Has the Water Rights Trading Policy Improved Water Resource Utilization Efficiency?

Abstract

Implementing natural resource protection systems and improving regional water resource utilization efficiency are effective ways to resolve the contradiction between economic development and water resource poverty. To this end, this paper establishes a Difference-in-Difference (DID) model to analyze the impact of water rights trading pilot policies (WET) in 271 prefecture-level cities in China from 2006 to 2023 on water resource utilization efficiency (WEE). The research results indicate that (1) WET significantly improved WEE, while confirming the robustness of this effect; (2) WET exhibit significant heterogeneity in their policy effects on WEE, reflecting pronounced differences between northern and southern cities in terms of geographical location and water resource endowment. In cities with abundant water resources, this promotional effect is even more pronounced; (3) market vitality and water conservation benefits can positively promote the impact of WET through regulatory mechanisms. Based on this, expanding the pilot cities for water rights trading policies and enhancing market vitality can effectively improve WEE and alleviate the current situation of water resource poverty in the region.

1. Introduction

Water resources are fundamental natural resources essential for human survival and development, as well as strategic economic resources. Human life cannot exist without water, and neither can sustainable social development [1,2]. Many countries on Earth are still experiencing water shortages, while some regions are suffering from flooding. Improper allocation of water resources can severely undermine the sustainable development of human life [3]. How to overcome water poverty has always been a key public issue of global concern. Water shortages directly lead to a decline in food production, posing a serious threat to human life and safety. Therefore, in recent years, how to conserve water resources and rationally plan their use has been identified as an important part of achieving sustainable development goals [4].

In recent years, as the global population and human productive activities have continued to expand, the demand for water resources—a vital element in production—has grown increasingly significant. China, being one of the world’s most populous nations, also faces the challenge of water scarcity. In 2023, the national per capita comprehensive water consumption was 419 m³, and the water consumption per 10,000 yuan of GDP (at current prices) was 46.9 m³. Furthermore, the uneven distribution of water resources across time and space has long been a key issue that the Chinese government has sought to address [5]. Coastal cities have abundant water resources, while western regions such as Gansu suffer from water shortages and droughts throughout the year [6]. Balancing water resource allocation across regions is crucial to the long-term stability of the Chinese government and the well-being of its people [7]. However, previous policies have focused on how to redistribute water resources, with little research on measuring water resource utilization efficiency (WEE). Only by achieving rational water resource utilization can water resource policies be established to effectively improve utilization efficiency.

In recent years, many scholars have directly attributed WEE to agricultural irrigation efficiency, lacking research on urban water supply utilization efficiency. Shah et al. (2023) [8] pointed out that the “Agricultural Water Conversion Policy” introduced in 2012 can effectively utilize agricultural water resources efficiently and improve agricultural production efficiency [8,9]. However, there are still some issues that need to be addressed: First, whether the implementation of water resource management policies can influence WEE, and if so, what are the mechanisms and conditions that drive this influence?

To answer these questions, this study takes China’s 2014 water rights trading pilot policies as a starting point and uses water rights trading policies in 271 prefecture-level cities in China from 2006 to 2023 to examine the impact of these policies on WEE and the mechanisms behind this impact.

This paper makes three main contributions in terms of innovation: (1) Innovation in research perspective. By examining water resource utilization efficiency at the urban level, this study overcomes the limitations of previous analyses that focused solely on agricultural water use; (2) Innovation in Measurement Methods. While most studies employ data envelopment analysis (DEA) to measure efficiency indicators, this paper utilizes a stochastic frontier model based on a translog production function to assess water resource utilization efficiency; and (3) innovation in policy theory. Traditional water resource management has primarily focused on direct administrative regulation, whereas water rights trading policies are grounded in property rights theory. Water rights trading policies clarify the ownership of water resources, granting corresponding rights and responsibilities to water users, thereby enabling the rational allocation of water resources through market mechanisms. These policies enrich the theoretical framework of water resource management and provide valuable theoretical insights for the formulation and optimization of similar resource management policies.

2. Literature Review and Research Hypotheses

2.1. Institutional Background

Water resources are a public good that is non-competitive and non-exclusive, and everyone has equal rights to enjoy water resources. However, this characteristic also means that in some water-rich regions, the value of water resources is not fully utilized. For a long time, low water prices have further exacerbated the situation of water resource waste [10]. In order to improve this situation, relevant studies have pointed out that adjusting water resource prices is one of the important methods for improving the efficiency of water resource allocation [11]. By establishing a core pricing mechanism based on “reasonable pricing” and “targeted subsidies and water conservation incentives,” water use efficiency and water conservation rates can be effectively improved [4].

Against this backdrop, the Chinese government has gradually introduced and continuously improved WET. The implementation of this policy aims to optimize the allocation of water resources through market mechanisms and resolve the issue of uneven distribution of water resources among different regions and water users. Specifically, water rights trading policies clarify the ownership of water resources, grant water users certain water rights, and allow these rights to be traded under certain conditions, thereby incentivizing water users to use water resources more efficiently and achieve sustainable water resource utilization. Over time, water rights trading policies have been refined and optimized in practice, gradually becoming an important policy tool for promoting the efficient use and rational allocation of water resources.

Research on water rights trading policies has always been a focus of attention among scholars. WET in China has undergone three main stages of transformation. In the first phase, farmers will mainly meet their own agricultural water needs and construct water-saving devices. The Chinese government has issued water conservation policies, such as the 1987 Yellow River Water Resources Allocation Plan.

The second phase was from 2000 to 2013, when water use began to be constrained by water resource prices [12]. People are quite sensitive to changes in water rights prices. When the government incorporates environmental and resource externalities into water rights pricing, people’s water conservation behavior will change positively [13]. This is because China not only sets a fixed price per square meter for water resources but also adopts a tiered pricing system. Once the basic water consumption per household is exceeded, a penalty mechanism will be implemented, and a new pricing method will be applied to the excess water consumption. Therefore, changes in water prices will positively stimulate residents’ water conservation behavior. Research shows that comprehensive agricultural water price reform can raise farmers’ awareness of water conservation and promote sustainable agricultural development [14].

The third phase began in 2014 with the designation of WET. Research on WET in China had been ongoing since before 2000, and Zhejiang Province’s water rights trading was the first water rights transaction in China. Additionally, based on the actual annual water supply volume, Yiwu City will pay Dongyang City a comprehensive management fee of 0.1 yuan per cubic meter [15]. Since 2014, China has successively established seven provinces and regions, including Ningxia, Jiangxi, Hubei, Inner Mongolia, Henan, Gansu, and Guangdong, to launch water rights pilot programs.

2.2. Measuring Water Resource Efficiency

The second is research on water resource utilization efficiency. Liu et al. (2025) [16] showed in their research that water resources are a fundamental element of human life and socioeconomic development and that their rational use is crucial to the sustainable development of society and the economy. Current scholars mainly study the efficiency of agricultural and industrial water use [17]. This situation highlights the urgency of scientifically measuring WEE in order to better assess and optimize the allocation and use of water resources, thereby providing a basis for decision-making to achieve sustainable use of regional water resources. Wei et al. (2021) [18] calculated the agricultural water use efficiency of nine provinces in the Yellow River basin from 2008 to 2017 and empirically concluded that economic development and water resource endowment have a positive impact on agricultural water use efficiency. Climate change and industrial economic development in today’s world are exacerbating water scarcity. Although some countries are using advanced desalination technology to alleviate water shortages, this has not been effective in changing the current situation [19]. Therefore, studying the factors that affect water resource quality is extremely important for improving WEE.

Regarding the measurement of water resource utilization efficiency, most scholars have only considered the efficiency of agricultural water use at a single level. However, water resources are not only used in agricultural production; they are indispensable in industrial and urban economic development. Currently, there are two main methods for measuring water resource utilization efficiency: DEA and stochastic frontier approach (SFA) [20]. DEA efficiency measurement has always been the mainstream method. In recent years, the DEA measurement theory system has been continuously updated, from the simple CCR model to the SBM model with the addition of non-expected output. Yan et al. (2024) [21] pointed out that although the DEA model is a relatively mature measure of efficiency, it often has many constraints and is a relatively complex model.

The random frontier method is a technique for analyzing the technical efficiency of production functions, proposed by Broek et al. (1980) [22] and Aigner et al. (1977) [23]. Random frontier analysis originated from the Douglas production function model, but the traditional Douglas production function is limited by its constraint of constant returns to scale. Transcendental logarithmic production functions can more accurately reflect changes in efficiency over time. When measuring WEE, considering that WEE changes over time, a time-varying model of transcendental logarithmic production functions is introduced. Given that water resource utilization efficiency varies over time and that random errors are unavoidable in the measurement process, the WEE measured by the stochastic frontier analysis is more robust. Therefore, this study proposes to use the stochastic frontier analysis method to measure efficiency.

Current research on the impact of WET on WEE. Many scholars have linked water rights trading to agricultural development, such as Ref. [24], who only considered its impact on agricultural water use, lacking a comprehensive perspective. Auci and Pronti (2023) [1] also focused solely on agricultural water use, noting that high water use efficiency in agriculture can alleviate regional water scarcity and improve farm economic performance, with agricultural water use being the primary driver of water resource consumption. Shah et al. (2023) [8] utilized data from 31 provinces in China to assess agricultural water use efficiency, concluding that the implementation of water rights trading policies led to a significant improvement in agricultural water use efficiency. A large body of literature has failed to comprehensively explore the effects of WET from the perspective of urban water supply, instead focusing solely on measuring WEE from the perspective of agricultural water use. WET can significantly improve agricultural water use efficiency, but their effects on urban water supply remain to be tested.

2.3. Theory and Hypothesis

2.3.1. Direct Impact of WET on WEE

WET are important policies for achieving water resource management and can effectively promote the rational allocation of water resources. Investigating the impact of WET on water resource utilization efficiency essentially involves analyzing how water resource property rights trading affects people’s water use behavior.

First, WET can enhance WEE by increasing water resource usage fees. Water resource usage fees are a key factor directly influencing WEE. Water resource management policies aim to improve WEE by implementing tiered pricing for water resources, thereby altering water usage behavior [25]. This approach of using water prices to incentivize residents’ water usage behavior can directly promote water-saving practices [26]. When water usage exceeds the quota, imposing surcharges on the excess volume encourages residents to proactively adopt water-saving behaviors. Additionally, WET can drive innovation in agricultural water-saving technologies. Such policies incentivize regions to develop water-saving devices and provide subsidies to regions that adopt water-saving innovations [4]. Water-saving devices can be widely applied in agriculture and industry, enhancing production efficiency while achieving efficient water resource utilization. Therefore, this paper proposes Hypothesis 1.

Hypothesis 1.

WET can significantly improve water resource utilization efficiency.

2.3.2. Heterogeneity of WET Impacts on WEE

Due to differences in water resource distribution across different regions, including variations in precipitation, land water storage capacity, and economic development levels between cities, the policy impact of WET will vary depending on these differences, thereby affecting the WEE.

From a direct perspective, China’s water resources are unevenly distributed spatially. The western regions have unique topography, which naturally poses challenges for water storage. The Yungui Plateau exhibits karst topography, where rainfall readily seeps into the ground, making surface water storage and utilization challenging [8]. Additionally, in regions like the Taklamakan Desert, where vast expanses of desert and Gobi dominate the surface, water retention is similarly difficult. Due to natural geographical conditions and people’s long-standing lack of awareness about water scarcity, there are differing perspectives on water conservation behaviors. Therefore, WET will produce heterogeneous effects in different geographical locations [27].

Water resources exhibit significant regional disparities in availability. Western regions often face water scarcity, with many areas being desert-like. In contrast, coastal cities, which have abundant water resources, have developed waterways and consequently enjoy higher economic levels. There are substantial differences in water resource endowments across regions. Due to the unique geographical location of cities, WET may have heterogeneous effects on water resource utilization efficiency. Additionally, whether the study area is located in an old industrial base may also result in heterogeneity. Industrial water use has long been the primary driver of water resource consumption, and industrial bases increase water resource consumption. Therefore, this paper proposes the second hypothesis, Hypothesis 2:

Hypothesis 2.

The impact of WET on WEE is heterogeneous.

2.3.3. Mechanisms Through Which WET Affects WEE

The role of WET in WEE is complex. How does the positive impact between the two come into play? Regarding the mechanism by which WET affects efficiency, this paper analyzes three channels from the perspective of water resource consumption.

WET allows water resources to be legally transferred and sold, operating under both government regulation and market mechanisms. Therefore, the effectiveness of WET is inevitably linked to market mechanisms. This paper uses market vitality to explore the underlying mechanisms. When market vitality is strong, resource elements can be reasonably allocated according to market demand, reducing unnecessary resource waste. Additionally, as market vitality reaches a high level, the number of water users continues to increase, thereby enhancing market competition to a certain extent. Market vitality plays an indispensable role in the WET by reasonably allocating water resources.

The second channel is water-saving benefits, measured by the volume of water reused within the city. First, from a technical perspective, the volume of reused water in cities indicates improvements in water-saving technology; economically, the increase in water-saving benefits enhances the revenue of water users, leading to changes in water resource utilization efficiency [28]; finally, from an institutional perspective, the benefits derived from water-saving behaviors demonstrate the effectiveness of the current water resource management policies. Therefore, water-saving benefits exert a regulatory effect on the impact of WET on WEE, thereby reinforcing the role of such policies.

Channel three is technological innovation output. Technological innovation investment has been incorporated into the study as a controlled variable [29], but does technological innovation output influence the policy effectiveness of water rights trading pilot programs? Technological innovation output represents improvements made by cities in water conservation; the more technological innovation output there is, the more effective water-saving devices become [30].

Hypothesis 3.

Market vitality and water-saving benefits play a positive moderating role in the impact of WET on WEE.

3. Method

3.1. Sample Selection and Data Sources

Since 2014, China has placed significant emphasis on the implementation of water rights trading pilot policies, establishing pilot programs in seven provinces and autonomous regions including Ningxia, Jiangxi, Hubei, Inner Mongolia, Henan, Gansu, and Guangdong. These seven pilot provinces encompass a total of 79 prefecture-level cities, which constitute the treatment group in this study. To establish a valid control group, we began with all prefecture-level cities in China as the initial sample, then excluded cities from the Tibet and Xinjiang regions due to severe data gaps. After this adjustment, the remaining non-pilot cities total 192. The final research sample comprises 271 prefecture-level cities in China. The study period spans from 2006 to 2023. The economic data used were primarily sourced from the China City Statistical Yearbook, China Urban Construction Statistical Yearbook, and various regional statistical yearbooks. Missing values in the dataset were addressed using linear interpolation.

3.2. Measurement Model Setting and Variable Measurement

3.2.1. Measurement Model

This study uses the 79 prefecture-level cities within the seven provinces where the Chinese government has established water rights trading pilot programs as the demonstration area, with the remaining regions serving as a control group where the policy has not been implemented. A Difference-in-Difference (DID) dummy variable is generated by multiplying the water rights trading pilot variable and the time variable to construct a classic DID econometric model.

$${WET}_{it}={\beta }_0+{\beta }_{1}MAD{Z}_{it}+{\beta }_{2}Contro{l}_{it}+{\theta }_i+{\eta }_t+{\epsilon }_{it}$$

(1)

where i denotes region and t denotes year; the dependent variable WET represents water resource utilization efficiency; Control represents control variables that affect WEE; ε is an unobservable random factor; θ and η represent city effects and year effects in the model, respectively.

3.2.2. Variable Measurement

WEE is defined as achieving the maximum benefit with minimal water supply and other inputs. In this study, urban water supply volume, water supply investment, and regional labor are selected as input variables, while urban GDP is employed as the output variable. GDP serves as the most comprehensive and authoritative indicator for measuring the final economic output of all resident units within a region. Therefore, examining the “maximum GDP output per unit of water resource input” reflects the comprehensive contribution rate of water resources to economic and social development at the macro level [31]. This approach aligns closely with the macro-strategic objective pursued by policymakers of “promoting development through water resources and supporting development with water resources.” While most existing literature utilizes DEA to measure water resource utilization efficiency, DEA has certain limitations and exhibits a “black-box nature,” meaning it cannot detail the allocation mechanisms of various resources. Consequently, this study adopts the Stochastic Frontier Analysis model. The specific functional form of the model is as follows:

$$Y_{it}=F\left(X_{it},\beta \right)\mathrm{e}\mathrm{x}\mathrm{p}\left(V_{it}-U_{it}\right)$$

(2)

where i represents the cross-sectional unit, t represents time, and Y represents output; F(X_it, α) is the production frontier, X_it denotes the input factor vector, and β is the vector of estimated coefficients; Vit is a general random disturbance term, assumed to be independent of U_it, used to represent the impact of statistical errors and various random factors on frontier output; U_it ≥ 0 is a time-varying technical inefficiency term, measuring relative production efficiency levels. This model is subject to random noise and technical inefficiency factors, making it difficult to achieve the optimal frontier level in actual production.

Since the traditional Douglas production function cannot change the proportions of various factors, it has limitations. Therefore, when using the stochastic frontier model, a time-varying model that goes beyond the logarithmic production function is used to measure water resource utilization efficiency. The time-varying model is consistent with the nature of water resource utilization efficiency changing over time, so it is more reasonable to use the time-varying model in this paper, and the model has also passed the test. The specific function formula is as follows:

$$\begin{array}{c}ln{Y}_{it}={\alpha }_0+{\beta }_K\left(ln{K}_{it}\right)+{\beta }_L\left(ln{L}_{it}\right)+{\beta }_W\left(ln{W}_{it}\right)+{\beta }_{t}t\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+0.5{\beta }_{KL}\left(ln{K}_{it}ln{L}_{it}\right)+0.5{\beta }_{KW}\left(ln{K}_{it}ln{W}_{it}\right)+0.5{\beta }_{KK}\left(ln{K}_{it}^2\right)\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+0.5{\beta }_{LL}(ln{L}_{it}{)}^2+0.5{\beta }_{WW}(ln{W}_{it}{)}^2+0.5{\beta }_u{t}^2+{\beta }_{Kt}t\left(ln{K}_{it}\right)\\ +{\beta }_{Lt}t(ln{L}_{it})+{\beta }_{Wt}t(ln{W}_{it})+v_{it}-u_{it}\hspace{1em}\end{array}$$

(3)

where Yit denotes the economic output of city i in year t, K, L, and W represent the city’s water supply investment, labor input, and urban water supply volume, respectively. Vit is the composite error term, where vᵢ is the random error term, representing the impact of random factors on the firm’s production frontier, assumed to follow Vit ~ N(0, σᵥ²); Uit is the inefficiency error term, representing the distance between the firm’s actual output and the production frontier, i.e., technical efficiency loss, assumed to be independent of vᵢ and distributed as Uit ~ N⁺(μᵢ, σᵤ²). WEE of city i in period t can be calculated by dividing the expected value of the actual water resource utilization output in period t by the expected value of output under the condition of full technical efficiency (Uit = 0), as shown in Equation (4):

$$W{EE}_{it}={\displaystyle \frac{{}_E(Y_{it}{U}_{it},X_{it})}{{}_E(Y_{it}{U}_{it}=0,X_{it})}}=\exp \left(-U_{it}\right)$$

(4)

To minimize model error, this paper incorporates variables that may affect WEE into the econometric model. This paper incorporates variables that may influence WEE into the econometric model: (1) Economic Development Level (Eco): This paper uses the logarithm of per capita regional GDP to measure economic development level. Per capita regional GDP to some extent reflects the scale of economic development and can represent the economic development status of the region [32]; (2) Technological progress (Tec): The logarithm of technology expenditure is used as a measure of technological progress. Regional expenditure on scientific and technological innovation can indirectly demonstrate the region’s emphasis on science and technology and its relatively high technological level. In the process of achieving sustainable green development, science and technology are important prerequisites for regional water-saving technological innovation and water-saving device optimization; (3) Industrial structure (Indu): The proportion of tertiary industry added value to regional GDP is used as an indicator of industrial structure. A more reasonable industrial structure indicates more rational resource allocation in the region, thereby improving WEE; (4) Urbanization level (City): The proportion of urban population to the total urban population is used as an indicator of urbanization level. The rapid development of urbanization has placed significant pressure on China’s natural resources, as urban expansion has encroached on more land. As the economy develops, more people are moving to cities, causing urban peripheries to expand continuously. Urbanization has also increased demand for water resources; (5) Employment Structure (Lab): expressed as the proportion of workers in the tertiary sector.

Table 1. Descriptive statistical analysis of control variables.

Variables	Sample Size	Mean	Sd	Min	Max	VIF
WEE	4698	95.707	7.870	62.688	100.00
WRT	4698	0.170	0.376	0.000	1.000	1.050
Lab	4698	0.572	0.153	0.22	1.020	1.380
Ind	4698	41.160	9.797	20.70	70.020	1.930
Eco	4698	10.550	0.715	8.80	12.020	3.750
City	4681	0.532	0.163	0.20	0.941	2.990
Tec	4698	10.094	1.654	6.26	14.440	2.510

presents the descriptive analysis of the main variables.

According to the results in Table 1, the mean value of the dependent variable WEE is 95.707%, indicating that the WEE of most cities is currently high. The minimum value is only 62.688%, which is significantly lower than the maximum value, indicating that some regions have relatively low WEE and inadequate water resource allocation. The mean value of WET is 0.170, indicating that the current pilot programs are limited to a small number of regions, consistent with the fact that only seven provinces have implemented such policies in reality. Finally, the descriptive analysis of control variables shows that the standard deviations of all control variables are far smaller than their means, indicating that the data is of good quality and relatively stable. Among them, the standard deviations of employment structure, economic development level, and urbanization level are the smallest, indicating that these three sets of data exhibit greater stability compared to industrial structure and technological progress. VIF indicates whether there is correlation between variables.

4. Empirical Research

4.1. Benchmark Regression Results

To verify whether WET can improve WEE, this paper employs a DID model. In the regression analysis presented in

Table 2. Benchmark regression analysis.

Variables	(2)	(1)	(3)	(4)
	Model (2)	Model (1)	Model (3)	Model (4)
	WEE	WEE	WEE	WEE
WET	9.2099 ***	0.2736 **	0.0894 ***	0.5111 ***
	(0.3772)	(0.1367)	(0.0331)	(0.1423)
Tec			0.1224 ***	−0.3028 ***
			(0.0309)	(0.0843)
Eco			1.4416 ***	0.9603 ***
			(0.1499)	(0.2740)
Indu			−0.0502 ***	−0.0736 ***
			(0.0052)	(0.0108)
Urba			−2.1592 ***	−0.9063
			(0.5276)	(1.0899)
Labor			3.4214 ***	3.9104 ***
			(0.3183)	(0.4933)
Constant	94.1403 ***	95.6601 ***	80.5212 ***	89.8391 ***
	(0.1530)	(0.0374)	(1.3702)	(2.8213)
Observations	4698	4698	4676	4675
City	NO	YES	NO	YES
Year	YES	YES	YES	YES
R-squared	0.124	0.933	0.931	0.939

Note: ***, ** and * respectively indicate that the regression result is significant at the level of 1%, 5% and 10%, and the coefficient in brackets is standard error.

, four different fixed-effects models were designed to ensure the robustness of the results. Models (1) to (3), respectively, do not control for time effects or city effects, control only for city effects, and control only for time effects, respectively. The regression results obtained from these models are identical to those from Model (4). From the regression analysis table in Table 2, Hypothesis 1 can be verified, indicating that the WET significantly and positively improved WEE. Model (4) demonstrates that the sustained implementation of WET exerts a statistically significant positive impact on regional WEE. The coefficient of WET is 0.5111, indicating that regions implementing this policy have improved their WEE by 0.5111% compared to non-implementing regions. From an economic significance perspective, this efficiency gain translates to a 0.5111% reduction in water consumption per 10,000 yuan of GDP, representing tangible resource conservation. These findings confirm that the water rights trading policy not only achieves statistical significance but also generates substantive water-saving benefits, bearing clear practical implications for sustainable water resource management. The underlying mechanism for this policy effect lies in the market value assigned to water resources through WET, which enables transactions and leasing of water rights across regions and river basins. This market-based approach effectively addresses spatial disparities in water resource endowments. By facilitating market circulation of water resources, WET promotes optimal allocation and consequently enhances utilization efficiency.

From the perspective of controlling variables, while economic development level and employment structure had a positive promotional effect, all other variables had an adverse impact on WEE. Among these, economic development level had a positive influence on WEE, with a coefficient value of 0.9603. This is because cities with higher economic levels place greater emphasis on resource allocation, generating more benefits, which in turn promotes the WEE. As shown in Table 2, the industrial structure has a negative impact on WEE. The tertiary sector in the industrial structure has relatively lower water consumption and pays relatively lower water fees. Therefore, without price constraints, the development of the tertiary sector may neglect WEE, leading to an adverse impact on WEE. Technological progress is theoretically expected to have a positive impact on WEE, but here it shows a negative impact. This may be due to the need for increased investment in technological innovation in the early stages, which requires water resources as testing conditions, thereby producing a negative impact. The significant negative coefficients for technological progress and industrial structure warrant careful interpretation. For industrial structure, this phenomenon may reflect inadequate water pricing mechanisms within the rapidly expanding tertiary sector, failing to provide sufficient incentives for water conservation. Regarding technological progress, the negative correlation likely stems from a time-lag effect wherein R&D investments precede the realization of actual efficiency gains. These findings emphasize the necessity for implementing complementary policies rather than suggesting that structural optimization or technological advancement adversely affects WEE.

The parallel trend test must be satisfied before using the DID model, which requires an assessment of changes in WEE before and after the establishment of the demonstration area. Figure 1 shows the corresponding parallel trend chart. As shown in Figure 1, WEE did not undergo significant changes before the policy implementation. In the current, a1 to a9 periods after the policy implementation, the coefficients are significant, and the policy effects are significantly positive, consistent with the baseline regression. The parallel trend analysis indicates that the WET effectively stimulated improvements in WEE.

4.2. Heterogeneity Analysis

4.2.1. Analysis of Urban Geographical Heterogeneity

In addition to examining the impact of WET on WEE across the entire region, this paper also conducts a heterogeneity analysis from three perspectives to examine whether WET have different policy effects in different regions.

The Yangtze River Economic Belt spans 11 provinces in China, accounting for over 40% of the country’s GDP. It is not only the economic hub of the nation but also a crucial ecological barrier. This paper conducts separate regressions for the Yangtze River Economic Belt and non-Yangtze River Economic Belt regions.

Table 3. Geographic heterogeneity.

Variables	(1)	(2)	(3)	(4)
	Yangtze River Economic Belt	Non-Yangtze River Economic Belt	North	South
WET	1.6136 ***	−0.0831	0.0997	0.7042 ***
	(0.2525)	(0.1829)	(0.2330)	(0.1899)
Tec	−0.5311 ***	−0.2684 **	−0.3978 ***	−0.1559
	(0.1516)	(0.1053)	(0.1320)	(0.1114)
Eco	0.4282	1.0638 ***	1.9251 ***	0.3217
	(0.5178)	(0.3492)	(0.4387)	(0.4131)
Indu	−0.0914 ***	−0.0603 ***	−0.0932 ***	−0.0384 ***
	(0.0197)	(0.0137)	(0.0167)	(0.0138)
Urban	3.2110	−3.7907 ***	−7.2123 ***	6.4884 ***
	(1.9605)	(1.3524)	(1.5540)	(1.5456)
Labor	2.5248 ***	5.1060 ***	4.5127 ***	3.6676 ***
	(0.6385)	(0.7380)	(0.7850)	(0.5878)
Constant	97.2719 ***	88.7077 ***	84.1795 ***	89.9688 ***
	(5.8283)	(3.5078)	(4.2923)	(4.4256)
Observations	1878	2797	1997	2678
R-squared	0.947	0.935	0.935	0.946
between-group differences	−1.697 ***		0.352 ***

Note: ***, ** and * respectively indicate that the regression result is significant at the level of 1%, 5% and 10%, and the coefficient in brackets is standard error. The p-value for coefficient differences is used to test the significance of differences in new quality productivity levels between groups, calculated using Fisher’s combined test (500 samples).

presents the results of the sample classification. For cities located along the periphery of the Yangtze River Economic Belt, the coefficient of WET is 1.6136 and exhibits strong significance, indicating that WEE in cities along the Yangtze River Economic Belt is significantly influenced by WET. In contrast, the coefficient of the core explanatory variable for cities outside the Yangtze River Economic Belt is −0.0831. This suggests that WET has not achieved the desired policy effects in cities outside the Yangtze River Economic Belt, and further exploration of complementary policies to enhance WEE is needed. Cities in the Yangtze River Economic Belt have a core explanatory variable coefficient of 1.6136, which is greater than the coefficient of the baseline regression. This indicates that implementing WET in the Yangtze River Economic Belt not only achieves policy effects but also demonstrates more favorable policy outcomes, with a greater magnitude of improvement in WEE.

Considering the practical significance of WET and the unique geographical location of the Yangtze River Economic Belt, it is important to note that this region serves as a crucial agricultural production zone in China, resulting in a naturally higher demand for water resources. However, to protect the natural resources and environment of the Yangtze River Economic Belt, the Chinese government has implemented targeted ecological and environmental protection policies for this region, such as the Yangtze River Protection Law and ecological compensation policies. These measures have led to significant differences in policy outcomes compared to non-Yangtze River Economic Belt regions.

Table 3, columns 3 and 4, provide a detailed description of the impact of WET on WEE in southern and northern regions. The core explanatory variables exhibit a positive influence in terms of their sign. However, only the southern regions passed the inspection. In the subgroup regression, the coefficient of WET for the southern region is 0.7042, indicating that WET has a significant policy effect in the southern region, positively promoting the effective improvement of WEE, with an efficiency increase of 0.7042% compared to regions that did not implement WET. Next, we provide corresponding explanations for these differences based on actual conditions. First, in terms of economic development, the southern region is more developed, and accordingly, southern cities have higher levels of marketization, with price mechanisms effectively guiding relevant enterprises to engage in water-saving behaviors and technological innovations in water-saving devices. Second, we analyze the differences based on the distinct patterns of water resource utilization between the northern and southern regions. Southern regions have an advantage in terms of original water resource reserves, with abundant water resources in water-rich areas. Southern regions place greater emphasis on policy objectives related to water quality and ecological protection, guiding local enterprises to proactively avoid high pollution costs and high consumption, thereby improving WEE. In northern regions, where water resources are scarce, water rights trading is subject to rigid quota restrictions and lacks flexibility, resulting in limited policy promotion effects. Finally, from the perspective of industrial structure. Southern regions are dominated by industry and services, with rapid adoption of water-saving technologies and sensitivity to water rights trading prices. In contrast, northern regions have a higher agricultural sector, with farmers as the primary users of water resources. Agricultural irrigation water-saving technologies are influenced by regional infrastructure and farmer acceptance, leading to delayed policy improvement effects. Therefore, WET is more effective in southern regions than in northern regions, reflecting the geographical heterogeneity between the two regions.

4.2.2. Analysis of Water Resource Heterogeneity

Based on the theory of factor endowments, regions abundant in factor resources often leverage their inherent advantages to gain more significant competitive opportunities during rapid economic development. Therefore, this study conducts a heterogeneity analysis of the policy effects based on water resource endowment. Precipitation serves as the source of terrestrial water resources. Whether surface runoff, groundwater recharge, or soil moisture, their most fundamental initial origin, largely unaffected by large-scale human intervention, is atmospheric precipitation. Thus, evaluating a region’s natural water resource endowment from the source of the water cycle makes precipitation the most direct indicator. Accordingly, this paper adopts precipitation as the metric for defining humid regions. Furthermore, grouping samples based on precipitation levels helps verify the previously proposed explanation regarding the regional heterogeneity in the causes of precipitation abundance differences between the north and south. This study employs average precipitation as the classification criterion, dividing 271 cities into humid and non-humid regions, to further investigate the differential impacts of the Water Rights Trading Policy on Water Use Efficiency across regions with varying precipitation conditions.

Table 4. Analysis of water resource endowment heterogeneity.

Variables	(1)	(2)
	Wet Areas	Non-Wet Area
WET	0.8809 ***	0.2346
	(0.2043)	(0.2098)
Tec	−0.1325	−0.3061 **
	(0.1140)	(0.1212)
Eco	0.6842	1.1356 ***
	(0.4365)	(0.3717)
Indu	−0.0237 *	−0.1095 ***
	(0.0141)	(0.0161)
Urban	6.9303 ***	−5.8574 ***
	(1.6161)	(1.4638)
Labor	5.0289 ***	3.2036 ***
	(0.7207)	(0.6750)
Constant	84.3182 ***	92.1552 ***
	(4.6609)	(3.6847)
Observations	2271	2364
R-squared	0.950	0.938
between-group differences	−0.646 **

Note: ***, ** and * respectively indicate that the regression result is significant at the level of 1%, 5% and 10%, and the coefficient in brackets is standard error. The p-value for coefficient differences is used to test the significance of differences in new quality productivity levels between groups, calculated using Fisher’s combined test (500 samples).

presents the results of two separate regression analyses conducted on two distinct groups of regions classified based on precipitation levels, in accordance with the factor endowment theory. As shown in the first column of the table, the core explanatory variable coefficient for the humid region is 0.8809, while the coefficient for the non-humid region in the second column is 0.2346. Although the core explanatory variables for both groups are positive, the absolute value for the humid region is greater than that for the non-humid region, indicating that the policy effects of water rights trading are more pronounced in humid regions. The reasons for this may be attributed to several factors. Similarly to southern regions, humid areas have ample precipitation, and WET in these regions focus more on market-based allocation, where price signals can effectively stimulate improvements in water resource utilization. In contrast, in non-humid regions, WET are primarily government-allocated quotas, with limited market transaction freedom, making it difficult for the market mechanisms of WET to function effectively, and the price signal-driven improvements in WEE are not prominent. Therefore, in the implementation of WET in the future, it is particularly important to establish alternative resource management policies in non-humid regions to effectively enhance WEE.

4.2.3. Analysis of Industrial Development Heterogeneity

When examining the impact of WET on WEE, it is methodologically necessary to categorize the sample into old industrial bases and non-old industrial bases. This classification is grounded in the close relationship between regional water consumption patterns and distinctive industrial structures as well as varying levels of industrialization. For the heterogeneity analysis concerning “industrial development foundation,” precise definitional criteria for “old industrial base” must be established. This study identifies old industrial bases according to the officially designated list of prefecture-level cities outlined in the “National Plan for the Adjustment and Transformation of Old Industrial Bases (2013–2022)” issued by the State Council of China. These bases are characterized by their substantial historical contributions to industrial development, a disproportionately high share of traditional industries within their economic structure, and unique challenges in industrial transformation and upgrading. Adopting this classification standard ensures both the objectivity of our heterogeneity analysis and its relevance to policy frameworks. By distinguishing between old industrial bases and their counterparts, this research effectively examines the heterogeneous effects of WET across these distinct regional categories [33].

In

Table 5. Analysis of heterogeneity in industrial development foundations.

Variables	(1)	(2)
	Old Industrial Base	Non-Old Industrial Base
WET	0.6498 ***	0.3250
	(0.1753)	(0.2387)
Tec	−0.2381 **	−0.3127 **
	(0.1035)	(0.1356)
Eco	1.0942 ***	0.9639 **
	(0.3305)	(0.4419)
Indu	−0.0526 ***	−0.1114 ***
	(0.0136)	(0.0170)
Urban	−0.5148	−2.6266 *
	(1.4219)	(1.4220)
Labor	3.8592 ***	2.9803 ***
	(0.6004)	(0.7153)
Constant	86.8812 ***	92.3395 ***
	(3.5035)	(4.4431)
Observations	3142	1533
R-squared	0.934	0.956
between-group differences	0.325

Note: ***, ** and * respectively indicate that the regression result is significant at the level of 1%, 5% and 10%, and the coefficient in brackets is standard error. The p-value for coefficient differences is used to test the significance of differences in new quality productivity levels between groups, calculated using Fisher’s combined test (500 samples).

, the first column represents samples from old industrial bases, and the second column represents samples from non-old industrial bases. The core explanatory variables for the two columns are 0.6498 and 0.3250, respectively, both indicating a positive promotional effect. The promotional effect is greater for old industrial bases, and only the samples from old industrial bases passed the significance test, while the samples from non-old industrial bases did not pass the significance test, indicating that the policy effect is not significant. Theoretically, old industrial bases primarily focus on heavy industries such as steel and chemicals, with lagging water conservation renovation projects, potentially facing significant resistance to WET. Non-old industrial bases, on the other hand, are dominated by emerging industries with higher WEE and more flexible markets, making WET more likely to exert a promotional effect. However, the test results show that the policy effects in old industrial bases are more significant. This may be because old industrial bases, being dominated by heavy industries, are subject to stricter environmental performance evaluations by the government, such as carbon emission reduction targets and wastewater reduction indicators. The government uses administrative coercive measures to force heavy industrial enterprises to participate in WET, and imposes stricter penalties when water consumption exceeds limits. Additionally, a higher proportion of these enterprises are state-owned, making them more responsive to policy directives. They can swiftly implement transaction requirements and enhance environmental performance, thereby driving the water rights trading policy to improve WEE. In contrast, non-traditional industrial bases primarily develop high-tech industries, where water use efficiency is already high, leaving limited room for marginal improvements through water rights trading, resulting in less noticeable effects. Therefore, traditional industrial bases demonstrate superior effectiveness in enhancing WEE through WET compared to non-traditional industrial bases.

4.3. Placebo Test

To avoid the influence of unobservable factors on the research results, this paper used random sampling 500 times to generate interaction terms to construct the treatment group and conducted a placebo test, as shown in Figure 2. The true estimated coefficient in this study is 0.5111, which differs significantly from the spurious estimated coefficient. The spurious estimated coefficients are primarily distributed around the 0 value in a normal distribution, and most p-values are greater than 0.1. Therefore, the results pass the placebo test. This indicates that the study results were not influenced by uncontrollable factors, and the findings on the impact of WET on WEE are reliable, with robust conclusions.

4.4. Robustness Test

4.4.1. Propensity Score Matching (PSM)

To ensure more reasonable comparison results between cities exposed to policy shocks and those not exposed, and to mitigate potential endogeneity issues, this study re-estimates the results using the PSM-DID method. This approach avoids potential selection bias while further examining the WET on urban WEE. Drawing on previous research, a 1:3 nearest neighbor matching method is employed in the PSM process to enhance the reliability of sample selection. The PSM-DID results are presented in Column (2) of the table above. After controlling for non-random factors and other policy influences, the establishment of pilot zones had no significant changes in the direction or coefficient of the impact on urban WEE. The coefficient of WET is 0.5243, consistent with the results from the baseline regression.

4.4.2. Entropy Matching

Similarly to PSM, entropy balancing matching can also make the selection of samples more reasonable. Entropy balancing matching typically eliminates differences in control variables by assigning different weights to each sample, ensuring that the control variables are similar at the first or higher-order moments. Reducing differences in control variables indicates that the primary factor influencing the final results is the impact of WET. The results of entropy matching are shown in column (3) of

Table 6. Robustness test.

Variables	PSM-DID	Entropy Matching
	WEE	WEE
WET	0.5243 ***	0.4296 ***
	(0.1428)	(0.1460)
Tec	−0.3115 ***	−0.1415
	(0.0845)	(0.0965)
Eco	0.9606 ***	0.7447 **
	(0.2749)	(0.3158)
Indu	−0.0730 ***	−0.0714 ***
	(0.0109)	(0.0121)
Urba	−0.8896	0.9287
	(1.1127)	(1.3396)
Labor	3.9192 ***	4.0566 ***
	(0.4939)	(0.5644)
Constant	89.8789 ***	89.1568 ***
	(2.8322)	(3.2524)
Observations	4657	4675
City	YES	YES
Year	YES	YES
R-squared	0.939	0.938

Note: ***, ** and * respectively indicate that the regression result is significant at the level of 1%, 5% and 10%, and the coefficient in brackets is standard error.

, with a coefficient of 0.4296, which remains positively significant. This indicates that the positive impact of WET remains valid.

4.4.3. Dual Machine Learning Model

To address potential model misspecification and enhance the robustness of our findings, we employ the dual machine learning (DML) approach as an alternative estimation strategy. Unlike traditional parametric models that rely on strict functional form assumptions, DML effectively handles high-dimensional controls and captures complex nonlinear relationships while providing valid statistical inference.

The DML framework combines Neyman-orthogonal scores with cross-fitting to mitigate overfitting biases. We first specify the partially linear DML model as follows:

$${WEE}_{it}={\theta }_0{WET}_{it}+g\left(X_{it}\right)+U_{it}$$

(5)

$$\begin{array}{ccc}E\left(U_{it}∣{WET}_{\mathrm{i}\mathrm{t}},X_{\mathrm{i}\mathrm{t}}\right)& =& 0\end{array}$$

(6)

Among these, ${WEE}_{it}$ denotes the WEE of the i-th city in year t; ${WET}_{\mathrm{i}\mathrm{t}}$ denotes the treatment variable, representing the policy variable “water rights trading”; ${\theta }_0$ is the coefficient of ${WET}_{\mathrm{i}\mathrm{t}}$, reflecting the policy effect. Here, $X_{\mathrm{i}\mathrm{t}}$ differs from its meaning in the baseline regression; dual machine learning adopts a more comprehensive approach, with control variables often being high-dimensional. The specific form $\widehat{g}\left(X_{it}\right)$ requires estimation via machine learning algorithms. Transforming Equations (5) and (6) yields the core explanatory variable coefficient estimates: ${WET}_{\mathrm{i}\mathrm{t}}$ denotes the treatment variable, representing the policy variable “water rights trading”; ${\theta }_0$ is the coefficient of WET, reflecting the policy’s effectiveness. Here, $X_{it}$ differs in meaning from the benchmark regression. Dual machine learning adopts a more comprehensive approach, often involving high-dimensional control variables. Machine learning algorithms are required to estimate the specific form $\widehat{g}\left(X_{it}\right)$. By transforming Equations (5) and (6), the core explanatory variable coefficient estimator is obtained as:

$${\widehat{\theta }}_0={\left({\displaystyle \frac{1}{n}}\sum _{i\in I,t\in T} {WET}_{it}^2\right)}^{-1}{\displaystyle \frac{1}{n}}\sum _{i\in I,t\in T} {WET}_{it}\left(Y_{it+1}-\widehat{g}\left(X_{it}\right)\right)$$

(7)

Here, a dual machine learning model replaces the baseline regression mentioned above, with samples split in a 1:4 ratio [16].

Table 7. Replacement regression model.

Variables	(1)	(2)	(3)	(4)	(5)
	Random Forest	Random Forest	Lasso Regression	Gradient Boosting	Neural Network
	WEE	WEE	WEE	WEE	WEE
WET	1.5998 ***	1.6092 ***	1.5792 ***	1.6020 ***	1.6327 ***
	(0.1106)	(0.1107)	(0.1128)	(0.1112)	(0.1139)
Constant	0.0374	0.0399	0.0644	0.0570	0.0422
	(0.0620)	(0.0621)	(0.0623)	(0.0624)	(0.0624)
Control variable terms	YES	YES	YES	YES	YES
Control variable quadratic term	NO	YES	YES	YES	YES
Observations	4676	4676	4676	4676	4676

Note: ***, ** and * respectively indicate that the regression result is significant at the level of 1%, 5% and 10%, and the coefficient in brackets is standard error.

presents four different dual machine learning models. The first column shows the results of the random forest algorithm without the quadratic term of the control variables. The second to fifth columns present the results of the random forest, lasso regression, gradient boosting, and neural network algorithms, respectively, all of which include the quadratic term of the control variables. Including the quadratic term enhances the model’s estimation performance, accounting for nonlinear relationships. The regression coefficient for the Random Forest algorithm in the second column of Table 7 is 1.6092, consistent with the benchmark regression results, indicating a positive and significant impact. Therefore, it can be concluded that the WET shock positively stimulates improvements in WEE. Through validation using multiple machine learning algorithms, the regression results remain significantly robust.

4.5. Mechanism Analysis

To explore the pathways through which WET influence WEE, a mechanism analysis was conducted from three dimensions: market vitality, water-saving benefits, and technological innovation outputs.

4.5.1. Market Vitality

The regression results for market vitality as a moderating variable are presented in

Table 8. Mechanism analysis.

Variables	Market	Wcb	Tec-Out
	WEE	WEE	WEE
WET	0.3323 **	0.4589 **	2.1302 **
	(0.1618)	(0.2010)	(0.8985)
market	−0.0168 *
	(0.0087)
Market × WET	0.0510 ***
	(0.0197)
Wcb		−0.0258 ***
		(0.0086)
Wcb × WET		0.0227 **
		(0.0105)
Tec-out			−1.0154 ***
			(0.1998)
Tec-out × WET			−0.1891 *
			(0.1013)
Tec	−0.3077 ***	−0.4334 ***	−0.2710 ***
	(0.0845)	(0.1105)	(0.0824)
Eco	0.9675 ***	1.4301 ***	1.1912 ***
	(0.2757)	(0.3515)	(0.2791)
Indu	−0.0743 ***	−0.0774 ***	−0.0710 ***
	(0.0109)	(0.0124)	(0.0107)
Urba	−0.9610	−2.1091	−0.9562
	(1.0965)	(1.3479)	(1.0632)
Labor	3.9416 ***	3.5113 ***	3.5098 ***
	(0.4952)	(0.6006)	(0.4900)
Constant	89.8977 ***	87.6278 ***	96.1966 ***
	(2.8491)	(3.4448)	(2.9754)
Observations	4675	3241	4675
R-squared	0.939	0.943	0.940

Note: ***, ** and * respectively indicate that the regression result is significant at the level of 1%, 5% and 10%, and the coefficient in brackets is standard error.

. The coefficient for the core explanatory variable, 0.3323, differs little from the benchmark regression value of 0.511, and both are positively significant. Additionally, the interaction coefficient between market vitality and the core explanatory variable is 0.0510, indicating that market vitality positively promotes the improvement of WEE when WET take effect. Market vitality thus exerts a positive moderating effect. This corresponds to the previously proposed hypothesis and validates it.

Market vitality is measured by the number of newly registered water-related enterprises in the region. Market vitality can stimulate the market mechanism function of water rights trading policies. Enhanced market vitality indicates intense market competition, enabling the market mechanism of water rights trading policies to function effectively. Only when market vitality is enhanced can various resource elements circulate freely, thereby improving the efficiency of resource allocation. WET allocate water resources within regions and across industries according to demand, promoting the rational allocation of water resources. Single-sided government-imposed regulations are insufficient to enhance WEE; only by combining with market mechanisms can WET better achieve their intended effects. Through theoretical analysis and regulatory effect model testing, it is concluded that market vitality plays a positive regulatory role in enhancing WEE through WET.

4.5.2. Water Conservation Benefits

The regression results showing the impact of WET on WEE, with water-saving benefits as a moderating variable, are presented in Table 8. The coefficient of WET is 0.4589, indicating that when the water-saving effect is incorporated as a moderating variable, the policy effect still exerts a significant positive promotional influence. Furthermore, examining the interaction term between water-saving benefits and policy variables reveals that water-saving benefits serve as a viable moderating variable. The coefficient of the interaction term is 0.0227 and passes statistical testing, indicating that water-saving benefits amplify the promotional effect of the policy.

Water conservation benefits are an important indicator for measuring the reuse of water resources in cities. WET enable the rational allocation of water resources in the market, allowing water volumes to be bought and sold according to the needs of different regions and industries, thereby assigning market value to water resources. After the policy is implemented, water users such as enterprises become highly sensitive to changes in water prices, especially under a tiered pricing model, which imposes severe economic penalties on water users who waste water resources. Therefore, water conservation benefits can effectively play a positive regulatory role in promoting WEE through water rights trading.

4.5.3. Technological Innovation Output

Beyond market vitality and water-saving benefits, this paper also analyzes other potential impact pathways from the perspective of technological innovation. The regulatory effects of technological innovation outputs are presented in Table 7, where technological innovation outputs are measured using the logarithm of urban wastewater treatment volume. The volume of wastewater treated within a region represents the amount of unavailable water resources that cities need to treat. A decrease in wastewater treatment volume indicates a higher level of technological innovation output in water conservation within the region, suggesting that more water resources are being utilized effectively and wastewater volume is reduced. Therefore, the technological innovation output measured by wastewater volume is a negative indicator, meaning that higher wastewater volume indicates lower technological innovation output, and vice versa. The regression results for the moderating effect of technological innovation output are presented in Table 7. The coefficient for the policy effect, the core explanatory variable, is 2.1302 and remains positively significant, the main effect of the policy remains unchanged.

When the negative indicator of technological output level—measured by wastewater discharge volume—is incorporated into the adjustment model, the negative coefficient of the interaction term (−0.1891) indicates that wastewater discharge volume mitigates the positive effects of this policy. The implementation of WET indicates that regions with higher wastewater discharge volumes experience weaker efficiency gains from these policies. This is because technological innovation enables the development of water-saving devices. Following the implementation of water rights trading policies, under the dual pressures of stringent government regulations and market pricing, an increasing number of enterprises are placing greater emphasis on reducing wastewater discharge. However, the use of wastewater treatment volume to measure technological innovation output serves as a negative indicator, with the interaction term coefficient being negative. Therefore, in the path analysis examining how WET influence WEE, technological innovation output exerts a positive reinforcing effect.

4.6. Spatial Effects Analysis

Traditional difference-in-differences models generally assume no spillover effects of treatment. However, in reality, water resources can flow freely across regions. Due to potential siphon effects and spatial spillover effects, the implementation of water rights trading policies may impact the water use efficiency of neighboring cities. To further analyze the magnitude of such spatial spillovers, this paper establishes a Spatial Durbin Model to examine the effects of water rights trading policies on the water use efficiency of surrounding cities. The model is specified as follows:

$${WEE}_{it}=\rho W_{ij}{WEE}_{it}+{\alpha }_1{WET}_{it}+\beta X_{it}+{\alpha }_2{W}_{ij}{WET}_{it}+{\beta }_2{W}_{ij}{WET}_{it}+{\mu }_i^{\prime }+{\nu }_t^{\prime }+{\epsilon }_{it}^{\prime }$$

(8)

Among these, W_ij represents the economic-geographic weight matrix. To simultaneously account for the influence of both geographic and economic factors, this study employs a nested geoeconomic matrix, specifically constructed as the product of an inverse geographic distance matrix and an economic distance matrix. ρ denotes the spatial autocorrelation coefficient. The remaining variables are consistent with those in the benchmark model. Based on the results of the Hausman test, Wald test, and LR test in

Table 9. Model specification.

LR test	SDM can be simplified to SAR	232.91 ***	0.000	SDM
	SDM can be simplified to SEM	224.97 ***	0.000	SDM
Wald test	SDM can be simplified to SAR	150.82 ***	0.000	SDM
	SDM can be simplified to SEM	218.07 ***	0.000	SDM
Hausman test		25.55 ***	0.000	fixed effect

Note: ***, ** and * respectively indicate that the regression result is significant at the level of 1%, 5% and 10%, and the coefficient in brackets is standard error.

, the Spatial Durbin Model with two-way fixed effects is selected as the most appropriate approach for measuring the spillover effects of the water rights trading policy pilot on the water use efficiency of neighboring cities.

Table 10. Test results of SDM and the outcomes of decomposing the spatial spillover effect.

	(1)	(2)	(3)	(4)	(5)
Variables	Main	Wx	LR_Direct	LR_Indirect	LR_Total
WET	0.4117 **	1.7946 **	0.4315 **	3.7229 ***	4.1545 ***
	(0.1695)	(0.7569)	(0.1712)	(1.3792)	(1.2953)
Tec	−0.0870	−4.3734 ***	−0.1205 **	−8.3355 ***	−8.4560 ***
	(0.0626)	(0.4176)	(0.0599)	(1.1400)	(1.1339)
Eco	0.3413 *	10.1525 ***	0.4335 **	19.5322 ***	19.9658 ***
	(0.1877)	(1.1150)	(0.1758)	(2.8308)	(2.7890)
Indu	−0.0479 ***	−0.3782 ***	−0.0507 ***	−0.7514 ***	−0.8021 ***
	(0.0076)	(0.0513)	(0.0074)	(0.1260)	(0.1249)
Urba	0.4287	15.6211 ***	0.5367	29.4206 ***	29.9574 ***
	(0.5878)	(4.0085)	(0.5595)	(7.9813)	(7.9301)
Labor	3.4472 ***	−0.2848	3.4800 ***	2.9540	6.4340
WET	(0.3802)	(3.4467)	(0.3761)	(6.6663)	(6.6648)
ρ	0.4690 ***
	(0.0699)
sigma2_e	3.5746 ***
	(0.0745)
City	YES
Year	YES
Observations	4878
R-squared	0.109
Number of cityid	271

Note: ***, ** and * respectively indicate that the regression result is significant at the level of 1%, 5% and 10%, and the coefficient in brackets is standard error.

presents the spatial spillover effects of the water rights trading policy and their decomposition results. The results show that the spatial autocorrelation coefficient ρ for water use efficiency is 0.4690 and significantly positive, indicating the presence of spatial autocorrelation in water use efficiency. Since the estimated coefficients of the explanatory variables in the Spatial Durbin Model cannot directly reflect their impact on the explained variable, the partial differentiation method is applied to decompose these effects into direct and indirect effects, as shown in columns (3) and (4) of Table 10. In terms of direct effects, the implementation of the water rights trading policy pilot has a significantly positive effect on improving water use efficiency within the implementing city itself, which is consistent with the results obtained using the traditional difference-in-differences model in the previous analysis. Furthermore, the estimated indirect effects are also significantly positive, indicating that the implementation of the water rights trading policy generates positive spatial spillover effects on the water use efficiency of neighboring cities. This suggests that among geographically proximate cities with similar economic development levels, the effectiveness driven by policy regulation in pilot cities simultaneously promotes the improvement of water use efficiency in surrounding cities.

5. Discussion

Water resources are a public good, and everyone has the right to enjoy them. The low price of water resources has led to waste and overconsumption by users. In 2014, the Chinese government issued a WET that imposed constraints and reasonable controls on water prices through water rights transfers and other measures. Similarly to many research findings, this paper uses a quantitative model to verify that the WET has promoted improvements in WEE [34]. This study makes several significant innovations relative to prior literature. First, while most literature focuses on agricultural WEE or industrial WEE, this study innovates the research perspective by calculating WEE from the perspective of comprehensive urban water supply, overcoming the limitations of analyses confined to single sectors. Second, whereas policy effectiveness in WET studies is often measured at the provincial level [35]. This study provides an innovation in assessment granularity by conducting a prefecture-level city analysis, offering more localized and detailed insights. Third, regarding measurement methodology, while many studies employ DEA or simple water intensity calculations, this study introduces a methodological innovation by utilizing a stochastic frontier analysis (SFA) model incorporating three input factors to calculate WEE, which accounts for stochastic noise and time-varying efficiency. It has been verified that WET can enhance water users’ awareness of water conservation, thereby improving WEE. This not only validates the reliability of some previous research findings but also, through its innovations, provides more nuanced insights for the future refinement of WET.”

After verifying the conclusions of previous literature, this paper examines whether WET exhibits heterogeneity by considering factors such as a city’s geographical location and water resource endowment. Cities are categorized into southern and northern regions, as well as whether they are part of the Yangtze River Economic Belt, with geographical location determining the inherent conditions of water resources. Cities are further classified into old industrial bases and non-old industrial bases based on their industrialization characteristics. Old industrial bases include the Liaoning Central-Southern Industrial Base and the Pearl River Delta Industrial Base, among others. Their economic development primarily relies on industrialization. Since industrial water use accounts for a significant portion of total urban water consumption, implementing WET in old industrial bases yields notable results, as their economic development requires substantial water resource inputs.

Other countries are also implementing WET to enhance WEE. Wheeler et al. (2020) highlighted that in Australia’s Murray-Darling Basin, a global benchmark for large-scale water rights trading, the cap-and-trade system reallocates water resources to higher-value uses, thereby enhancing regional economic and environmental resilience [36]. Unlike the bottom-up, market-driven development models in Australia and the United States, China adopts a top-down pilot model implemented through strong policy directives [37]. This approach enables rapid policy deployment and facilitates systematic data collection across pilot regions. While mature water markets typically rely on long-established principles of prior appropriation or riparian rights, China is simultaneously constructing its property rights system and market infrastructure. Moreover, the Chinese model emphasizes government oversight and ecological redlines to prevent market failures and ensure that water rights trading does not compromise environmental flows or social equity—a challenge also observed in other countries. Therefore, implementing WET requires careful consideration of multiple layers, including policy design, market mechanisms, and ecological protection.

Although the WET has had a relatively stable impact on WEE, the article emphasizes that the WET has led to increased water conservation awareness among users by regulating water prices [38]. Due to insufficient survey data on users’ water conservation awareness, we are unable to explore the mechanisms and pathways through which water conservation awareness operates. Future research should employ theoretical analysis and questionnaire surveys to collect data on users’ water conservation awareness, thereby further examining, from both theoretical and empirical perspectives, the pathways and mechanisms through which users’ water conservation awareness influences WET and WEE. Additionally, the classification of regions into humid and non-humid zones in this study could be enhanced by incorporating broader hydrological and drought indicators, as well as analyzing GDP based on sectoral breakdowns to examine the impact of water rights trading policies on water resource use efficiency. Limited by data availability and article length constraints, these aspects will be addressed in future research to further bridge these gaps.

6. Conclusions and Policy Recommendations

Given that water resources possess the characteristics of a public good and exhibit uneven distribution across regions, this has led to irrational water allocation, low resource utilization efficiency, and diminished water conservation awareness among residents. This study employs the Stochastic Frontier Analysis method to measure WEE across 271 Chinese cities from 2006 to 2023. A double difference model was employed to validate changes induced by WET. Empirical analysis demonstrates that the policy significantly enhances WEE, while multiple models and matching samples were used to conduct robustness tests on the conclusions. The results consistently demonstrate that this policy effect exhibits strong robustness, and parallel trend tests confirm that the impact stems specifically from the implementation of WET.

This study demonstrates through theoretical and econometric empirical analysis that policies exhibit regional variations based on geographical location. To ensure the effectiveness of WET, policymakers should establish additional incentives to enhance water use efficiency and improve water resource management systems, thereby enabling regional policies to achieve their objectives effectively. The pilot program of WET should be expanded in a categorized and stratified manner. Given the more pronounced policy effects in southern regions and humid areas, these regions can be prioritized as key zones for deepening and expanding the WET. For northern and non-humid regions, while advancing WET, it is crucial to strengthen complementary support policies, such as increasing investments in water-saving infrastructure and implementing differentiated subsidy schemes, to overcome the constraints imposed by natural water resource endowment. Governments should focus on fostering and enhancing market vitality within the water rights trading market. This can be achieved by simplifying administrative approval procedures, lowering market entry barriers, and encouraging the participation of diverse market entities. A vibrant market environment will enable price signals to function more effectively, thereby amplifying the positive impact of WET on WEE. Policy should aim to strengthen the water-saving benefits for water users. This includes establishing a reward and punishment mechanism linked to water-saving performance and providing financial subsidies or tax incentives for enterprises that adopt advanced water-saving technologies and utilize reclaimed water. This enhances the economic incentives for water users to conserve water, forming a positive feedback loop between policy implementation and technological advancement.”

At the same time, the Chinese government needs to further refine WET and promote the rational allocation of water resources. Equally important is the need to consider the impact of weather changes on water resources in the near future. The current climate and weather patterns are also undergoing constant changes, with drought conditions emerging in southern regions in recent years [39]. In fact, the threat of climate change and worsening pollution pose new challenges to maintaining the sustainable development of essential water resources. Significant disparities in resource endowments and economic development across different regions of China exacerbate these challenges.