The Impact of Property Tax Expectations on Household Asset Allocation

Abstract

Rational asset allocation is central to household wealth accumulation. This paper employs data derived from the 2019 China Household Finance Survey to methodically examine the influence of property tax expectations on the asset allocation decisions of households. This study finds that the expectation of property tax positively influences both the probability of holding risky financial assets and their proportion of household assets. The percentage of housing assets constitutes a substantial negative moderating factor affecting the relationship between property tax expectation and household investments in risky financial assets. The positive effect of property tax expectations is more significant in eastern China. A definite expectation of property tax causes a precautionary saving effect among households. This study highlights that forming reasonable expectations about property taxes can help households adjust their investment portfolios in advance, diversify their asset allocation, and mitigate the impact of changes in property tax policies.

1. Introduction

Effective asset allocation is crucial for household wealth accumulation. Classic portfolio theory suggests that households should participate in financial markets regardless of their risk preferences. Nonetheless, the phenomenon of limited participation remains pervasive among households, as many individuals exhibit a preference for retaining risk-free assets, such as bank deposits, while largely refraining from engaging in investments within risky financial markets. The scholarly literature has endeavored to examine this issue through multiple lenses, including the costs associated with participation, individual subjective attitudes, and underlying background risks. However, a comprehensive understanding of the limited participation phenomenon has yet to be fully explained [1,2,3]. In light of slowing consumer growth and gaps in retirement savings, it is increasingly important to guide households toward reasonable participation in risky financial assets to increase their income from investments. Moreover, due to housing reforms and urbanization, China’s real estate market has experienced prolonged prosperity, leading to soaring housing prices. The property tax discussed in this paper refers to a tax on privately owned residential properties, which was introduced in a pilot in Shanghai and Chongqing in 2011. It serves to foster healthy development in the real estate market and has been repeatedly mentioned in government reports. The reform practices of developed economies such as the United States and Europe based on property tax systems have shown that property taxes not only affect the equilibrium of the real estate market but also impact the economic decisions of micro entities. However, economic conditions have made the selection of cities for expanding the pilot uncertain, leading to diverse expectations among households regarding property taxes.

Given the two-fold nature of housing as both an investment asset and a consumption good, expectations of property taxes significantly influence household decisions on risky financial investment. Property taxes can alter housing prices, prompting households to adjust their expected returns on housing investments and consider substituting them with other assets. For households with substantial housing assets, property taxes increase uncertainty in disposable income, potentially crowding out investments in risky financial assets through the route of background risk. On this basis, this study utilizes data derived from the 2019 China Household Finance Survey to conduct a comprehensive analysis of how expectations regarding property taxes influence the asset allocation choices made by households. It offers theoretical and practical insights to encourage households to reasonably adjust their asset allocation, diversify their sources of income, and accumulate wealth.

2. Literature Review

Currently, there is a limited body of research that has directly investigated the influence of property tax expectations on the asset allocation decisions of households. The most relevant research focuses on two areas: the influence of housing on risky household financial investment decisions and the effect of property taxes on housing prices. This paper reviews the existing research in these areas.

Existing scholarly investigations into the influence of housing on household asset allocation choices have primarily concentrated on determining whether the effect of housing assets on investment decisions is primarily governed by the “crowding out effect” or the “wealth effect”. Household investment in risky financial assets incurs transaction costs, and the variation in wealth among households leads to differences in their capacity to absorb investment losses and manage investment expenses [1,4,5]. As a significant component of household wealth, housing assets in countries with rising housing price expectations generally reflect a “wealth effect”, where increased housing net worth raises household participation in risky financial markets [6,7,8]. However, some studies have proposed alternative viewpoints from the perspective of background risk, arguing that the risks faced by households in risk financial asset investments include not only financial asset investment risks but also risks associated with income stability, expenditure, and liquidity constraints [9,10,11]. Households with housing assets must manage the investment risks arising from housing price fluctuations while also addressing liquidity risks due to the substantial portion of wealth tied up in housing assets. To control their overall risk levels, households with housing assets may avoid taking on financial risk, which demonstrates the “crowding-out effect” [12,13]. Empirical research on the relation between housing assets and household participation in risk financial markets shows that, compared with the wealth effect, housing assets generally reveal a stronger “crowding-out effect” on household risk financial asset investment [14,15,16]. Additionally, some studies have attempted to differentiate between the impacts of housing liabilities and housing net worth, explaining the dual effects of housing assets on household asset allocation. Related studies have shown that commitment expenditures related to housing ownership lead agents to hold their financial assets in a secure form [17]. While housing liabilities suppress household investment in risky financial assets, an increase in housing net worth encourages greater investment in such assets [18,19,20].

A contemporary discourse exists concerning the influence of property taxes on housing prices, with principal points of disagreement centering around the perspectives of the “benefit tax”, the “capital tax”, and the “new” view. Some early studies consider property taxes as a capital tax, assuming a fixed supply of land, free capital flow, and constant return rates. Under these conditions, they find that capital would bear no tax burden, and property taxes would ultimately be passed on to consumers, resulting in rapid housing price increases [21,22]. Some scholars believe that property tax increases the holding cost for homebuyers, which in turn will lower housing prices [23,24]. Other scholars have analyzed the impact of property taxes from the angle of benefit tax and propose that local governments, driven by competition, are motivated to attract population inflows by providing public goods. Thus, property taxes would enhance local public spending, ultimately reflecting the value of housing assets [25,26]. Related studies have found that while an increase in property taxes directly suppresses housing prices, the resultant escalation in public expenditure exerts a favorable influence on housing prices, thereby rendering the overall impact of property taxes on housing prices effectively neutral [27]. In the “new” view of property taxes’ impact, regional differences in property taxes may lead to capital misallocation and generate the “profit tax effect” and “excise tax effect”. On the one hand, property taxes reduce overall capital returns and curb speculative demand. On the other hand, capital flows from high-tax regions to low-tax regions, changing the productivity of production factors and leading to variations in housing prices between the two areas [28,29]. Although there is no consensus on the theoretical discussion of the impact of property taxes on housing prices, subsequent empirical studies mostly find that property taxes have a suppressive effect on housing prices [30,31,32,33]. As research has deepened, some researchers have also examined the potential heterogeneous effects of property taxes, suggesting that the impact of property taxes on housing prices varies under different tax rate designs, housing characteristics, and market structures [34,35,36,37,38].

When reviewing present studies on the economic impact of property taxes, it has been found that most research has taken a macroeconomic perspective relying on theoretical models or experiences from pilot cities. However, for countries like China, where property taxes have not been widely implemented due to uncertainties in expanding the scope of pilot cities, the impact of property taxes on households in non-pilot cities remains largely reflected in expectations. Few studies have examined whether differences in expectations regarding property taxes influence household decisions on asset allocation. Therefore, this paper provides a systematic examination of the effect of property tax expectations on household asset allocation using household data to further elucidate the investment behavior of households.

3. Hypothesis Development

Similar to the dual effects of housing assets on risky financial assets, including the “wealth effect” and “crowding-out effect”, the expectation of property taxes also impacts household asset allocation through three distinct pathways: changes in the expected returns on housing assets; uncertainty in the value of existing housing assets; and changes in disposable income due to taxation.

From the perspective of changes in expected returns on housing assets. As stated in the Theory of the “capital tax” view and the “new” view, the imposition of property taxes significantly alters the holding costs of housing assets and leads to a marked decrease in their investment returns [21,28,39,40]. When making investment decisions, households consider the expected returns and relative relationships of housing assets, risk financial assets, and risk-free financial assets [41,42,43,44]. Therefore, regardless of whether households own housing assets, differences in expectations about whether property taxes will be levied can lead them to have varying expectations about the returns on housing investments. This results in differences in households’ willingness to speculate on housing and creates varying asset allocation decisions among households through the “asset substitution effect”. Therefore, this paper proposes the following hypothesis:

H1. 

The expectation that property taxes will be levied has a positive effect on households’ investment in risky financial assets.

Nevertheless, owing to disparities in the valuation of housing assets possessed by households and the design of the property tax system, the influence of property tax expectations on asset allocation is likely to vary among different households. From the perspective of alteration in the value of existing housing assets, the imposition of property taxes alters the value of housing assets, resulting in greater wealth uncertainty for households holding more housing assets. When making investment decisions in risky financial assets, households not only consider financial risk but also assess their overall risk exposure. Higher background risk will crowd out household investments in risky financial assets [45,46,47]. Thus, when households with significant housing wealth anticipate the imposition of property taxes in the future, the greater uncertainty in their wealth levels will force them to avoid risky financial assets to maintain a manageable level of overall risk. From the perspective of disposable income uncertainty, the imposition of property taxes will also lead to a significant decrease in disposable income for households with substantial housing assets. In contrast to risk-free financial assets like bank deposits, risky financial assets are characterized by elevated transaction costs. These costs encompass not only the time associated with gathering information and making informed decisions but also the fees related to the execution of investment transactions [1,5,48,49]. Significant differences in income levels lead to varying investment decisions in risky financial assets among households. Therefore, for households with more housing assets, the expectation that a property tax will be levied may lead to a decrease in their expected disposable income, subsequently reducing their investment in risky financial assets. Thus, this paper proposes the following hypothesis:

H2. 

The proportion of housing assets will negatively moderate the impact of property tax expectations on households’ investment in risky financial assets.

4. Method

4.1. Data and Variables

The data utilized in this study are derived from the China Household Finance Survey 2019 (CHFS 2019), which was carried out by the Survey and Research Center for China Household Finance at the Southwestern University of Finance and Economics. This survey collected financial information at both the household and individual levels across 29 provinces (including autonomous regions and municipalities) in China, providing a solid foundation for research on household asset allocation [50]. As this paper focuses on the impact of property tax expectations, samples from cities that have already been designated as pilot areas (Chongqing and Shanghai) are excluded from the baseline tests. After excluding samples with missing key variables, the final dataset consists of 15,310 valid household samples.

Based on the CHFS 2019 questionnaire and the literature’s research experience [51], this study selects “risky financial asset holdings (Risk)” and the “proportion of risky financial assets (Risk_p)” as the core dependent variables to examine the effect of property tax expectations on the probability and proportion of household investments in risky financial assets. The risky financial assets considered in this paper primarily include stocks, funds, financial products, financial derivatives, non-RMB assets, and gold. For robustness checks, this paper examines a scenario in which stocks are the only risky financial asset.

The essential explanatory variable in this paper is “property tax expectations”, which is assigned based on household responses to the question, “When do you expect property taxes to be implemented in the future?”. Households answering “No” are assigned a value of 0, whereas those responding “Within 1 year”, “Within 2 to 3 years”, “Within 3 to 5 years”, or “More than 5 years” are assigned a value of 1. Robustness tests also examine how different settings of the explanatory variable affect the research conclusions.

In selecting control variables, this paper includes variables considered influential in household risk financial asset investment decisions based on the literature. This encompasses a collection of variables pertaining to the characteristics of members (e.g., gender, health status, age, age squared, years of education, financial literacy, risk attitude), a set of household characteristic variables (e.g., household size, income, net assets, number of houses held), and provincial characteristic variables. The specific variable settings are presented in

Table 1. Descriptive statistics.

Variables	Variable Description	N	Mean	SD	Min	Max
Risk	If the household holds any risky financial assets (stocks, financial management products, funds, non-government bonds, financial derivatives, foreign currency assets, and gold), assign a value of 1; otherwise, assign a value of 0	15,310	0.142	0.349	0	1
Risk_p	The proportion of risky financial assets in total financial assets	15,310	0.0653	0.247	0	1
Stock	If the household holds stocks, assign a value of 1; otherwise, assign a value of 0	15,310	0.0799	0.227	0	1
Stock_p	The proportion of stocks in total financial assets	15,310	0.0276	0.131	0	1
Tax	Dummy variable. Assign values based on responses to the question, “When do you expect property taxes to be implemented in the future?”. Households answering “No” are assigned a value of 0, whereas those responding “Within 1 year”, “Within 2 to 3 years”, “Within 3 to 5 years”, or “More than 5 years” are assigned a value of 1	15,310	0.713	0.453	0	1
Gender	Dummy variable. If the householder is male, assign a value of 1; If the householder is female, assign a value of 0	15,310	0.710	0.454	0	1
Age	Householder’s age in 2019	15,310	53.41	14.35	18	97
Age2	The square term of age	15,310	3059	1571	324	9409
Edu	Householder’s years of education	15,310	10.71	3.801	1	22
Health	The self-assessed health status of the householder. Assign values based on responses to the question, “How would you rate your current health condition compared to your peers?” The responses “Very Good”, “Good”, “General”, “Poor”, and “Very Poor” will be assigned values of “1”, “2”, “3,” “4”, and “5”, respectively.”	15,310	2.569	0.953	1	5
Marriage	Dummy variable. If the householder is married, assign a value of 1; otherwise, assign a value of 0	15,310	0.847	0.360	0	1
Size	Number of household members	15,310	3.037	1.410	1	15
Att	Risk aversion: assign values based on responses to the question, “If you had a sum of money to invest, which investment project would you prefer?” The responses “High risk, high return”, “Slightly higher risk, slightly higher return”, “Average risk, average return”, “Slightly lower risk, slightly lower return”, and “Unwilling to take any risk” will be assigned values of “1”, “2”, “3,” “4”, and “5”, respectively	15,310	4.224	1.079	1	5
Fin	Financial literacy: assign values based on responses to the question, “How much attention do you typically pay to economic and financial information?”. The responses “Never pay attention”, “Rarely pay attention”, “Sometimes pay attention”, “Often pay attention”, and “Always pay attention” will be assigned values of “1”, “2”, “3”, “4”, and “5”, respectively	15,310	1.965	1.020	1	5
House	The number of homes owned by a household	15,310	1.116	0.672	0	10
Income	Total annual household income/10,000	15,310	10.27	12.22	0	87.86
Wealth	(Household assets − Household liabilities)/10,000	15,310	134.2	191.5	−13.05	1140
Tax_c	Average “Tax” variable’s value within the municipality in which the household resides	15,310	0.289	0.106	0.222	0.750
House_p	The proportion of housing assets in total assets	15,310	0.611	0.324	0	1
Savings	(Household income−Household consumption)/Household income. The value is set to 0 when it is less than 0.	15,310	0.203	0.254	0	0.988

Note: The term “householder” in the CHFS 2019 means the main decision-maker in the household rather than the official householder registered in the household registration system.

.

In Table 1, a very small number of households have negative net assets. This occurs when their total liabilities exceed their total assets, which can be attributed to factors such as educational debt among young households, medical expenses, or a decline in housing asset values for households with mortgage obligations. In the robustness tests, this paper presents the results from subsample tests that exclude households with zero net assets, negative net assets, and zero income.

Table 2. Correlation matrix of key variables and VIF test.

Variables	Tax	Risk	Age	Age2	Edu	Gender	Marriage	VIF
Tax	1							1.08
Risk	0.122 ***	1
Age	−0.112 ***	−0.032 ***	1					46.62
Age2	−0.108 ***	−0.035 ***	0.988 ***	1				45.60
Edu	0.197 ***	0.272 ***	−0.380 ***	−0.363 ***	1			1.55
Gender	0.006	−0.038 ***	−0.042 ***	−0.050 ***	0.027 ***	1		2.23
Marriage	−0.003	0.027 ***	−0.056 ***	−0.096 ***	0.047 ***	0.303 ***	1	1.32
Health	−0.096 ***	−0.076 ***	0.267 ***	0.255 ***	−0.247 ***	−0.051 ***	−0.058 ***	1.14
Size	−0.019 **	−0.031 ***	−0.216 ***	−0.238 ***	−0.051 ***	0.182 ***	0.345 ***	1.36
Att	−0.188 ***	−0.245 ***	0.303 ***	0.289 ***	−0.259 ***	−0.033 ***	0.009	1.30
Fin	0.175 ***	0.320 ***	−0.112 ***	−0.111 ***	0.305 ***	0.027 ***	0.027 ***	1.26
House	0.040 ***	0.163 ***	−0.021 ***	−0.042 ***	0.119 ***	0.045 ***	0.141 ***	1.40
Income	0.130 ***	0.273 ***	−0.165 ***	−0.160 ***	0.337 ***	0.026 ***	0.116 ***	1.53
Wealth	0.112 ***	0.363 ***	−0.020 **	−0.025 ***	0.305 ***	−0.025 ***	0.069 ***	2.04
Variables	Health	Size	Att	Fin	House	Income	Wealth
Health	1
Size	−0.049 ***	1
Att	0.136 ***	−0.081 ***	1
Fin	−0.122 ***	0.012	−0.352 ***	1
House	−0.090 ***	0.187 ***	−0.092 ***	0.114 ***	1
Income	−0.169 ***	0.166 ***	−0.219 ***	0.219 ***	0.301 ***	1
Wealth	−0.128 ***	0.048 ***	−0.172 ***	0.228 ***	0.439 ***	0.504 ***	1

Note: ** p < 0.05, *** p < 0.01.

illustrates the correlation matrix pertaining to the principal variables utilized in this study and the results of the variance inflation factor (VIF) test. The findings reveal a noteworthy positive correlation between property tax expectations and household investment in risky financial assets, thereby providing partial support for Hypothesis 1. Additionally, with the exceptions of age and age squared, the correlations among the main variables are all less than 0.8, and the VIF values are less than 10, suggesting that there are no severe multicollinearity issues among these variables. As for the variables “Age” and “Age²”, given the large sample size and the statistical significance of their regression coefficients in empirical research, it can be assumed that the results are not significantly impacted by multicollinearity. Therefore, considering that numerous studies on household asset allocation have examined the nonlinear relationship between age and risky financial assets investment, this paper chooses to include both “Age” and “Age²” as control variables and reports the regression results excluding the Age² variable in the robustness tests.

4.2. Model

In the benchmark regression, since “Risk” is a binary variable, considering the estimation issues caused by the characteristics of the variable, this paper applies the logit model to explore the effects of property tax expectations on households’ holdings of risky financial assets. The specific model settings are as follows:

$$Pr\left({Risk}_i=1\right)={\displaystyle \frac{exp\left({\beta }_0+{\beta }_1{Tax}_i+{\beta }_2{Controls}_i+{\beta }_3{Prov}_i\right)}{1+exp\left({\beta }_0+{\beta }_1{Tax}_i+{\beta }_2{Controls}_i+{\beta }_3{Prov}_i\right)}}$$

(1)

Among them, i represents the i-th household. “Tax”, which is property tax expectations, represents the core explanatory variable. “Risk = 1” suggests that the household holds risky financial assets. “Controls” represents the set of control variables, which includes “Gender”, “Age”, “Age²”, “Edu”, “Health”, “Marriage”, “Size”, “Att”, “Fin”, “House”, “Income”, and “Wealth”. “Prov” represents the set of dummy variables representing provincial characteristics.

In discussions regarding the proportion of risky financial assets, since the “Risk_p” variable is censored, this paper uses the Tobit model, which is more suitable for this type of data, to inspect the impact of property tax expectations on the proportion of risky financial assets. The specific model settings are as follows:

$$\begin{gathered} {{risk\_p}_i}^{*}={\beta }_0+{\beta }_1{Tax}_i+{\beta }_2{Controls}_i+{\beta }_3{Prov}_i+{\epsilon }_i, \\ {Risk\_p}_i=\mathit{max}(0,{{risk\_p}_i}^{*}) \end{gathered}$$

(2)

Among them, “risk_p*” stands for the true value of the proportion of risky financial assets, and “Risk_p” represents the observed value of the proportion of risky financial assets. ε represents the random error term. Other variables are set the same as those in the logit model.

In the robustness test involving changes to the explained variables, this paper replaces the “Risk” variable in Model (1) and the “Risk_p” variable in Model (2) with the “Stock” variable and “Stock_p” variable, and the revised Models (1) and (2) are then re-run for analysis. In the robustness test about changes in explanatory variables, this paper changes the setting method of the “Tax” variable in Models (1) and (2). Other robustness tests do not change the variable settings in the Model (1) and Model (2).

5. Results

5.1. Benchmark Regression Results

Table 3. The impact of property tax expectations on risky financial assets investment.

	(1)	(2)	(3)	(4)
Variables	Risk		Risk_p
Tax	0.112 ***	0.030 ***	0.103 ***	0.033 ***
	(14.74)	(4.54)	(13.93)	(4.94)
Age		0.008 ***		0.007 ***
		(6.19)		(4.90)
Age2		−0.000 ***		−0.000 ***
		(−4.79)		(−3.37)
Edu		0.014 ***		0.014 ***
		(16.53)		(14.41)
Gender		−0.034 ***		−0.032 ***
		(−5.95)		(−5.33)
Marriage		0.020 **		0.014
		(2.45)		(1.54)
Health		−0.000		−0.002
		(−0.10)		(−0.50)
Size		−0.011 ***		−0.013 ***
		(−5.07)		(−5.81)
Att		−0.032 ***		−0.034 ***
		(−14.60)		(−12.54)
Fin		0.051 ***		0.055 ***
		(22.02)		(18.19)
House		0.008 *		−0.008 *
		(1.91)		(−1.93)
Income		0.001 ***		0.001 ***
		(4.73)		(5.60)
Wealth		0.000 ***		0.000 ***
		(12.85)		(9.30)
Prov		Yes		Yes
N	15,310	15,310	15,310	15,310
Pseudo R2	0.020	0.259	0.034	0.379

Note: The logit model reports average marginal effects with the z-value in parentheses. The Tobit model reports regression coefficients with the t-value in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

presents the regression outcomes for the relationship between property tax expectations and household investment in risky financial assets. In Columns 1 and 2, the dependent variable is “Risk”, and a logit model is employed to explore the impact of property tax expectations on the likelihood of holding risky financial assets. In Columns 3 and 4, the dependent variable is “Risk_p,” and a Tobit model is used to explore how property tax expectations affect the investment intensity of risky financial assets.

The regression analysis presented in Table 3 demonstrates that the average marginal effect and regression coefficient for the primary explanatory variable, “Tax”, is statistically significant, exceeding 0 at the 1% significance level. This finding suggests that households anticipating the imposition of property taxes are more inclined to partake in risky financial asset investments and maintain a larger proportion of such assets within their portfolios. For example, according to the regression results in Column 2, controlling for other factors, households expecting property taxes to be levied are 3.0% more likely to hold risky financial assets compared to other households. This impact accounts for 21% of the average probability of holding risky financial assets in the sample (14.2%), which has definite economic implications and provides support for Hypothesis 1.

With respect to the other control variables, the regression results show an inverted U-shaped relation between the householder’s age and investment in risky financial assets. A boost in a household’s number of years of education stimulates investment in risky financial assets, whereas a larger household size suppresses participation in such investments. Risk-averse households are inclined to avoid risky financial assets, whereas increased financial literacy inspires households to hold more risky financial assets. Higher income and wealth levels positively affect the holding of risky financial assets. These results are in correspondence with the literature. However, given potential endogeneity issues, this paper does not delve deeply into the explanations of the control variables’ results.

A few control variables yielded unexpected results. Notably, the average marginal effect of the gender variable was negative, contradicting the traditional view that males prefer risks more than females. However, this result aligns with the empirical findings in much of the literature. This inconsistency could stem from the fact that, after controlling for several economic variables, the gender characteristics may have been partially accounted for [7,45,51]. Furthermore, the health variable exhibited significant differences in both significance levels and regression coefficients across different models. These variations may result from the impact of health on household risky financial markets investment being explained by income and wealth factors.

5.2. Robustness Test

To analyze the possible impact of overly broad definitions of risky financial assets on the research conclusions, this paper focuses specifically on the case where risky financial assets are defined solely as stocks and where the core explanatory variables are changed to “Stock” and “Stock_p”. Specifically, this paper replaces the “Risk” variable in Model (1) and the “Risk_p” variable in Model (2) with the “Stock” variable and “Stock_p” variable, respectively. The revised models (1) and (2) are then re-run for further analysis. The corresponding results are shown in

Table 4. Robustness test (stock).

	(1)	(2)	(3)	(4)
Variables	Stock		Stock_p
Tax	0.059 ***	0.014 ***	0.074 ***	0.020 ***
	(10.06)	(2.76)	(9.36)	(2.86)
Controls		Yes		Yes
Prov		Yes		Yes
N	15,310	15,310	15,310	15,310
Pseudo R2	0.017	0.267	0.030	0.431

Note: The logit model reports average marginal effects with the z-value in (). The Tobit model reports regression coefficients with the t-value in (). *** p < 0.01.

. Wherein “Controls” represents the set of control variables, “Prov” represents the set of provincial characteristic variables, and “Yes” means the variable set is controlled. The terms “Controls”, “Prov”, and “Yes” in the following text also have the same meaning. The (1) and (3) columns report the regression results without the inclusion of control variables and provincial characteristic variables, while the (2) and (4) columns present the regression results after adding all the control variables and provincial characteristic variables used in Table 3. In Table 4, the results of regression from each column indicate that the average marginal effect (regression coefficient) and significance level of the “Tax” variable have not changed significantly compared with the results in the corresponding columns of Table 3. Thus, Hypothesis 1 remains robust.

In Table 3, the core explanatory variable was initially defined as “Assign a value of 0 to households that respond ‘No’ and assign a value of 1 to households that respond ‘Within 1 year’, ‘Within 2 to 3 years’, ‘Within 3 to 5 years’ or ‘More than 5 years’.” Given that households that responded with “More than 5 years” may not differ significantly from those that responded “No” in terms of their actual property tax expectations, this paper adjusts the settings of explanatory variables to perform a robustness test. Specifically, the new setting of the “Tax” variable is “Assign a value of 0 to households responding ‘No’ or ‘More than 5 years’ and a value of 1 to households responding ‘Within 1 year’, ‘Within 2 to 3 years’ or ‘Within 3 to 5 years’.” After that, this paper re-runs Model (1) and Model (2) to conduct the robustness test. The results of this robustness test are presented in

Table 5. Robustness test (changes in explanatory variable).

	(1)	(2)	(3)	(4)
Variables	Risk		Risk_p
Tax	0.050 ***	0.013 **	0.048 ***	0.016 ***
	(8.67)	(2.52)	(8.52)	(2.96)
Controls		Yes		Yes
Prov		Yes		Yes
N	15,310	15,310	15,310	15,310
Pseudo R2	0.006	0.257	0.010	0.377

Note: The logit model reports average marginal effects with the z-value in parentheses. The Tobit model reports regression coefficients with the t-value in parentheses. ** p < 0.05, *** p < 0.01.

. Among them, “Controls” and “Prov” carry the same meanings as those in Table 4. As indicated by the regression results in each model column, although there is a slight decrease in the value and significance of the coefficients after the adjustment of explanatory variables, the “Tax” variable still has a notable positive effect on the probability and proportion of households holding risky financial assets. The reduction in coefficients may be because households that respond “More than 5 years” still hold the expectation that property taxes will be levied.

Since Chongqing and Shanghai became pilot cities in 2011, samples from these cities were excluded from the benchmark regression. To test the potential impact of this sample exclusion on the research conclusions, this paper conducts a robustness check by using the full sample of households before exclusion. The corresponding results are presented in columns 1 and 2 of

Table 6. Robustness test (sample variation).

	(1)	(2)	(3)	(4)
Variables	Risk	Risk_p	Risk	Risk_p
Tax	0.027 ***	0.031 ***	0.032 ***	0.036 ***
	(4.39)	(5.04)	(4.67)	(5.34)
Controls	Yes	Yes	Yes	Yes
Prov	Yes	Yes	Yes	Yes
N	16,810	16,810	14,770	14,770
Pseudo R2	0.266	0.393	0.256	0.381

Note: The logit model reports average marginal effects with the z-value in parentheses. The Tobit model reports regression coefficients with the t-value in parentheses. *** p < 0.01.

. Wherein except for the difference in sample size, the model and variable settings in Columns 1 and 2 are consistent with those in Columns 2 and 4 of Table 3. The value and significance of the average marginal effect (regression coefficient) of the core explanatory variable “Tax” do not differ significantly from the results in Table 3. Therefore, excluding the samples from Chongqing and Shanghai does not affect the robustness of the research conclusions in this paper. The results reported in Columns 3 and 4 of Table 6 present the subsample analysis after excluding households with non-positive net assets and zero income. Apart from changes in the sample size, the model specifications and variables in Column 3 and Column 4 are consistent with those in Column 2 and Column 4 of Table 3. The findings indicate that the regression results for the “Tax” variable do not significantly differ from the corresponding columns in Table 3, confirming the robustness of the conclusions.

To analyze possible endogeneity issues in benchmark regression, this paper employs an instrumental variable. The relative results are shown in

Table 7. Regression results of endogenous treatment.

	(1)	(2)	(3)	(4)
Variables	Risk	Stock	Risk_p	Stock_p
Tax	1.492 ***	1.696 ***	0.290 ***	0.261 ***
	(5.75)	(4.82)	(5.97)	(4.99)
Controls	Yes	Yes	Yes	Yes
Prov	Yes	Yes	Yes	Yes
Wald	0.000	0.000	0.000	0.000
Ar	0.000	0.000	0.000	0.000
N	15,310	15,310	15,310	15,310

Note: Reports regression coefficients with the z-value in parentheses. *** p < 0.01.

, and Columns 1 and 2 use a two-step IV-Probit model, whereas Columns 3 and 4 use a two-step IV-Tobit model. Each model includes all the control variables used in column 2 of Table 3. The instrumental variable utilized in this paper is the “Tax_c” variable. The rationale for the instrumental variable setting is twofold. First, the average property tax expectations in the city reflect other households’ expectations about the likelihood that property taxes will be levied, which significantly influences the household’s own expectations through social interaction. Second, the average property tax expectations in the city are not directly related to the household’s risky financial asset investment. From the results in Table 7, after the application of the instrumental variable, the regression coefficient associated with the “Tax” variable continues to exhibit a statistically significant value exceeding 0 at the 1% significance level, thereby providing empirical support for Hypothesis 1. Additionally, the weak instrument test rejects the possibility of weak instruments.

The potential issue of multicollinearity between the “Age” variable and “Age²” variable and the potential impact of the “House” variable on the conclusions is discussed. This paper separately excludes the “Age²” variable and the “House” variable from the control variables to conduct a robustness test. The first and second columns in

Table 8. Robustness test (remove variables).

	(1)	(2)	(3)	(4)
Variables	Risk	Risk_p	Risk	Risk_p
Tax	0.031 ***	0.034 ***	0.030 ***	0.033 ***
	(4.61)	(4.97)	(4.55)	(4.93)
Controls (Remove Age2)	Yes	Yes
Controls (Remove House)			Yes	Yes
Prov	Yes	Yes	Yes	Yes
N	15,310	15,310	15,310	15,310
Pseudo R2	0.257	0.377	0.258	0.379

Note: The logit model reports average marginal effects with the z-value in parentheses. The Tobit model reports regression coefficients with the t-value in parentheses. *** p < 0.01.

present the regression results after removing the “Age²” variable, while the other models and variable settings remain consistent with those in the second and fourth columns of Table 3. The third and fourth columns report the regression results after excluding the “House” variable, with the other models and variable settings also consistent with the second and fourth columns of Table 3. The results show that the regression results in each column are not significantly different from the corresponding results in Table 3, confirming that the conclusion that the “Tax” variable suppresses household holdings of risky financial assets remains robust.

5.3. Further Discussion

5.3.1. Moderating Effect of Housing Asset Proportion

This paper proposes that the imposition of property taxes will lead to changes in the wealth and disposable income levels of households with a greater proportion of housing assets. This effect dampens the positive impact of property tax expectations on households’ investments in risky financial assets. To verify Hypothesis 2, this paper examines the moderating effect of the “House_p” variable, with the results presented in

Table 9. Moderating effect of housing asset proportion.

	(1)	(2)	(3)	(4)
Variables	Risk		Risk_p
Tax × House_p	−0.013	−0.017	−0.046 **	−0.057 ***
	(−0.57)	(−0.94)	(−2.13)	(−2.87)
Tax	0.109 ***	0.037 ***	0.121 ***	0.061 ***
	(6.96)	(2.89)	(7.32)	(4.10)
House_p	−0.082 ***	−0.086 ***	−0.107 ***	−0.122 ***
	(−4.05)	(−5.05)	(−5.53)	(−6.60)
Controls		Yes		Yes
Prov		Yes		Yes
N	15,310	15,310	15,310	15,310
Pseudo R2	0.020	0.259	0.034	0.379

Note: The logit model reports average marginal effects with the z-value in parentheses. The Tobit model reports regression coefficients with the t-value in parentheses. ** p < 0.05, *** p < 0.01.

. Specifically, this paper incorporates the “House_p” variable and the interaction term between the “House_p” and “Tax” variables into Models (1) and (2). The models are then re-estimated to conduct analysis. The control variables in each column are set the same as the corresponding columns in Table 3. The regression results show that the interaction term is significantly negative only in Models 3 and 4. The findings indicate that the percentage of housing assets significantly negatively moderates the influence of property tax expectations on the proportion of risky financial assets possessed by households. Compared with households with a lower proportion of housing assets, those with a higher proportion of housing assets experience a lesser impact from property tax expectations on their allocation to risky financial assets, thus partially supporting Hypothesis 2.

5.3.2. Heterogeneity

Due to differences in economic trajectories, the eastern regions and first-tier cities of China have experienced a prolonged period of rapid growth in housing asset prices. To investigate whether the impact of property tax expectations may exhibit regional heterogeneity, this study groups the samples based on the following criteria: eastern regions/non-eastern regions; first-tier and new first-tier cities/second-tier and lower-tier cities. Then, the model from Table 3 is re-run to explore regional differences. The regression results are shown in

Table 10. Discussion on regional heterogeneity.

	(1)	(2)	(3)	(4)
Variables	Risk		Risk
	Eastern Region	Non-Eastern Region	First-tier/New First-tier Cities	Second-tier cities and below
Tax	0.047 ***	0.017 **	0.058 ***	0.017 **
	(3.96)	(2.18)	(3.90)	(2.48)
Controls	Yes	Yes	Yes	Yes
Prov	Yes	Yes	Yes	Yes
N	6243	9067	4847	10,463
Pseudo R2	0.439	0.217	0.257	0.224

Note: The logit model reports average marginal effects with the z-value in parentheses. ** p < 0.05, *** p < 0.01.

. The results indicate that the value and significance of the “Tax” variable are consistently greater in eastern regions (first-tier/new first-tier cities) than in non-eastern regions (second-tier cities and below). This disparity may stem from the “asset substitution effect” caused by property taxes being more pronounced in eastern regions or first-tier cities, which have sustained high housing asset investment returns over the long term.

5.3.3. Property Tax Expectations and Savings Rate

In the previous discussion, this paper investigated the adverse moderating influence of housing asset proportions on the relationship between households’ expectations regarding property taxes and their investments in risky financial assets. Unlike risky financial asset investments, property tax expectations affect households’ precautionary savings primarily through uncertainty. Specifically, property tax expectations impact households’ uncertainty about their income and wealth levels, leading to increased precautionary savings. This paper further discusses the impact of property tax expectations on households’ precautionary savings. The results are presented in

Table 11. Discussion on savings rate.

	(1)	(2)
Variables	Savings	Savings
Tax	0.092 ***	0.033 ***
	(10.65)	(4.94)
Controls		Yes
Prov		Yes
N	15,310	15,310
Pseudo R2	0.007	0.379

Note: Reports regression coefficients with the t-value in parentheses. *** p < 0.01.

, except that the dependent variable is replaced with the “Savings”; the other model settings are the same as those in Columns 3 and 4 of Table 3. According to the results, the coefficient for the “Tax” variable is notably greater than 0 at the 1% level. Considering the results in Column 2, compared with households that believe that property taxes will not be levied, those that believe that property taxes will indeed be levied have a 3.3% higher savings rate. This finding indicates that households increase their precautionary savings in response to uncertainties in wealth and disposable income due to the expected collection of property taxes.

6. Conclusions

This paper utilizes data derived from the China Household Finance Survey conducted in 2019 to investigate the influence of anticipated property taxes on households’ decisions regarding investments in risky financial assets. The results reveal a marked positive correlation between the expectation of property tax implementation and both the likelihood and the proportion of households holding such risky financial assets. This conclusion remains robust after changes in variable settings, sample variation, and endogeneity tests. The proportion of housing assets significantly negatively moderates the impact of property tax expectations on the proportion of risky financial assets. Regional heterogeneity analysis reveals that the influence of property tax expectations on household financial asset investment risk is more pronounced in eastern China. Further discussion of household savings decisions reveals that households believe that property taxes may be levied to engage in precautionary savings.

These conclusions offer several policy implications. First, related departments should reasonably coordinate the pace of property tax pilots based on the economic context. This paper finds that property tax expectations significantly influence household risk financial investment decisions. When households believe that the property tax will be levied, they will adjust their asset allocation in advance. Therefore, helping households form reasonable expectations about property taxes can reduce the impact of the actual implementation of property tax policies on households. This is important for household wealth accumulation and the smooth implementation of property tax reforms. However, the expectation that property taxes will be levied may lead households to increase precautionary savings to mitigate income and wealth uncertainties, which could hinder consumer demand and economic development. Hence, it is crucial to carefully manage the timing of the property tax pilot process, ensuring that the policy has cross-cycle and countercyclical effects and avoiding tightening expectations during an economic recovery.

Second, pilot cities and tax policies should be carefully selected and designed. The design of property tax policies in pilot cities is a key factor influencing non-pilot regions’ expectations of the incidence of taxation. For households with significant housing assets, the expectation that property taxes will be levied can significantly impact their expected wealth and disposable income levels. This may prevent them from investing in risky financial assets and lead to reduced household consumption. Therefore, the expansion of pilot cities and policy design should consider the demonstration effect. Based on the experiences of previous pilot cities such as Chongqing and Shanghai, policies need to consider urban economic conditions and household asset structures to effectively regulate the long-term effects of property taxes while mitigating potential negative economic impacts.

Third, there is a balance between taxes and government spending. Both the actual start of property tax collection and the expected collection of property taxes can lead to household concerns about wealth levels and disposable income uncertainty, prompting them to reduce consumption and engage in precautionary savings. To alleviate the adverse economic effects of precautionary savings, the government should balance property tax collection with public spending. While increasing tax revenue, the government should also enhance public services and social security levels to reduce household concerns about future uncertainties and stimulate consumption.

While this paper’s research background and data are specific to China, the conclusions drawn can also offer valuable insights for other countries. In any country considering the implementation or adjustment of property tax, households may alter their investment behavior in response to tax expectations, particularly in countries where real estate constitutes a significant portion of household assets. For governments seeking to develop or modify property tax policies, this paper provides a useful reference for managing public expectations to avoid sharp fluctuations in household investment and consumption behaviors due to anticipated changes in property tax policies.

Considering that the data used in this paper are derived from China and that the research primarily focuses on examining the impact of property tax expectations on household investments in risky financial assets, this paper may have certain limitations. Future research could delve deeper into the heterogeneous effects of property tax expectations on household asset allocation across different countries with varying economic structures, explore the effects of property tax expectations on households’ allocation of other assets, and investigate the differences in impacts arising from various categories of tax expectations.