Exploring the Triple Dividend Effect and Threshold Effect of Environmental Protection Tax: Evidence from Chinese Listed Companies

Abstract

This study uses financial data from 872 Chinese listed companies (2018–2022). It tests the triple dividend effect and threshold effect of China’s environmental protection tax (EPT) using high-dimensional fixed effects models and panel threshold models. We document that (1) EPT creates an environmental dividend for Chinese listed companies. It significantly reduces pollution emissions. A 1-unit tax increase reduces LnTPPE by 2.5%. (2) EPT creates a significant innovation dividend. It forces enterprises to improve the quality of authorized patents. A 1-unit tax increase raises patent technological complexity by 0.79%. (3) EPT creates an economic dividend. It significantly improves firm performance. A 1-unit tax increase raises relative corporate revenue by 38.1%. (4) EPT exerts significant threshold effects on micro-level triple dividend outcomes among Chinese listed companies. A heterogeneity analysis shows significant differences in threshold effects between non-heavily polluting and heavily polluting industries. This study confirms that China’s EPT generates a micro-level triple dividend effect alongside coexisting threshold effects for listed companies. This provides literature references for China to design and implement differentiated policies and offers a quantitative empirical case for implementing globally sustainable EPT strategies.

1. Introduction

Global climate crises and ecological deterioration are intensifying. The Porter Hypothesis [1] posits that appropriate environmental regulation can stimulate corporate innovation and enhance long-term competitiveness. Environmental taxes correct pollution behaviors through policy signals, theoretically improving ecological conditions, stimulating green technology innovation, and promoting economic growth [2,3]. As a crucial policy tool balancing economic growth and ecological protection, environmental taxes are widely regarded as vital pathways for sustainable development [4,5]. From the 1990s environmental tax reforms pioneered by Nordic countries to China’s implementation of the environmental protection tax law (EPT) in 2018, replacing its previous pollutant discharge fee system, significant variations in policy effectiveness have emerged since EPT’s 2018 enactment. Its potential multiple benefits and drawbacks have attracted extensive academic attention. Research shows divergent views on EPT’s “inhibitive effect” and “promotive effect” [6]. Environmental taxation’s inhibitory effect on carbon emissions is widely confirmed [7]. Chenghao et al. [8] verified EPT’s promotive effect on China’s regional economy. Deng et al. [9] identified a double dividend achieved through enhanced corporate resilience and carbon reduction. Kettner et al. [10] studied Austria’s carbon pricing revenue recycling mechanisms, demonstrating their potential for achieving a triple dividend: emission reduction, economic growth, and consumption equity. Pereira et al. [11] noted that appropriately using tax revenues for tax reductions and energy efficiency investments can achieve a triple dividend: emission reduction, economic promotion, and fiscal improvement. However, actual schemes miss opportunities by failing to optimize revenue usage.

The concept of the “double dividend effect” was proposed [12]. Its continuous development led to exploring the “triple dividend effect” [13,14]. Based on the above practical and theoretical backgrounds, this study defines EPT as achieving three simultaneous objectives: improving environmental quality (environmental dividend), promoting corporate green technology innovation (innovation dividend), and advancing economic development (economic dividend).

Current research limitations highlight the complexity of studying triple dividend mechanisms. There are three theoretical gaps: First, most studies focus solely on one or two dividend dimensions, fragmenting the intrinsic linkages among environmental, innovation, and economic effects. Second, traditional models over-rely on linear assumptions; most research examines only emission reduction effects [15,16], neglecting cross-period impacts on innovation quality (e.g., patent complexity) and firm value. Threshold effect research is fragmented [17,18], ignoring EPT’s threshold effects—for instance, the Porter Hypothesis cannot explain innovation effects when tax burden exceeds critical points. Third, research on industrial heterogeneity lags, lacking quantitative evidence on regulatory response differences between heavily and non-heavily polluting industries. Studies on policy transmission mechanisms remain fragmented, failing to establish a unified analytical framework for environmental taxes’ synergistic effects on pollution control, technological innovation, and value creation.

In practical research, theoretical shortcomings manifest in two aspects: First, existing studies simplistically treat EPT policy effects as exogenous shocks for empirical analysis [3,8,19,20], without constructing multidimensional empirical and robustness analyses. Second, policy implementation is studied as a “one-size-fits-all” strategy with singular evaluation dimensions. This theory–practice gap urgently necessitates building an integrated micro-mechanism analysis system.

China’s EPT, implemented in 2018, adopts differentiated tax rates and tax reduction measures. This design enhances the policy flexibility of EPT and provides a unique environment for studying corporate triple dividend mechanisms.

This study proposes five sets of hypotheses. H1-H3 verify the micro-level triple dividends of EPT (environmental emission reduction/green innovation/economic performance improvement), H4 reveals the nonlinear threshold effect of the triple dividends (environmental/innovation/economic dimensions critical points), and H5 confirms that the threshold responses of non-heavily polluting industries and heavily polluting industries are significantly different.

The distinctive contribution lies in establishing the first integrated analytical framework combining triple dividends, threshold effects, and industrial heterogeneity, providing a micro-empirical engine for precision regulation of environmental policy transitions.

This study is structured as follows: Section 1: EPT theory and practical background, research gaps, research questions, and innovations. Section 2: Developing the analytical framework and proposing research hypotheses. Section 3: Data sources (Chinese listed company panel data), variable definitions, model specification. Section 4: Analysis of triple dividend effects and threshold effects, robustness checks, and heterogeneity analysis. Section 5: Hypothesis validation, theoretical and practical contributions, limitations, and future research directions. Section 6: Summary of findings.

2. Literature Review and Theoretical Hypotheses

2.1. Literature Review on EPT’s Environmental Dividend

Research on EPT’s environmental dividend mainly focuses on the scale effect and structural effect of pollution reduction. Recent studies confirm EPT’s emission reduction effect [21,22,23]. Others study EPT’s role in promoting corporate ESG [24,25,26]. From a micro-enterprise research perspective and methodology, the majority of studies employ the difference-in-differences (DID) method to establish micro-enterprise empirical research [9,27,28]. These studies have empirically demonstrated, to a certain extent, the policy effectiveness of environmental taxes in controlling pollution reduction, achieving the environmental dividend effect. Theoretically, they primarily rely on the green dividend within the double dividend theory to explore how environmental goals are achieved [9,29,30]. In 1991, British economist David [13] formally proposed the “double dividend”. Bovenberg [14] further empirically demonstrates that environmental taxes affect economic development through two channels, yielding a double dividend.

In summary, the environmental dividend in this study stems from this origin. Current research has two limitations: (1) Over-reliance on linear models ignores the “threshold compensation” effect of EPT on pollution control. (2) The micro-level transmission mechanism is unclear. There is no empirical evidence on how EPT drives pollution control decisions through cost internalization.

2.2. Literature Review on EPT’s Innovation Dividend

Research on EPT’s innovation dividend has shifted from the “cost inhibition” paradigm to the “innovation compensation” paradigm [1,31]. In the context of environmental taxation, the balance between the cost effect and innovation offset effect manifests as a dynamic mechanism. EPT constrains enterprises through an environmental cost signaling mechanism, internalizing pollution externalities to mitigate the negative impacts of taxation [32,33]. This mechanism spurs enterprises to intensify investments in green technology R&D, leading to synergistic improvements in technological innovation and environmental performance [34,35]. By leveraging the “innovation offset effect,” EPT incentivizes enterprises to enhance their green technology innovation (GTI) capabilities, thereby neutralizing incremental environmental compliance costs. At the policy level, implementing tax credit transfer mechanisms can alleviate corporate tax burdens, with endogenous revenue growth directly strengthening GTI capacity. Furthermore, empirical studies indicate that green taxation exhibits no statistically significant impact on overall innovation, environmental innovation, or non-environmental innovation in developed economies in the short term [36]. Some studies [37,38,39,40] suggest EPT drives innovation through an “R&D subsidy effect,” but do not distinguish patent quality dimensions. Breakthrough research [41,42] found EPT can improve corporate patent technology, but lacks examination of technological synergy (e.g., cross-field patent portfolios). The recent literature [19] constructed a “knowledge spillover index,” proving EPT increases the proportion of cross-group patents via a “technological catch-up effect.” Gaps remain: measurement of technological complexity lags; traditional patent classification cannot capture patent depth; and industry heterogeneity mechanisms are unknown.

2.3. Literature Review on EPT’s Economic Dividend

Some studies [43,44,45] empirically prove EPT can promote economic growth. Studies also point to the double dividend effect (coexistence of environmental and economic dividends) [9,46,47,48], and the feasibility of a triple dividend [10,11]. Some even suggest a quadruple dividend effect for carbon taxes [49]. However, few studies examine the triple dividend effect of China’s EPT.

2.4. Literature Review on EPT’s Threshold Effect

Studies on threshold effects or nonlinear relationships of EPT [3,17,18,49,50] primarily focus on developed economies and OECD countries, with minimal research examining EPT impacts within individual nations and scarce investigation into its micro-level corporate effects. Classic research [51] proposed the “single threshold model,” showing EPT’s pollution suppression effect strengthens significantly beyond a threshold, but did not consider multidimensional threshold variables. Research [52] found EPT has U-shaped relationships with energy transition under different conditions. Current research neglects how EPT effects influence threshold effects or nonlinear relationships at the micro-enterprise level, with mechanisms of sectoral heterogeneity among micro-level enterprises remaining unclear.

2.5. Theory and Hypotheses

Building upon Porter’s Hypothesis [1], the double dividend theory of environmental taxation [43], this study examines the triple dividend effect and threshold effect of China’s EPT. It reveals the micro-level mechanisms through which environmental taxes influence corporate performance via multiple pathways. By internalizing pollution costs through policy signals [3,21,22,23], it directly suppresses emissions, spurs corporate green technology innovation [34,35], and ultimately achieves scale economy effects [10,11,43,44,45]. Therefore, Hypotheses 1, 2, and 3 are proposed:

H1. 

The implementation of EPT has significantly reduced corporate pollution emission levels.

H2. 

EPT has significantly improved green technological innovation.

H3. 

EPT has a significant positive effect on corporate performance.

Hansen [51] pioneered the systematic analytical framework for fixed-effect panel threshold models, establishing the methodological foundation for related research. This framework has been widely applied and significantly extended in subsequent studies. The recent literature [53,54,55] continues to deepen its applications, focusing particularly on identifying and estimating more complex conditional threshold effects, with threshold theory continuing to evolve. Therefore, to verify the nonlinear relationships of the aforementioned EPT triple dividend effects in this study, Hypothesis 4 is proposed:

H4a. 

EPT’s pollution reduction effect has a threshold effect.

H4b. 

EPT’s technological innovation effect has a threshold effect.

H4c. 

EPT’s firm performance enhancement effect has a threshold effect.

EPT imposes cost transmission pressures, causing enterprises with heavily polluting versus non-heavily polluting industries characteristics to exhibit differing resilience capacities. This leads to significant divergence in their thresholds and effect intensities in responding to policy shocks. Thus, Hypothesis 5 is proposed in this study:

H5. 

The threshold effects of EPT’s triple dividends differ significantly between non-heavily polluting industries and heavily polluting industries.

By matching China’s corporate pollution database, financial data, and patent data, this study provides micro-level evidence for the “Porter Effect” of EPT and extends the double dividend to a triple dividend effect. It offers a theoretical basis and practical insights for optimizing EPT design and promoting the sustainable development of enterprises under the “Dual Carbon” goals.

3. Research Design

3.1. Data

The raw data comes from annual reports of listed companies. It was compiled and calculated by the “PPMANDATA” platform, sourced from the PPMANDATA database. Second, EPT data was collected. This involved selecting and extracting EPT data published by China’s A-share listed companies. The data covers the period from 2018 to 2022. Companies were classified based on whether they belong to heavily polluting industries. Companies with excessive missing values were excluded. Some missing values were filled using a regression prediction model. Some variables were sourced from the CSMAR database. Invention patent data came from the China National Intellectual Property Administration. Companies classified as (ST), (PT), and (*ST) were also excluded. Finally, a sample of 872 companies was obtained, totaling 4360 observations.

3.2. Variables

Table 1. Variable definitions and measurements.

Variable Types	Variable Name	Symbol	Explanation and Calculation Methods	Data Source
Core explained variable	Total phosphorus pollution equivalent	LnTPPE	Total phosphorus pollution equivalent, add 1 and take the natural logarithm	PPMANDATA
	Quality of authorized patents	QAP	The calculation method will be described in detail below	PPMANDATA
	Total operating income	LnTOI	Total operating income, natural logarithm	PPMANDATA
Core explanatory variables	Annual actual payment of environmental protection tax	LnAEPT	Annual actual EPT payment, natural logarithm	PPMANDATA
Explained variable	Water body comprehensive pollution equivalent	LnWCPE	Water body comprehensive pollution equivalent, add 1 and take the natural logarithm	PPMANDATA
	Quality of patent application	PAQ	The calculation method will be described in detail below	PPMANDATA
	Total profit	LnTP	Total profit, natural logarithm	PPMANDATA
Explanatory variable	Annual payment of environmental protection tax is required	LnREPT	Annual EPT payable, natural logarithm	PPMANDATA
Threshold variables	Total pollution equivalent	LnTPE	Overall pollution equivalent, add 1 and take the natural logarithm	PPMANDATA
Control variable	The age of a listed company	LnAge	The natural logarithm of the listing year plus 1	PPMANDATA
	Operating profit growth rate	OPGR	Operating profit growth rate based on the previous year	PPMANDATA
	Operating profit ratio	OPR	Operating profit divided by total profit	PPMANDATA
	Cash ratio	Cash	(Cash and marketable securities)/current liabilities	CSMAR
	Comprehensive tax rate	CTR	The sum of taxes payable, taxes and surcharges minus tax refunds, divided by total profit	PPMANDATA
	R&D investment amount	R&D	The natural logarithm of R&D investment	PPMANDATA
	Number of employees	LnNE	The natural logarithm of the number of employees in the enterprise	PPMANDATA
	Financial leverage	FL	Total debt to shareholders’ equity ratio	PPMANDATA
	ROA	ROA	Ratio of net profit to total assets	CSMAR
	Net intangible assets	LnNIA	Initial cost—accumulated amortization—impairment provision, finally take the natural logarithm	CSMAR
	Financial expenses	LnFE	Financial expenses take natural logarithm	PPMANDATA
	Shareholding ratio of top ten shareholders	SRTTS	The ratio of the total number of shares held by the top 10 shareholders to the total share capital	PPMANDATA
	Net profit	LnNP	The final profit of the company after deducting all costs, expenses, taxes, etc., is taken as the natural logarithm	PPMANDATA
	Total pollution equivalent	LnTPE	The chemical oxygen demand and ammonia nitrogen emissions in industrial wastewater, sulfur dioxide and nitrogen oxide emissions in industrial waste gas are standardized and converted into a unified pollution equivalent number, and the total is added with 1 to take the logarithm	PPMANDATA
	Debt-to-asset ratio	DAR	Total liabilities to total assets ratio	PPMANDATA

contains the names of the variables, their explanations and calculation methods, and the data source used. As introduced and demonstrated by Rao et al. [56], Shen & Zhang [57], and Zhou et al. [58], the study incorporates a comprehensive set of control variables that are carefully selected to summarize the operating and financial health of listed companies. The selection of these control variables is well supported in the literature.

This study uses the quality of patent applications as a proxy variable for the innovation dividend. It innovates in measuring both applied patent quality and granted patent quality. First, this study collects the main classification codes of patents held by listed companies. China uses the IPC patent classification system. The format is “Section—Class—Subclass—Main Group—Subgroup,” for example, “A01B01/00.” Simply counting the number of unique main classification codes in a firm’s patents does not accurately reflect internal differences in patent quality. It may overestimate patent quality. For example, Company A has three patents with main classification codes: A01B01/01, A01B01/02. Company B has three patents with main classification codes: A01B02/00, B02C03/00. Both companies have the same number of unique classification codes. However, Company A’s patents only utilize variations within the single main group A01B01. Company B’s patents utilize two distinct main groups: A01B02 and B02C03. Clearly, Company B’s patents demonstrate a broader and likely more advanced application of technology. Therefore, Company B has higher patent quality.

To address this, this study defines corporate patent quality at the main group level, drawing on the calculation concept of the Herfindahl Index:

$$Paten{t}_{it}= 1-\sum {\left({\displaystyle \frac{X_{mit}}{X_{it}}}\right)}^2$$

(1)

where Xmit is the cumulative number of invention and utility model patent applications by firm i in main group m up to year t; Xit is the cumulative number of all patent applications by firm i across all main groups up to year t.

A higher value of Patent_it indicates a higher quality of the firm’s patent applications. The same logic applies to measuring the quality of granted patents. In data processing for this metric, we only included invention patents and utility model patents. This is for two reasons: 1. Design patents use a completely different classification system from invention and utility model patents. They cannot be calculated using the same method. 2. Compared to invention and utility model patents, design patents reflect relatively lower independent innovation capability. They do not adequately represent the complexity of knowledge applied during patent creation. Therefore, design patents were excluded.

3.3. Descriptive Statistics of Variables

Table 2. Descriptive statistics.

Variable	Obs	Mean	Std. Dev.	Min	Max
LnTPPE	4360	6.777	0.213	6.292	7.185
QAP	4360	0.922	0.092	0.000	0.996
LnTOI	4360	22.525	1.511	18.389	28.83
LnAEPT	4360	14.519	1.54	9.968	20.892
LnWCPE	4360	0.142	0.002	0.134	0.149
PAQ	4360	0.924	0.089	0.000	0.996
LnTP	4360	19.881	1.596	12.804	26.086
LnREPT	4360	14.698	1.539	10.143	21.066
LnTPE	4360	0.148	0.002	0.143	0.153
LnAge	4360	2.741	0.368	1.946	3.434
OPGR	4360	−0.838	36.560	−1909.852	535.324
OPR	4360	1.011	2.452	−21.704	155.451
Cash	4360	0.478	0.574	0.004	7.609
CTR	4360	0.296	2.411	−108.161	59.813
R&D	4360	19.180	1.275	11.125	24.410
LnNE	4360	8.438	1.195	3.367	13.253
FL	4360	1.525	4.634	−6.365	270.994
ROA	4360	0.030	0.081	−1.856	0.759
LnNIA	4360	19.665	1.595	11.64	26.314
LnFE	4360	18.148	1.612	9.053	24.049
SRTTS	4360	53.53	13.744	8.78	97.93
LnNP	4360	19.724	1.574	12.506	25.823
DAR	4360	0.472	0.18	0.047	0.994

contains 4360 observations. The number of observations for all variables is 4360. This indicates the dataset is complete with no missing values. The data is described in detail as follows:

Water Pollution Equivalents: Comprehensive water pollution equivalent and total pollution equivalent show very low volatility.

Patent Quality: High quality but contains zero values. The relatively small standard deviation indicates overall high patent quality among sample listed companies, with a relatively concentrated distribution.

Financial Scale: Operating revenue, total profit, and net profit are substantial. The mean values of the logarithms of operating revenue, total profit, and net profit, when converted back to original values, are very large. This shows the sample mainly consists of large- and medium-sized enterprises. The relatively large standard deviations indicate significant size differences between Chinese listed companies.

Environmental Tax Burden: The mean actual paid EPT (14.519) is slightly lower than the mean payable EPT (14.698). This suggests that sample Chinese listed companies, overall, exhibit slight tax avoidance or benefit from tax incentives. The large standard deviation indicates significant differences in environmental tax burden between Chinese listed companies. The wide range between max and min values further confirms substantial variation in EPT burden across Chinese listed companies.

Financial Performance and Risk: The OPGR fluctuates greatly, and there are significant differences between different companies. Cash ratio, average cash short-term debt repayment capacity is acceptable, there are tax incentives/penalties in the comprehensive tax rate, and R&D investment is relatively concentrated. ROA: Mean 0.030, standard deviation 0.081. Min −1.856, Max 0.759. Overall profitability is low and highly volatile, with significant losses in some Chinese listed companies.

Financial Leverage and Risk: Financial Leverage: Mean 1.525 appears acceptable, but the standard deviation is very large (4.634). Min −6.365 indicates negative shareholder equity with a large absolute value. Max 270.994 shows liabilities far exceeding assets.

Asset–Liability Ratio: Mean 47.2%, within the commonly considered reasonable range of 40–60%.

Financial Expenses: Mean 18.148, standard deviation 1.612, indicating significant variation in interest burdens among Chinese listed companies.

These descriptive insights lay the foundation for a more detailed examination of the impact of EPT on corporate micro-performance and preliminarily reveal the mysteries of corporate environmental management, financial stability, and operations, laying the foundation for in-depth empirical exploration.

3.4. Triple Dividend Effect Model Construction

The two-way fixed effects (TWFE) model is a robust empirical tool for testing policy effects. It is a standard and powerful method for handling panel data. It controls unobservable time-invariant heterogeneity and time trends. This makes it highly suitable for evaluating the effect of the EPT, effectively mitigating omitted variable bias.

Therefore, this study directly uses the TWFE model to verify the triple dividend effect of the EPT. It also controls for firm, year, and whether the firm belongs to a heavily polluting industry. Three TWFE models are constructed as follows:

$$Y_{ijt}= {\alpha }_0+ {\alpha }_1{X}_{ijt}+ {{\displaystyle \sum }}_{k=1}^n{\alpha }_{k}contro{l}_{i,t} + {\mu }_i+{\gamma }_t+ {\varphi }_j+ {\epsilon }_{itj}$$

(2)

$$W_{ijt}={\beta }_{0t}+{\beta }_1{X}_{ijt}+{{\displaystyle \sum }}_{k=1}^n{\beta }_{k}contro{l}_{i,t}+{\mu }_i+{\gamma }_t+{\varphi }_j+{\epsilon }_{itj}$$

(3)

$$Z_{ijt}={\delta }_0+{\delta }_1{X}_{ijt}+{{\displaystyle \sum }}_{k=1}^n{\delta }_{k}contro{l}_{i,t}+{\mu }_i+{\gamma }_t+{\varphi }_j+{\epsilon }_{itj}$$

(4)

where Y_ijt is the dependent variable for firm i in industry j in year t; α is the constant term; EPT_it is the core explanatory variable (EPT); μ_i represents the individual fixed effect (firm-specific); λ_t represents the time fixed effect (year-specific); θ_j represents the fixed effect for whether the firm belongs to a heavily polluting industry; and ε_ijt is the random error term.

Based on the above models, empirical analysis is conducted. Regressions are run both with and without control variables.

3.5. Threshold Effect Model Construction

This study models based on Hansen’s static panel threshold regression model. It explores the nonlinear relationship between the EPT and the triple dividend effects on Chinese listed companies. Existing studies confirm divergent impacts of GDP and industrialization on emission distributions [59]. According to the empirical evidence of Deng et al. [17], pollution emissions are used as the nonlinear threshold of taxation and green technology. The theoretical mechanism stems from compliance cost-to-profit elasticity. In this study, pollution equivalent directly determines EPT’s tax base (Article 7 of China’s environmental protection tax law). When pollution equivalent resides in the transition zone, EPT compels clean production to achieve cost savings, ultimately enhancing corporate profits. Therefore, this study uses the total pollution emission equivalent as the threshold of all dividend effects, with the symbol LnTPE. Hypothesis: For a specific threshold value γ, $\sigma$, $\theta$, when LnTPE ≤ $\mathsf{\gamma }/\sigma /\theta$ and when LnTPE > $\mathsf{\gamma }/\sigma /\theta$, the impact of EPT on the triple dividends is significantly different. The econometric model is constructed as shown in Equations (5)–(7).

$$Y_{it}={\alpha }_0+{\alpha }_{11}LnTP{E}_{it}\times I\left(LnTP{E}_{it}\le {\gamma }_1\right)+{\alpha }_{12}LnTP{E}_{it}\times I\left({\gamma }_1<LnTP{E}_{it}\le {\gamma }_2\right)+\cdots$$

$$+{\alpha }_{1n}LnTP{E}_{it}\times I({\gamma }_{n-1}<LnTP{E}_{it}\le {\gamma }_n)+{\alpha }_1{K}_{i,t-m}+{\alpha }_2{Q}_{it}+{\epsilon }_{itj}$$

(5)

$$W_{it}={\beta }_0+{\beta }_{11}LnTP{E}_{it}\times I\left(LnTP{E}_{it}\le {\sigma }_1\right)+{\beta }_{12}LnTP{E}_{it}\times I\left({\sigma }_1<LnTP{E}_{it}\le {\sigma }_2\right)+\cdots$$

$$+{\beta }_{1n}LnTP{E}_{it}\times I({\sigma }_{n-1}<LnTP{E}_{it}\le {\sigma }_n)+{\beta }_1{W}_{i,t-m}+{\beta }_2{Q}_{it}+{\epsilon }_{itj}$$

(6)

$$Z_{it}={\delta }_0+{\delta }_{11}LnTP{E}_{it}\times I\left(LnTP{E}_{it}\le {\theta }_1\right)+{\delta }_{12}LnTP{E}_{it}\times I\left({\theta }_1<LnTP{E}_{it}\le {\theta }_2\right)+\cdots$$

$$+{\delta }_{1n}LnTP{E}_{it}\times I({\theta }_{n-1}<LnTP{E}_{it}\le {\theta }_n)+{\delta }_1{Z}_{i,t-m}+{\delta }_2{Q}_{it}+{\epsilon }_{itj}$$

(7)

where $Y_{it}$, $W_{it}$, and $Z_{it}$ represent the environmental dividend threshold effect, innovation dividend threshold effect, and economic dividend threshold effect for firm i in period t, respectively; ${\alpha }_0$, ${\beta }_0$, and ${\delta }_0$ represent the individual effect; I(·) is the indicator function (it equals 0 if the inequality inside the brackets is false and 1 if it is true. Threshold Value One: ${\gamma }_1$, ${\gamma }_2$,…, ${\gamma }_n$. Threshold Value Two: ${\sigma }_1$, ${\sigma }_2$,…, ${\sigma }_n$. Threshold Value Three: ${\theta }_1$, ${\theta }_2$,…, ${\theta }_n$. These threshold values are unknown and are estimated during the regression process); $Q_{it}$ represents the control variables; $\epsilon$ is the random disturbance term; ${\alpha }_{11}$, ${\alpha }_{12}$,…, ${\alpha }_{1n}$; ${\beta }_{11}$, ${\beta }_{12}$,…, ${\beta }_{1n}$; and ${\delta }_{11}$, ${\delta }_{12}$,…, ${\delta }_{1n}$ represent the coefficients of the EPT’s effect on the firm’s triple dividends when the threshold variable falls into different threshold value intervals.

4. Results

4.1. Environmental Dividend Effect

Table 3. Regression results of environmental dividend effect.

Variables	Benchmark Regression	Variables	Replacing Core Explanatory Variable	Variables	Replacing Explained Variable
	(1)		(2)		(3)
	LnTPPE		LnTPPE		LnWCPE
LnAEPT	−0.0250 ***	LnREPT	−0.0251 ***	LnAEPT	−0.0002 **
	(0.0080)		(0.0080)		(0.0001)
CTR	0.0031 ***	CTR	0.0031 ***	CTR	0.0001 **
	(0.0010)		(0.0010)		(0.0001)
FL	0.0011 ***	FL	0.0011 ***	FL	−0.0001 ***
	(0.0002)		(0.0002)		(0.0001)
DAR	−0.1180 **	DAR	−0.1180 **	DAR	0.0002
	(0.0586)		(0.0586)		(0.0005)
OPGR	0.0001 **	OPGR	0.0001 **	OPGR	0.0001 **
	(0.0001)		(0.0001)		(0.0001)
OPR	0.0006 *	OPR	0.0006 *	OPR	−0.0001 ***
	(0.0004)		(0.0004)		(0.0004)
LnFE	0.0014	LnFE	0.0014	LnFE	0.0002
	(0.0039)		(0.0039)		(0.0004)
Cash	−0.0013	Cash	−0.0014	Cash	0.0002 *
	(0.0096)		(0.0096)		(0.0086)
Constant	7.1670 ***	Constant	7.1730 ***	Constant	0.1440 ***
	(0.1380)		(0.1400)		(0.0013)
ID FE	YES	ID FE	YES	ID FE	YES
Year FE	YES	Year FE	YES	Year FE	YES
Industry FE	YES	Industry FE	YES	Industry FE	YES
Observations	4360	Observations	4360	Observations	4360
R-squared	0.299	R-squared	0.299	R-squared	0.464

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1. We report robust standard errors in parentheses and cluster robust standard errors at the individual level.

contains the regression results of the environmental dividend effect. From Table 3, benchmark regression column (1), the coefficient for LnAEPT is significantly negative in the LnTPPE model. Including control variables, controlling for individual fixed effects, time fixed effects, and fixed effects for heavily polluting versus non-heavily polluting industries, and using standard errors clustered at the individual level, the coefficient of EPT is significantly negative. Therefore, the existence of an environmental dividend effect is confirmed, supporting H1.

4.2. Innovation Dividend Effect

From

Table 4. Regression results of innovation dividend effect.

Variables	Benchmark Regression	Variables	Replacing Core Explanatory Variable	Variables	Replacing Explained Variable
	(4)		(5)		(6)
	QAP		QAP		PAQ
LnAEPT	0.0079 ***	LnREPT	0.0079 ***	LnAEPT	0.0077 **
	(0.0030)		(0.0030)		(0.0033)
R&D	0.0086 ***	R&D	0.0086 ***	R&D	−0.0035
	(0.0016)		(0.0016)		(0.0022)
LnTPE	2.2970 **	LnTPE	2.2970 **	LnTPE	−1.2540
	(1.1050)		(1.1050)		(1.0540)
LnNIA	0.0061 *	LnNIA	0.0061 *	LnNIA	0.0060 *
	(0.0034)		(0.0034)		(0.0035)
LnAge	0.0370	LnAge	0.0370	LnAge	0.0574
	(0.0326)		(0.0326)		(0.0385)
FL	−0.0003	FL	−0.0003	FL	00003
	(0.0001)		(0.0001)		(0.0001)
LnFE	−0.0009	LnFE	−0.0009	LnFE	−0.0002
	(0.0012)		(0.0012)		(0.0021)
SRTTS	0.0002	SRTTS	0.0002	SRTTS	−0.0004 *
	(0.0002)		(0.0002)		(0.0002)
DAR	0.0121	DAR	0.0121	DAR	0.0010
	(0.0200)		(0.0200)		(0.0232)
ID FE	YES	ID FE	YES	ID FE	YES
Year FE	YES	Year FE	YES	Year FE	YES
Industry FE	YES	Industry FE	YES	Industry FE	YES
Constant	0.0892	Constant	0.0873	Constant	0.8140 ***
	(0.2030)		(0.2040)		(0.2160)
Observations	4360	Observations	4360	Observations	4360
R-squared	0.579	R-squared	0.579	R-squared	0.542

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1. We report robust standard errors in parentheses and cluster robust standard errors at the individual level.

, benchmark regression column (4), the LnAEPT coefficient is significantly positive in the QAP model. Including control variables, controlling for individual fixed effects, time fixed effects, and fixed effects for heavily polluting versus non-heavily polluting industries, and using standard errors clustered at the individual level, the coefficient of EPT is significantly positive. This confirms that EPT enhances corporate green technology innovation quality through the “Innovation Compensation Effect”. It fully verifies the double dividend mechanism of the “Porter Hypothesis”: the transmission path from emission reduction dividend to innovation dividend. Therefore, the existence of the innovation dividend effect is verified. We support H2.

4.3. Economic Dividend Effect Test

Based on

Table 5. Regression results of economic dividend effect.

Variables	Benchmark Regression	Variables	Replacing Core Explanatory Variable	Variables	Replacing Explained Variable
	(7)		(8)		(9)
	LnTOI		LnTOI		LnTP
LnAEPT	0.3810 ***	LnREPT	0.3810 ***	LnAEPT	0.1860 ***
	(0.0265)		(0.0264)		(0.0574)
R&D	0.0277 ***	R&D	0.0276 ***	R&D	0.1250 ***
	(0.0084)		(0.0084)		(0.0281)
LnNIA	0.1600 ***	LnNIA	0.1600 ***	LnNIA	0.1310 **
	(0.0355)		(0.0355)		(0.0527)
LnFE	0.0222 ***	LnFE	0.0222 ***	LnFE	−0.0018
	(0.0065)		(0.0065)		(0.0201)
SRTTS	0.0019 *	SRTTS	0.0019 *	SRTTS	0.0056 **
	(0.0010)		(0.0010)		(0.0027)
DAR	0.3820 *	DAR	0.3820 *	DAR	0.6610 *
	(0.213)		(0.213)		(0.366)
LnAge	−0.0657	LnAge	−0.0668	LnAge	0.0108
	(0.1560)		(0.156)		(0.4760)
FL	−0.0001	FL	−0.0001	FL	−0.0433 **
	(0.0006)		(0.0006)		(0.0174)
ROA	0.2670	ROA	0.2680	ROA	0.1590
	(0.4210)		(0.4210)		(0.4120)
LnTPE	1.7010	LnTPE	1.7190	LnTPE	15.1800
	(3.4600)		(3.4560)		(12.4000)
ID FE	YES	ID FE	YES	ID FE	YES
Year FE	YES	Year FE	YES	Year FE	YES
Industry FE	YES	Industry FE	YES	Industry FE	YES
Constant	12.5500 ***	Constant	12.4800 ***	Constant	9.4340 ***
	(1.0500)		(1.0500)		(2.5580)
Observations	4360	Observations	4360	Observations	4360
R-squared	0.982	R-squared	0.982	R-squared	0.790

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1. We report robust standard errors in parentheses and cluster robust standard errors at the individual level.

, benchmark regression column (7), the LnAEPT coefficient is significantly positive in the LnTOI model. Including control variables, controlling for individual fixed effects, time fixed effects, and fixed effects for heavily polluting versus non-heavily polluting industries, and using standard errors clustered at the individual level, the coefficient of EPT is significantly positive. This confirms the existence of an economic dividend from EPT. This breaks the misconception that “environmental protection harms profitability”. Therefore, the existence of the economic dividend effect is verified. We support H3.

4.4. Robustness Tests for Triple Dividend Effects

To verify the robustness of the benchmark regression results, we conducted a series of tests, including replacing the core explanatory variable, replacing the explained variable, and conducting placebo tests.

(1)

Replacing core explanatory variable

Replace the explanatory variable from LnAEPT to LnREPT. The regression results are shown in Table 3, column (2), Table 4, column (5), and Table 5, column (8). The coefficients are significant at the 1% level. It verifies again that the benchmark regression results of this study are robust.

(2)

Replacing explained variable

As shown in Table 3, column (3), replace the total phosphorus pollution equivalent in the environmental dividend effect with the comprehensive water pollution equivalent. The regression coefficient is significantly negative at the 5% level, confirming the environmental dividend effect. As shown in Table 4, column (6), replace QAP in the innovation dividend effect with PAQ. The regression coefficient is significantly positive at the 5% level, confirming the innovation dividend effect. As shown in Table 5, column (9), replace LnTOI in the economic dividend effect with LnTP. The regression coefficient is significantly positive at the 1% level, confirming the economic dividend effect. These results show that the benchmark regression results of this study are robust.

(3)

Placebo test

The placebo test is shown in Figure 1. This test is designed to verify the robustness of the EPT’s impact on the triple dividend effect. The experimental group is randomly simulated 1000 times. Most of the estimated p-values exceed 0.1, which is not statistically significant. The estimated values follow a normal distribution. The placebo estimated coefficient is significantly different from the benchmark regression coefficient. This result shows that the results of this study are not accidental and are not affected by other policies or random factors.

4.5. Triple Dividend Endogeneity Treatment

To alleviate potential endogeneity issues, the first lag of the core explanatory variable was used as an instrumental variable and added to the model regression [60,61,62]. The regression results are shown in

Table 6. Robustness tests for triple dividend effects.

Variables	(10)	Variables	(11)	Variables	(12)
	LnTPPE		QAP		LnTOI
LnAEPT	−0.0241 ***	LnAEPT	0.0077 **	LnAEPT	0.3290 ***
	(0.0081)		(0.0031)		(0.0213)
ID FE	YES	ID FE	YES	ID FE	YES
Year FE	YES	Year FE	YES	Year FE	YES
Industry FE	YES	Industry FE	YES	Industry FE	YES
Constant	7.2210 ***	Constant	0.0765	Constant	9.4480 ***
	(0.1740)		(0.2000)		(0.9130)
Observations	4360	Observations	4360	Observations	4360
R-squared	0.299	R-squared	0.579	R-squared	0.987

Notes: *** p < 0.01, ** p < 0.05, We report robust standard errors in parentheses and cluster robust standard errors at the individual level.

, columns (10) to (12). After including the instrumental variable, the main regression coefficients are significant at the 1%, 5%, and 1% levels, respectively. This indicates that the EPT has a triple dividend effect on Chinese listed companies. It further confirms the acceptance of H1, H2, and H3.

4.6. Threshold Effect Test

Using Stata 18.0 software, the threshold effects of the EPT on the environmental dividend, innovation dividend, and economic dividend were tested separately. The Bootstrap method was employed with 300 repetitions to confirm the number of thresholds. The results of the threshold effect test are shown in

Table 7. Threshold test results and threshold values.

Threshold Variable	Category	F-Statistic	p-Value	10%	5%	1%
Environmental Dividend (LnTPE)	Single threshold	884.75	0.0000	34.5873	38.9440	44.1797
	Double threshold	278.22	0.0000	12.9546	15.7757	19.3122
	Triple threshold	118.88	0.5033	150.8915	159.6774	177.4139
Innovation Dividend (LnTPE)	Single threshold	65.66	0.0000	22.6174	24.9470	31.0908
	Double threshold	11.89	0.1900	14.8077	17.2050	22.0554
	Triple threshold	5.77	0.6900	14.4380	17.1205	24.1672
Economic Dividend (LnTPE)	Single threshold	54.78	0.0167	41.8945	45.6052	57.1272
	Double threshold	30.10	0.0167	21.1265	25.0061	32.5627
	Triple threshold	11.40	0.3367	16.2771	18.9824	27.8074

. It can be seen from Table 7 that EPT has a double threshold effect on the environmental dividend effect, a single threshold effect on the innovation dividend, and a double threshold effect on the economic dividend effect.

4.7. Threshold Effect Regression Results

According to the threshold effect test results, this study uses a double threshold model to estimate the threshold effect of EPT on environmental dividends and economic dividends, and uses a single threshold model to estimate the threshold effect of innovation dividends. The results are shown in

Table 8. Threshold results.

Threshold Variable	Category	Threshold	F-Statistic	p-Value	10%	5%	1%
Environmental Dividend (LnTPE)	Double threshold	0.1461	883.68	0.0000	31.9842	37.5101	40.7290
		0.1481	278.71	0.0000	14.2754	16.5016	22.3186
Innovation Dividend (LnTPE)	Single threshold	0.1469	65.66	0.0000	21.6471	24.6141	29.6358
Economic Dividend (LnTPE)	Double threshold	0.1482	54.78	0.0167	39.5499	44.9802	56.7865
		0.1493	30.10	0.0200	21.8293	23.9698	33.6916

and

Table 9. Regression results of threshold model.

Variables	(13)	Variables	(14)	Variables	(15)
	LnTPPE		QAP		LnTOI
${\gamma }_i$ ≤ 0.1461	−0.0386 ***	${\sigma }_i$ ≤ 0.1469	0.0069 **	${\theta }_i$ ≤ 0.1482	0.3700 ***
	(0.0076)		(0.0029)		(0.0260)
0.1461 < ${\gamma }_i$ ≤0.1481	−0.0289 ***	${\sigma }_i$ > 0.1469	0.0081 ***	0.1482 < ${\theta }_i$ ≤ 0.1493	0.3740 ***
	(0.0076)		(0.0029)		(0.0261)
${\gamma }_i$ > 0.1481	−0.0202 ***			${\theta }_i$ > 0.1493	0.3800 ***
	(0.0075)				(0.0265)
Constant	7.153 ***	Constant	0.461 ***	Constant	13.300 ***
	(0.1410)		(0.0780)		(0.9660)
Observations	4360	Observations	4360	Observations	4360
Number of id	872	Number of id	872	Number of id	872
R-squared	0.228	R-squared	0.048	R-squared	0.513

Notes: Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05.

.

Table 8 shows under the double threshold for the variable LnTPE. The F-statistic of the first threshold value (0.1461) of environmental dividend is 883.68, significant at the 1% level. The F-statistic of the second threshold value (0.1481) of environmental dividend is 278.71, significant at the 1% level. The F-statistic of the single threshold value (0.1469) of innovation dividend is 65.66, significant at the 1% level. The F-statistic of the first threshold value (0.1482) of economic dividend is 54.78, significant at the 5% level. The F-statistic of the second threshold value (0.1493) of economic dividend is 30.10, significant at the 5% level.

Table 9 shows the regression results of the triple dividend threshold effect. Specifically, Environmental Dividend (Table 9, column 13): When LnTPE ≤ 0.1461, the coefficient of EPT is significantly negative. A one-unit increase in EPT leads to a significant decrease of approximately 0.0386 units in LnTPPE, representing the strongest inhibitory effect. When 0.1461 < LnTPE ≤ 0.1481, the coefficient of EPT is significantly negative. EPT still significantly inhibits pollution emission, but the inhibitory effect weakens. When LnTPE > 0.1481, the coefficient of EPT is significantly negative. This indicates that for enterprises with inherently higher pollution levels, EPT still significantly inhibits their pollution emission, but the inhibitory effect further weakens.

Innovation Dividend (Table 8, column 20): When LnTPE ≤ 0.1469, the coefficient of EPT is significantly positive. For enterprises with lower pollution levels, a one-unit increase in EPT significantly increases QAP by approximately 0.0069 units. When LnTPE > 0.1469, the coefficient of EPT is significantly positive. This indicates that for enterprises with higher pollution levels, a one-unit increase in EPT significantly increases QAP by approximately 0.0081 units, and the increase magnitude is slightly higher than for the low-pollution group.

Economic Dividend (Table 8, column 21): When LnTPE ≤ 0.1482, the coefficient of EPT is significantly positive. A one-unit increase in EPT leads to a significant increase of approximately 0.3700 units in LnTOI. When 0.1482 < LnTPE ≤ 0.1493, the coefficient of EPT is significantly positive. EPT still significantly increases operating income, and the increase magnitude is slightly higher than for the low-pollution group. When LnTPE > 0.1493, the coefficient of EPT is significantly positive. This indicates that for enterprises with higher pollution levels, a one-unit increase in EPT significantly increases operating income by approximately 0.3800 units, and the increase magnitude is the highest. In summary, this study verified H4.

Figure 2 shows the Likelihood Ratio (LR) function graph for the threshold effects of the triple dividends on Chinese listed companies. It can be seen that the threshold value for the firm’s triple dividends crosses the LR value (LR = 7.35). These threshold values are significant at the 95% confidence level. This indicates that the EPT has a threshold effect on all three dividends (environmental, innovation, economic) for Chinese listed companies.

4.8. Robustness Test for Threshold Effects

To determine if the threshold regression results are robust, this study continues to use the method of replacing explanatory variables for re-estimation and replaces LnAEPT with LnREPT for regression analysis. The results are shown in

Table 10. Threshold effect robustness test.

Variables	(16)	Variables	(17)	Variables	(18)
	LnTPPE		QAP		LnTOI
${\gamma }_i$ ≤ 0.1461	−0.0384 ***	${\sigma }_i$ ≤ 0.1469	0.0070 **	${\theta }_i$ ≤ 0.1482	0.3700 ***
	(0.0076)		(0.0029)		(0.0259)
0.1461 < ${\gamma }_i$ ≤ 0.1481	−0.0289 ***	${\sigma }_i$ > 0.1469	0.0082 ***	0.1482 > ${\theta }_i$ ≤ 0.1493	0.3740 ***
	(0.0076)		(0.0029)		(0.0261)
${\gamma }_i$ > 0.1481	−0.0203 ***			${\theta }_i$ > 0.1493	0.3800 ***
	(0.0075)				(0.0264)
Constant	7.1580 ***	Constant	0.4590 ***	Constant	13.2200 ***
	(0.1420)		(0.0782)		(0.9660)
Observations	4360	Observations	4360	Observations	4360
Number of id	872	Number of id	872	Number of id	872
R-squared	0.228	R-squared	0.048	R-squared	0.513

Notes: Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05.

.

Overall, the estimated coefficients from the threshold regression still exhibit remarkable consistency, further validating the robustness of the threshold regression results.

4.9. Threshold Effect Endogeneity Treatment

To further alleviate potential endogeneity issues, the first lag of the core explanatory variable was used as an instrumental variable and added to the threshold regression. The threshold regression results are shown in

Table 11. Endogenous treatment of threshold effect.

Variables	(19)	Variables	(20)	Variables	(21)
	LnTPPE		QAP		LnTOI
${\gamma }_i$ ≤ 0.1461	−0.0388 ***	${\sigma }_i$ ≤ 0.1469	0.0064 **	${\theta }_i$ ≤ 0.1482	0.3180 ***
	(0.0076)		(0.0029)		(0.0209)
0.1461 < ${\gamma }_i$ ≤ 0.1481	−0.0291 ***	${\sigma }_i$ > 0.1469	0.0076 ***	0.1482 < ${\theta }_i$ ≤ 0.1493	0.3210 ***
	(0.0075)		(0.0029)		(0.0209)
${\gamma }_i$ > 0.1481	−0.0204 ***			${\theta }_i$ > 0.1493	0.3250 ***
	(0.0075)				(0.0211)
Constant	7.1410 ***	Constant	0.4380 ***	Constant	10.6200 ***
	(0.1750)		(0.0784)		(0.8170)
Observations	4360	Observations	4360	Observations	4360
Number of id	872	Number of id	872	Number of id	872
R-squared	0.228	R-squared	0.048	R-squared	0.649

Notes: Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05.

, columns (19) to (21). After including the instrumental variable, the threshold effect regression coefficients are all significant. This indicates that the EPT has a threshold effect on the triple dividends.

4.10. Heterogeneity Analysis: Heavily Polluting vs. Non-Heavily Polluting Industries

This study further explores differences in threshold effect identification by conducting heterogeneity tests of the threshold effects. With reference to studies by Zhou et al. [63] and Zor [64], and according to the List of Environmental Verification Industry Classifications for Listed Companies issued by the Ministry of Environmental Protection of the People’s Republic of China (now the Ministry of Ecology and Environment) on 24 June 2008, 19 sub-industries within mining, manufacturing, and the production and supply of electricity, heat, gas, and water are classified as heavily polluting industries. Consequently, this study categorizes the sample into heavily polluting and non-heavily polluting industries based on the latest industry classification codes issued by the China Securities Regulatory Commission (CSRC). We then conduct heterogeneity tests.

4.10.1. Heterogeneity of Environmental Dividend Threshold Effect: Heavily Polluting vs. Non-Heavily Polluting Industries

The threshold effects and regression results for the environmental dividend in non-heavily polluting industries and heavily polluting industries are shown in

Table 12. Test results of threshold effect of environmental dividend heterogeneity in non-heavily polluting and heavily polluting industries.

Environmental Dividend Double Threshold	Non-Heavily Polluting Industries
	Threshold	F-Statistic	p-Value	10%	5%	1%
	0.1462	621.93	0.0000	33.8913	37.7737	43.1968
	0.1493	164.22	0.0000	12.4986	14.6075	18.0003
	Heavily Polluting Industries
	Threshold	F-Statistic	p-Value	10%	5%	1%
	0.1461	285.09	0.0000	12.9017	14.9197	20.7654
	0.1481	66.79	0.0000	9.5923	12.0635	18.3548

and

Table 13. Regression results of the threshold effect of environmental dividend heterogeneity in non-heavily polluting and heavily polluting industries.

Variables	Non-Heavily Polluting Industries	Variables	Heavily Polluting Industries
	(22)		(23)
	LnTPPE		LnTPPE
${\gamma }_i$ ≤ 0.1462	−0.0337 ***	${\gamma }_i$ ≤ 0.1461	−0.0539 ***
	(0.0091)		(0.0156)
0.1462 < ${\gamma }_i$ ≤ 0.1493	−0.0218 **	0.1461 < ${\gamma }_i$ ≤ 0.1481	−0.0454 ***
	(0.0090)		(0.0156)
${\gamma }_i$ > 0.1493	−0.0128	${\gamma }_i$ > 0.1481	−0.0359 **
	(0.0090)		(0.0155)
Constant	6.9560 ***	Constant	7.5590 ***
	(0.1710)		(0.2740)
Observations	3005	Observations	1295
Number of id	601	Number of id	259
R-squared	0.226	R-squared	0.229

Notes: Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05.

.

Non-heavily polluting industries: When the environmental dividend threshold variable is below the threshold value of 0.1462, the regression coefficient of EPT on environmental dividend is significantly negative at the 1% level, indicating the strongest inhibitory effect. When exceeding the threshold of 0.1493, the coefficient becomes statistically insignificant, reflecting a weakened inhibitory effect.

Heavily polluting industries: When the environmental dividend threshold variable is below the threshold value of 0.1461, the regression coefficient of EPT on environmental dividend is significantly negative at the 1% level. The inhibitory effect weakens when exceeding the threshold values. Regardless of the EPT level, it has a negative effect on the water pollution level of both types of industries. However, the effect is more sensitive for heavily polluting industries. The suppression effect becomes stronger for them.

There is a significant heterogeneous difference compared to non-heavily polluting industries. This indicates a significant change in the heterogeneous impact of the EPT on the first environmental dividend threshold for heavily polluting industries.

Figure 3 shows the Likelihood Ratio (LR) function graphs for the environmental dividend threshold effect in non-heavily polluting industries and heavily polluting industries. The double threshold value for the environmental dividend in both non-heavily polluting industries and heavily polluting industries crosses the critical LR value (LR = 7.35). This indicates the threshold value is significant at the 95% confidence level. Therefore, the EPT has a double threshold effect on the environmental dividend for both non-heavily polluting industries and heavily polluting industries.

4.10.2. Heterogeneity of Innovation Dividend Threshold Effect: Heavily Polluting vs. Non-Heavily Polluting Industries

The threshold effects and regression results for the innovation dividend in non-heavily polluting industries and heavily polluting industries are shown in

Table 14. Test results of threshold effect of innovation dividend heterogeneity in non-heavily polluting and heavily polluting industries.

	Non-Heavily Polluting Industries
Innovation Dividend Single Threshold Effect	Threshold	F-Statistic	p-Value	10%	5%	1%
	0.1469	58.06	0.0000	18.3296	20.2424	25.2858
	Heavily Polluting Industries
	Threshold	F-Statistic	p-Value	10%	5%	1%
	0.1460	18.33	0.0333	14.8221	17.0088	22.9545

and

Table 15. Regression results of threshold effect of innovation dividend heterogeneity in non-heavily polluting and heavily polluting industries.

Variables	Non-Heavily Polluting Industries	Variables	Heavily Polluting Industries
	(24)		(25)
	QAP		QAP
${\sigma }_i$ ≤ 0.1469	0.0062 *	${\sigma }_i$ ≤ 0.1460	0.0074
	(0.0035)		(0.0061)
${\sigma }_i$ > 0.1469	0.0076 **	${\sigma }_i$ > 0.1460	0.0089 *
	(0.0035)		(0.0060)
Constant	0.5150 ***	Constant	0.3660 **
	(0.0868)		(0.1720)
Observations	3005	Observations	1295
Number of id	601	Number of id	259
R-squared	0.049	R-squared	0.049

Notes: Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

.

Non-heavily polluting industries: When the innovation dividend threshold variable is below 0.1469, the regression coefficient of EPT on innovation dividend is positive but statistically significant only at the 10% level, indicating the weakest stimulative effect. When exceeding the threshold, the coefficient becomes significant at the 5% level with an enhanced stimulative effect.

Heavily polluting industries: When the innovation dividend threshold variable is below 0.1460, the regression coefficient of EPT on innovation dividend is positive but statistically insignificant. When exceeding the threshold, the coefficient gains significance at the 10% level, demonstrating a strengthened stimulative effect. Specifically, the effect is more sensitive for heavily polluting industries. The promotion effect becomes stronger for them. There is a significant heterogeneous difference compared to non-heavily polluting industries.

Figure 4 shows the Likelihood Ratio (LR) function graphs for the innovation dividend threshold effect in non-heavily polluting industries and heavily polluting industries. The single threshold value for the innovation dividend in both non-heavily polluting industries and heavily polluting industries crosses the critical LR value (LR = 7.35). This indicates the threshold value is significant at the 95% confidence level. Specifically, EPT exhibits statistically distinguishable single threshold effects on the innovation dividend for both non-heavily and heavily polluting industries, with industry-specific threshold values.

4.10.3. Heterogeneity of Economic Dividend Threshold Effect: Heavily Polluting vs. Non-Heavily Polluting Industries

The threshold effects and regression results for the economic dividend in non-heavily polluting industries and heavily polluting industries are shown in

Table 16. Test results of threshold effect of economic dividend heterogeneity in non-heavily polluting and heavily polluting industries.

	Non-Heavily Polluting Industries
Economic Dividend Double Threshold Effect	Threshold	F-Statistic	p-Value	10%	5%	1%
	0.1481	36.96	0.0600	31.1280	37.9827	42.1181
	NO	14.54	0.2067	17.4473	21.6709	26.4534
	Heavily Polluting Industries
	Threshold	F-Statistic	p-Value	10%	5%	1%
	NO	24.42	0.1533	26.3754	31.6434	36.2200
	NO	7.64	0.7267	17.7372	20.1788	24.1648

and

Table 17. Regression results of the threshold effect of economic dividend heterogeneity in non-heavily polluting and heavily polluting industries.

Variables	Non-Heavily Polluting Industries	Variables	Heavily Polluting Industries
	(26)		(27)
	LnTOI		LnTOI
${\theta }_i$ ≤ 0.1481	0.3620 ***	${\theta }_i$ ≤ NO	0.3350 ***
	(0.0297)		(0.0306)
0.1481 < ${\theta }_i$ ≤ NO	0.3660 ***	NO < ${\theta }_i$ ≤ NO	0.3410 ***
	(0.0298)		(0.0306)
${\theta }_i$ > NO	0.3720 ***	${\theta }_i$ > NO	0.3480 ***
	(0.0303)		(0.0310)
Constant	13.5900 ***	Constant	13.4800 ***
	(1.2320)		(1.0290)
Observations	3005	Observations	1295
Number of id	601	Number of id	259
R-squared	0.473	R-squared	0.631

Notes: Robust standard errors in parentheses, *** p < 0.01.

.

Non-heavily polluting industries: When the economic dividend threshold variable is below 0.1481, the regression coefficient of EPT on economic dividend is significantly positive at the 1% level, representing the weakest stimulative effect. Beyond this threshold, a monotonic enhancement of the stimulative effect is observed, though no double threshold effect exists.

Heavily polluting industries: EPT demonstrates a general enhancement effect on economic dividend; however, no statistically significant threshold effect is detected.

Specifically, when exceeding the threshold value, the effect is more sensitive for non-heavily polluting industries. The promotion effect becomes stronger for them. There is a significant heterogeneous difference compared to heavily polluting industries. This indicates a significant change in the heterogeneous impact of the EPT on the economic dividend threshold for non-heavily and heavily polluting industries.

Figure 5 shows the Likelihood Ratio (LR) function graphs for the economic dividend threshold effect in non-heavily and heavily polluting industries. The single threshold value for the economic dividend in non-heavily polluting industries crosses the critical LR value (LR = 7.35). This indicates the threshold value is significant at the 95% confidence level. However, no statistically significant threshold effect on economic dividend has been detected for heavily polluting industries. Specifically, EPT demonstrates statistically distinct heterogeneous effects on the economic dividend threshold response between non-heavily and heavily polluting industries.

5. Discussion

5.1. Triple Dividend Hypothesis Validation Analysis

This study uses panel data from 872 Chinese listed companies (2018–2022). It verifies the triple dividend effect and threshold effect of the EPT using two-way fixed effects and threshold models. The results fully support hypotheses H1 to H5. They reveal the synergy mechanism between environmental regulation and corporate development.

Environmental Dividend (H1): The EPT significantly suppresses the LnTPPE. The coefficient is −0.0250 (p < 0.01). A 1-unit tax increase reduces LnTPPE by 2.5%. This aligns with Porter’s Hypothesis [1]. It shows that EPT strengthens emission reduction incentives. This matches studies finding that environmental taxes suppress pollution [3,21,22,23].

Innovation Dividend (H2): The elasticity of QAP to EPT is 0.0079 (p < 0.01). A 1-unit tax increase raises patent technological complexity by 0.79%. This breaks the traditional “pollution avoidance” theory. It reveals the micro-mechanism of the innovation compensation effect. It aligns with studies finding that EPT promotes innovation [31,34,35,65]. High-tax industries file cross-group patents (e.g., “B02C03 solid waste treatment” and “A01B02 emission reduction”). This diversifies risk.

Economic Dividend (H3): The elasticity of LnTOI to EPT is 0.381 (p < 0.01). A 1-unit tax increase raises LnTOI complexity by 38.1%. This confirms the “pollution control—market competitiveness” transformation path. It matches studies finding that EPT promotes economic growth and firm performance [10,11,43,44,45].

Robustness tests strengthen the conclusions. After replacing core variables and conducting placebo tests, the triple dividend effects remain significant. After instrumental variable methods, endogeneity issues did not distort the results. Therefore, H1, H2, and H3 have been confirmed.

5.2. Triple Dividend Threshold Effect Hypothesis Analysis

Using Hansen’s panel threshold model, significant threshold effects exist for the triple dividends. This verifies the nonlinear nature of H4a, H4b, and H4c.

Environmental Dividend Threshold: When LnTPE ≤ 0.1461, EPT exerts its strongest inhibitory effect. The emission reduction effect progressively weakens across the subsequent intervals of 0.1461 < LnTPE ≤ 0.1481 and LnTPE > 0.1481. This demonstrates that EPT generates significant emission reduction effects for enterprises at all pollution levels, with the magnitude diminishing as firms’ own pollution intensity increases. Enterprises with lower pollution levels exhibit greater sensitivity to EPT, resulting in stronger emission reduction outcomes. Conversely, heavily polluting firms demonstrate reduced sensitivity and weaker inhibitory effects. This pattern aligns with the marginal abatement cost theory in environmental economics [66], whereby low-pollution firms achieve cost-effective reductions through operational adjustments, while high-pollution enterprises face steep cost curves for substantial abatement due to technological constraints and infrastructure lock-in [18].

Innovation Dividend Threshold: When LnTPE ≤ 0.1469, EPT significantly enhances QAP for low-pollution enterprises. For firms with LnTPE > 0.1469, the innovation-stimulating effect becomes marginally stronger. Specifically, EPT exerts a statistically significant positive impact on corporate innovation quality, with a slightly more pronounced incentive effect observed in highly polluting firms. This phenomenon arises because heavily polluting enterprises face greater environmental compliance pressure. To ensure operational viability and meet emission standards, they exhibit stronger motivation to pursue breakthrough innovations that offset rising compliance costs [36]. These findings substantiate the positive aspect of the Porter Hypothesis, wherein properly designed environmental regulations can stimulate innovation competitiveness.

Economic Dividend Threshold: When LnTPE ≤ 0.1482, EPT’s coefficient exhibits statistical significance with a positive sign. This positive effect progressively strengthens across the intervals 0.1482 < LnTPE ≤ 0.1493 and LnTPE > 0.1493, where the average marginal effect increases from 1.0% to 1.6%. The enhancement effect reaches its most pronounced magnitude when LnTPE exceeds 0.1493. EPT demonstrates statistically significant positive impacts on corporate operating revenue. Critically, this revenue-enhancing effect intensifies monotonically as firms’ own pollution levels increase. This “strong nonlinearity” shows that low-pollution industries gain market premiums via green certification. High-pollution industries experience “suppress first, boost later.” Short-term pollution control costs rise, but long-term circular economy optimizes cost structure, enhancing performance [2,3,17,49,50]. For enterprises with pollution equivalent below 0.1493, provide interest-subsidized technology transition loans, including partial People’s Bank of China relending rate reductions. For medium–high polluters exceeding industry-average pollution intensity per unit output, activate tiered tax rates, penalties for exceeding thresholds, and reductions for compliance.

Heterogeneity analysis shows that the environmental dividend threshold effect is more significant for heavily polluting industries. Its threshold (0.1461, 0.1481) is lower than non-heavily polluting industries (0.1462, 0.1493). This reflects their urgency for pollution control. The threshold effect of the innovation dividend is also more reflected in heavily polluting industries. EPT is more sensitive to heavily polluting industries, with a lower threshold (0.1460) than that of non-heavily polluting industries (0.1469), and a stronger stimulating effect. However, the threshold effect of economic dividends is reflected in non-heavy pollution industries. EPT has a single threshold effect on non-heavy pollution industries, but no threshold effect on heavy pollution industries.

This study provides evidence that significantly different coefficients, significantly different threshold points, and passed statistical tests robustly support the existence of substantial, statistically significant differences in the mechanisms by which EPT affects environmental, innovation, and economic benefits between heavily polluting and non-heavily polluting industries.

Clear and economically logical potential mechanisms underlie these differences: First, pollution emission intensity constitutes the core mechanism for the environmental benefit difference. Second, sectoral disparities in technological opportunities and absorptive capacity are key to the innovation benefit difference. Third, the economic benefit difference primarily stems from the Porter effect (innovation compensation and efficiency enhancement).

This study effectively confirms the significant heterogeneity of EPT effects across different industries. Understanding these differences and their underlying mechanisms is crucial for formulating more targeted and effective environmental–economic policies.

5.3. Theoretical Contributions

This study achieves groundbreaking advances in environmental taxation theory:

Develops an analytical framework examining EPT’s nonlinear threshold effects on the triple dividend (environmental pollution reduction, innovation stimulation, economic enhancement). This model deciphers the micro-mechanisms through which EPT influences the triple dividend, transcending the traditional “cost-performance” dichotomy of the Porter Hypothesis. It incorporates novel dimensions—technological complexity and knowledge spillovers—into EPT theory.

Empirically validates the marginal abatement cost theory and environmental policy incentive theory in environmental economics, pioneering a nonlinear policy evaluation paradigm distinct from conventional linear assessment models.

Establishes an industry heterogeneity model revealing statistically distinct thresholds. China-specific classification of non-heavily/heavily polluting enterprises exhibits significant nonlinear threshold differentials. This uncovers industry-specific patterns of environmental policy efficacy, providing quantitative evidence for environmental taxation theory while advancing environmental regulation from “universal policy” toward “precision governance” frameworks.

5.4. Practical Contributions

This study’s findings will provide policymakers with three practical tools:

• Policy Implementation: Establish a “Dynamic Threshold Adjustment Mechanism.” Increase tax intensity for high-value-added polluters to enhance emissions reduction sensitivity. Implement an “EPT-Carbon Trading” linkage mechanism for large labor-intensive polluters to achieve precise matching of policy instruments and strengthen innovation linkages. Continuously raise tax intensity for high-pollution, high-emission industries to drive dual improvements in pollution reduction and performance. Specifically, implement stepped EPT rates for heavily polluting industries above the threshold. This enables precise regulatory control, reduces enforcement costs (avoiding “one-size-fits-all” enforcement), creates channels for green supply chain certification services, and leverages policy steering effects.
• Enterprise Management: Create a “Green Technology Investment Decision Model.” Guide industries to identify optimal technological pathways. Prioritize end-of-pipe treatment technologies for low-emission industries. Implement circular economy models for enterprises exceeding thresholds. Strategic advice for managers in heavily polluting industries is as follows: deploy abatement technologies (e.g., cleaner production equipment) proactively as EPT approaches the environmental threshold to avoid sharp cost increases from stricter regulation; increase green technology investments after exceeding the threshold; develop by-product recycling technologies; and pursue green certifications to enhance market premiums. Strategic advice for managers in non-heavily polluting industries: accelerate green technology R&D when EPT surpasses the innovation benefit threshold; establish joint ESG innovation labs with universities; patent environmental technologies; and for economic benefit strategies, develop environmental management tools to gain returns through asset-light models.

Regulatory agencies can implement the following measures to optimize EPT design: Implement differentiated threshold-based supervision: Strengthen enforcement intensity for heavily polluting industries, while focusing on innovation guidance for non-heavily polluting industries. Establish dynamic policy instruments: For heavily polluting industries, introduce stepped pollution discharge fees linked with carbon trading mechanisms. For non-heavily polluting industries, distribute innovation vouchers (redeemable for R&D subsidies when EPT exceeds thresholds). Create cross-industry collaboration mechanisms: Establish technology alliances between heavily polluting and non-polluting enterprises, permitting the purchase of innovation-driven emission reductions. Provide targeted incentives: Offer green credit and certification channels to enterprises achieving the economic threshold.

5.5. Research Limitations

The empirical analysis has limitations: Methodologically, this study employs conventional panel threshold models, where the non-adoption of dynamic threshold models may cause dynamic bias in estimation. Instrumental variables are not sought in historical and geographical conditions, and only lagged explanatory variables in the time series are used as instrumental variables, which may not completely solve the endogeneity problem. Data: Limited by listed company disclosure, emerging variables like ESG investment and digital tech penetration were not included. This may underestimate EPT’s effect. Mechanism Analysis: Tech transmission paths rely on secondary patent data. Lacks field observation of micro-behaviors like R&D personnel flow and industry–university–research collaboration. Policy Evaluation: Does not distinguish regional environmental capacity differences. May overestimate policy dividends. Industry Classification: Industries are simply classified as heavily/non-heavily polluting. Does not consider sub-industries (e.g., new vs. traditional energy) and their heterogeneous responses. Duration: The long-term persistence of EPT effects needs more than 5 years of observation. This may affect the completeness of policy impact assessment.

5.6. Future Research Directions

• Build a dynamic threshold model. Study how EPT can enhance the efficiency of tax collection and administration.
• Use satellite remote sensing (e.g., night light data) and IoT sensors to build real-time monitoring networks. Develop minute-level policy effect assessment systems.
• Conduct cross-country comparative studies. Analyze optimal EPT–carbon price linkage ratios in OECD countries. Explore “Pollution Equivalent Banking” systems. Allow industries to accumulate emission reduction credits through tech upgrades. Provide a Chinese solution for the sustainable development of global climate governance.

6. Conclusions

This study shows that the environmental protection tax (EPT) generates synergistic effects of economic, environmental, and technological innovation dividends for Chinese listed companies. It partially validates the Porter Hypothesis.

• Environmental Dividend Achieved: The EPT significantly suppresses pollution emissions. A 1-unit tax increase reduces LnTPPE by 2.5%. This verifies the tax’s precise emission reduction capability.
• Significant Technological Innovation Dividend: The EPT forces Chinese listed companies to improve patent application and grant quality. A 1-unit tax increase raises patent technological complexity by 0.79%. It drives green innovation through increased patent complexity and cross-field technology synergy.
• Economic Dividend Confirmed: The EPT significantly improves companies’ performance. A 1-unit tax increase raises relative corporate revenue by 38.1%.
• Heterogeneity analysis shows significant differences in threshold effects between non-heavily polluting and heavily polluting industries.

This study confirms that China’s EPT generates a micro-level triple dividend effect alongside coexisting threshold effects for listed companies. This provides literature references for China to design and implement differentiated policies and offers a quantitative empirical case for implementing globally sustainable EPT strategies.