From Investment to the Environment: Exploring the Relationship between the Coordinated Development of Two-Way FDI and Carbon Productivity under Fiscal Decentralization

Abstract

The governance exerted by governments plays a pivotal role not only in driving local economic advancement but also in bolstering environmental management and enhancing Carbon Productivity (CP). This paper investigates the impact of two-way Foreign Direct Investment (FDI) coordination development (DFDI) on China’s CP from the perspective of fiscal decentralization (FD). Utilizing panel data from 30 Chinese provinces spanning 2006–2020, we apply a Spatial Error Model to discern that DFDI effectively elevates CP. However, an excessively high degree of FD constrains the potential environmental performance benefits that FDI might offer. Further analysis using a Dynamic Threshold Model reveals a significant dynamic non-linearity in the impact of DFDI on CP under the threshold effect of FD. In contrast to Inward FDI (IFDI), China’s Outward FDI (OFDI) actually impedes the enhancement of CP. Our results underscore that well-calibrated FD can align economic growth with environmental sustainability. This study offers insights into policy frameworks fostering sustainable development in China and similar economies. It indicates that tailored policies are essential to mitigate the diverse environmental impacts of different FDI flows, supporting sustainable investment practices.

1. Introduction

Economic advancement is frequently accompanied by the intensive utilization of non-renewable energy resources, consequently leading to a substantial increase in greenhouse gas emissions [1,2,3,4,5,6,7]. Amidst the dual challenges of a global economic downturn and climate change, enhancing CP has emerged as an effective strategy to mitigate economic and environmental pressures [8]. In response to these challenges, China has adopted a strategy of innovative development through a combination of domestic and international economic cycles. This strategy emphasizes sustainable economic growth through improved energy efficiency [9]. However, reducing carbon emissions and enhancing CP is a daunting task for China, the world’s largest developing country. Compared to developed nations, developing countries face a delicate balance between economic growth and carbon emission reduction [10]. Achieving this balance requires not only continuous innovation but also robust government support.

The literature on FDI and its environmental impact presents two principal hypotheses. The “Pollution Haven” hypothesis, originating from Baumol and Oates [11], posits that developed countries, due to stringent environmental regulations and high compliance costs, tend to shift high-pollution industrial activities to less developed nations through OFDI. This shift exacerbates environmental degradation in these recipient countries, a phenomenon supported by research from Baek, Muhammad et al. and Zhang et al. [12,13,14]. Conversely, the “Pollution Halo” hypothesis, proposed by Zarsky [15], suggests that IFDI can introduce advanced environmental production technologies through spillover effects, ultimately enhancing the green production efficiency of host countries [16,17].

In the context of global economic integration, the symbiotic relationship between IFDI and OFDI is increasingly crucial in unlocking the potential of both domestic and global economic cycles. This interaction not only promotes the effective integration of domestic and international resources but also significantly enhances capabilities in research and development innovation and talent acquisition [18]. Indeed, CP, which considers the relationship between economic growth and carbon emissions, provides a more precise metric for assessing a region’s ability to achieve sustainable development [19]. In China, the DFDI has become an important means to advance CP. The role of strong governmental policy support and effective implementation in this process requires further study to provide more comprehensive insights into sustainable development.

Furthermore, the government plays a pivotal role in attracting foreign investment, stimulating local economic growth, and significantly influencing environmental governance [20,21]. Studies indicate that implementing FD can improve the efficiency of resource allocation and motivate local governments to enhance environmental standards, thereby improving regional environmental quality [22]. Conversely, literature also shows that FD can lead to detrimental competition among local governments, resulting in environmental deterioration [23,24]. It appears that the relationship between FD and environmental quality may not be linear. In the initial stages of FD, local governance structures may opt for bottom-line competition to promote local economies at the expense of local environmental sustainability [25]. However, in the later stages, local governments may follow a top-line competition approach to foster competition and improve environmental conditions [26]. This highlights the need for a multifaceted analysis of FD’s impact on CP and the non-linear effects on environmental sustainability.

This study aims to explore the separate and combined impacts of the DFDI and FD on regional CP in China. Despite existing research attempts to elucidate the relationship between DFDI, FD, and CP, a definitive understanding remains elusive. The specific impact of the interaction between DFDI and FD on CP and how changes in FD modulate the effect of the coordinated flow of international production factors on CP are subjects of ongoing debate. Moreover, the varying impacts of FDI components on CP, due to differences in samples and methodological approaches, present challenges for government policy formulation and are central to our inquiry.

Our research makes marginal contributions in several aspects: Firstly, by constructing a spatial error model, we discover that the coordinated development of two-way FDI substantially enhances CP in China, while excessive fiscal decentralization limits the potential benefits of FDI on CP enhancement. Secondly, the dynamic threshold model used in our study assesses the impact of FD at different threshold levels on the relationship between DFDI and CP, providing a more nuanced and dynamic analytical perspective. This offers guidance on how governments can utilize fiscal decentralization to achieve optimal pathways for enhancing CP, avoiding estimation biases inherent in traditional static threshold models. Lastly, we delve into the impact mechanisms of the sub-indicators (IFDI and OFDI) of DFDI on CP. This deepens our understanding of effective strategies to boost CP and provides a robust theoretical and empirical foundation for formulating effective policy measures.

The remainder of this paper is organized as follows: Section 2 provides a literature review and theoretical hypotheses; Section 3 introduces the methodology and data utilized; Section 4 analyzes the empirical results; Section 5 offers further discussion; and Section 6 concludes with a discussion on the policy implications and directions for future research.

2. Literature Review and Theoretical Hypotheses

2.1. Carbon Productivity

The concept of CP serves as a crucial indicator to assess the relationship between carbon emissions and economic efficiency, reflecting how a country or region manages and controls carbon emissions while achieving economic growth. Originally proposed by Kaya and Yokobori [27], it is commonly measured by the ratio of total CO₂ emissions to Gross Domestic Product (GDP) [28], with the aim of evaluating the economic benefits obtained from the process of carbon emissions [29]. Through a meticulous review of the existing research, we have identified two main research directions concerning CP: Firstly, the focus has been on the measurement methods of CP. For instance, the SBM model has been applied to assess the level of CP in China [30] while Zhu et al. [31] utilizing data from 89 countries to evaluate carbon emission intensity. A series of studies have also adopted the Malmquist index method to measure CO₂ emission efficiency [32,33,34,35,36]. Secondly, research has concentrated on exploring the factors affecting CP. Previous studies have primarily focused on industrial structure [37,38], technological progress [39,40,41,42], the degree of trade openness [43,44], economic development [45,46], energy structure [47,48], and other aspects.

2.2. Coordinated Development of Two-Way Foreign Direct Investment

In the economic system shaped by globalization, developing countries can enhance their outward investment capabilities by attracting FDI [49]. Pan et al. [50], based on transnational panel data, found that the stronger the host country’s absorption capacity for foreign investment and the larger its market size, the more effectively it can promote the development of outward direct investment. The Investment Development Path (IDP) theory further emphasizes that, as economies develop, the interaction between IFDI and OFDI also gradually strengthens. Under the conditions of an open economy, this implies that the coordinated development of two-way FDI will inevitably affect the environmental pollution status of a country or region.

2.3. Coordinated Development Level of Two-Way FDI and Carbon Productivity

This study defines the “level of coordinated development of two-way FDI” as the coordinated development resulting from the technology spillover effects of IFDI and the reverse technology spillover effects of OFDI. Although earlier studies have explored the impacts of IFDI and OFDI on carbon emissions from various dimensions, the conclusions have been inconsistent [51]. Some research suggests that IFDI in developed countries may turn developing countries into pollution havens, leading to a dramatic increase in carbon emissions [52,53,54]. However, IFDI also introduces energy-saving and environmentally friendly advanced green production technologies, enhancing the technological level and industrial structure of the host countries, thereby positively impacting carbon emission reduction [55,56]. Studies by Wang and Zhang, Demena and Afesorgbor [52,57] also confirmed the pollution halo effect of IFDI. Against this backdrop, China’s Belt and Road Initiative has spurred research into the environmental impacts of Chinese OFDI, with its effects on carbon emissions showing an inverted U-shaped relationship, subject to temporal and regional variations [58]. Li et al. and Mahadevan and Sun [59,60] and found that the reverse technology spillover effect of Chinese OFDI might be associated with the introduction of low-carbon technologies or reduced export carbon emissions. According to Grossman and Krueger [61], FDI can affect the host country’s environment under an open economic system, including but not limited to carbon emissions and CP, through scale effects, structural effects, and technology spillover effects [41,62,63,64].

In summary, although existing literature has analyzed the impact of IFDI and OFDI on carbon emissions separately and discussed the role of FDI on CP, most studies have been conducted independently. For example, Long et al. [65] used a spatial generalized three-stage least squares (GS3SLS) method based on Chinese provincial panel data to analyze the comprehensive impact of IFDI on CP. In contrast, Pan et al. [54], by calculating the total factor carbon productivity (TFCP) for various provinces in China, tested the positive impact of OFDI’s reverse technology spillover effect on total factor productivity.

2.4. Fiscal Decentralization and Carbon Productivity

FD, as a pivotal factor influencing governmental policy choices and behavioral patterns, has a particularly pronounced effect on local environmental governance and carbon emission performance. Current research delves deeply into the relationship between FD and environmental governance, which is generally categorized into two perspectives: The first viewpoint suggests that FD may lead to a “race to the bottom,” where local governments, in an effort to attract investment, may lower environmental standards, undoubtedly exacerbating carbon emissions [66,67,68]. This perspective highlights the potential for competitive deregulation, where fiscal autonomy can incentivize localities to engage in detrimental environmental competition. On the other hand, another viewpoint posits that FD enhances the autonomy of local governments, enabling them to implement environmental policies more flexibly and effectively, which could be conducive to carbon reduction [69,70,71]. This standpoint underscores the empowerment of local authorities, suggesting that decentralization can foster more tailored and responsive environmental management strategies at the local level. Clearly, these perspectives reflect the complex and nuanced interplay between FD and environmental management, indicating that the outcomes of decentralization are contingent on how the balance between local autonomy and central oversight is struck and managed.

FDI is believed to be closely linked to CP. However, many studies fail to consider the potential coordinated interaction between IFDI and OFDI. Although the relationship between FD and CP has been extensively researched, the specific conclusions about this relationship remain inconsistent. Existing research has not fully examined the potential nonlinear effects of FD. Additionally, it is worth noting that Wang [72] used a political economy model to predict the non-monotonic impact of devolving fiscal authority on inflows of foreign direct investment. China is currently at a moderate level of power devolution, and local governments are very active in attracting foreign direct investment.

2.5. Hypothesis

Recent studies have delved into the intricate interplay between fiscal policies, FDI, and their environmental impacts, particularly concerning carbon emissions. Long et al. [65] discovered that while local FDI positively influences CP, its effect in surrounding areas can be adverse. Khan et al. [70] observed that FD impacts CO₂ emissions in a unidirectional manner, underscoring a complex relationship between economic factors and environmental outcomes. Crucially, Shan et al. [67] revealed the non-linear influences of FD, indicating that while its linear aspect exacerbates carbon emissions, the non-linear aspect can mitigate them. These findings highlight the necessity for a nuanced understanding of the environmental impacts of fiscal policies. Further, Iqbal et al. and Feng et al. [73,74] emphasized the intricate relationship between economic growth strategies, such as export diversification and environmental taxation, and their environmental consequences. Elheddad et al. [75] underscored the importance of FD and local government competition in FDI on the effectiveness of energy subsidies. Additionally, Xu and Li. [76] utilized spatial analysis to assess the spillover effects of FD on CO₂ emissions, advocating for optimized fiscal policies for effective emission reduction. These studies collectively emphasize the significance of considering both linear and non-linear dynamics in fiscal and environmental policies for balancing economic growth with sustainable development.

Notably, Wang [72] used a politico-economic model to predict the non-monotonic impact of devolved fiscal authority on foreign direct investment inflows. This underscores the importance of examining how FD changes the coordinated flow of international production factors and impacts CP. Based on these theoretical analyses and practical considerations, our study aims to address four scientific questions: (1) can the DFDI and FD enhance China’s CP?; (2) within the context of China’s FD framework, does the interplay between these two factors still positively affect CP?; (3) how do different components of FDI influence CP in China?; and (4) can a higher level of FD alter the dynamic relationship between DFDI and CP?

This paper seeks to address these research gaps by providing an in-depth analysis of DFDI and FD on CP, with a particular focus on the spatial and dynamic non-linear impacts. Our study contributes novel insights by constructing a spatial error model and adopting a dynamic threshold model, offering a comprehensive understanding of these complex relationships.

3. Methodology and Data

3.1. Econometric Model

3.1.1. Benchmark Model

Based on the analytical model by Jiang and Zhao [77], this paper constructs the following baseline regression model to examine the effects of DFDI and FD on CP:

$${lnCP}_{i,t}={\beta }_0+{\beta }_{1}ln{DFDI}_{it}+{\beta }_2{Fiscal}_{it}+{\beta }_{3}ln{DFDI}_{it}\times {Fiscal}_{it}+{\beta }_4{X}_{it}+{\gamma }_i+{\epsilon }_{it}$$

(1)

where ${CP}_{i,t}$ is the CP of province $i$ in year $t$; ${DFDI}_{i,t}$ represents the level of coordinated development of two-way FDI; ${Fiscal}_{it}$ represents FD. We measure FD using two indicators: fiscal autonomy and revenue FD. ${DFDI}_{i,t}\times {Fiscal}_{i,t}$ is the interaction term of the two indicators, which is used to test for the moderating effect model. By including this interaction term, we aim to capture how the impact of DFDI on CP changes when considered in conjunction with different levels of FD. $X_{i,t}$ includes a series of control variables; ${\gamma }_i$ captures unobserved province-specific effects that account for time-invariant regional characteristics such as geographical location, climate, and culture; and ${\epsilon }_{it}$ is a random disturbance term. Finally, ${\beta }_1-{\beta }_4$ are the unknown coefficients.

3.1.2. The Spatial Econometric Model

In the field of regional economics, traditional econometric models often overlook the potential correlation among variables across regions, which can lead to biased estimations. To address this limitation, this paper introduces spatial econometric models that aim to reduce estimation errors caused by inappropriate empirical methods. According to the suggestions by Elhorst [78], there are three main ways to set spatial effects, leading to three types of spatial econometric models: Spatial Lag Model (SLM), Spatial Error Model (SEM), and Spatial Durbin Model (SDM). The selection of the appropriate spatial econometric model is generally based on the Lagrange Multiplier (LM) test and the Wald test. The respective model expressions are as follows:

$$y_{it}={\alpha }_{it}+\rho W{y}_{it}+x_{it}{\beta }_{it}+{\epsilon }_{it}$$

(2)

$$y_{it}={\alpha }_{it}+x_{it}{\beta }_{it}+u_{it}$$

(3)

$$u_{it}=\lambda W{u}_{it}+v_{it}$$

(4)

$$y_{it}={\alpha }_{it}+\rho W{y}_{it}+W{x}_{it}{\gamma }_{it}+x_{it}{\beta }_{it}+{\epsilon }_{it}$$

(5)

In the equations above, $y_{it}$ represents the dependent variable, $x_{it}$ represents the explanatory variables, $W$ is the known spatial weight matrix, ${\epsilon }_{it}$ and $v_{it}$ are the random error terms, $\rho$ and $\gamma$ are the spatial lag parameters.

3.1.3. Spatial Weight Matrix

In order to control for the spatial spillover effects of CP, this paper constructs a Geographic distance weighted matrix (W1):

$$W_{ij}^d=\left\{\begin{array}{c}\frac{1}{d_{ij}} , i\ne j\\ 0 , i=j\end{array}\right.$$

(6)

where $d_{ij}$ represents the straight-line distance between the provincial capitals of $i$ and $j$,$\frac{1}{d_{ij}}$ indicates the attenuation degree of the geographical distance correlation between provinces. The closer the geographic distance, the greater the weight obtained.

3.1.4. Spatial Autocorrelation Analysis

In spatial econometrics, the Moran’s I index is a measure of spatial autocorrelation, commonly used to determine the degree of correlation among values located in geographic space [79,80]. The index ranges from −1 (indicating perfect dispersion or negative spatial autocorrelation) to +1 (indicating perfect correlation or positive spatial autocorrelation). A value of 0 indicates a random spatial pattern, with no discernible autocorrelation. The Moran’s I is calculated by assessing the attribute values of spatial units in conjunction with a spatial weights matrix. In the context of this research, utilizing the global Moran’s I to examine whether CP has significant spatial autocorrelation is crucial as it could reveal the extent to which CP levels are influenced and dependent on each other across different regions.

$$Mora{n}^{\prime }s I=\frac{n\sum _{i=1}^n\sum _{j=1}^n{W}_{ij}(x_i-\overline{x})(x_j-\overline{x})}{S^2{\sum }_{i=1}^n{\sum }_{j=1}^n{w}_{ij}}$$

(7)

Here, $S^2={\sum }_{i=1}^n(x_i-{\overline{x})}^2,\overline{x}=\frac{1}{n}{\sum }_{i=1}^n{x}_i$.

The spatial correlation between provinces can be tested through the local Moran’s I, as follows:

$$I_i=Z_i\times {\sum }_{j=1}^n{w}_i{z}_j$$

(8)

where, $z_i=x_i-\overline{x}$ and $z_j=x_j-\overline{x}$ represent the deviations of observed values from the mean value, respectively.

3.1.5. Dynamic Panel Threshold Model

To further examine the impact of the DFDI under different degrees of FD on CP, this paper introduces a dynamic threshold regression model. Traditional static threshold models, when processing data and estimating models, automatically divide the data into different intervals, which may lead to biased estimation results. They also ignore the time lag factors of the dependent variables, which can cause endogeneity issues. To address these problems, this paper, drawing on the approach of Wu et al. [81], establishes the following dynamic threshold panel model.

$$\begin{gathered} {lnCP}_{i,t}={\beta }_0+{\beta }_{1}ln{CP}_{it-1}+{\beta }_2{lnDFDI}_{it}·I({Fiscal}_{it} \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} \le \theta )+{\beta }_3{lnDFDI}_{it}·I({Fiscal}_{it}>\theta )+{\beta }_4{X}_{i,t}+{\gamma }_i+{\epsilon }_{it} \end{gathered}$$

(9)

In Equation (9), ${CP}_{it-1}$ denotes the first lag term of ${CP}_{i,t}$. The threshold variable is ${Fiscal}_{it}$, which represents the FD. $I\left(·\right)$ refers to the indicator function, if the arguments in parentheses are valid, then its value will be 1, otherwise will be 0. $\theta$ is the specific threshold value. The meanings of other parameters are the same as those in Formula (1).

3.2. Variables and Data

3.2.1. Explained Variable

The definition of CP, according to Kaya and Yokobori [27], refers to the ratio of gross domestic product (GDP) to carbon dioxide emissions over a certain period. As the National Bureau of Statistics and various provincial statistical yearbooks do not directly publish data on carbon emissions, the CP data used in this paper are estimated. One of the main challenges of this paper is the calculation of provincial carbon dioxide emissions. The calculation of provincial CO₂ emissions is conducted using the methodology provided by the Intergovernmental Panel on Climate Change (IPCC). The equation is formulated as follows:

$${CO}_2=\sum _{i=1}^n{CO}_2=\sum _{i=1}^n{E}_i\times {NCV}_i\times {CC}_i\times {COF}_i\times \frac{44}{12}$$

(10)

In Equation (10), ${CO}_2$ represents the total carbon dioxide emissions and $i$ denotes the type of end-use energy. This paper selects eight types of energy as the end-use energy, namely coal, coke, crude oil, gasoline, kerosene, diesel, fuel oil, and natural gas, to reduce the errors caused by primary energy division methods. $E_i$ is the consumption volume of the $i\text{\_}th$ type of end-use energy; ${NCV}_i$ represents the average low heating value of the $i\text{\_}th$ energy type; ${COF}_i$ is the carbon oxidation factor for the $i\text{\_}th$ type of energy; ${CC}_i$ denotes the carbon emission factor of the $i\text{\_}th$ energy, which reflects the carbon content per unit of calorific value; and the numbers 44 and 12 represent the ratio of the molecular weight of carbon dioxide to that of carbon elements, respectively.

Following Sun and Du [82], this paper measures the provincial-level CP as follows:

$${Cpd}_{it}={Gd}_{it}/{Cd}_{it}$$

(11)

In Equation (11), $Cpd$ represents the provincial CP, $Gd$ stands for the provincial gross domestic product, and $Cd$ is the total amount of carbon dioxide emissions at the provincial level, which is estimated from the consumption of the eight types of energy.

3.2.2. Key Explanatory Variable

When constructing the two-way indicator for IFDI and OFDI, we employed various methods, but none sufficiently captured the interactive level of two-way FDI. We ultimately adopted the measurement approach from Huang et al. [83], using a coupling system model to measure the coordinated development level of two-way FDI as follows:

$$C_{it}\left(IO\right)=\frac{{IFDI}_{it}\times {OFDI}_{it}}{{{(\alpha IFDI}_{it}\times {\beta OFDI}_{it})}^{\gamma }}$$

(12)

In the formula, IFDI and OFDI respectively represent the stock of IFDI and OFDI for province $i$ in year $t$, both measured in billion USD. The data are sourced from regional Statistical Yearbooks and China’s Statistical Bulletin of Foreign Direct Investment. The weights for IFDI and OFDI, denoted as α and β, are both set to 0.5. The symbol γ serves as an adjustment coefficient, which, according to past research, is set to 2. The calculation of the coupling degree indicates the interaction level between IFDI and OFDI, utilizing the coordination index to measure the overall state of system development. The introduction of the coupling degree indicator is as follows:

$$\mathrm{T}=\left[C_{it}\left(IO\right) \right.\frac{{IFDI}_{it}\times {OFDI}_{it}}{2}{\left. \right] }^{\frac{1}{2}}$$

(13)

Subsequently, by incorporating the coupling degree formula (12) into the coordination development index formula (13), we propose a method for calculating the degree of DFDI:

$${DFDI}_{it}=D_{it}\left(IO\right)=\left[C_{it}\right.\left(IO\right)\times \frac{{IFDI}_{it}+{OFDI}_{it}}{2}{\left. \right] }^{\frac{1}{2}}= \left[ \right.\frac{{IFDI}_{it}\times {OFDI}_{it}}{({IFDI}_{it}+{OFDI}_{it})/2}{\left. \right] }^{\frac{1}{2}}$$

(14)

3.2.3. Moderating Variable and Threshold Variable

In our analytical model, the degree of FD serves both as a moderating and a threshold variable. To measure FD, we follow the approach of Wu and Heerink [84], using fiscal autonomy (the ratio of public budget revenue to public budget expenditure) and revenue FD (the relative size of local fiscal incomes compared with central incomes) as metrics. The formulas for these indicators are as follows:

$$Fiscal\text{\_}1=\frac{Public budget revenue in year t of city i}{Public budget expenditure in year t of city i}$$

(15)

$$Fiscal\text{\_}2=\frac{{Local fiscal income per capita}_{it}}{{Local fiscal income per capita}_{it}+{Central fiscal income per capita}_{it}}$$

(16)

3.2.4. Control Variables

To ensure the precision of our estimations and to mitigate the biases due to omitted variable bias, this study includes the following control variables which may influence CP: 1. Economic Development Level (pgdp) is measured by per capita Gross Domestic Product (GDP). Considering the Environmental Kuznets Curve (EKC) hypothesis, which posits a relationship between economic growth and CO₂ emissions [85,86], we include both pgdp and its square to test the validity of this hypothesis in the context of Chinese provinces. 2. Openness to Trade (open) is quantified as the ratio of total imports and exports to regional GDP. 3. Population Density (pd) is the ratio of permanent residents to the administrative area. 4. Human Capital Level (hc) is represented by the number of college students per 10,000 people. 5. Infrastructure Level (inf) is evaluated by the ratio of total road mileage to administrative area size.

Moreover, acknowledging that industrial structure and energy consumption patterns are significant determinants of regional environmental quality [87], the study uses the ratio of tertiary to secondary industry added value to measure the advancement of industrial structure and the proportion of coal consumption in total energy consumption to reflect the cleanliness of the energy structure.

The data for these analyses primarily come from the China Statistical Yearbook and the China Energy Statistical Yearbook. The sample includes panel data from 30 Chinese provinces (excluding Tibet, Hong Kong, Macau, and Taiwan) from the years 2006 to 2020. During data processing, all FDI amounts denominated in dollars were converted into billions of RMB using the annual exchange rate of the US dollar to RMB. Moreover, all economic indicators were deflated to the base year of 2003 using the Consumer Price Index (CPI). To eliminate potential heteroscedasticity in the analysis and to ensure the robustness of the estimations, all variables, except for per capita GDP, were log-transformed, and missing values were imputed using linear interpolation.

4. Empirical Results Analysis and Discussion

4.1. Spatial Autocorrelation Test

To accurately assess the spatial correlation of CP across 30 provinces in China, this study initially employed the global Moran’s I index for analysis, with results presented in

Table 1. 2006–2020 Global Moran’s I Index for CP.

Year	Moran’I	Z	p-Value
2006	0.312	2.924	0.004
2007	0.312	2.949	0.003
2008	0.306	2.854	0.004
2009	0.310	2.877	0.004
2010	0.312	2.892	0.004
2011	0.305	2.830	0.005
2012	0.294	2.730	0.006
2013	0.300	2.773	0.006
2014	0.295	2.732	0.006
2015	0.329	3.003	0.003
2016	0.350	3.175	0.002
2017	0.368	3.317	0.001
2018	0.376	3.379	0.001
2019	0.376	3.383	0.001
2020	0.379	3.397	0.001

.

The analysis from Table 1 indicates that from 2006 to 2020, the global Moran’s I values remained above zero, passing the significance test at the 1% level. Notably, the Moran’s I index for CP fluctuated around 0.3 from 2006 to 2015. However, starting in 2016, there was a gradual upward trend in the index. This shift corresponds with the Chinese government’s increased efforts in environmental protection and the implementation of a series of emission reduction policies in recent years. Specifically, both the 13th Five-Year Plan and China’s Nationally Determined Contributions (NDC) have focused on targets to reduce carbon intensity.

Subsequently, the study leveraged Moran’s I scatter plots to further examine the spatial heterogeneity of CP and analyze the local spatial agglomeration characteristics of the provinces. As depicted in Figure 1, the scatter plots for the representative years of 2006 and 2020 clearly revealed that most Chinese provinces are situated in the first and third quadrants. This pattern of high-high and low-low clustering significantly reflects the local agglomeration of CP.

To visually display the distribution pattern of CP across Chinese provinces, we showcased a spatial distribution map for the 30 provinces (as shown in Figure 2). The map demonstrates significant disparities in CP among the provinces. For instance, the economically developed eastern regions have a notably higher CP compared to the western provinces, and the southern regions outpace the northern ones. Overall, this spatial distribution mirrors the state of regional economic development in China.

In summary, there is a significant spatial correlation between the CP of Chinese provinces, with high-productivity areas often adjacent to one another, and the same applies to areas with low productivity. This lays the groundwork for our subsequent regression analysis using spatial econometric models.

4.2. Space Suitability Test and Empirical Results

Before conducting the empirical regression analysis, our study meticulously selected the model by applying LM, Wald, LR, and Hausman tests, adhering to the criteria suggested by LeSage [88] that emphasize stability and the significance level of the Lagrange Multiplier (LM). The results presented in

Table 2. LM, Wald, LR, and Hausman test results.

Matrix	Geographic Distance Weighted Matrix
Tests	LM	Robust LM	wald	LR
SEM	56.872 ***	36.675 ***	181.11 ***	172.79 ***
SAR	23.279 ***	2.082	120.44 ***	114.56 ***
Hausman test	−57.010

Note: *** indicates statistical significance at the 1% levels.

reveal that all tests, except for the Robust LM (lag), are significant at the 1% level, suggesting that the Spatial Error Model (SEM) is more appropriate for regression analysis than the Spatial Lag Model (SLM). Moreover, the Hausman test yielded a negative statistic, supporting the validity of the random effects assumption. Therefore, this paper primarily utilizes the SEM under the random effects assumption for estimation purposes.

As shown in

Table 3. FE and SEM estimation results.

Variables	(1)	(2)	(3)	(4)	(5)
	FE	SEM2	SEM3	SEM4	SEM5
lndfdi	1.081 ***	0.516 **	1.242 ***	0.457 **	1.114 ***
	(5.279)	(2.520)	(4.299)	(2.301)	(4.955)
Fiscal_1		0.005 ***	0.013 ***
		(2.771)	(4.528)
lndfdi × Fiscal_1			−0.141 ***
			(−3.548)
Fiscal_2				0.048 ***	0.127 ***
				(5.956)	(8.630)
lndfdi × Fiscal_2					−0.998 ***
					(−6.270)
pgdp	0.070 ***	0.070 ***	0.071 ***	0.072 ***	0.067 ***
	(13.897)	(14.972)	(15.268)	(15.984)	(15.189)
pgdp2	−0.002 ***	−0.002 ***	−0.002 ***	−0.002 ***	−0.001 ***
	(−7.685)	(−8.005)	(−7.432)	(−8.608)	(−5.844)
open	0.556	1.270 ***	1.233 ***	1.327 ***	1.097 ***
	(1.244)	(3.553)	(3.505)	(3.833)	(3.335)
pd	−0.144 ***	−0.124 ***	−0.107 ***	−0.131 ***	−0.104 ***
	(−3.105)	(−4.835)	(−4.228)	(−5.406)	(−4.434)
hc	0.218 ***	0.143 ***	0.136 ***	0.142 ***	0.125 ***
	(12.244)	(8.093)	(7.811)	(8.318)	(7.588)
inf	0.108 ***	0.124 ***	0.116 ***	0.118 ***	0.103 ***
	(4.246)	(5.923)	(5.604)	(5.831)	(5.348)
is	−0.031 ***	0.006	0.009	0.010	0.021 **
	(−2.756)	(0.655)	(0.925)	(1.069)	(2.408)
es	−0.080 **	−0.110 ***	−0.099 ***	−0.111 ***	−0.125 ***
	(−2.088)	(−3.277)	(−2.998)	(−3.425)	(−4.039)
_cons	1.418 ***	1.639 ***	1.521 ***	1.650 ***	1.510 ***
	(5.503)	(10.169)	(9.423)	(10.660)	(10.139)
λ		0.570 ***	0.579 ***	0.563 ***	0.622 ***
		(11.788)	(12.119)	(11.420)	(13.125)
sigma2_e		0.001 ***	0.001 ***	0.001 ***	0.001 ***
		(13.936)	(13.942)	(13.968)	(13.847)
N	450	450	450	450	450
R2	0.954	0.595	0.609	0.607	0.627

Note: *** and ** indicates statistical significance at the 1% and 5% levels, respectively; the value in parentheses represents t-statistics.

, the regression coefficients for the level of DFDI are positive and significant at the 1% and 5% levels, corroborating the positive impact of DFDI on the CP of Chinese provinces. Excluding interaction terms and using the SEM as an example, an increase of one standard deviation in the level of DFDI results in an approximately 0.9% increase in CP (calculated as 0.516 × 0.019 or 0.457 × 0.019).

Regarding the FD variable Fiscal_1, its regression coefficient is positive and significant at the 1% level in columns (2) and (3), indicating that FD in China has a stimulating effect on regional CP [89]. Similar outcomes are derived when using a different FD indicator, Fiscal_2, as shown in columns (4) and (5). This suggests that a higher degree of fiscal autonomy may enhance the efficiency of resource allocation and the effective management of carbon emissions through enabling local governments to formulate policies tailored to local conditions, thereby improving CP.

Additionally, the regression results in columns (3) and (5) show that the interaction term coefficients for DFDI development and FD are negative and statistically significant. This finding suggests that the positive impact of DFDI on CP can vary with the degree of FD across regions. Such a contradictory or adverse effect is notable, as it may stem from inconsistencies in policy, regulation, or enforcement standards due to varying degrees of FD; this could influence FDI decision-making and limit the potential environmental performance benefits. For instance, regions with lower environmental standards may attract more FDI, but these investments may not employ the most advanced eco-friendly technologies. Moreover, this could reflect a more complex policy milieu where the effect of FDI is influenced by how local governments utilize their fiscal autonomy. In some cases, local authorities might prioritize attracting FDI at the expense of environmental standards, thereby weakening improvements in CP.

The estimated coefficients for the control variables also offer insightful findings. The coefficient of the linear term for per capita GDP is positive, while the coefficient of the quadratic term is negative, indicating an inverted U-shaped relationship between economic growth and CP, which further corroborates the Environmental Kuznets Curve (EKC) hypothesis [89]. The regression coefficient for economic openness is significantly positive at the 1% level, suggesting that increased openness to international trade and capital flows can enhance regional CP. The impact of population density is significantly negative, implying that higher population density could lead to increased energy consumption and, consequently, carbon emissions. Both infrastructure and human capital have positive effects on CP, suggesting that China has managed to balance economic development with environmental protection to some extent, as regions with a concentration of talents often manage to reduce carbon emissions. The industrial structure does not significantly enhance CP, indicating that China is still in the early stages of industrial restructuring. The coefficient for energy structure is positive and statistically significant, confirming that an energy-intensive economic production model is not conducive to improving CP.

4.3. Robustness test

• Replacing spatial weight matrix

To address potential biases due to different types of spatial matrices, this study constructed a new economic geographical spatial weight matrix based on the per capita GDP of each province. The results, as shown in columns (1) and (2) of

Table 4. Robustness check.

Variables	Economic Geography Weight Matrix		FDI_flow		CP_DEA
	(1)	(2)	(3)	(4)	(5)	(6)
lndfdi	0.825 ***	0.563 ***	0.017 ***	0.014 ***	0.023 ***	0.018 ***
	(3.867)	(2.674)	(5.950)	(4.863)	(5.574)	(4.257)
Fiscal_1	0.004 **		0.013 **		0.006 ***
	(2.147)		(2.417)		(3.028)
Fiscal_2		0.053 ***		0.037 ***		0.049 ***
		(6.126)		(4.587)		(6.275)
pgdp	0.070 ***	0.070 ***	0.065 ***	0.067 ***	0.070 ***	0.072 ***
	(13.783)	(14.398)	(14.110)	(14.836)	(14.822)	(15.861)
pgdp2	−0.002 ***	−0.002 ***	−0.002 ***	−0.002 ***	−0.002 ***	−0.002 ***
	(−7.363)	(−7.728)	(−7.400)	(−7.916)	(−7.926)	(−8.572)
open	1.401 ***	1.471 ***	1.059 ***	1.159 ***	1.295 ***	1.338 ***
	(3.629)	(3.952)	(3.047)	(3.412)	(3.604)	(3.851)
pd	−0.111 ***	−0.119 ***	−0.117 ***	−0.121 ***	−0.125 ***	−0.132 ***
	(−4.319)	(−4.907)	(−4.565)	(−4.978)	(−4.862)	(−5.446)
hc	0.187 ***	0.183 ***	0.141 ***	0.139 ***	0.143 ***	0.143 ***
	(11.028)	(11.294)	(8.054)	(8.203)	(8.110)	(8.364)
inf	0.099 ***	0.097 ***	0.104 ***	0.101 ***	0.129 ***	0.121 ***
	(4.316)	(4.457)	(5.076)	(5.014)	(6.147)	(5.953)
is	−0.013	−0.007	0.010	0.012	0.006	0.010
	(−1.201)	(−0.692)	(1.038)	(1.321)	(0.635)	(1.034)
es	−0.108 ***	−0.115 ***	−0.085 ***	−0.091 ***	−0.108 ***	−0.110 ***
	(−3.122)	(−3.440)	(−2.631)	(−2.888)	(−3.193)	(−3.382)
_cons	1.339 ***	1.381 ***	1.519 ***	1.533 ***	1.673 ***	1.672 ***
	(8.058)	(8.692)	(9.464)	(9.867)	(10.396)	(10.838)
λ	0.684 ***	0.685 ***	0.578 ***	0.568 ***	0.575 ***	0.561 ***
	(10.360)	(10.562)	(12.016)	(11.567)	(11.874)	(11.175)
sigma2_e	0.001 ***	0.001 ***	0.001 ***	0.001 ***	0.001 ***	0.001 ***
	(14.274)	(14.297)	(13.927)	(13.954)	(13.919)	(13.957)
N	450	450	450	450	450	450
R2	0.561	0.588	0.580	0.601	0.595	0.607

Note: *** and ** indicates statistical significance at the 1% and 5% levels, respectively; the value in parentheses represents t-statistics.

, demonstrate that the signs and statistical significances of the regression outcomes are consistent with the baseline regression (Table 3 column (1)), with only minor numerical adjustments. This confirms the robustness of our empirical results.

• Replacing key explanatory variable

After reconstructing the DFDI flow data, the analysis revealed that its positive influence on CP remains significant, as illustrated in columns (3) and (4) of

Table 5. Analysis results of FDI components.

Variables	IFDI		OFDI
	(1)	(2)	(3)	(4)
IFDI	0.023 ***	0.013 *
	(3.467)	(1.895)
OFDI			−0.032 ***	−0.039 ***
			(−3.002)	(−3.843)
Fiscal_1	0.005 ***		0.007 ***
	(2.596)		(3.734)
Fiscal_2		0.045 ***		0.055 ***
		(5.367)		(7.148)
pgdp	0.072 ***	0.073 ***	0.071 ***	0.073 ***
	(15.632)	(16.468)	(15.118)	(16.332)
pgdp2	−0.002 ***	−0.002 ***	−0.002 ***	−0.002 ***
	(−8.431)	(−8.825)	(−7.776)	(−8.453)
open	1.310 ***	1.356 ***	0.996 ***	0.940 ***
	(3.708)	(3.932)	(2.697)	(2.649)
pd	−0.131 ***	−0.134 ***	−0.109 ***	−0.112 ***
	(−5.117)	(−5.564)	(−4.218)	(−4.623)
hc	0.147 ***	0.145 ***	0.142 ***	0.141 ***
	(8.371)	(8.505)	(8.133)	(8.372)
inf	0.135 ***	0.127 ***	0.120 ***	0.106 ***
	(6.630)	(6.390)	(5.774)	(5.328)
is	0.007	0.010	0.011	0.016 *
	(0.756)	(1.101)	(1.219)	(1.768)
es	−0.088 ***	−0.098 ***	−0.098 ***	−0.103 ***
	(−2.644)	(−2.995)	(−2.999)	(−3.277)
_cons	1.679 ***	1.676 ***	1.576 ***	1.557 ***
	(10.507)	(10.887)	(9.811)	(10.143)
λ	0.576 ***	0.566 ***	0.627 ***	0.616 ***
	(12.241)	(11.648)	(14.051)	(13.171)
sigma2_e	0.001 ***	0.001 ***	0.001 ***	0.001 ***
	(13.962)	(13.983)	(13.858)	(13.878)
N	450	450	450	450
R2	0.583	0.603	0.605	0.613

Note: *** and * indicates statistical significance at the 1% and 10% levels, respectively; the value in parentheses represents t-statistics.

. This suggests that both flow and stock data of FDI contribute positively to CP, reflecting sensitivity to external shocks and policy changes. By comparing the outcomes of both datasets, the robustness of our findings was further corroborated.

• Replacing dependent variable

Moreover, this study recalculates the CP using the DEA model, following the methodology proposed by Fan et al. [8]. We have considered a composite measure encompassing desired outputs (regional Gross Domestic Product, with data indexed to the base year of 2006), undesired outputs (total CO₂ emissions), as well as inputs such as capital (total investment in fixed assets), labor (end-of-year employment numbers), and energy (total energy consumption). The regression results of the refined CP index, as presented in columns (5) and (6) of Table 5, confirm that the positive influence of DFDI remains significant even after replacing the dependent variable’s measure. This bolsters the robustness of the research findings.

5. Further Discussion

5.1. DFDI Subsample Analysis

Table 5 presents the estimated results of two sub-indicators under the coordinated development level of two-way FDI. Specifically, the regression coefficients in columns (1) and (2) are positive at the 1% and 10% significance levels, indicating that an increase in IFDI is conducive to enhancing local CP. This is because the inflow of international capital is not merely a simple transfer of funds but also a medium for the transmission of advanced technology and management expertise. However, the analysis in columns (3) and (4) reveals that the regression coefficients for OFDI are significantly negative, suggesting that an increase in OFDI actually suppresses CP. This adverse effect may stem from the limited nature of resources, with investment inherently seeking high returns, leading capital to favor investment opportunities with higher returns. Consequently, if a substantial amount of capital is invested abroad, it may lead to a reduction in industrial investment in the local and surrounding areas, which is detrimental to the economic development and progress of these regions. Furthermore, this pattern of capital flow restricts technological innovation, negatively impacting the nation’s carbon production efficiency.

From these results, we can conclude that the strategic layout of two-way FDI should place greater emphasis on balance. While attracting IFDI, it is also crucial to prudently promote the healthy development of OFDI to ensure that it can bring about a reverse spillover of technology and management knowledge to the home country, thereby achieving a true “win-win” situation.

5.2. Results of Dynamic Panel Threshold Regression

Technically, Seo et al. [90] devised specific computational commands for the first-differenced generalized method of moments (GMM) and the asymptotic variance estimators proposed by Seo and Shin [91]. Compared to the traditional xthreg command, the xthenreg command they introduced offers more consistent and asymptotically normal estimates. Importantly, they also introduced a new, more efficient bootstrap algorithm for testing the presence of a threshold effect, which has significant advantages over the nonparametric independent and identically distributed bootstrap method initially proposed by Seo and Shin [91]. Furthermore, considering the peculiarity of the kink, they employed a constrained GMM estimation approach.

Table 6. Results of dynamic panel threshold regression.

	Lower Regime	Upper Regime	Overall	Post-Estimation Tests
L(1)lncp	0.831 ***	0.129 ***	0.906 ***	Kink	0.011 *
	(0.025)	(0.019)	(0.005)		(0.006)
lndfdi	−0.945 **	2.061 ***	0.660 ***	Threshold indicator	3.926 ***
	(0.396)	(0.493)	(0.105)		(0.651)
Fiscal_1	0.020 ***	−0.013 *	−0.013 **	95% Conf. Interval	[2.650, 5.202]
	(0.005)	(0.007)	(0.006)	AR(1) (p-value)	0.006
_cons	−0.312			AR(2) (p-value)	0.189
	(0.032)			Hansen J (p-value)	0.289
				Linearity test (p-value)	0.000
L(1)lncp	0.987 ***	−0.114 **	0.935 ***	Kink	0.120 ***
	(0.056)	(0.058)	(0.006)		(0.015)
lndfdi	−0.681 **	1.125 ***	1.042 ***	Threshold indicator	1.114 ***
	(0.326)	(0.383)	(0.102)		(0.098)
Fiscal_2	−0.791 ***	0.728 ***	0.088 ***	95% Conf. Interval	[0.922, 1.306]
	(0.206)	(0.216)	(0.016)	AR(1) (p-value)	0.006
_cons	−0.254 **			AR(2) (p-value)	0.189
	(0.112)			Hansen J (p-value)	0.289
				Linearity test (p-value)	0.000

Notes: Standard errors in parentheses. ***, ** and * denote significant at the 1%, 5% and 10% levels, respectively.

reports the regression results of Equation (9), reflecting the nonlinear relationship between the DFDI and CP. To verify the positive effect of DFDI on CP, this study uses the Seo et al. [90] dynamic threshold panel regression model to analyze the panel data of 30 provinces in China from 2006 to 2020. The upper part of Table 6 presents the regression analysis results using the FD indicator Fiscal_1 as the threshold variable, while the lower part conducts analysis based on Fiscal_2 as the threshold variable.

The regression findings demonstrate a significant threshold effect between DFDI development and CP, with threshold values at 3.926 and 1.114, both significant at the 1% level. The p-values for the linearity tests also reject the null hypothesis of no threshold effect, indicating the presence of a nonlinear relationship between the variables. Additionally, the p-value for the kink test is significant, thus avoiding potential issues with kinked relationships.

Under the threshold of Fiscal_1, there is a U-shaped relationship between DFDI and CP. When FD, Fiscal_1, is less than or equal to 3.926, DFDI development suppresses the enhancement of CP. Conversely, when Fiscal_1 exceeds 3.926, DFDI significantly fosters the growth of CP. The results with Fiscal_2 as the threshold variable align with those of Fiscal_1, further affirming the robustness of the findings. Overall, this study statistically substantiates that DFDI significantly promotes the improvement of CP. Moreover, the Hansen test confirms the validity of the instrumental variables used in the study. The robustness of the model is ascertained using Arellano–Bond autoregressive tests, with no evidence found of overidentification restrictions. All tests indicate the absence of AR(1) and AR(2) autocorrelations in the model, hence supporting the empirical results as robust and unbiased.

Further analysis revealed the specific factors that influence the relationship between DFDI and CP at different levels of FD. For instance, in environments with lower FD, FDI might tend to flow into areas with lax environmental standards, which may boost economic output in the short term but does not aid in enhancing CP. Conversely, in situations of higher FD, local governments might be better able to effectively utilize fiscal incentives, such as tax breaks, to attract and promote FDIs that offer environmentally friendly technologies and management practices.

6. Conclusion and Policy Implications

This research has extensively explored the impact of Foreign Direct Investment (FDI) coordination development (DFDI) on carbon productivity (CP) in China, within the framework of fiscal decentralization (FD). Our study, grounded in panel data from 30 Chinese provinces over the period 2006–2020, employs a Spatial Error Model and a Dynamic Threshold Model to provide a nuanced understanding of this relationship.

The key findings of our analysis are threefold. First, we established that DFDI effectively enhances CP, as evidenced by the spatial error model. This underscores the role of FDI in promoting not just economic development but also in enhancing environmental management and CP. However, our analysis also revealed that an excessive level of FD can limit the potential environmental performance benefits provided by foreign investment. Secondly, we employed a dynamic threshold model to further analyze the impact of DFDI on CP under the effect of the FD threshold. Our results indicate a significant dynamic non-linear relationship, emphasizing the need for a balanced approach in FD to harness the full potential of DFDI in improving CP. Lastly, contrary to IFDI, we found that China’s OFDI actually impedes the improvement of CP. This distinction is crucial for policy formulation, as it highlights the different roles played by IFDI and OFDI in China’s CP landscape.

The significance of this study lies in its comprehensive examination of the interplay between DFDI and FD in the context of CP. By highlighting the dynamic non-linear impacts and the spatial characteristics of this relationship, our research fills a gap in the existing literature, which has lacked a consensus on the interaction between DFDI and FD and their collective impact on CP. Moreover, the differentiated impacts of IFDI and OFDI provide a more granular understanding of FDI components on CP, a consideration vital for the formulation of precise and effective governmental policies. Based on the above empirical results and research conclusions, the following policy suggestions are proposed to simultaneously accommodate economic development and environmental protection:

Firstly, in the context of domestic and international dual circulation development, it is more important to plan the flow of two-way FDI rationally so that its layout is more rationalized. Moreover, the introduction of foreign capital should be more regulated to prevent the entry of highly polluting enterprises. Instead, encourage and attract green innovative technology companies to join, and through technological spillover, accelerate the low-carbon transformation of enterprises and improve CP.

Secondly, concerning the adverse impact of OFDI on China’s CP, the Chinese government should guide OFDI to transfer to other countries with comparative advantages. They should optimize the allocation of resources for production factors such as capital and labor, reduce the proportion of industries with high pollution, high carbon emissions, and low added value, and focus resources on developing high-tech, high-value-added, and green industries. Additionally, appropriate incentive policies should be devised to attract outward investments aimed at acquiring advanced foreign technology, to improve the backward technology spillover effect of OFDI, and thereby promote domestic technological progress and industrial upgrading.

Thirdly, local governments can make rational use of fiscal autonomy, providing tax reductions, subsidies, or other fiscal incentives to foreign investments with green or low-carbon technologies. Through DFDI, local governments can promote international technological exchange and cooperation, introduce advanced low-carbon technologies and management experience, and support local enterprises to invest abroad and spread green technology globally. However, it is also necessary to establish a comprehensive environmental quality-monitoring network and assessment system to ensure that FDI projects comply with environmental protection requirements and to adjust environmental policies and measures in a timely manner.

Admittedly, our analysis is confined to the context of DFDI and FD’s impact on CP within the territories of China. Given the unique economic, institutional, and environmental dynamics of China, our findings should be applied to other regions or countries with caution. Despite these limitations, this study’s implications possess a degree of external validity that extends beyond the Chinese provinces analyzed. The dynamic non-linear relationships and the spatial characteristics of FDI’s influence on environmental outcomes are concepts that can be applied to broader contexts, especially in developing countries grappling with similar challenges of balancing economic development with environmental sustainability.

For future research, it would be beneficial to replicate this study in diverse economic and environmental frameworks to further validate the generalizability of our findings. Research could also expand to analyze the role of government policy in mediating the effects of FDI on environmental indicators, not just CP, but also on other dimensions of ecological impact. Furthermore, our study utilized provincial-level data, which provides a broad overview but may not capture the nuances of FD at different administrative tiers. A more granular approach, such as examining city-level data, could uncover insights into the micro-level effects of fiscal policies and FDI on CP. Exploring these variations at a finer scale is an avenue for improvement that our future studies will aim to address.