Impact of Foreign Direct Investment on Green Total Factor Productivity: New Evidence from Yangtze River Delta in China

Abstract

China has entered a period of high-quality development. As an important feature of high-quality development, green total factor productivity (GTFP) has attracted much attention. With the opening-up and economic globalization, the Yangtze River Delta, one of the strongest and most technological regions in China, has been attracting an increasing amount of foreign direct investment (FDI). This study investigates if FDI is conducive to GTFP under the constraints of specific resources and a specific environment, which has important practical significance for the utilization of FDI in the Yangtze River Delta and China. Through a literature review and sorting the current FDI in the Yangtze River Delta, the GTFP and its decomposition indicators of 27 cities from 2004 to 2019 are calculated based on their energy consumption and pollution. Using the fixed-effects model and threshold model of panel data, this study tests whether FDI promotes GTFP and whether a nonlinear impact of FDI on GTFP exists. It finds that (1) the GTFP of most cities in the Yangtze River Delta improved during the sample period, but their annual growth declined. Technology is the dominant factor affecting the growth of GTFP. (2) FDI in the Yangtze River Delta has increased, and the investment structure has improved, but the distribution among cities is uneven. (3) The scale and quality of FDI have a positive impact on GTFP, which supports the “Pollution Halo” hypothesis. Economics, education, networks, and trade openness can promote the growth of GTFP, while environmental regulation, government intervention, and industrialization have a negative impact. (4) The quality of FDI, economics, the industrial structure, the environmental regulation, and the internet are each a significant single threshold characteristic for the impact of FDI on GTFP. When one of these factors is lower than a certain threshold, FDI has less impact on GTFP. When one exceeds a certain threshold, FDI’s positive promotion effect on GTFP significantly improves. Based on the analysis, this study offers some suggestions. The government should improve the FDI selection mechanism based on realities, make appropriate environmental regulatory policies, strengthen the construction of networks, and improve the “Internet+” effect on productivity.

1. Introduction

China’s accession to the WTO and the implementation of the “bringing in” policy have continuously improved our country’s level of opening up to the outside world, and the scale of foreign direct investment (FDI) by multinational companies in China has continued to increase. Against the background of the global decline in foreign direct investment for four consecutive years up to 2019, China’s attraction of foreign capital still grew against the trend, with the actual use of foreign capital reaching 138.14 billion US dollars, ranking at the forefront of the world [1]. The Yangtze River Delta region, with its economic conditions, resource endowment, location advantages, and policy support, has attracted a large influx of FDI, becoming a key bridge for “circulation” between China and other parts of the world under the global economic slowdown and intensifying tensions with the United States. The inflow of foreign capital can not only directly fill the funding gap for the economic growth of the Yangtze River Delta region but also promote regional economic growth by increasing employment opportunities, bringing technology spillover, and improving the level of human capital. Furthermore, due to the positive external effects brought by foreign capital, FDI is more conducive to attracting high-quality domestic and foreign enterprises to the Yangtze River Delta region, enhancing the vitality of the region’s economic growth [2]. Of course, the inflow of foreign capital will not only bring technological progress and industrial upgrading but also, with the upgrading of the industrial structure of developed countries, heavily polluting, high-energy-consuming enterprises will also transfer from abroad, bringing about negative impacts such as resource consumption and environmental pressure. Therefore, whether the Yangtze River Delta region will become a “pollution haven” for foreign capital or confirm the “pollution halo” hypothesis, and whether a large influx of FDI is truly beneficial to the improvement of green total factor productivity under resource and environmental constraints, is still a controversial issue.

1.1. Calculation of Green Total Factor Productivity (GTFP)

Total factor productivity (TFP) is the ratio of total output to the sum of input production factors in a certain period of time and is used to measure the contributions of technological innovation and the improvement of production processes to overall productivity [3,4,5]. As environmental deterioration, energy shortages, and other problems are becoming prominent, there is a consensus that economic development should not depend solely on increasing investments in resource factors while ignoring their impacts on the ecological system, so sustainable development is the trend [6,7]. In this new era, the conventional TFP is no longer suitable for measuring the quality of China’s sustainable economic development [8,9]. Therefore, some scholars have added the constraints of resources and the environment [10,11,12], i.e., energy consumption and environmental pollution, to the measurement model and have produced a new indicator, GTFP, for measuring the growth of the green economy. A common practice is to treat resource factors as input indicators in an accounting framework, but no consensus has been reached on how to treat the factor of environmental pollution [13,14,15]. Some scholars, such as Domazlicky et al. [16], hold the view that the pollution due to production should be deemed an underutilized factor of production and that the amounts of discharged pollutants actually represent an aggregation of various resources that have not been converted into effective output, so pollutant discharge should be classified as an input indicator. However, this line of thinking belittles the actual damage to the environment caused by pollution, whose sole classification as waste is tantamount to a failure to recognize the harmful nature of pollution. Another group of scholars represented by Nanere et al. [17] divide the output factors into good outputs and bad outputs. As a by-product of industrial production, the environmental pollution factor is classified as a bad output and an unexpected output, and, in this way, it has been incorporated into the accounting framework of GTFP. This approach is more realistic and is the way most scholars currently treat the environmental pollution factor [18,19,20].

At present, the calculation models used by scholars [21,22] include the following: (1) the output distance function (DF) and Malmquist index. DF can be used to calculate multiple outputs from multiple inputs to measure the distance between a production unit and the production frontier when environmental pollution and energy consumption are taken into account. The Malmquist index model can be used to measure the dynamic efficiency of time series data and can also be used to decompose the comprehensive efficiency index into the technical efficiency change index and technical progress index [23,24]. (2) The directional distance function (DDF) and Malmquist–Luenberger (ML) index: to cope with increases in expected outputs coexisting with decreases in unexpected outputs resulting from multiple inputs and to overcome the radial characteristics of DF, Chung & Fare [25] put forward an analytical model based on the DDF and formulated the ML index to measure GTFP, which incorporates the environmental factor. (3) The Slacks-based measure (SBM)-DDF model and ML index: there are problems related to angle sections and radial variations, so traditional methods for data-envelopment analysis (DEA) often yield GTFP measurements that deviate significantly from reality [26,27]. To solve this problem, Tone [28] proposed a non-radial and non-angular SBM-DEA model, which can not only solve the problem of unexpected outputs but can also solve the problem of slack variables in radial models. Using an SBM-DEA model, Fare [29] constructed a non-radial and non-angular DDF model (SBM-DDF) that incorporated the slackness of variables and achieved a higher accuracy in efficiency assessment.

1.2. Environmental Impacts of Foreign Direct Investment (FDI)

Many scholars have conducted extensive research on the environmental impacts of FDI and have considered whether its continuous growth would increase the environmental burdens on the host country. There are currently two hypotheses: (1) the pollution refuge hypothesis—FDI leads to increased environmental pollution; (2) the pollution halo hypothesis—FDI helps reduce environmental pollution. The pollution refuge hypothesis was first put forward by Walter [30]. Worldwide capital flows allow polluting industries to migrate across borders more easily. Developed countries have more draconian environmental regulations, so their multinational enterprises reduce high execution costs by transferring their highly polluting industries via FDI to developing countries, which become pollution refuges. Many scholars [31,32] have tested this hypothesis with regard to different host countries and industries. Jiang et al. [33] tested the correlation between energy consumption and FDI by using panel data from 1990–2011 in 65 countries. Their results show positive correlations in all except the low-income countries, i.e., increased FDI increases energy consumption and pollution.

The pollution halo hypothesis was first put forward by Zarsky [34]. Foreign capital from developed countries or regions can bring advanced technologies and management expertise to underdeveloped countries and regions. This spillover effect can boost the development of environmental conservation technologies in the host countries and thus improve their environmental quality. Tang and Tan [35] analyzed the FDI, carbon emissions, and economic development from 1976 to 2009 in Vietnam, then concluded that FDI was conducive to a reduction in carbon emissions. Abdouli et al. [36] investigated the correlation between environmental quality and FDI in Middle Eastern and North African (MENA) countries by using panel data and a vector autoregressive model, and concluded that FDI could reduce industrial pollution in these countries.

1.3. FDI’s Effects on TFP and GTFP

FDI affects a host country’s economic growth in two main ways [37]. First, FDI provides high-quality capital that stimulates investment and production. Second, the advanced production technologies and management expertise brought by FDI can drive the enterprises in the host country to improve their production efficiency, thus promoting economic growth. Some empirical studies have shown that FDI produces a positive technology spillover effect in the host country, i.e., the advanced technologies and management expertise brought by multinational enterprises can promote TFP. Ahmed [38] used a time series regression model to analyze the spillover effects of FDI and concluded that FDI inflow boosted Malaysia’s economic growth by driving investment in production. Some scholars, such as Kukulski and Ryan [39], believe that FDI’s ability to cause positive spillover effects depends on the institutional environment, potential technological gaps, and threshold conditions in the host country. Orlic et al. [40] used a dynamic panel model to study the spillover effects of foreign companies on the TFP of local manufacturing enterprises and proved that the local enterprises benefited from investments by foreign companies in upstream knowledge-intensive services and downstream manufacturing industries.

Researching sustainability, quality, and other issues in economic development, many scholars have shifted their focus to FDI’s effect on GTFP. Empirical studies in this area have yielded many inconsistent results due to the differences in the research methods, research scopes, and indicators used. Some studies [41,42] have shown FDI’s effects on GTFP to be positive. For example, Daddi et al. [43] have shown that inflows of foreign capital created a good competitive atmosphere, which stimulates technological innovation, reduces energy consumption, and decreases the discharge of pollutants by the host country’s enterprises. Yoon et al. [44] pointed out that FDI inflows could indirectly improve income levels in the host country. They postulated that citizens with increased incomes would have higher requirements for environmental quality and influence enterprises to implement green transformation. Some studies [45,46] have concluded that the effects of FDI on GTFP were significantly negative. Rafindadi et al. [47] pointed out that, if developing countries blindly pursued economic growth but neglected environmental protection, then the developed countries would prefer to relocate their highly polluting factories to these countries. Such relocations would not be conducive to the development of green technologies and the growth of GTFP.

The following facts have been found in the relevant literature. First, most scholars [48,49,50] have adopted nonparametric methods to calculate GTFP by regarding resource consumption as an input variable and the environmental pollution factor as an unexpected output variable in their accounting systems. Calculations of China’s GTFP vary significantly because of differences in sample scopes, pollution indicators, and calculation methods. Second, scholars [51,52,53] have reached similar conclusions in their theoretical analyses of the influencing mechanisms by which FDI affects GTFP. Empirical studies on how FDI influences the traditional TFP have yielded significantly different conclusions because of differences in perspective. Few studies have been conducted to investigate the influence of capital flow on GTFP at the city level. Third, investigations of the effects of FDI, such as the threshold effect, on GTFP have gradually shifted from linear relationships to nonlinear relationships. However, few relevant theoretical and empirical studies can be found at present, so the selection of threshold variables is still constrained. Finally, most studies [54,55,56] regard FDI as homogeneous capital and ignore the qualitative differences between different foreign investments. In sum, most studies [57,58,59] have analyzed only the influences of the scale of FDI on TFP but have paid little attention to the influence of the quality of FDI on GTFP.

This study tackled the question of whether capital inflow could improve GTFP. Twenty-seven cities in the Yangtze River Delta were chosen as samples. After analyzing the FDI in the region, we estimated the GTFP and its decomposed indicators of each city’s energy consumption and environmental pollution for 2004–2019. Then, we analyzed the estimation results from different perspectives. Next, we applied a fixed-effect model to the panel data to empirically test if the scale and quality of FDI had improved GTFP in the Yangtze River Delta. We used the panel threshold model to empirically test the threshold effect of FDI on GTFP and explore its nonlinear effects. Finally, we formulated several suggestions regarding strategies for attracting investment and promoting the growth of the green economy.

2. Methodology and Data

2.1. Measurement of GTFP of Cities in Yangtze River Delta

2.1.1. Measurement Method and Model Setting

Among the methods for calculating GTFP, DEA has the advantage of eliminating the need to determine the specific function form in advance. When non-radial and non-directional DDF is combined with the ML index, the model can not only decompose the efficiency index but can also take the negative output into account while calculating the positive output. This method fully involves the outputs of economic growth, energy consumption, and environmental pollution in accordance with the sustainable development policy of the government. Therefore, this study chose the DEA-ML index method to estimate GTFP by constructing a DDF, then combining the DEA method with the ML index to calculate the green ML (GML) while considering the unexpected output and its decomposed indexes, which are the green technical change index (GTC) and green technical efficiency change index (GEC).

(1) DDF

Firstly, the production possibility set, which includes input, output, and unexpected output, is constructed. Assume that, in the periods of $\mathrm{t}=\mathrm{1,2},\dots ,\mathrm{T}$, there are $\mathrm{k}=1,2,\dots ,\mathrm{K}$ regions or cities, whose production activities involve t $\mathrm{m}=\mathrm{1,2},\dots ,\mathrm{M}$ factors of production $\mathrm{X}$, $\mathrm{n}=\mathrm{1,2},\dots ,\mathrm{N}$ expected outputs $\mathrm{Y}$, and $\mathrm{r}=\mathrm{1,2},\dots ,\mathrm{R}$ unexpected outputs $\mathrm{B}$. Then, the input-output vector of city k in year t can be expressed as (${\mathrm{x}}_{\mathrm{k}}^{\mathrm{t}},{\mathrm{y}}_{\mathrm{k}}^{\mathrm{t}},{\mathrm{b}}_{\mathrm{k}}^{\mathrm{t}}$) and the production possibility set can be expressed as $\mathrm{P}\left(\mathrm{x}\right)=\left\{\left(\mathrm{y},\mathrm{b}\right):\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{b}\mathrm{l}\mathrm{e} \mathrm{o}\mathrm{u}\mathrm{t}\mathrm{p}\mathrm{u}\mathrm{t} \mathrm{y} \mathrm{f}\mathrm{o}\mathrm{r} \mathrm{i}\mathrm{n}\mathrm{p}\mathrm{u}\mathrm{t} \mathrm{x}\right\}$. Thus, the following equation can be obtained:

$$\begin{array}{c}{\mathrm{P}}^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}}\right)=\{\left({\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}}\right): {\sum }_{\mathrm{k}=1}^{\mathrm{K}}{\mathrm{z}}_{\mathrm{k}}^{\mathrm{t}}{\mathrm{y}}_{\mathrm{k}\mathrm{n}}^{\mathrm{t}}\ge {\mathrm{y}}_{\mathrm{k}\mathrm{n}}^{\mathrm{t}},\mathrm{n}=1,\dots ,\mathrm{N}; {\sum }_{\mathrm{k}=1}^{\mathrm{K}}{\mathrm{z}}_{\mathrm{k}}^{\mathrm{t}}{\mathrm{b}}_{\mathrm{k}\mathrm{r}}^{\mathrm{t}}={\mathrm{b}}_{\mathrm{k}\mathrm{r}}^{\mathrm{t}},\mathrm{r}=1,\dots ,\mathrm{R};\\ {\sum }_{\mathrm{k}=1}^{\mathrm{K}}{\mathrm{z}}_{\mathrm{k}}^{\mathrm{t}}{\mathrm{x}}_{\mathrm{k}\mathrm{m}}^{\mathrm{t}}\le {\mathrm{x}}_{\mathrm{k}\mathrm{m}}^{\mathrm{t}},\mathrm{m}=1,\dots ,\mathrm{M}; {\sum }_{\mathrm{k}=1}^{\mathrm{K}}{\mathrm{z}}_{\mathrm{k}}^{\mathrm{t}}=1, {\mathrm{z}}_{\mathrm{k}}^{\mathrm{t}}\ge 0,\mathrm{k}=1,\dots ,\mathrm{K}\} \end{array}$$

(1)

where ${\mathrm{z}}_{\mathrm{k}}^{\mathrm{t}}$ represents the weight of a cross-section observation value and the constraint ${\sum }_{\mathrm{k}=1}^{\mathrm{K}}{\mathrm{z}}_{\mathrm{k}}^{\mathrm{t}}=1$ indicates that the production possibility frontier is subject to variable returns to scale (VRS). However, the constraint assumes that the inputs and outputs have strong disposability. Weak disposability of the unexpected outputs, i.e., their reduction, decreases the expected outputs. This paper follows the practice of Kuosmanen (2005) of decomposing parameter z into the coefficient $\mathsf{\mu }$ of the disposable part and the coefficient $\mathsf{\lambda }$ of the unchanged part [13]:

$$\begin{gathered} \begin{array}{c}{\mathrm{P}}^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}}\right)=\{\left({\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}}\right): {\sum }_{\mathrm{k}=1}^{\mathrm{K}}{\mathsf{\lambda }}_{\mathrm{k}}^{\mathrm{t}}{\mathrm{y}}_{\mathrm{k}\mathrm{n}}^{\mathrm{t}}\ge {\mathrm{y}}_{\mathrm{k}\mathrm{n}}^{\mathrm{t}},\mathrm{n}=1,\dots ,\mathrm{N}; {\sum }_{\mathrm{k}=1}^{\mathrm{K}}{\mathsf{\lambda }}_{\mathrm{k}}^{\mathrm{t}}{\mathrm{b}}_{\mathrm{k}\mathrm{r}}^{\mathrm{t}}={\mathrm{b}}_{\mathrm{k}\mathrm{r}}^{\mathrm{t}},\mathrm{r}=1,\dots ,\mathrm{R};\\ \end{array} \\ {\sum }_{\mathrm{k}=1}^{\mathrm{K}}\left({\mathsf{\lambda }}_{\mathrm{k}}^{\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{k}}^{\mathrm{t}}\right){\mathrm{x}}_{\mathrm{k}\mathrm{m}}^{\mathrm{t}}\le {\mathrm{x}}_{\mathrm{k}\mathrm{m}}^{\mathrm{t}},\mathrm{m}=1,\dots ,\mathrm{M};{\sum }_{\mathrm{k}=1}^{\mathrm{K}}\left({\mathsf{\lambda }}_{\mathrm{k}}^{\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{k}}^{\mathrm{t}}\right)=1, \\ {\mathrm{z}}_{\mathrm{k}}^{\mathrm{t}}\ge 0,\mathrm{k}=1,\dots ,\mathrm{K}\} \end{gathered}$$

(2)

Then, we construct the following DDF:

$$D_0^t\left({x^t,y}^t,b^t;g^y,g^b\right)=max\left\{\right(\beta :\left(y^t,b^t\right)+\beta g\in P^t\left(x^t\right)$$

(3)

where $\mathrm{g}=\left({\mathrm{g}}^{\mathrm{y}},{\mathrm{g}}^{\mathrm{b}}\right)$ is the increase in the direction vector of the expected outputs and unexpected outputs, and β is the maximum possibility that the expected outputs increase proportionally and the unexpected outputs decrease proportionally for a given input.

Finally, the DDF of individual k in period t is used to solve the following linear programming:

$$\begin{gathered} {\mathrm{D}}_0^{\mathrm{t}}\left({\mathrm{x}}_{\mathrm{k}}^{\mathrm{t}},{\mathrm{y}}_{\mathrm{k}}^{\mathrm{t}},{\mathrm{b}}_{\mathrm{k}}^{\mathrm{t}};\mathrm{g}\right)=\max \mathsf{\rho } \\ \mathrm{s}.\mathrm{t}.{\sum }_{\mathrm{k}=1}^{\mathrm{K}}{\mathsf{\lambda }}_{\mathrm{k}}^{\mathrm{t}}{\mathrm{y}}_{\mathrm{k}\mathrm{n}}^{\mathrm{t}}\ge \left(1+\mathsf{\beta }\right){\mathrm{y}}_{\mathrm{k}\mathrm{n}}^{\mathrm{t}},\mathrm{n}=1,\dots ,\mathrm{N} \\ {\sum }_{\mathrm{k}=1}^{\mathrm{K}}{\mathsf{\lambda }}_{\mathrm{k}}^{\mathrm{t}}{\mathrm{b}}_{\mathrm{k}\mathrm{r}}^{\mathrm{t}}=\left(1-\mathsf{\beta }\right){\mathrm{b}}_{\mathrm{k}\mathrm{r}}^{\mathrm{t}},\mathrm{r}=1,\dots ,\mathrm{R} \\ {\sum }_{\mathrm{k}=1}^{\mathrm{K}}{(\mathsf{\lambda }}_{\mathrm{k}}^{\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{k}}^{\mathrm{t}}){\mathrm{x}}_{\mathrm{k}\mathrm{m}}^{\mathrm{t}}\le \left(1-\mathsf{\beta }\right){\mathrm{x}}_{\mathrm{k}\mathrm{m}}^{\mathrm{t}},\mathrm{m}=1,\dots ,\mathrm{M} \\ {\sum }_{\mathrm{k}=1}^{\mathrm{K}}{(\mathsf{\lambda }}_{\mathrm{k}}^{\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{k}}^{\mathrm{t}}){\mathrm{x}}_{\mathrm{k}\mathrm{m}}^{\mathrm{t}}\le \left(1-\mathsf{\beta }\right){\mathrm{x}}_{\mathrm{k}\mathrm{m}}^{\mathrm{t}},\mathrm{m}=1,\dots ,\mathrm{M} \\ {\sum }_{\mathrm{k}=1}^{\mathrm{K}}{(\mathsf{\lambda }}_{\mathrm{k}}^{\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{k}}^{\mathrm{t}})=1,{\mathrm{z}}_{\mathrm{k}}^{\mathrm{t}}\ge 0,\mathrm{k}=1,\dots ,\mathrm{K} \end{gathered}$$

(4)

(2) Malmquist–Luenberger (ML) exponential model

Applying the DDF containing unexpected outputs to the ML exponential model, we can obtain the formula of the efficiency index:

$$\mathrm{G}\mathrm{M}\mathrm{L}={\{{\displaystyle \frac{1+{\mathrm{D}}_0^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}},{\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}};{\mathrm{g}}^{\mathrm{t}}\right)}{1+{\mathrm{D}}_0^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}+1},{\mathrm{y}}^{\mathrm{t}+1},{\mathrm{b}}^{\mathrm{t}+1};{\mathrm{g}}^{\mathrm{t}+1}\right)}}\times {\displaystyle \frac{1+{\mathrm{D}}_0^{\mathrm{t}+1}\left({\mathrm{x}}^{\mathrm{t}},{\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}};{\mathrm{g}}^{\mathrm{t}}\right)}{1+{\mathrm{D}}_0^{\mathrm{t}+1}\left({\mathrm{x}}^{\mathrm{t}+1},{\mathrm{y}}^{\mathrm{t}+1},{\mathrm{b}}^{\mathrm{t}+1};{\mathrm{g}}^{\mathrm{t}+1}\right)}}\}}^{{\displaystyle \frac{1}{2}}}$$

(5)

$$\mathrm{G}\mathrm{E}\mathrm{C}={\displaystyle \frac{1+{\mathrm{D}}_0^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}},{\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}};{\mathrm{g}}^{\mathrm{t}}\right)}{1+{\mathrm{D}}_0^{\mathrm{t}+1}\left({\mathrm{x}}^{\mathrm{t}+1},{\mathrm{y}}^{\mathrm{t}+1},{\mathrm{b}}^{\mathrm{t}+1};{\mathrm{g}}^{\mathrm{t}+1}\right)}}$$

(6)

$$\mathrm{G}\mathrm{T}\mathrm{C}={\displaystyle \frac{1+{\mathrm{D}}_0^{\mathrm{t}+1}\left({\mathrm{x}}^{\mathrm{t}},{\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}};{\mathrm{g}}^{\mathrm{t}}\right)}{1+{\mathrm{D}}_0^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}},{\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}};{\mathrm{g}}^{\mathrm{t}}\right)}}\times {\displaystyle \frac{1+{\mathrm{D}}_0^{\mathrm{t}+1}\left({\mathrm{x}}^{\mathrm{t}+1},{\mathrm{y}}^{\mathrm{t}+1},{\mathrm{b}}^{\mathrm{t}+1};{\mathrm{g}}^{\mathrm{t}+1}\right)}{1+{\mathrm{D}}_0^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}+1},{\mathrm{y}}^{\mathrm{t}+1},{\mathrm{b}}^{\mathrm{t}+1};{\mathrm{g}}^{\mathrm{t}+1}\right)}}$$

(7)

$$\mathrm{G}\mathrm{M}\mathrm{L}=\mathrm{G}\mathrm{E}\mathrm{C}\times \mathrm{G}\mathrm{T}\mathrm{C}$$

(8)

where the GML value ${\mathrm{G}\mathrm{T}\mathrm{F}\mathrm{P}}_{\mathrm{t}}^{\mathrm{t}+1}$ is the GTFP change index of individual k from period t to period $\mathrm{t}+1$. $\mathrm{G}\mathrm{E}\mathrm{C}$ and $\mathrm{G}\mathrm{T}\mathrm{C}$ are the green technology efficiency change index and green technology change index, respectively. Measurement values of GML, GEC, and GTC greater or smaller than 1 indicate increased or decreased GTFP, green technology efficiency, and green technological progress, respectively.

2.1.2. Variable Selection and Data Sources

(1) Input indicators

Labor input: number of employed persons in city i at the end of year t.

Capital input: stock of fixed capital determined by the perpetual inventory method. The capital investment in city i in year t can be expressed as ${\mathrm{K}}_{\mathrm{i},\mathrm{t}}=\left(1-\mathsf{\beta }\right){\mathrm{K}}_{\mathrm{i},\mathrm{t}-1}+{\mathrm{I}}_{\mathrm{i},\mathrm{t}}$, where ${\mathrm{I}}_{\mathrm{i},\mathrm{t}}$ is the increase in the investments in the social fixed assets. As per the factor prices, the fixed assets investment price index of each city’s province for 2004 is used to correct the total social fixed assets investment. I is the base period capital stock ${\mathrm{K}}_0={\mathrm{I}}_0/\left({\mathrm{g}}_{\mathrm{i}}+\mathsf{\beta }\right)$ of city i, where ${\mathrm{g}}_{\mathrm{i}}$ is the geometric average growth rate of the actual investment during the study period. β is the depreciation rate of fixed assets, which is set to 0.96% as per Shan Haojie (2009) [60].

Energy input: the consumption of coal, oil, and natural gas in each country or province. Because of the lack of statistical data for individual cities, we used the annual electricity consumption stated in the statistical yearbooks of each city.

(2) Output indicators

Expected output: real GDP of each city excluding price factors. Taking 2004 as the base period, we used the GDP deflator of each city’s province to convert the city’s nominal GDPs of the study period.

Unexpected output: assessments of environmental pollution based on single pollutants are not comprehensive. The DEA-ML method allows multiple unexpected outputs to be taken into account for the calculation of efficiency, but the pollutants should satisfy the weak disposability required by Equation (4), so we selected industrial SO₂ emissions, industrial soot emissions, and industrial wastewater discharge in each city for assessing environmental pollution as per the practice of most Chinese scholars. To ensure the integrity of the data, missing data of the pollution emissions in individual cities were supplemented by interpolation.

The descriptive statistics of the indicators included in the GML accounting system are shown in

Table 1. Descriptive statistics of GML measurement indicators.

Indicator Type	Variable Name	Unit	Observed Value	Average	Standard Deviation	Minimum Value	Maximum Value
	Labor input	10,000	432	91.96	109.78	6.38	730.46
Input indicator	Capital stock	100 million yuan	432	8700.05	7993.41	179.69	42,129.12
	Energy consumption	100 million kWh	432	301.18	311.94	11.43	1568.58
Expected output	Real GDP	100 million yuan	432	1641.74	1804.69	92.43	10,785.16
	SO2 emissions	1 million tons	432	574.67	607.44	13.84	4963.77
Unexpected output	Soot emissions	100 tons	432	264.33	205.63	7.92	1314.33
	Wastewater discharge	100 tons	432	158.08	166.63	4.86	857.35

. All data for the study period were obtained from the relevant statistical yearbooks of 27 cities in the Yangtze River Delta, China City statistical yearbooks, Jiangsu Province statistical yearbooks, Zhejiang Province statistical yearbooks, and Anhui Province statistical yearbooks.

2.2. Empirical Analysis of FDI’s Effects on GTFP

After analyzing and measuring the effects of FDI on GTFP, we conducted an empirical test to measure the linear effects of FDI quantity and FDI quality on the GTFP of the 27 cities in the Yangtze River Delta region separately by using their panel data for 2004–2019. We also conducted a threshold linear regression of foreign investment quality, economic development, industrial structure, internet penetration rate, etc., to reveal the nonlinear influence of the current level of foreign capital’s direct utilization on GTFP.

2.2.1. Linear Regression Model

The basic form of the international research and development (R&D) spillover model, the Coe and Helpman model (C–H model) of TFP, is as follows:

$${\mathrm{l}\mathrm{n}\mathrm{T}\mathrm{F}\mathrm{P}}_{\mathrm{i}}={\mathsf{\beta }}_{\mathrm{i}0}+{\mathsf{\beta }}_{\mathrm{i}\mathrm{d}}{\mathrm{l}\mathrm{n}\mathrm{S}}_{\mathrm{i}\mathrm{d}}+{\mathsf{\beta }}_{\mathrm{i}\mathrm{f}}{\mathrm{l}\mathrm{n}\mathrm{S}}_{\mathrm{i}\mathrm{f}}$$

(9)

where i represents different countries or regions, d represents the interior of country/region i, and f represents the exterior of country/region i. Thus, ${\mathrm{S}}_{\mathrm{i}\mathrm{d}}$ represents the R&D spillover stock originating from China and ${\mathrm{S}}_{\mathrm{i}\mathrm{f}}$ represents the R&D spillover stock from abroad. A basic paradigm for scholars to study technology spillover, the C–H model is based on the view that the technology spillover effects of investment have important impacts on the technological progress of a host country or region. Inspired by the C–H model and the results of previous studies, we constructed the following basic linear model:

$${\mathrm{l}\mathrm{n}\mathrm{G}\mathrm{T}\mathrm{F}\mathrm{P}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{F}\mathrm{D}\mathrm{I}}_{\mathrm{i}\mathrm{t}}+\mathsf{\beta }{\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathrm{u}}_{\mathrm{i}}{+\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(10)

$${\mathrm{l}\mathrm{n}\mathrm{G}\mathrm{T}\mathrm{F}\mathrm{P}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{F}\mathrm{D}\mathrm{I}\mathrm{Q}}_{\mathrm{i}\mathrm{t}}+\mathsf{\beta }{\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathrm{u}}_{\mathrm{i}}{+\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(11)

where i represents the city (I = 1, 2,…27), t represents the period (t = 1, 2,…16), GTFP is green total factor productivity, FDI represents the scale of directly utilized foreign investment, FDIQ represents the quality of FDI, ${\mathrm{u}}_{\mathrm{i}}$ represents the individual effects of the city, ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$ is a random disturbance term, and ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$ is a control variable. The level of economic development (ECO), level of industrial structure (IND), internet penetration rate (NET), level of openness to trade (OPEN), environmental regulation level (EVO), and level of government financial intervention (GOV) were selected as control variables. Equations (10) and (11) can be extended to:

$$\begin{gathered} {lnGTFP}_{it}={\beta }_0+{\beta }_1{FDI}_{it}+{\beta }_2{ECO}_{it}+{\beta }_3{IND}_{it}+{\beta }_4{NET}_{it}+{ \beta }_5{OPEN}_{it}+ \\ {\beta }_6{EVO}_{it}+{\beta }_7{GOV}_{it}+u_i{+\epsilon }_{it} \end{gathered}$$

(12)

$$\begin{gathered} {lnGTFP}_{it}={\beta }_0+{\beta }_1{FDIQ}_{it}+{\beta }_2{ECO}_{it}+{\beta }_3{IND}_{it}+{\beta }_4{NET}_{it}+{\beta }_5{OPEN}_{it} \\ +{\beta }_6{EVO}_{it}+{\beta }_7{GOV}_{it}+u_i{+\epsilon }_{it} \end{gathered}$$

(13)

2.2.2. Nonlinear Regression Model

We tested the nonlinear influence of FDI on GTFP by using the individual fixed-effect panel threshold model of Hansen (1999) [61]. Taking a single threshold as an example, the basic form of the model is:

$$\left\{\begin{array}{c}{\mathrm{Y}}_{\mathrm{i}\mathrm{t}}={\mathsf{\alpha }}_{\mathrm{i}}+{\mathrm{X}}_{\mathrm{i}\mathrm{t}}^{\prime }{\mathsf{\beta }}_1+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}, \mathrm{i}\mathrm{f} {\mathrm{q}}_{\mathrm{i}\mathrm{t}}\le \gamma \\ {\mathrm{Y}}_{\mathrm{i}\mathrm{t}}={\mathsf{\alpha }}_{\mathrm{i}}+{\mathrm{X}}_{\mathrm{i}\mathrm{t}}^{\prime }{\mathsf{\beta }}_2+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}},\mathrm{i}\mathrm{f} {\mathrm{q}}_{\mathrm{i}\mathrm{t}}>\gamma \end{array}\right.$$

(14)

where i = 1, 2,..., N; t = 1, 2,..., T, ${\mathrm{q}}_{\mathrm{i}\mathrm{t}}$ is the threshold variable; $\mathsf{\gamma }$ is the threshold value; ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}^{\prime }$ is an exogenous explanatory variable; and ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$ is independently and identically distributed but unrelated to ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}^{\prime }$. Thus, the basic single-threshold model is:

$${lnGTFP}_{it}=\alpha X_{it}+{\alpha }_1{FDI}_{it}·I\left(q_{it}\le \gamma \right)+{\alpha }_2{FDI}_{it}·I\left(q_{it}>\gamma \right)+u_i{+\epsilon }_{it}$$

(15)

where $\mathrm{I}$(·) is an indicative function. As discussed for the influence mechanism in previous sections, the threshold variables in this study are ${\mathrm{F}\mathrm{D}\mathrm{I}\mathrm{Q}}_{\mathrm{i}\mathrm{t}},{\mathrm{E}\mathrm{C}\mathrm{O}}_{\mathrm{i}\mathrm{t}}$, ${\mathrm{I}\mathrm{N}\mathrm{D}}_{\mathrm{i}\mathrm{t}}$, ${\mathrm{E}\mathrm{V}\mathrm{O}}_{\mathrm{i}\mathrm{t}}$, and ${\mathrm{N}\mathrm{E}\mathrm{T}}_{\mathrm{i}\mathrm{t}}$, while ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$ is the control variable group. After the addition of the threshold variables, the threshold model becomes:

$${lnGTFP}_{it}=\mu X_{it}+{\mu }_1{FDI}_{it}·I\left({FDIQ}_{it}\le \sigma \right)+{\mu }_2{FDI}_{it}·I\left({FDIQ}_{it}>\sigma \right)+u_i^0+{\epsilon }_{it}^0$$

(16)

$${lnGTFP}_{it}=\alpha X_{it}+{\alpha }_1{FDI}_{it}·I\left({ECO}_{it}\le \gamma \right)+{\alpha }_2{FDI}_{it}·I\left({ECO}_{it}>\gamma \right)+u_i^1+{\epsilon }_{it}^1$$

(17)

$${lnGTFP}_{it}=\lambda X_{it}+{\lambda }_1{FDI}_{it}·I\left({IND}_{it}\le k\right)+{\lambda }_2{FDI}_{it}·I\left({IND}_{it}>k\right)+u_i^2+{\epsilon }_{it}^2$$

(18)

$${lnGTFP}_{it}=\phi X_{it}+{\phi }_1{FDI}_{it}·I\left({EVO}_{it}\le \eta \right)+{\phi }_2{FDI}_{it}·I\left({EVO}_{it}>\eta \right)+u_i^3+{\epsilon }_{it}^3$$

(19)

$${lnGTFP}_{it}=\rho X_{it}+{\rho }_1{FDI}_{it}·I\left({NET}_{it}\le \delta \right)+{\rho }_2{FDI}_{it}·I\left({NET}_{it}>\delta \right)+u_i^4+{\epsilon }_{it}^4$$

(20)

2.3. Variable Selection and Data Processing

2.3.1. Explained Variable

As can be seen from the measurement results provided earlier in this paper, the calculation obtained from the DEA method is a kind of relative efficiency and the GML index is the year-on-year change index of GTFP. To truly reflect the economic growth of a given year, we took the GTFP of 2004 as the base number 1 and calculated the GTFP values of subsequent years by performing multiplication with the GML index.

2.3.2. Explanatory Variables

(1) Amount of FDI: The original figures of the sample cities in the study period are recorded in US dollars. The FDI figures used in this paper are RMB figures converted from USD according to the average exchange rate in the study period. (2) Quality of FDI (FDIQ): this value is calculated in the following way. First, the sum of the FDI amounts in RMB of the sample cities is divided by the number of FDI projects in that year. Then, the figure is modified according to the proportion of FDI in the tertiary industry characterized by low pollution and low resource consumption, including high-tech industries, services, renewable energy industries, etc.

2.3.3. Control Variables and Threshold Variables

(1) Economic development level (ECO): the actual value of the per capita GDP of each city after the exclusion of the price factor. ECO directly affects the TFP level of a region. In addition, the introduction of FDI may promote the formation of unbalanced industrial structures in economically backward cities and may also increase the number of polluting enterprises, thus restricting the growth of the green economy. However, the negative effects brought by the introduction of foreign capital are relatively small in cities with a high ECO. Therefore, we should consider not only the influence of economic development on GTFP but also the possible threshold constraints of the ECO in FDI’s effects on GTFP.

(2) Industrial structure (IND): the ratio of the added value of the secondary industry to the GDP of a region. The secondary industry is the industry with the highest energy consumption and pollutant discharge, so when the economy of a certain region is dominated by the secondary industry, the growth of GTFP may be hindered. The progress in industrial structure upgrading has changed the industrial structure of the Yangtze River Delta region. With the decline in the weight of the secondary industry and the increase in the weight of the tertiary industry, the latter has been attracting more FDI from the secondary industry. This trend reduces the consumption of natural resources and eases environmental pressures. Therefore, there may be a threshold constraint on the effects of the industrial structure on the growth of GTFP.

(3) Network penetration rate (NET): this study used the ratio of the number of internet users in each city to the total number of households in each region to measure the internet penetration rate. As a new technology, the internet not only promotes science and technology but also improves the efficiency of technological transformation, thus improving TFP. In addition, the popularization of the internet promotes the exchange and dissemination of information, which facilitates the introduction of FDI, makes the negotiation between enterprises more efficient, and lowers transaction costs. Poor network infrastructure will restrict the efficiency of economic activities and generate errors in information transmission, which would lead to mistakes in resource allocation. Therefore, there may be a threshold constraint on FDI’s effects on GTFP growth.

(4) Environmental regulation (EVO): measured by the effects of controlling pollutant discharge or the costs of controlling environmental pollution. This study used the sewage treatment rate of each city as the index of environmental regulation. On the one hand, overly lax environmental regulation lowers the entry threshold of foreign capital but raises the probability of attracting high-energy-consumption and high-pollution enterprises. On the other hand, excessively draconian regulations increase compliance costs for foreign-funded enterprises. Therefore, we should consider not only the two-way influence of environmental regulation on GTFP but also the possible threshold constraint on FDI’s effects on GTFP growth.

(5) Human capital level (EDU): represented by the logarithmic number of college students in the total population. EDU not only represents the skill level and overall quality of workers but also labor productivity. Moreover, EDU has a positive external effect because a high EDU would promote the generation, dissemination, and transformation of new knowledge and technologies, thus improving the GTFP.

(6) Level of openness to trade (OPEN): represented by the logarithmic total import and export volumes in RMB. A higher OPEN leads to stronger spillover effects from FDI. However, extensive trade growth could also increase resource wastage and environmental pollution, thus restricting GTFP.

(7) Level of government intervention (GOV): the proportion of fiscal expenditures in the city budget in terms of GDP. GOV can directly reflect the degree of government intervention in social and economic activities, as well as its strength in guiding resource allocation.

The data for empirical analysis were obtained from China City Statistical Yearbooks, China City Construction Yearbooks, and the statistical yearbooks of the 27 sample cities for 2004–2019. The empirical analysis was conducted with Stata 15.1 software. The descriptive statistics of all the variables are shown in

Table 2. Descriptive statistics of variables.

Variables	Observations	Average	Standard Deviation	Minimum Value	Maximum Value
GTFP	432	0.950	0.310	0.158	2.709
FDI	432	1.383	1.909	0.005	13.162
FDIQ	432	0.329	0.482	0.007	3.716
EVO	432	0.830	0.188	0.052	0.998
GOV	432	0.124	0.049	0.050	0.280
EDU	432	0.505	0.925	0.052	2.542
ECO	432	3.140	2.288	0.590	19.880
NET	432	0.722	0.557	0.030	3.134
IND	432	0.510	0.078	0.270	0.750
OPEN	432	0.661	1.764	0.162	1.044

.

3. Results and Discussion

3.1. Calculation Results and Analysis

We used the MATLAB R2018b software to calculate the GML, GEC, and GTC of the 27 cities in the Yangtze River Delta for 2004–2019 and conducted a comprehensive analysis of their variation characteristics and the distribution of GTFP values using both longitudinal and transverse analyses.

(1) Longitudinal analysis

The calculated values were subjected to longitudinal analysis along the timeline.

Table 3. GML and its decomposed indexes in Yangtze River Delta from 2004 to 2019.

Year	GEC	GTC	GML
2004–2005	0.977	0.973	0.950
2005–2006	1.059	0.968	1.025
2006–2007	1.161	0.826	0.959
2007–2008	0.914	1.135	1.037
2008–2009	1.235	0.768	0.954
2009–2010	1.043	0.931	0.970
2010–2011	1.000	0.957	0.959
2011–2012	1.010	0.967	0.977
2012–2013	0.968	0.942	0.919
2013–2014	1.016	0.933	0.951
2014–2015	1.003	0.992	0.995
2015–2016	0.988	1.146	1.132
2016–2017	0.969	1.129	1.095
2017–2018	0.969	1.049	1.016
2018–2019	1.068	0.967	1.032
Average	1.022	0.975	0.997

shows these values, as well as the geometric mean values of GTFP, for the study period. As the efficiency values calculated by the DEA-ML method are dynamic indexes, the years shown in the table are interval values. Generally speaking, the geometric average growth rates of the GML, GEC, and GTC are −0.3%, 2.2%, and −2.5%, respectively. These figures indicate that the GML indexes of the 27 cities in the Yangtze River Delta region declined as a whole during the study period. The GEC was strengthened, but the GTC declined, as it was higher than 1 during 2007–2008 and 2015–2018 while being lower than 1 in the other years. The distributions of the GML and GTC are largely consistent over the years and indicate that the GML has increased mainly as a result of improvements in green technology rather than in green technology efficiency. The efficiency of green technology transformation is still a bit low given that China is in an era emphasizing innovation-driven development.

Figure 1 shows the fluctuations in and variation trends of GTFP. The study period can be divided into three phases, roughly corresponding to the 11th, 12th, and 13th Five-year Plans of China. In 2004–2010 (before and during the 11th Five-Year Plan), GTFP fluctuates in a wide range; in 2011–2015 (the 12th Five-Year Plan), GTFP is stable; in 2016–2019 (the early and middle parts of the 13th Five-Year Plan), GTFP fluctuates, with the ML index being above one each year.

As regards the variation trends, the GML hovers at about 1 in 2004–2008 and GTFP exhibits an upward trend. Affected by the financial crisis in 2008, the expected output GDP in the Yangtze River Delta region fell sharply, the GML dropped below 1, and GTFP fell sharply. In 2009, China’s government implemented a four-trillion-yuan economic stimulus plan. As a result, the weight of heavy industry in the economy increased and GTFP remained below 1 until 2013. During this period, the Yangtze River Delta region experienced an acceleration of industrial development while its output growth was accompanied by high energy consumption and high pollutant emissions, which led to a significant decline in GTFP as regards resources and the environment. During 2008–2014, improvements in green technology efficiency outpaced the progress in technology, probably because the enthusiasm of enterprises for R&D declined in the aftermath of the economic crisis. Meanwhile, the scale of the manufacturing industry increased rapidly while the government strengthened its control of environmental protection and energy conservation. During the 12th Five-Year Plan and after 2011, the government began to use a “Green Development Index” for green technology efficiency. The growth rate of GTFP began to rise in 2014 and maintained a positive growth rate after 2015. The policy of upgrading the industrial structure and transforming the economy into a high-quality development mode promoted green technology and increased GTFP.

(2) Transverse analysis

A transverse analysis was conducted to compare the values of the GML and its decomposed indexes for the 27 sample cities.

Table 4. Values of GML and its decomposed indexes of cities in Yangtze River Delta region.

Rank	City	GEC	GTC	GML
1	Shanghai	1.023	1.070	1.068
2	Hangzhou	1.026	1.023	1.050
3	Wenzhou	1.022	1.015	1.038
4	Hefei	1.036	0.993	1.028
5	Shaoxing	1.023	1.002	1.026
6	Nantong	1.025	1.001	1.025
7	Jiaxing	1.034	0.986	1.020
8	Chizhou	1.037	0.978	1.015
9	Nanjing	1.035	0.976	1.011
10	Yancheng	1.035	0.974	1.007
11	Ningbo	1.029	0.977	1.005
12	Xuancheng	1.016	0.989	1.005
13	Changzhou	1.039	0.965	1.003
14	Taizhou, Jiangsu	1.025	0.977	1.001
15	Zhenjiang	1.037	0.964	1.000
16	Yangzhou	1.030	0.969	0.999
17	Taizhou, Zhejiang	1.015	0.983	0.998
18	Zhoushan	1.027	0.969	0.995
19	Suzhou	1.025	0.958	0.994
20	Wuhu	1.029	0.963	0.991
21	Wuxi	1.037	0.951	0.986
22	Chuzhou	1.023	0.962	0.985
23	Anqing	1.020	0.957	0.977
24	Tongling	1.009	0.945	0.954
25	Jinhua	0.992	0.959	0.951
26	Huzhou	0.986	0.925	0.912
27	Maanshan	0.972	0.910	0.885
	Average	1.022	0.975	0.997

shows the values of the GML from 2004 to 2019 in descending order. The values of 15 cities are greater than 1, which means that most cities had improved GTFP to varying degrees.

The city with the largest GML value is Shanghai, followed by Hangzhou, Wenzhou, Hefei, Shaoxing, and Nantong. The GML, GEC, and GTC values of these six cities are all greater than 1, thereby indicating that their increases in GTFP had been driven by both green technological progress and green efficiency improvement. The GML and GEC values of Jiaxing, Chizhou, Nanjing, Yancheng, Ningbo, Xuancheng, Changzhou, Taizhou Jiangsu, and Zhenjiang are also greater than 1, but their GTC values are smaller than 1, which means that the GTFP growth in these cities had been driven by improvements in green technology efficiency, rather than the direct contributions of green technological progress. The GEC values of Yangzhou, Taizhou Zhejiang, Zhoushan, Suzhou, Wuhu, Wuxi, Chuzhou, Anqing, and Tongling are greater than 1, but their GML and GTC values are smaller than 1, which means that the efficiency of green technology in these cities had improved, but their declines in GTC had directly slowed the growth of GTFP. The GML, GEC, and GTC values of Jinhua, Huzhou, and Maanshan are all below 1, thus indicating that deterioration in green technology and technical efficiency have led to the negative growth of GTFP.

We followed the example of Yu and Wei [62] and classified the sample cities into four types according to their GML value: GTFP high-growth cities (GML ≥ 1.2), GTFP medium-growth cities (1.1 ≤ GML < 1.2), GTFP low-growth cities (1 ≤ GML < 1.1), and GTFP negative-growth cities (GML < 1).

Table 5. Statistics of the GTFP growth types of the cities in Yangtze River Delta region from 2004 to 2019.

Year	Number of GTFP High-Growth Cities	Number of GTFP Medium-Growth Cities	Number of GTFP Low-Growth Cities	Number of GTFP Negative-Growth Cities
2004–2005	1	0	9	17
2005–2006	2	1	16	8
2006–2007	0	0	18	9
2007–2008	2	2	20	3
2008–2009	0	1	8	18
2009–2010	0	0	10	17
2010–2011	0	1	4	22
2011–2012	0	0	9	18
2012–2013	1	0	5	21
2013–2014	1	0	4	22
2014–2015	0	2	9	16
2015–2016	6	8	11	2
2016–2017	1	6	18	2
2017–2018	1	4	10	12
2018–2019	1	3	16	7

shows the statistics of the growth types of the cities in different periods.

The proportions of negative-growth and low-growth cities are large and their fluctuation ranges are wide. The changes in their numbers are obviously complementary, and the study period can be divided into three phases accordingly. The first phase is from 2004 to 2008. During 2004–2005, 17 cities experienced negative growth of GTFP as a consequence of extensive economic growth. During 2005–2008, this situation significantly improved. The number of negative-growth cities decreased to three in 2008, mainly because the government had strengthened environmental regulation, energy conservation, and emission reduction in 2006, which promoted sustainable growth. In the second phase, from 2009 to 2014, the number of cities with negative growth of their GTFP increased sharply and peaked at 22 because the Yangtze River Delta region intensified efforts to attract foreign investment after the outbreak of the international economic crisis. As a result, some high-energy-consumption and high-pollution projects with quick returns were introduced into the Yangtze River Delta region. They increased environmental pollution, thus slowing down the growth of GTFP. In the third phase, from 2014 to 2019, the number of cities with negative growth of their GTFP began to decline because of intensified efforts to protect the environment, conserve energy, and reduce emissions as per the 12th Five-Year Plan.

The numbers of medium-growth and high-growth cities are basically stable, and their variation trends are consistent with those of cities with low growth rates. It is worth noting that the number of medium-growth cities and the number of high-growth cities reached their highest values of 8 and 6, respectively, in 2016, thus indicating that the innovation-driven strategy achieved remarkable results and most cities in the Yangtze River Delta region achieved significant improvements in GTFP.

3.2. Regression Results and Analysis

3.2.1. Linear Regression Analysis

(1) Multiple collinearity test

Table 6. Results of multiple collinearity test.

Variables	FDI	FDIQ	EVO	GOV	EDU	ECO	NET	IND	OPEN
FDI	1.000
FDIQ	−0.050	1.000
EVO	0.237	0.293	1.000
GOV	0.141	0.392	0.215	1.000
EDU	0.385	0.049	0.376	−0.089	1.000
ECO	0.450	−0.196	0.114	−0.232	0.353	1.000
NET	0.458	0.054	0.452	0.115	0.381	0.310	1.000
IND	−0.268	0.050	−0.040	−0.401	0.073	0.107	−0.401	1.000
OPEN	0.673	−0.184	0.447	−0.229	0.505	0.498	0.672	−0.095	1.000

shows the correlation coefficients between the variables in the model. The correlation coefficients between all variables are smaller than 0.8.

(2) Variable stationarity test

The time dimension T = 16 is approximately equal to the cross-section dimension N = 27, so it is feasible to perform unit root tests on the variables. The HT test, same unit root LLC test, and different unit root IPS tests were used to test the stationarity of the variables. The test results are shown in

Table 7. Results of variable stationarity test.

Variables	HT Test	LLC Test	IPS Test
GTFP	−4.478 *** (0.000)	−2.876 ** (0.002)	−3.291 *** (0.001)
FDI	−0.163 (0.435)	−4.003 *** (0.000)	−1.869 ** (0.031)
FDIQ	−1.478 *	−4.404 ***	−4.256 ***
	(0.069)	(0.000)	(0.000)
ECO	−9.473 *** (0.000)	−5.332 *** (0.000)	−3.882 *** (0.000)
IND	−2.491 ** (0.006)	−6.011 *** (0.000)	1.148 (0.874)
EVO	−2.764 ** (0.003)	−9.501 *** (0.000)	−3.159 *** (0.001)
EDU	−1.935 ** (0.027)	−4.995 *** (0.000)	−3.591 *** (0.000)
GOV	−3.327 *** (0.000)	−2.608 ** (0.004)	0.315 (0.624)
OPEN	−2.172 ** (0.015)	−4.655 *** (0.000)	−2.923 ** (0.002)

Note: ***, **, and * represent significance at the levels of 1%, 5%, and 10%, respectively; the figures in brackets are p-values.

. Most variables pass all three tests, and all variables pass at least two tests, thereby indicating that the series is stationary.

(3) Model form test

Before conducting the empirical analysis, we used the Hausman test to choose between a random effect model and a fixed-effect model. The results are shown in

Table 8. Results of model form test.

Model	Test Method	Null Hypothesis	p Value
(10)	Hausman test	The random effect model is better.	0.000
(11)	Hausman test	The random effect model is better.	0.000

. The p-value obtained from the Hausman test using Equations (10) and (11) is 0, which means a significant rejection of the original hypothesis. Therefore, the panel fixed-effect model is superior to the random effect model.

(4) Regression analysis

We used the panel fixed-effect model to perform a linear regression of the effects of FDI on GTFP in the cities in the Yangtze River Delta.

Table 9. Effects of FDI scale and FDI quality on GTFP.

	(10) Fixed Effect Model	(11) Fixed Effect Model
FDI	0.059 *** (0.013)
FDIQ		0.054 ** (0.021)
EVO	−0.239 ** (0.086)	−0.213 * (0.091)
GOV	−1.245 ** (0.456)	−0.730 (0.448)
EDU	0.068 ** (0.038)	0.064 ** (0.037)
ECO	0.103 *** (0.005)	0.102 *** (0.005)
NET	0.027 *** (0.028)	0.020 ** (0.027)
IND	−0.122 * (0.106)	−0.464 * (0.199)
OPEN	0.015 *** (0.026)	0.032 ** (0.029)
_cons	−0.315 (0.143)	0.137 (0.174)
N	432	432
R-sq	0.596	0.589
F-statistic	82.69	78.03

Note: ***, **, and * represent significance at the levels of 1%, 5%, and 10%, respectively; the figures in brackets are standard deviations; the below is the same.

shows that the coefficients of the FDI quantity and FDI quality are positive and significant at the levels of 1% and 5%, respectively, which indicates that the utilization of FDI has had significantly positive effects on the GTFP of the cities in the Yangtze River Delta, thus confirming the pollution halo hypothesis. The introduction of advanced environmental protection technology and equipment along with FDI has alleviated the environmental pollution in the Yangtze River Delta. Meanwhile, the government has paid more attention to the structure of FDI and supervising polluting enterprises. The above factors, combined, have promoted inflows of FDI into resource-saving and environment-friendly enterprises, thus effectively reducing pollution, promoting green technology, and improving green technology efficiency.

Besides the key explanatory variables FDI and FDIQ, the control variables have also affected the GTFP in the Yangtze River Delta region to some extent. The ECO is positive and significant at the level of 1%, thereby indicating that the growth of the per capita GDP has promoted the growth of the green economy and has supported the environmental Kuznets curve. The reason for this may be that the growth of the per capita GDP represents improvements in the incomes and living standards of the residents, who will then have requirements for higher-quality living environments and higher purchasing power to pay for environmentally friendly products with relatively higher prices. The enhancement of the public’s awareness of environmental protection will also encourage enterprises to adopt a more pro-environmental image. To satisfy the environmental protection aspirations of consumers, enterprises must pay more attention to social responsibility, utilize more environment-friendly technologies, and supply more environment-friendly products.

The EVO is negative and significant at the level of 5%, which indicates that improvements in environmental regulation could reduce the pressure on the environment but restrain the development of enterprises, especially in the manufacturing industry. Under the restrictions of environmental protection law, the compliance costs of industrial enterprises with high energy consumption and high pollution can be very high and dampen their enthusiasm for R&D, and are thereby not conducive to promoting technological progress.

The GOV is negative and significant at the level of 1%, which indicates that excessive financial intervention by the government has an inhibitory effect on the growth of GTFP because frequent interventions in economic activities disrupt the independent R&D of enterprises, thus hindering improvements in their R&D capabilities and technological progress.

The EDU is positive and significant at the level of 5%, which indicates that improvements in human capital significantly promote growth in GTFP. A higher EDU means that more well-educated workers are participating in economic activities. Not only have they mastered much knowledge and advanced technologies, but they can also acquire, utilize, and disseminate technologies more efficiently, thus promoting technological progress and improving technical efficiency. In addition, people with high levels of education are more capable of adopting energy conservation and environmental protection concepts and technologies, thus promoting GTFP.

The NET is positive and significant at the level of 1%, which indicates a significant promoting effect on GTFP. The development of networks can improve productivity, reduce enterprises’ operational costs, and promote technological progress through industrial integration.

The IND is negative and significant at 10%. As also suggested by studies at the national and provincial levels, the IND of the Yangtze River Delta region has an inhibitory effect on GTFP. With a rise in the IND, energy consumption and pollution in the manufacturing industry become more prominent, which is not conducive to improving GTFP.

The OPEN is positive and significant at 1% and 5%, which indicates a positive effect on the growth of the green economy. A high OPEN means more opportunities to interact with leading foreign enterprises, which would provide more advanced technologies and management expertise, thus promoting GTFP.

(5) Robustness test

Flaws in the variable selection and data processing methods would affect the reliability of the regression results, so we used the following three methods to test their robustness.

Method 1: Exclude samples of some years from the original sample set and reuse the data of the 27 cities in the Yangtze River Delta for 2008–2019 to make estimations;

Method 2: The effects of the FDI scale and FDI quality cannot be felt immediately, i.e., the FDI in one period may affect GTFP in the next period, so it is reasonable to shift the explanatory variable, GTFP, to the next period and redo the regression;

Method 3: Change the calculation method for the key explanatory variables. First, substitute the FDI flow with the actual proportion of FDI in the city’s GDP for the current year and substitute the FDI quality index with the amount of actually used FDI per project in the current year. Then, use the treatment described earlier to convert the original figures into RMB according to the average exchange rate of the current year.

The results of the robustness test are shown in

Table 10. Results of robustness test.

	Model (10)			Model (11)
	Test 1	Test 2	Test 3	Test 1	Test 2	Test 3
FDI	0.096 *** (0.017)	0.089 *** (0.014)	0.010 * (0.005)
FDIQ				0.048 ** (0.015)	0.001 ** (0.001)	0.031 * (0.012)
EVO	−0.292 * (0.161)	−0.020 * (0.097)	−0.169 * (0.091)	−0.213 * (0.168)	−0.125 (0.102)	−0.121 (0.092)
GOV	−1.242 * (0.539)	−1.275 ** (0.465)	−0.892 * (0.450)	−0.661 (0.556)	−0.802 * (0.482)	−0.630 (0.451)
EDU	0.026 (0.058)	0.086 * (0.101)	0.093 * (0.038)	0.100 * (0.062)	0.110 ** (0.043)	0.114 ** (0.039)
ECO	0.135 *** (0.013)	0.101 *** (0.005)	0.104 *** (0.005)	0.137 *** (0.013)	0.100 *** (0.005)	0.102 *** (0.005)
NET	0.139 *** (0.034)	0.103 *** (0.030)	0.039 (0.028)	0.055 * (0.032)	0.017 (0.028)	0.018 (0.027)
IND	−0.221 (0.272)	−0.461 (0.236)	−0.401 * (0.206)	−0.510 * (0.262)	−0.368 * (0.213)	−0.279 (0.198)
OPEN	0.113 ** (0.036)	0.077 ** (0.029)	0.047 (0.028)	0.045 * (0.037)	0.032 (0.031)	0.038 (0.028)
_cons	0.238 (0.234)	−0.079 (0.179)	0.190 (0.174)	0.320 (0.243)	0.103 (0.184)	0.031 (0.184)
N	324	432	432	324	432	432
R-sq	0.4030	0.583	0.601	0.355	0.547	0.602
F value	91.60	93.07	86.27	84.28	82.08	87.21

Note: ***, **, and * represent significance at the levels of 1%, 5%, and 10%, respectively; the figures in brackets are standard deviations; the below is the same.

. The conclusions obtained from the regression by the above methods are consistent with the conclusions drawn earlier in this paper, thereby indicating that the conclusions are stable.

3.2.2. Nonlinear Regression Analysis

(1) Threshold effect test

The first step of the nonlinear regression is to test if the model exhibits the threshold effect. If there are thresholds, it is necessary to further determine their number. The method uses Stata SE15.1 to perform the threshold self-sampling test by using the FDI quality, level of economic development, industrial structure, environmental regulations, and internet penetration rate as the threshold variables. All the values were obtained from 300 samplings by the bootstrap method. The results of the threshold quantity tests using Equations (16)–(20) are summarized in

Table 11. Threshold effect estimation and test results.

Threshold Variable	Threshold Type	Threshold Value	F-Value	p-Value	1% Critical Value	5% Critical Value	10% Critical Value
	Single	0.352 *	14.92	0.033	16.651	21.822	34.454
FDIQ	Double	0.191	8.44	0.340	12.590	14.442	18.937
	Triple	1.805	26.29	0.140	27.903	33.622	42.617
ECO	Single	7.020 **	49.78	0.027	33.454	39.173	61.927
	Double	4.730	14.95	0.403	26.904	34.589	125.824
	Triple	0.950	11.07	0.746	110.097	145.921	173.573
IND	Single	0.360 ***	57.07	0.000	27.913	35.438	42.607
	Double	0.520	9.79	0.506	29.273	38.355	52.348
	Triple	0.560	5.20	0.860	25.369	34.790	51.493
EVO	Single	0.944 **	38.86	0.003	20.172	23.596	30.447
	Double	0.965	11.51	0.406	21.745	26.261	36.373
	Triple	0.301	5.94	0.733	15.519	17.698	23.164
NET	Single	1.536 **	36.12	0.010	19.813	23.353	35.235
	Double	0.427	5.19	0.783	16.775	21.614	27.644
	Triple	0.4779	6.63	0.8000	19.184	22.372	26.565

Note: ***, **, and * represent significance at the levels of 1%, 5%, and 10%, respectively; the below is the same.

.

The level of economic development passes the single-threshold test at the 1% significance level; the industrial structure, environmental regulations, and internet penetration rate pass the single-threshold test at the 5% significance level; the FDI quality passes the single-threshold test at the 10% significance level. The results of the double-threshold and triple-threshold tests of the five models are not significant and indicate that FDIQ, ECO, IND, EVO, and NET have only one threshold.

(2) Threshold regression analysis

The regression results of the panel threshold model are shown in

Table 12. Estimation results of panel threshold parameters.

Threshold Variables [Threshold Values]	FDIQ [0.169]	ECO [7.02]	IND [36.42]	EVO [94.42]	NET [1.536]
NET	0.037 (0.027)	0.079 *** (0.026)	0.113 *** (0.026)	0.063 ** (0.027)	0.013 (0.031)
ECO	0.103 *** (0.005)	0.099 *** (0.004)	0.099 *** (0.004)	0.099 *** (0.004)	0.100 *** (0.004)
GOV	−0.925 ** (0.450)	−1.319 *** (0.402)	−1.618 *** (0.413)	−1.204 ** (0.418)	−1.184 ** (0.420)
EVO	−0.110 (0.092)	−0.283 ** (0.081)	−0.186 ** (0.081)	−0.206 ** (0.083)	−0.166 ** (0.083)
IND	0.025 (0.213)	0.327 *** (0.202)	0.683 *** (0.205)	0.566 ** (0.207)	0.161 (0.208)
EDU	0.102 *** (0.037)	0.093 ** (0.033)	0.098 ** (0.033)	0.096 ** (0.034)	0.109 ** (0.034)
OPEN	0.061 ** (0.029)	0.010 ** (0.026)	0.065 ** (0.026)	0.073 ** (0.027)	0.064 ** (0. 027)
Fdi0	0.006 (0.016)	0.063 *** (0.012)	0.091 *** (0.012)	0.056 *** (0.013)	0.076 *** (0.012)
Fdi1	0.062 *** (0.019)	0.122 *** (0.013)	0.037 *** (0.014)	0.101 *** (0.012)	0.119 *** (0.013)

Note: *** and ** represent significance at the levels of 1% and 5%, respectively; the figures in brackets are standard deviations; the below is the same.

. FDI has a nonlinear influence on GTFP. The FDI inflow in the Yangtze River Delta region has a significant promoting effect on the growth of GTFP, but this promotion varies significantly under different threshold conditions.

When the FDI quality index is below the threshold value of 0.169, the amount of actually utilized foreign capital has no significant effect on the growth of GTFP, thus indicating that the technology spillover effect of FDI is not significant when the scale of FDI is small. When the FDI quality index exceeds the threshold value of 0.169, FDI has a significant promoting effect on GTFP at the level of 1%, which indicates that the scale effect due to the aggregation of foreign capital can promote the growth of the green economy.

When the level of economic development is lower than the threshold of 7.02, the positive influence coefficient of FDI with respect to GTFP is 0.063 and significant at the level of 1%. When the level of economic development is higher than the threshold, the influence coefficient of FDI is 0.122 and FDI has a significantly stronger positive effect on GTFP in this interval.

When the industrial structure index is lower than the threshold of 36.42%, FDI has a significant positive effect on GTFP at the level of 1%. When the urban industrial structure index exceeds 36.4%, the positive effect of FDI on GTFP is still significant at the level of 1% but is lower when the industrial structure index is low.

The single threshold of environmental regulation is 94.42%. When the level of environmental regulation represented by the urban sewage treatment rate is lower than the threshold, the influence coefficient of FDI with respect to GTFP is positive, but the value is small. When the coefficient crosses the threshold, the positive effect of FDI is stronger and the improvement in environmental regulation can enhance the promoting effect of FDI on GTFP.

The threshold of the internet penetration rate is 1.536. When the internet penetration rate is lower than the threshold, the effect of FDI on GTFP is smaller than when the internet penetration rate is above the threshold, thus indicating that the promoting effect of FDI on GTFP is limited when the internet penetration rate is not high. When the internet penetration rate exceeds the threshold, the promoting effect of FDI on the growth of the green economy is enhanced.

In summary, the promoting effect of FDI on GTFP growth is more salient in cities with a higher quality of foreign capital utilization than in cities with higher levels of economic development, lower industrial structure indexes, stricter environmental regulations, and higher network penetration rates. The 27 cities in the Yangtze River Delta can be divided into different groups according to the average values of their threshold variables of the last five years, as shown in

Table 13. Grouping of cities according to different threshold values.

Index	Quantity	Cities in the Strong Promotion Zone
High quality of foreign capital utilization	24	Shanghai, Nanjing, Wuxi, Changzhou, Suzhou, Nantong, Yangzhou, Zhenjiang, Taizhou Jiangsu, Hangzhou, Wenzhou, Ningbo, Jiaxing, Huzhou, Zhoushan, Taizhou Zhejiang, Hefei, Wuhu, Maanshan, Tongling, Anqing, Chuzhou, Chizhou, Xuancheng
High levels of economic development	1	Shanghai
Lower industrial structure indexes	6	Shanghai, Nanjing, Hangzhou, Wenzhou, Zhoushan, Hefei
Stricter environmental regulations	20	Shanghai, Nanjing, Wuxi, Changzhou, Suzhou, Yangzhou, Zhenjiang, Hangzhou, Wenzhou, Ningbo, Huzhou, Shaoxing, Jinhua, Taizhou Zhejiang, Hefei, Maanshan, Anqing, Chuzhou, Chizhou, Xuancheng
High network penetration rates	11	Shanghai, Nanjing, Wuxi, Changzhou, Suzhou, Zhenjiang, Hangzhou, Ningbo, Jiaxing, Huzhou, Jinhua

.

All cities, except for Yancheng, Jinhua, and Shaoxing, crossed the threshold for the quality of foreign capital utilization in the past five years, thus demonstrating the significant promoting effect of FDI on GTFP. Only Shanghai crossed the threshold for the level of economic development. For the other cities, there is still room for improving FDI’s effect on GTFP. Only the six cities of Shanghai, Nanjing, Hangzhou, Wenzhou, Zhoushan, and Hefei have proportions of secondary industry lower than the threshold for industrial structure, so FDI promotes the growth of GTFP in these cities. For the other cities, it is feasible to enhance FDI’s effect on GTFP through industrial transformation and upgrading. Most of the cities have crossed the threshold for environmental regulation, thus indicating that it is already at a high level in the Yangtze River Delta region. However, it is necessary to control the intensity of environmental regulation and prevent the inhibitory effects of draconian environmental regulations on the development of enterprises. For the network penetration rate, 11 cities have exceeded the threshold. For the other cities, further expansion of their internet infrastructures is required to enhance their effects on the transformation and upgrading of traditional enterprises for improving GTFP.

4. Conclusions and Recommendations

4.1. Conclusions

This study was conducted to explore the effects of directly utilized foreign investment (FDI) on green total factor productivity (GTFP). First, the data-envelopment analysis (DEA) method and directional distance function (DDF) method based on the Malmquist–Luenberger productivity index model were used to measure the GTFP values of 27 cities in the Yangtze River Delta for 2004–2019. We also analyzed the variation patterns of FDI and GTFP. In a subsequent empirical analysis, the linear influence of FDI on GTFP was tested by a panel fixed-effect model, then the nonlinear influence of FDI on GTFP was tested by a single-threshold panel regression model. The key conclusions of this study are as follows.

First, the GTFP in the Yangtze River Delta region grew with a declining annual average growth rate. Specifically, the green technology efficiency increased by 2.2%, but the green technology declined by 2.5%. The average GTFP variation indexes of 15 cities were greater than 1 and the fluctuations in the proportion of cities with negative growth were attenuated. Hence, most cities in the Yangtze River Delta experienced growth in GTFP to varying degrees and achieved green economic growth.

Second, the scale of FDI in the Yangtze River Delta was steadily increasing during the study period but has entered a slow growth period in recent years. Meanwhile, the structure of foreign investment has been significantly improved, as reflected by the fact that the tertiary industry has surpassed the secondary industry to become the largest recipient of FDI in the cities in the Yangtze River Delta. At present, there is a great imbalance among the cities in the Yangtze River Delta in terms of FDI received. For cities with more developed economies, better investment environments, higher degrees of trade openness, stronger policy support, and stronger location advantages, the scale of FDI is larger. With the decline in the proportion of foreign investment in the manufacturing industry and the increase in the amount of investment per project, the quality of foreign investment in the analyzed cities has increased in recent years.

Third, an empirical analysis of the influence of the scale and quality of actually utilized foreign investment on GTFP in the 27 cities found that FDI had salient positive effects on GTFP. The Yangtze River Delta region has not become a pollution refuge for foreign investment, thus confirming the pollution halo hypothesis. In addition, the level of regional economic development, human capital, the internet penetration rate, and trade openness have had salient positive effects on GTFP, whereas environmental regulation, government intervention, and industrial structure have had salient negative effects.

Fourth, the regression results of the panel threshold model had shown that the quality of foreign investment, level of regional economic development, industrial structure, level of environmental regulation, and internet penetration rate had single thresholds for the effects of FDI on GTFP. When a threshold variable is below a certain value, FDI has little influence on GTFP. When the variable exceeds a certain threshold value, the promoting effect of FDI on GTFP is significantly enhanced. At present, most of the 27 cities in the Yangtze River Delta have not crossed the thresholds for economic development, industrial structure optimization, and internet infrastructure construction. Also, several cities have not crossed the thresholds of foreign capital quality and environmental regulation, thus indicating that it is necessary to improve the level of economic development in the Yangtze River Delta region, optimize its industrial structure, intensify network construction, attract high-quality long-term investment, and formulate reasonable environmental regulations in order to enhance the promoting effect of FDI on GTFP.

4.2. Policy Recommendations

4.2.1. Comprehensively Improving Trade Openness for Foreign Investment and Its Quality

To enhance the promoting effect of FDI on GTFP, the Yangtze River Delta region should improve its trade openness, adopt the more aggressive promotion of investment, and make use of its advantages in resources, location, and policies to attract more foreign investment in order to realize the strategic goal of developing the region and radiating benefits over the whole country. In addition, it is necessary to improve the quality of FDI while increasing its scale, which can be achieved by raising market access barriers and increasing policy incentives. First, before the approval of a foreign investment project, it is necessary to conduct a comprehensive evaluation of its sources of funds, investor qualifications, potential risks, and possible benefits by using a scientific and comprehensive foreign-capital quality evaluation system to ensure that the project would not have negative impacts on the environment. Efforts should be made to improve the environment-related market access mechanism for foreign capital and the existing environmental protection policies, strictly control speculative foreign capital and foreign capital inflows aimed at transferring industries with high pollution and high energy consumption into China, and guide foreign capital to flow into industries with low energy consumption and low pollution. Second, while market access restrictions are imposed, it is necessary to formulate reasonable incentives and preferential policies for investments in some fields such as modern services and high technology, as well as to give priority to the introduction of foreign-invested projects that emphasize sustainable development and bring advanced technologies and high levels of management.

4.2.2. Introducing Foreign Investment Based on Local Conditions

At present, the scale of FDI in the 27 cities in the Yangtze River Delta region is more than half that of the whole country, but there is a serious imbalance in distribution, which means that the GTFP-promoting effects of FDI differ between cities. The government should fully consider the local levels of economic development, the current situations of foreign capital utilization, and the capabilities of absorbing the technology spillover brought about by FDI, then implement a differentiated investment promotion strategy. Priority should be given to the FDI flowing into the local advantageous industries in order to promote the green and coordinated development of the regional economy. Cities, such as Shanghai, Hangzhou, and Suzhou, with high levels of foreign investment, should place emphasis on adjusting and optimizing the structure of foreign investment, improving the quality of foreign investment, accelerating the transformation and upgrading of foreign investment projects, and accumulating practical experience to drive the expansion and optimization of foreign investment in the whole Yangtze River Delta region. Other cities in the region should establish reasonable and appropriate investment-management regimes based on their own industrial advantages and the advantage of their location in the Yangtze River Delta region. Emphasis should be placed on expanding the scale of foreign investment and attracting more high-quality foreign-funded enterprises to gain more opportunities for learning advanced production technologies and management models. Through these measures, the promoting effect of foreign investment on GTFP can be maximized.

4.2.3. Implementing Moderate Environmental Regulation and Applying Prudent Government Intervention

Under different environmental regulations and levels of government intervention, the effects of foreign investment on the GTFP of the 27 cities in the Yangtze River Delta exhibit different characteristics. The environmental regulations and government financial intervention will directly affect the growth of GTFP. It is necessary to formulate differentiated environmental regulations, implement differentiated government intervention measures based on local conditions, such as the level of economic development, and control the regulation intensity in a reasonable range. Exercising such prudence can avoid the mistakes of blindly increasing the intensity of environmental regulation or blindly strengthening financial intervention. Specifically for cities that have not crossed the threshold of environmental regulation, the local governments should moderately increase the intensity of environmental regulation or seek ways to improve the effects of environmental regulations to urge all enterprises to reduce their pollutant discharge, improve their resource utilization efficiency, and innovate their production technologies, thus improving the level of green technology. For cities that have crossed the threshold of environmental regulation, the local governments should dynamically adjust their environmental policies according to the economy-promoting effects of the existing environmental regulations in order to avoid the mistake of blindly increasing regulation intensity.

4.2.4. Intensify the Construction of Network Infrastructure

The construction of network infrastructure is conducive to the growth of GTFP, as the effects of FDI on GTFP are stronger in cities with higher internet penetration rates. The Yangtze River Delta region, as a pioneering region for advanced technology and industrial integration, should make full use of the infiltration effects that internet technology has on traditional industries. Firstly, it is feasible to improve the efficiency of traditional industrial supply chains through the optimization of information flows, which improves the efficiency of green technology. Secondly, incentives should be provided to actively guide the integration of the internet with local industries that have low energy consumption, low pollution, and high added value, as well as cultivate emerging industries empowered by the internet to realize the transition of the role of the internet from improving green efficiency to promoting technological progress. Thirdly, internet development strategies should be formulated according to local conditions. For cities with internet penetration rates lower than the regional threshold, it is necessary to increase investment in internet infrastructure. For cities with internet penetration rates higher than the regional threshold, more attention should be paid to encouraging innovative applications of the internet, cultivating talents in the realm of science and technology, and promoting the development of green science and technology.

4.3. Study Limitations and Directions for Future Research

First, the selection of some variables was constrained by the availability of data. In particular, the selection of the FDI quality index should be improved. We plan to explore the construction of FDI quality evaluation systems at the city level in subsequent research. Second, the influencing factors of GTFP are complicated. This paper considers as many variables as possible, but there is also the possibility of there being missing important variables. We will consider the influencing factors as comprehensively as possible in subsequent research. Finally, we considered human capital, the network penetration rate, industrial structure, and environmental regulation when analyzing the threshold effect. There may be other factors that play a threshold role in FDI’s effects on GTFP which require further research.