allocation, the asset allocation effect of owning multiple houses is significant among young and middle-aged, middle-aged and elderly, and elderly households, with the results of 0.0292, 0.0206, and 0.0337 for the impact effect, respectively. Similarly, the asset allocation effect is strongest in older households. While the asset allocation effect of owning multiple houses in young urban households is not significant. This may be due to the fact that in the elderly urban households, on the one hand, the accumulation of wealth gradually reaches the highest and the pressure of life gradually decreases, with surplus funds and time more likely to be allocated to risky financial assets. On the other hand, factors such as the long-term accumulation of investment experience and the need for estate planning may also lead them to deepen their allocation to risky financial assets. But for young urban households who have been working for a short time, the asset allocation of housing is not significant after making multiple property allocations under the pressure of both work and family.

Table 13.

Regression results of Probit model for the sample with owning multiple houses under different age groups.

Variables	(1)	(2)	(3)	(4)
	young	young and middle-aged	middle-aged and elderly	elderly
houses	0.0059	0.0477***	0.0394***	0.0508***
	(0.0337)	(0.0176)	(0.0150)	(0.0182)
age	−0.0073*	−0.0026	−0.0056***	−0.0052***
	(0.0042)	(0.0017)	(0.0014)	(0.0010)
gender	−0.0435	0.0037	−0.0061	−0.0004
	(0.0295)	(0.0170)	(0.0139)	(0.0142)
edu	0.0043	0.0200***	0.0277***	0.0245***
	(0.0089)	(0.0049)	(0.0046)	(0.0039)
marriage	−0.0471	−0.0151	−0.0285	0.0009
	(0.0371)	(0.0291)	(0.0212)	(0.0172)
health	−0.1298	−0.0063	0.0319*	0.0302**
	(0.0798)	(0.0285)	(0.0166)	(0.0146)
riskpre	0.1452***	0.1363***	0.1572***	0.2470***
	(0.0370)	(0.0258)	(0.0316)	(0.0503)
lnincome	0.0678***	0.0393***	0.0309***	0.0505***
	(0.0125)	(0.0062)	(0.0055)	(0.0066)
lnasset	0.1015***	0.1241***	0.1141***	0.0674***
	(0.0164)	(0.0082)	(0.0061)	(0.0055)
Observations	1550	5222	6381	4563
Pseudo R-squared	0.0879	0.119	0.136	0.149

Table 14.

Regression results of Tobit model for multiple housing samples under different age groups.

Variables	(1)	(2)	(3)	(4)
	young	young and middle-aged	middle-aged and elderly	elderly
houses	0.0125	0.0292***	0.0206**	0.0337***
	(0.0197)	(0.0099)	(0.0086)	(0.0108)
age	−0.0025	−0.0010	−0.0018**	−0.0022***
	(0.0024)	(0.0010)	(0.0008)	(0.0006)
gender	−0.0314*	−0.0219**	−0.0185**	−0.0223**
	(0.0172)	(0.0094)	(0.0078)	(0.0087)
edu	0.0026	0.0116***	0.0185***	0.0234***
	(0.0052)	(0.0027)	(0.0026)	(0.0025)
marriage	−0.0348	−0.0118	−0.0092	0.0065
	(0.0216)	(0.0158)	(0.0114)	(0.0101)
health	−0.0714	−0.0147	−0.0020	0.0101
	(0.0484)	(0.0153)	(0.0091)	(0.0088)
riskpre	0.0706***	0.1015***	0.1085***	0.1472***
	(0.0223)	(0.0139)	(0.0161)	(0.0229)
lnincome	0.0267***	0.0114***	0.0077***	0.0124***
	(0.0065)	(0.0032)	(0.0029)	(0.0032)
lnasset	0.0425***	0.0606***	0.0507***	0.0288***
	(0.0092)	(0.0043)	(0.0032)	(0.0030)
Constant	−0.5098***	−0.7444***	−0.5704***	−0.3288***
	(0.1347)	(0.0712)	(0.0605)	(0.0595)
Observations	1550	5222	6381	4563
Pseudo R-squared	0.143	0.275	0.362	0.786

5.3. Regional heterogeneity analysis of the impact of housing on household risk financial asset allocation

To further investigate the differences in housing crowding-out and asset allocation effects among urban households in different regions, the sample of urban households is divided into three major economic regions, namely Eastern, Central, and Western, with reference to Zhou et al. [34]. East includes Liaoning, Hebei, Beijing, Tianjin, Shandong, Jiangsu, Shanghai, Zhejiang, Fujian, Guangdong, and Hainan. Central includes Heilongjiang, Jilin, Shanxi, Henan, Anhui, Jiangxi, Hubei, and Hunan. And West includes Shaanxi, Chongqing, Guizhou, Sichuan, Yunnan, Gansu, Qinghai, Inner Mongolia, Guangxi Zhuang, and Ningxia Hui Autonomous Region.

Table 15 shows the results of the model regression of risky financial asset allocations and allocation proportions for urban households owning only one house in different regions. The marginal effects on the probability of risky financial asset allocation in the Eastern, Central, and Western regions are −0.2072, −0.1865, and −0.1805, respectively. The crowding-out effect of owning only one house on the probability of risky financial asset allocation for urban households is the strongest in the Eastern region, followed by housing in the Central region, and the weakest one in the Western region. In terms of the proportion of risky financial asset allocation, the influence coefficients of owning only one house on urban households in the Eastern, Central, and Western regions are −0.0913, −0.0850, and −0.0734, respectively. Similarly, the housing crowding-out effect is weakest in the Western region where economic development is lagging behind. It can be seen that the crowding-out effect of owning one house in urban households is more significant in economically developed regions, which is due to the fact that in these regions, housing prices are rising rapidly, and most families will pour out everything they have to allocate one house. The absolute dominance of housing assets in the asset allocation structure of these households has led to a weakening of the liquidity of household funds, making them unable and unwilling to re-enter the capital markets for risky financial products and inhibiting the depth of their allocations in risky financial markets.

Table 15.

Regression results of the model for the sample with owning one house under different region groups.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	Eastern region	Central region	Western region	Eastern region	Central region	Western region
	Probit			Tobit
house	−0.2072***	−0.1865***	−0.1805***	−0.0913***	−0.0850***	−0.0734***
	(0.0208)	(0.0281)	(0.0279)	(0.0106)	(0.0131)	(0.0130)
age	−0.0045***	−0.0043***	−0.0043***	−0.0006***	−0.0013***	−0.0016***
	(0.0004)	(0.0005)	(0.0005)	(0.0002)	(0.0003)	(0.0003)
gender	−0.0071	−0.0013	−0.0052	−0.0196***	−0.0099	−0.0124
	(0.0132)	(0.0149)	(0.0146)	(0.0074)	(0.0088)	(0.0085)
edu	0.0204***	0.0159***	0.0252***	0.0169***	0.0112***	0.0150***
	(0.0039)	(0.0045)	(0.0043)	(0.0022)	(0.0028)	(0.0026)
marriage	−0.0091	−0.0393**	−0.0135	−0.0040	−0.0272**	0.0003
	(0.0169)	(0.0199)	(0.0190)	(0.0092)	(0.0108)	(0.0105)
health	0.0562***	0.0034	0.0049	0.0215**	−0.0091	−0.0084
	(0.0178)	(0.0178)	(0.0182)	(0.0098)	(0.0099)	(0.0099)
riskpre	0.1846***	0.0726**	0.1541***	0.1141***	0.0699***	0.0716***
	(0.0259)	(0.0324)	(0.0324)	(0.0134)	(0.0180)	(0.0163)
lnincome	0.0369***	0.0347***	0.0183***	0.0132***	0.0118***	0.0047*
	(0.0049)	(0.0061)	(0.0051)	(0.0025)	(0.0033)	(0.0027)
lnasset	0.1000***	0.0945***	0.0883***	0.0400***	0.0368***	0.0334***
	(0.0052)	(0.0067)	(0.0065)	(0.0026)	(0.0034)	(0.0034)
Constant				−0.4829***	−0.3366***	−0.2443***
				(0.0358)	(0.0470)	(0.0426)
Observations	7401	4287	4612	7401	4287	4612
Pseudo R-squared	0.135	0.127	0.138	0.302	0.570	0.610

Table 16 presents the model regression results of the risk financial asset allocation probability and allocation proportion of urban households with multiple houses in different regions for whether owning multiple houses. According to the regression results, the marginal effect of owning multiple houses on the probability of household risky financial asset allocation in the Eastern region and Western region is 0.0450 and 0.0504, respectively, which is significant at the significance level of 1 %. The marginal effect in the Central region is 0.0347, which is significant at the 90 % confidence level. The influence coefficients of housing on the proportion of risky financial assets allocation in each region are 0.0328, 0.0212, and 0.0391, respectively, and the influence results are all relatively significant, and the Central region has the least influence coefficient. Generally, the housing asset allocation effect is stronger in the Eastern and Western regions than in the Central region. This is because in the Eastern region of economically developed areas, multiple houses in addition to self-occupancy can bring higher rental income through renting, rising house prices will also bring appreciation of assets, coupled with the introduction of the Housing without Speculation policy, all of which will increase the willingness and depth of urban households to allocate risky financial products and tools. While in the Western region where the economy is relatively backward and the house prices are lower, these households will seek asset preservation and appreciation by investing in other risky financial assets in addition to purchasing houses.

Table 16.

Regression results of the model for the sample with owning multiple houses under different region groups.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	Eastern region	Central region	Western region	Eastern region	Central region	Western region
	Probit			Tobit
houses	0.0450***	0.0347*	0.0504***	0.0328***	0.0212**	0.0391***
	(0.0142)	(0.0184)	(0.0174)	(0.0079)	(0.0105)	(0.0099)
age	−0.0040***	−0.0040***	−0.0042***	−0.0005**	−0.0013***	−0.0015***
	(0.0004)	(0.0006)	(0.0005)	(0.0002)	(0.0003)	(0.0003)
gender	−0.0086	0.0050	−0.0076	−0.0228***	−0.0137	−0.0215**
	(0.0133)	(0.0157)	(0.0153)	(0.0074)	(0.0090)	(0.0086)
edu	0.0247***	0.0147***	0.0225***	0.0208***	0.0116***	0.0119***
	(0.0039)	(0.0047)	(0.0044)	(0.0022)	(0.0028)	(0.0026)
marriage	−0.0031	−0.0272	0.0083	−0.0010	−0.0185	0.0104
	(0.0184)	(0.0218)	(0.0212)	(0.0100)	(0.0116)	(0.0116)
health	0.0612***	0.0120	−0.0057	0.0165	−0.0032	−0.0143
	(0.0186)	(0.0193)	(0.0202)	(0.0101)	(0.0104)	(0.0105)
riskpre	0.1861***	0.0849***	0.1604***	0.1314***	0.0678***	0.0826***
	(0.0241)	(0.0323)	(0.0309)	(0.0127)	(0.0174)	(0.0157)
lnincome	0.0490***	0.0467***	0.0254***	0.0165***	0.0123***	0.0044
	(0.0053)	(0.0069)	(0.0057)	(0.0027)	(0.0035)	(0.0029)
lnasset	0.0987***	0.1142***	0.1143***	0.0419***	0.0436***	0.0460***
	(0.0058)	(0.0083)	(0.0079)	(0.0029)	(0.0041)	(0.0040)
Constant				−0.6570***	−0.5269***	−0.4733***
				(0.0421)	(0.0586)	(0.0546)
Observations	8154	4559	5003	8154	4559	5003
Pseudo R-squared	0.143	0.136	0.152	0.306	0.397	0.405

6. Conclusions and recommendations

This paper utilizes the 2019 CHFS data for urban households to construct a Probit model for the probability of risky financial financing allocation and a Tobit model for the proportion of risky financial asset allocation, and selects the average house price of the province where the urban household is located as an instrumental variable for IVProbit and IVTobit regression tests and risk preference as the mediator variable for mediation effect tests using the stepwise regression method. Through the above empirical research, the following main conclusions are drawn:

Urban households owning only one house exhibit a crowding-out effect on the risky financial asset allocation and a deeper impact on the allocation possibility. And the impact of owning only one house on the probability and proportion of risky financial asset allocation of urban household is mediated by the degree of risk preference. The impact of owning multiple houses on the risky financial asset allocation of urban households is characterized by the asset allocation effect. Among urban households owning houses, housing borrowing other than bank loans has a significant negative and dampening effect on the probability of risky financial asset allocation.

The effect of housing on risky financial asset allocation is income, age, and regionally heterogeneous. The crowding-out effect of owning one house is greater for middle-income urban households than for low-income and higher-income urban households, and the asset allocation effect of owning multiple houses is significant only among low-income and middle-income urban households. The crowding-out effect of owning one house gradually diminishes with the increase of age, and the asset allocation effect of owning multiple houses exists in urban households of young and middle-aged and above. The crowding-out effect of owning one house on the risky financial asset allocation is strongest in the economically developed Eastern region and weakest in the Western region with relatively backward in economic development. The asset allocation effect of owning multiple houses is stronger in the Eastern and Western regions than in the Central region.

In response to the findings of the study, it is possible to provide certain reference value for the government to introduce financial policies, and to guide financial institutions to offer differentiated financial products and investment services in response to the specific conditions of different households. Particularly, the following aspects should be addressed to improve the current situation. Firstly, guide households to rationally allocate financial assets to avoid asset homogenization. Secondly, regulate the real estate market and promote the development of the financial market. Thirdly, alleviate the crowding-out effect of household housing according to the characteristics of different income levels. Fourthly, meet the needs of households of different ages and introduce more targeted financial products and services. Lastly, coordinate regional development and narrow the differences in regional development levels.

CRediT authorship contribution statement

Lili Wu: Formal analysis, Data curation, Conceptualization, Funding acquisition, Investigation, Methodology, Software, Validation, Writing – original draft, Writing – review & editing. Hui Yu: Writing – review & editing, Resources, Project administration, Funding acquisition, Formal analysis.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.