The Impact of Place-Based Policies on Firm Performance: Evidence from China

Abstract

This study investigates the causal effect of the first round of China’s Great Western Development Strategy (GWDS) on the total factor productivity (TFP) of Chinese manufacturing firms employing the geographic regression discontinuity design. It uses the firm-level data from China’s Annual Survey of Industrial Firms (ASIF) database from 1998 to 2007. To follow the principle of the geographic regression discontinuity design and ensure the validity of our identification strategies, only firms within a 10 km radius on either side of the GWDS boundary were retained in the baseline regression. The main results include some of the following: (1) The GWDS increased the TFP of firms on the western side of the boundary in the range of 11.2% to 13.7%. (2) The main mechanisms of this improvement were identified as the reduction of a firm’s actual income tax rate and increased firm investment in high-quality human capital. (3) The GWDS has a greater impact on private firms, small firms, and labor-intensive firms. This study provides reliable evidence that place-based policies can promote the sustainable development of firms within the affected regions, and could serve as policy inspiration to alleviate regional development disparities in other developing countries.

1. Introduction

To promote employment growth in specific regions and ensure balanced development between regions, many countries have introduced place-based policies for specific regions [1,2,3]. Governments have implemented this important policy tool by way of tax incentives, financial subsidies, infrastructure construction, or specific preferential measures [4]. The existing literature on this topic largely focuses on developed countries, and there are relatively few studies that have evaluated the impact of place-based policies in developing countries [2,3,4]. Developing countries differ greatly from developed countries in terms of resource endowments, market conditions, and political systems. Therefore, place-based policies may have different effects in developing countries compared with those in developed countries.

This article focuses on China, the largest developing country in the world, which also experiences significant regional development disparities [5]. To ensure balanced regional development, the Chinese government has implemented numerous policies, such as the Northeast Revitalization Strategy and the Economic Development Zone Policy. Among these, the Great Western Development Strategy (GWDS) is the most widely implemented and longest-lasting regional policy with the most favorable tax incentives. Since 2001, a series of preferential measures has been gradually introduced as part of the GWDS, such as increasing the magnitude of transfer payments and reducing the income tax of some firms in the region covered by the program. These measures are designed to improve the local business environment and attract more capital and talent to the region. It is widely recognized that the development of firms underpins the development of a region, with the improvement of firm total factor productivity (TFP) being considered the continuous driving force of regional economic growth. Hence, whether the GWDS can help improve firm TFP in the benefitting area is a matter of interest.

Most of the literature on the effect of the GWDS is based on a macro perspective. These studies use macroeconomic data to examine the impact of the program on economic growth, industrial structure, and the environment [4,5,6]. However, relatively few studies have been conducted from the micro perspective. Further, in terms of research methods, most studies regard the GWDS as a “natural experiment,” and use the difference-in-differences method to evaluate the effects of this program. Given the significant differences between the affected areas (the west side of the GWDS boundary) and unaffected areas (the east side of the GWDS boundary), the above methods may fail to eliminate endogenous problems or ensure that the common trend assumption is met, resulting in bias in the estimated results [4]. The present study employs the geographic regression discontinuity (RD) design and uses firm-level data from China’s Annual Survey of Industrial Firms (ASIF) database to estimate the causal effects of the GWDS on firm TFP and explore the mechanisms of these effects.

This study makes three contributions: First, it provides reliable evidence that China’s place-based policies improve firm productivity, thereby enriching empirical research on place-based policies. Most of the place-based policy studies evaluate the effects of relevant policies in developed countries such as the United States and Europe [2,3,7]. In contrast, this study is about China—the largest developing country in the world—and focuses on the effect of China’s most vigorously implemented place-based policy thus far. Although there are significant differences between China and developed countries in terms of institutions and resources, this study found that place-based policies can still significantly promote the development of policy-affected firms in China, thereby promoting regional economic growth. Second, there is a paucity of literature on the impact of the GWDS on firms. Using both detailed firm-level geographic data and firm-level production data from Chinese manufacturing firms, a geographic regression discontinuity design is used to solve endogenous problems in empirical estimation. Hence, this study provides an unbiased evaluation of the causal effects of the first round of China’s GWDS on the TFP of Chinese manufacturing firms. Further, it explores the influencing mechanism of this program, which can be used to inform the third round of the GWDS as well as subsequent policies that have been implemented since 2020. Last, this study contributes to the design of empirical research based on geographic discontinuity. Empirical studies based on the geographic regression discontinuity design have gradually emerged over recent years, and have been applied in many fields, including economics, political science, and environmental science [8,9,10,11,12]. This study examines the impact of the GWDS on firm TFP by using geographic boundary discontinuity, thereby enriching the existing empirical literature based on a geographic discontinuity design.

2. Policy Background and Literature Review

2.1. Literature Review

Place-based policies are a series of preferential policies and measures formulated by various governments for a specific region. They usually target areas with backward development or poor performance, such as deteriorating central business districts or economically disadvantaged areas. The purpose of the policies is to stimulate local employment and economic development by providing special support in the form of tax subsidies and public investments [4]. However, can the place-based policies implemented by various governments achieve these goals? Can the supported areas achieve increased employment and sustained economic growth? Which individuals or firms can benefit from these policies? Are there other adverse consequences? To answer these questions, scholars have conducted empirical studies in recent years on the effects of such place-based policies.

Existing studies evaluating place-based policies largely focus on developed countries [2,3,4]. Busso et al. [1] conducted an empirical evaluation of the Federal Urban Empowerment Zone Program in the United States. The program consisted of a series of tax subsidies for designated areas. The authors found that the program significantly increased local employment and enhanced the wages and income of local workers while keeping price levels constant. Givord [13] studied the effect of the Zones Franches Urbaines (ZFU) in France, which aims to boost local businesses by exempting them from taxes. They found that the policy had a significant stimulative effect on business creation and employment in the treated areas. Kline and Moretti [7] studied the long-term impact of the Tennessee Valley Authority Program in the United States. Their results showed that the program significantly increased agricultural and manufacturing growth in the short term. In the longer term, the program negatively affected agricultural growth; however, it still significantly increased manufacturing growth and ultimately raised local income levels. Bronzini and Iachini [14] studied the effect of an R&D subsidy program implemented in northern Italy. Their results showed that the policy had no positive effect on firm investment for the entire sample. However, the policy triggered a lot of investment for small firms. Ehrlich and Seidel [15] used the spatial regression discontinuity design to evaluate the effect of temporary place-based subsidies in the West German Zonenrandgebiet. Their results indicated that the temporary place-based policy promoted lasting increases in local incomes and employment. Criscuolo et al. [16] studied the employment effect of the Regional Selective Assistance Program in the United Kingdom, which provided financial assistance to economically backward and financially strained areas. The authors found that the program succeeded in improving local employment. Luo et al. [17] studied the effect of Economic Development Priority Areas (EDPA), a tax reduction policy designed to promote regional economic development in the United States. They found that EDPA significantly reduced poverty rates and raised local housing prices in affected areas. However, the policy had no significant effect on population size or employment.

Recent years have seen an increase in studies on place-based policies in developing countries, mainly India and China. Chaurey [18] conducted an empirical evaluation of a location-based tax incentive scheme in India, which offered tax exemptions and capital subsidies to new and existing companies in two Indian states since 2003. It was found that the policy led to significant increases in employment, output, and capital. Hasan et al. [19] studied the Backward District Program in India, which aims to promote manufacturing in underdeveloped parts of the country. Their results showed that the program significantly increased firm entry and employment in the short term. Shenoy [20] evaluated the effect of a comprehensive policy for infrastructure construction, investment subsidies, and tax incentives in Uttarakhand in India, which has been implemented stage by stage since 2002. The results showed that the policy boosted local economic growth and increased benefits for residents. Lu et al. [3] evaluated the impact of the Special Economic Zones (SEZ) policy in China on local economic performance. The SEZ policy seeks to foster a cluster economy by attracting technologically advanced industrial facilities. Their results indicated that the SEZ policy had a positive impact on local capital input, employment, and wages, as well as the number of firms in the designated areas. Song et al. [21] studied the effect of the SEZ policy in China on foreign investment. Their results showed that the SEZ policy promoted foreign capital by improving institutional quality in affected areas. Jia et al. [4] evaluated the effect of China’s GWDS using county-level data based on the geographic regression discontinuity design. They found that the policy accelerated industrialization, thereby promoting local economic growth.

2.2. Policy Background

Like many other countries, China is faced with uneven regional economic development [5]. In the early stages of reform and opening up, China initiated market-oriented reforms. The country’s eastern region developed rapidly because of its geographical advantages and certain preferential policies given by the government at that time. In contrast, the western region, which is located further inland, experienced slower development. In the 1990s, China formed three economic belts in the eastern region, central region, and the western region, with the level of economic development decreasing from east to west. According to data from the National Bureau of Statistics, the per capita GDP of western China in 1999 was only USD 560, which was about half of that of eastern China, indicating a huge economic development gap between the east and the west [4]. To narrow the development gap between the regions and achieve balanced and coordinated development among the regions, China commenced the implementation of the GWDS in 2000. The program covers six provinces (Gansu, Guizhou, Yunnan, Shanxi, Sichuan, and Qinghai), five province-level autonomous regions (Guangxi, Ningxia, Inner Mongolia, Tibet, and Xinjiang), one municipality (Chongqing), and three additional autonomous prefectures (Enshi Tujia and Miao Autonomous Prefecture, Xiangxi Tujia and Miao Autonomous Prefecture, and Yanbian Korean Autonomous Prefecture).

The GWDS is the longest-lasting, most widely covering, and influential place-based policy in China [6]. To ensure the effective implementation of the program, the central government formulated several guiding documents and policy measures. In October 2000, the State Council issued the Notice on Several Policies and Measures for the Implementation of the Great Western Development Program (GF [2000] no. 33). This is the start of the first round of the GWDS. The Notice included three key preferential policies to be introduced in the western region from 2001 onward: (1) Increasing capital investment. This involved (i) increasing the proportion of central government construction funds used in the region; (ii) prioritizing construction projects in the region; (iii) increasing the magnitude of general transfer payments from central financial funds to the region; and (iv) increasing financial and credit support. (2) Improving the business environment. This involved (i) introducing preferential tax policies, with corporate income tax levied at a reduced rate of 15% from 2001 to 2010. With the approval of provincial governments, corporate income tax for domestic firms in ethnic autonomous areas could be further reduced or exemptions could be applied at regular intervals. (ii) preferential policies on land and mineral resources were also implemented in the region. (3) Attracting talent and expanding opening-up. This involved (i) further broadening the field of foreign investment, widening the channels of foreign investment, and intensively developing foreign trade, and (ii) introducing policies to attract and retain talent and encourage talent to establish businesses in the region.

The 11th, 12th, and 13th Five-Year Plan for the GWDS were issued in 2007, 2012, and 2017, respectively, launching the second round of the GWDS. In 2020, the Guidelines of the Central Committee of the Communist Party of China and the State Council on Promoting the Development of the Western Region in the New Era and Forming a New Pattern were issued, making a start on the third round of the GWDS. The second and third rounds represented the consolidation and continuation of the first round of the GWDS. The main purpose of this study is to examine the impact of the first round of the GWDS on firm TFP and its mechanisms. The reasons are as below: First, since more policies were added in the second and the third rounds of the GWDS from 2007 onward, if we estimate the effects of the recent round of the program, the effects may not be clean. The effect would be the superimposed effect of several rounds of policies; the net effect of the recent round cannot be separated. Second, since the second and third rounds continued the policies of the first round of the GWDS, the findings of our paper on the effect of the first round could provide references for the third round of the program as well as its follow-up measures. Third, due to the variable absence and data anomalies in the ASIF database after 2007, it is rarely used by scholars. Therefore, we mainly evaluate the effect of the first round of the GWDS (2001–2007).

3. Data and Methods

3.1. Data

This study used sample data from the ASIF database from 1998 to 2007. The data included samples of non-state-owned firms with an annual output value of more than RMB 5 million and samples from all state-owned firms. Chinese ASIF data have been widely used in academia because of their long-time span, extensive indicators, and large sample size. However, there are many obvious outliers in this database due to statistical errors and other reasons. Therefore, in this study, the data were cleaned according to the practice of Brandt et al. [22]. Samples with missing or negative values in indicators, such as in gross industrial output value, original value of capital, paid-in capital, liabilities, and intermediate input, were excluded. Samples that clearly did not conform to accounting principles were further excluded. Finally, samples with fewer than eight employees and abnormal start times were also excluded.

TFP is a measure of efficiency in production—how much output is obtained from a given set of inputs [23]. As an important index to measure the efficiency of firms, TFP is an important reflection of the core competitiveness of firms and directly determines their performance. Therefore, we used TFP as an indicator to measure firm performance in this paper. The TFP index of firms was constructed using detailed firm-level production data from the ASIF database. The calculation of TFP index used in the baseline regression in this study followed the practice of Levinsohn and Petrin [24]. The TFP index constructed based on the method of Olley and Pakes [25] was further used for robustness checks in this study. As the specific location information of each firm is reported in the ASIF database each year, this information was used to digitize information on geographic location and generate the latitude and longitude of each firm for every year.

The geographic data in this study are based on information that is publicly available on the Geospatial Data Cloud Platform of the Computer Network Information Center of the Chinese Academy of Sciences. A geographic information system software was used to determine the boundary separating the GWDS region from the non-GWDS region, as shown in Figure 1. Similarly to the study by Jia et al. [4], Inner Mongolia Autonomous Region and Yanbian Korean Autonomous Prefecture were excluded from the analysis in this study because they lie in northern or northeastern China and are distant from other areas covered by the GWDS. The areas to the left of the boundary fall within the GWDS-covered region (western side of the boundary), and the areas to the right of the boundary (eastern side of the boundary) are not covered by the GWDS. The distance of each firm to the nearest boundary was calculated using their latitude and longitude.

3.2. Research Design

This article examines the impact of the GWDS on firm productivity in the areas where the program was implemented. Thus far, China has implemented three rounds of the GWDS. Considering the availability of data, as mentioned earlier, this study only evaluated the effects of the first round of the GWDS (2001–2007).

To identify the effect of this program more accurately, the study followed Jia et al. [4] by using a geographic discontinuity design to estimate the causal effects of the GWDS on firm TFP. The running variable was the nearest distance from the firm to the boundary of the GWDS region. This design was based on the premise that samples located on either side of the geographic discontinuity should be highly similar in terms of various characteristics. This ensured that the samples without policy intervention serve as good counterfactual samples against those directly benefiting from the program. It was necessary to ensure that firms on either side of the GWDS boundary are highly similar before the implementation of the GWDS. To ensure characteristics are similar on both sides, only firms within a 10 km radius on either side of the boundary were retained in the baseline regression.

The baseline model in this study is as follows:

$$\begin{array}{l}{TFP}_{ijk}={\alpha }_1{GWDS}_{ijk}+{\alpha }_2{Distance}_{ijk}+{\alpha }_3{GWDS}_{ijk}\ast {Distance}_{ijk}+{\displaystyle \sum _{s=1}^{15}}{\gamma }_s{Seg}_s+{\mu }_j+{\sigma }_{ijk}\\ {s.t.\hspace{1em}-h\le Distance}_{ijk}\le h\end{array}$$

(1)

where ${TFP}_{ijk}$ indicates the log of the TFP of firm i of industry j in county k; ${GWDS}_{ijk}$ indicates whether the firm is located within the area implementing the GWDS (on the western side of the boundary). If yes, it is 1; otherwise, it is 0. ${Distance}_{ijk}$ is the nearest distance from the firm to the boundary. To account for the influence of industry on firm TFP, we control for industry fixed effects, ${\mu }_j$. Further, the boundary is divided into 15 segments of equal length and the fixed effects of each segment (${Seg}_s$) are controlled. ${Seg}_s$ (s = 1, 2, 3, 4, …, 15) is an indicator dummy, which takes 1 when the firm is closest to the segment (s), and 0 otherwise. The segment design was adopted in this study for the following reason: the running variable used in the research design is the nearest distance from the firm to the boundary of the GWDS region. Hence, two firm samples equidistant to the boundary but located in two counties that are distant from each other (with large latitude and longitude differences) could be compared in the regression process. However, these two samples may have large differences because they are not located on the side of the same boundary segment, and could introduce bias in the direct comparison. When the boundary segments were controlled, the sample comparison in each boundary segment could effectively avoid this situation. Theoretically, the more segments, the better. However, as the number of segments increased, the number of samples in each segment decreased, which may have affected the identification of the regression results. Determining the optimal number of segments is still a theoretical challenge. Following Jia et al. [4], we divided the boundary into 15 segments.

Given the existence of fixed effects in this regression discontinuity design, an ordinary least squares (OLS) regression of TFP was first carried out on the dummy variables of industry and boundary segment. This was carried out in line with the accepted literature (e.g., He et al. [26]), to obtain the residualized TFP after absorbing the fixed effects of industry and boundary segments. The residualized TFP was then used as the explained variable for further regression. To overcome the possible spatial correlation, the baseline regression clustered standard errors at the county level. The firm TFPs from 2001 to 2007 were used as the explained variable of the study. The coefficients of the study results should be interpreted as the average causal effects during the implementation period of the GWDS (2001–2007).

Robustness checks of the empirical results were carried out by setting different model specifications. Specifically, the following three specifications were used in this study: (1) a nonparametric RD model controlling the industry fixed effects and boundary segment fixed effects; (2) a nonparametric RD model controlling the industry-by-boundary segment fixed effects; and (3) a parametric RD model with a polynomial of the running variables.

3.3. Balance Test

The key hypothesis in the geographic regression discontinuity design is that pre-treatment variables other than the explained variable vary smoothly on both sides of the discontinuity. Firms located on the eastern side of the boundary should be similar in characteristics to those located on the western side, so that they can be used as a valid counterfactual comparison group. To this end, the time-invariant characteristic variables of firms within 10 km from both sides of the boundary as well as the time-varying characteristic variables with pre-2001 values were used to check the balance.

To verify this hypothesis, the following variables were used as the explained variables to conduct a balance test: firm establishment date, firm ownership type, log value of TFP, log value of labor, log value of capital, and log value of intermediate input. The results are reported in

Table 1. Balance test of variables within 10 km from either side of the boundary.

	Firm Establishment Date	State-Owned Firm	TFP	Labor (log)	Capital (log)	Intermediate Input (log)
	(1)	(2)	(3)	(4)	(5)	(6)
GWDS	6.910	0.085	−0.034	0.030	0.107	0.078
	(19.104)	(0.062)	(0.048)	(0.236)	(0.186)	(0.257)
Observations	3237	3256	814	814	814	814
R2	0.065	0.312	0.487	0.635	0.603	0.646

Note. The coefficients obtained by running an OLS regression of firm variables on the GWDS variable; distance between the firm and the boundary; and industry fixed effects and the boundary segment fixed effects. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors clustered at the county level.

. The time-invariant firm establishment date and firm ownership type are full sample data, namely, pooled panel data from 1998 to 2007. The time-varying log value of TFP, log value of labor, log value of capital, and log value of intermediate input are pooled panel data from 1998 to 2000. To be consistent with the baseline regression, the samples were also limited to those 10 km from the boundary in the balance test. Moreover, this study controlled for industry fixed effects, boundary segment fixed effects, and firm distance to the boundary, with the standard errors clustered at the county level. The results reveal that all key variables are balanced (see Table 1). The above balance test shows that firms crossing the boundary within a small bandwidth range are indeed relatively similar, which enhances confidence in the study’s identification strategy.

4. Empirical Results

4.1. Baseline Results

Before conducting the empirical analysis, the relationship between firm TFP and the nearest distance of the firm to the boundary was identified, as illustrated in Figure 2. The horizontal coordinate in Figure 2 is the distance of the firm to the boundary, while the vertical coordinate is the residualized TFP after absorbing the industry-by-boundary segment effects. The negative distance on the left side of the graph indicates that the firm is located on the eastern side of the GWDS boundary (i.e., the area not affected by the GWDS). The positive distance on the right side of the graph indicates that the firm is located on the western side of the GWDS boundary (i.e., the area affected by the GWDS). It can be clearly seen from Figure 2 that firm TFP jumps significantly at the boundary. The results in Figure 2 provide preliminary evidence that the GWDS has a positive effect on firm TFP. The relationship between the GWDS and firm TFP is discussed in greater detail through further empirical regression in this study.

To test the impact of the GWDS on firm TFP, firms within 10 km on either side of the GWDS boundary were used as the analysis samples.

Table 2. Impact of the GWDS on firm TFP.

	TFP
	(1)	(2)	(3)
Panel A: Industry fixed effects and boundary segment fixed effects
GWDS	0.129 ***	0.134 ***	0.152 ***
	(0.042)	(0.042)	(0.040)
Panel B: Industry-by-boundary segment fixed effects
GWDS	0.118 **	0.128 **	0.106 ***
	(0.036)	(0.036)	(0.036)
Observations	2400	2400	2400
Kernel	Triangle	Epanech	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at the county level. *** and ** indicate significance at the 1% and 5% levels, respectively.

presents the results of the baseline regressions. The results are reported under three different forms of the kernel density function, and are shown to be robust regardless of the kernel density function. In Panel A, the study controls for industry fixed effects and boundary segment fixed effects. The results reveal that the GWDS has significantly improved firm TFP, increasing the TFP of affected firms in the range of 13.8% ($e^{0.129}-1$) to 16.4% ($e^{0.152}-1$). Panel B estimates a more saturated model that controls for industry-by-boundary segment fixed effects. The results show that the GWDS has increased the TFP of affected firms in the range of 11.2% ($e^{0.106}-1$) to 13.7% ($e^{0.128}-1$). The results in Panel B are slightly lower than those in Panel A, but they are effectively very similar.

4.2. Effects within Firms

In line with He et al. [26], the specification of “difference-in-discontinuities” was adopted to study the effects of TFP dynamic changes within firms affected by the program before and after the implementation of the GWDS in 2001. The model of this study was specified as follows:

$$\begin{array}{ll}\overline{{TFP}_{ijkt}}=& {\alpha }_1{GWDS}_{ijk}+f\left({Dsitance}_{ijk}\right)+{GWDS}_{ijk}\ast f\left({Distance}_{ijk}\right)+{\alpha }_2{GWDS}_{ijk}\ast {Post01}_t\\ & +f\left({Dsitance}_{ijk}\right)\ast {Post01}_t+{GWDS}_{ijk}\ast f\left({Distance}_{ijk}\right)\ast {Post01}_t+{\sigma }_{ijkt}\\ s.t.& -h\le {Distance}_{ijk}\le h\end{array}$$

(2)

where ${Post01}_t$ is a time dummy variable, with a value of 1 after the implementation of the GWDS (2001–2007) and a value of 0 beforehand (1998–2000). $\overline{{TFP}_{ijkt}}$ indicates the residualized TFP of firm i of industry j in the county k in year t after absorbing the firm fixed effects, industry-by-year fixed effects, and boundary segment-by-year fixed effects.

One benefit of the above model specification is that it allowed the full use of the characteristics of the panel data structure to study changes in firm TFP. This made it possible to exclude the impacts of the endogenous location selection of firms on both sides of the boundary.

Table 3. Changes in firm TFP under the specification of “difference-in-discontinuities”.

	TFP
	(1)	(2)	(3)
GWDS	0.073 **	0.070 **	0.065 **
	(0.0259)	(0.029)	(0.028)
Firm Fixed Effects	Yes	Yes	Yes
Industry-by-year Fixed Effects	Yes	Yes	Yes
Boundary segment-by-year Fixed Effects	Yes	Yes	Yes
Observations	2519	2519	2519
Kernel	Triangle	Epanech	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. In line with Grembi et al. [27] and He et al. [26], “bias-corrected” coefficients are reported. ** indicates significance at the 5% level, respectively.

presents the results of the regression specified by the above “difference-in-discontinuities.” The results are reported with the second-order polynomial of the running variable of Distance; the results with the first-order polynomial are basically consistent with those of the second-order polynomial. It was found that the GWDS significantly improved TFP in affected firms, ranging from 6.72% ($e^{0.065}-1$) to 7.57% ($e^{0.073}-1$). All GWDS coefficients in Table 3 are significantly positive, indicating that the results are robust. To verify the validity of the regression, the subsample data from 1998 to 2000 were further used to conduct a parallel trend test. In this subsample, it was assumed that the “placebo GWDS” took place in 1999, and the effects of this policy were estimated. If the parallel trend was satisfied, there would be no effects of the “placebo GWDS”. As can be seen in

Table 4. Parallel trend test under the specification of “difference-in-discontinuities”.

	TFP
	(1)	(2)	(3)
Placebo GWDS	0.003	0.008	0.017
	(0.047)	(0.048)	(0.046)
Firm Fixed Effects	Yes	Yes	Yes
Industry-by-year Fixed Effects	Yes	Yes	Yes
Boundary segment-by-year Fixed Effects	Yes	Yes	Yes
Observations	717	717	717
Kernel	Triangle	Epanech	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. In line with Grembi et al. [27] and He et al. [26], “bias-corrected” coefficients are reported.

, the effects of the “placebo GWDS” were, as expected, almost equal to 0 and insignificant, indicating that the changes in firm TFP were indeed caused by the GWDS implemented in 2001.

4.3. Robustness Tests

To check the robustness of the baseline regression, this study discussed the three issues that have may affected the baseline results: (1) the endogenous location selection of firms, (2) the spillover effect of the GWDS, and (3) the specification form of the RD model.

4.3.1. Impact of the Endogenous Location Selection of Firms

The GWDS provides preferential measures for firms on the western side of the boundary, which may have resulted in some new firms preferring to be located on the western side of the boundary. However, some firms originally on the east side of the boundary may have also relocated to the western side of the boundary to take advantage of these preferential measures. When any of the above cases occurred, it affected the baseline regression results.

Table 3 shows the regression results after controlling for firm fixed effects. The results reflect the internal changes in firms before and after the implementation of the program rather than the results caused by the endogenous location selection of firms. To further exclude the impact of the endogenous location selection of new firms, we only retained firms that existed for five consecutive years, from 1999 to 2003. We then re-estimated the model (the nonparametric RD model with control variables and controlled fixed effects of the industry-by-boundary segment). The results shown in

Table 5. Effects of firms that existed for 5 consecutive years.

	TFP
	(1)	(2)	(3)
GWDS	0.093 *	0.107 *	0.126 **
	(0.055)	(0.057)	(0.055)
Industry-by-boundary segment Fixed Effects	Yes	Yes	Yes
Observations	376	376	376
Kernel	Triangle	Epanech	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at county level. ** and * indicate significance at the 5% and 10% levels, respectively.

are consistent with the baseline results. All of the firms in this study included large, non-state-owned firms with an annual output value higher than RMB 5 million, as well as all state-owned firms. The relocation of large non-state-owned firms requires high costs and takes a long time. State-owned firms rarely relocate because of their specific nature. Hence, only a few firms were expected to be relocated during the observation period. This study revealed that only 3.7% of the firms in the baseline regression samples (within 10 km on either side of the GWDS boundary) relocated. No firm relocated from the eastern side of the boundary to the western side during the observation period. This further confirms the robustness of the baseline results.

4.3.2. Spillover Effect

Another possible impact on the baseline results was the spillover effect near the GWDS boundary. For instance, since firms on both sides are close to each other, there may be competition in input and output markets. The preferential measures in the western region may disadvantage eastern firms in terms of market competition. These result in a spillover effect, overestimating the real policy effect. Hence, we generated “virtual eastern side firms” to replace the actual eastern side firms, following He et al. [26]. Specifically, we used the nearest neighbor matching method to find the most similar samples of “virtual eastern side firms” far from the boundary (more than 10 km) for each actual sample of eastern firms, using firm characteristic data preceding 2001. As “virtual eastern side firms” are not adjacent to the actual western side firms (i.e., those within a distance of 10 km from the boundary), there was no spillover effect found between the actual western side firms and “virtual eastern side firms.” As a robustness check, the actual western side firms and “virtual eastern side firms” were used as samples for the regression. The regression results are reported in

Table 6. Effect of “virtual eastern side firms”.

	TFP
	(1)	(2)	(3)
GWDS	0.067	0.068 *	0.066 *
	(0.047)	(0.040)	(0.034)
Industry-by-Boundary segment Fixed Effects	Yes	Yes	Yes
Observations	2124	2124	2124
Kernel	Triangle	Epanech	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at the county level. * indicate significance at 10% level, respectively.

. The effect is still positive, and the results in columns (2) and (3) are statistically significant. These results confirm that the baseline results are robust and only slightly affected by the spillover effect of firms.

4.3.3. Placebo Test

In this study, a placebo test was performed by artificially changing the regression discontinuity boundary. The GWDS boundary was moved 10 km to the east and 10 km to the west to generate new virtual boundary lines. If the effect of the GWDS is indeed based on the boundary, then the firms on either side of the placebo virtual boundary generated by the movement would have been located on one side of the real boundary, which should have no effect. We re-estimated the effect of the policy using firms within 10 km of either side of the new placebo virtual boundary. As can be seen in

Table 7. Effect of moving the boundary by 10 km.

	TFP
	(1)	(2)	(3)
Panel A: Moving the boundary 10 km to the west
GWDS	0.031	0.035	0.022
	(0.044)	(0.044)	(0.046)
Observations	2644	2644	2644
Panel B: Moving the boundary 10 km to the east
GWDS	−0.025	−0.029	−0.041
	(0.029)	(0.030)	(0.032)
Observations	3253	3253	3253
Kernel	Triangle	Epanech	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at the county level.

, the effect of the new placebo virtual boundary was insignificant. The effect was further re-estimated by moving the boundary 20 km to the east and 20 km to the west to generate additional virtual boundary lines. As can be seen in

Table 8. Effect of moving the boundary by 20 km.

	TFP
	(1)	(2)	(3)
Panel A: Moving the boundary 20 km to the west
GWDS	−0.013	−0.005	0.083
	(0.061)	(0.067)	(0.079)
Observations	2369	2369	2369
Panel B: Moving the boundary 20 km to the east
GWDS	0.017	0.019	0.014
	(0.028)	(0.026)	(0.031)
Observations	4244	4244	4244
Kernel	Triangle	Epanech	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at the county level.

, the effects of these new placebo virtual boundary lines were also insignificant. The above results again confirm the robustness of the baseline results in this study.

4.3.4. Additional Robustness Tests

Different bandwidths were used to test the robustness of the baseline results. Firms located within 5 km and 15 km from either side of the boundary were used for the regression. As can be seen in

Table 9. Regression results for different bandwidths.

	TFP
	(1)	(2)	(3)
Panel A: Firms within 5 km from either side of the boundary
GWDS	0.143 ***	0.141 ***	0.120 **
	(0.044)	(0.044)	(0.058)
Observations	1141	1141	1141
Panel B: Firms within 15 km from either side of the boundary
GWDS	0.124 ***	0.118 ***	0.099 ***
	(0.032)	(0.033)	(0.034)
Observations	4034	4034	4034
Kernel	Triangle	Epanech.	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at the county level. *** and ** indicate significance at the 1% and 5% levels, respectively.

, irrespective of whether the bandwidth was set to 5 km or 15 km, the results are consistent with the baseline results. Furthermore, the robustness of the baseline results was tested using a parametric regression. The results of the parametric model specification in

Table 10. Regression results under different parametric model specifications.

	TFP
	(1)	(2)	(3)
GWDS	0.118 ***	0.174 ***	0.170 ***
	(0.036)	(0.057)	(0.061)
Observations	2400	2400	2400
Polynomial	Linear	Quadratic	Cubic

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at the county level. *** indicates significance at 1% level, respectively.

are very close to the baseline results, irrespective of whether it is a for linear, quadratic, or cubic polynomial RD model. Moreover, a new TFP index was calculated based on the Olley and Pakes [25] (OP) method as the explained variable in this study. It was found that the results were still consistent with our baseline results, as can be seen in

Table 11. TFP calculated based on the OP method.

	TFP
	(1)	(2)	(3)
GWDS	0.123 **	0.122 **	0.107 *
	(0.058)	(0.060)	(0.063)
Industry-by-boundary segment Fixed Effects	Yes	Yes	Yes
Observations	1268	1268	1268
Kernel	Triangle	Epanech	Uniform

Note. TFP estimated using the method of Olley and Pakes [25]. Standard errors are clustered at the county level. ** and * indicate significance at the 5% and 10% levels, respectively.

. Finally, we estimated the effect of the GWDS while controlling for the firm-level pretreatment characteristics, including firm size, firm age, firm leverage, whether or not the firm is state-owned, and whether or not the firm is an exporter. As shown in

Table 12. Regression results of controlling for the firm-level characteristics.

	TFP
	(1)	(2)	(3)
GWDS	0.118 **	0.128 **	0.106 *
	(0.036)	(0.036)	(0.036)
Industry-by-boundary segment Fixed Effects	Yes	Yes	Yes
Observations	2400	2400	2400
Kernel	Triangle	Epanech	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at the county level. ** and * indicate significance at the 5%, and 10% levels, respectively.

, the results of additionally controlling for the firm-level characteristics are consistent with our baseline results, indicating that our results are robust with respect to the inclusion of firm-level control variables. All the above robustness tests confirmed that the baseline results in this study are reliable.

5. Further Analysis

5.1. Mechanism Analysis

Considering the preferential development measures of the GWDS, how do the firms affected by the program react and then improve their TFP? To answer this question, we first analyzed firm input–output indicators related to TFP and identified the causes of increased TFP. Second, we examined the impact of the GWDS on the actual income tax rate of firms. Last, we analyzed whether the effect of the GWDS has resulted in improved innovation levels and increased firm exports.

5.1.1. Input–Output Indicators

The input–output indicators of firms were analyzed to determine the actual causes of improved firm TFP. First, three input indicators were used: labor input, capital input, and intermediate input. Second, output (industrial value added) and profit were analyzed. Third, we examined whether the GWDS had improved the factor productivity of firms, namely, labor factor productivity (log value of output/labor force) and capital factor productivity (log value of output/capital stock). Last, higher wages indicate higher quality human capital; therefore, we analyzed wages per capita.

The results of the input–output factor analyses are presented in

Table 13. Impact of the GWDS on firm input and output.

	(1)	(2)	(3)
Panel A: Factor input
Labor input (log)	0.107	0.061	0.142
	(0.142)	(0.150)	(0.117)
Capital input (log)	0.089	0.039	−0.052
	(0.214)	(0.220)	(0.238)
Intermediate input (log)	0.602 **	0.585 **	0.636 ***
	(0.235)	(0.243)	(0.240)
Panel B: Factor output
Industrial added value (in RMB 10,000)	668.56 **	886.06 *	1443.8 *
	(302.26)	(492.99)	(768.64)
Profit (in RMB 10,000)	511.73 ***	558.61 ***	532.00 ***
	(160.96)	(153.80)	(171.92)
Panel C: Single factor productivity
Labor factor productivity (log)	0.851 ***	0.842 ***	0.734 ***
	(0.165)	(0.174)	(0.212)
Capital factor productivity (log)	0.597 ***	0.618 ***	0.660 ***
	(0.148)	(0.154)	(0.169)
Panel D: Wages
Per capita wage (log)	0.296 ***	0.309 ***	0.257 **
	(0.086)	(0.034)	(0.108)
Kernel	Triangle	Epanech	Uniform

Note. Standard errors are clustered at the county level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

. Regarding input, the GWDS has little impact on labor input and capital input, but it significantly increases the intermediate input of firms. From the perspective of output, the GWDS has significantly improved the output level of the affected firms, revealing a noticeable increase in industrial added value and profit level. The program improves both the labor and capital factor productivity of the firms, as observed from the analysis of firm factor productivity. To further explore the sources of factor productivity improvement, the input of human capital was analyzed. The GWDS has significantly promoted the employment of high-quality human capital by firms.

5.1.2. Actual Income Tax Rate of Firms

According to the provisions of the GWDS, in terms of tax incentives, for firms in the western region, the corporate income tax was be levied at a reduced rate of 15% from 2001 to 2010. Prior to 2007, other firms in the eastern region paid a nominal corporate income tax rate of 33%, which was more than double the 15% paid by firms in the western region. Thus, firms in the western region have greater tax incentives. The literature indicates that tax incentives can effectively improve firm TFP [28,29,30]. Hence, we tested whether the GWDS reduces the actual income tax of firms, thereby promoting firm TFP. We first generated the log value of the actual income tax rate (income tax payable/total profit). Then, we examined the impact of the GWDS on the actual income tax. As can be seen in

Table 14. Impact of the GWDS on actual income tax rate of firms.

	Actual Income Tax Rate of Firms (log)
	(1)	(2)	(3)
GWDS	−0.200 *	−0.192 *	−0.182 *
	(0.102)	(0.101)	(0.099)
Observations	2358	2358	2358
Kernel	Triangle	Epanech.	Uniform

Note. Standard errors are clustered at the county level. * indicates significance at 10% level, respectively.

, the GWDS has indeed significantly reduced the actual income tax rate of firms in the western region, which could promote firm TFP.

5.1.3. Firm Innovation and Exports

The positive role of firm innovation and technological progress on firm TFP has been demonstrated by many scholars [31,32,33]. The GWDS also emphasizes the important role of scientific and technological innovation in development. Therefore, this study investigated whether the GWDS has exerted a positive impact on firm innovation. For firm innovation, common measures are R&D investment [34], patent application [35], and the proportion of new product output value [36]. In the database used in this study, only 1.5% of the firms hold patents, while the vast majority do not. Moreover, there are some issues in the R&D data in the Chinese ASIF Database. Firms with an R&D investment value greater than 0 only account for 2.7% of the total number of samples. Given the data availability, the proportion of new product output value of manufacturing firms (new product output value/total industrial output value) was used to measure the level of firm innovation. As can be seen in Panel A of

Table 15. Impact of the GWDS on firm innovation and exports.

	TFP
	(1)	(2)	(3)
Panel A: Firm innovation (new product value/total assets)
GWDS	0.006	0.001	0.011
	(0.012)	(0.011)	(0.012)
Observations	2035	2035	2035
Panel B: Firm exports (log)
GWDS	0.114	0.446	0.206
	(0.535)	(0.483)	(0.459)
Observations	2035	2035	2035
Kernel	Triangle	Epanech.	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at the county level.

, the GWDS does not improve the innovation of firms. As for the impact of firm exports on firm TFP, numerous studies indicate that firm exports can improve firm TFP [37,38]. To verify whether or not the GWDS boosts firm exports, the log value of the export output value of firms was used as an indicator of firm exports. The results are shown in Panel B of Table 15. The results reveal that the GWDS does not boost firm exports.

5.2. Heterogeneity Analysis

This section explores whether or not different types of firms are affected differently by the GWDS. The exploration is based on firm ownership type, firm size, and type of industry in which the firm operates. First, we examine whether firms with different ownership types are affected differently by the program. The heterogeneous results are presented in

Table 16. Heterogeneous regression results of firm ownership.

	TFP
	(1)	(2)	(3)
Panel A: State-owned firms
GWDS	0.106 **	0.096 **	0.171 ***
	(0.044)	(0.049)	(0.036)
Observations	534	534	534
Panel B: Non-state-owned firms
GWDS	0.174 ***	0.170 ***	0.178 ***
	(0.058)	(0.057)	(0.059)
Observations	1866	1866	1866
Kernel	Triangle	Epanech.	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [21]. Standard errors are clustered at the county level. *** and ** indicate significance at the 1% and 5% levels, respectively.

. It is found that the GWDS has a greater impact on non-state-owned firms than on state-owned firms. This may be because state-owned firms are subject to less financing constraints, while non-state-owned firms are subject to greater financing constraints. Therefore, the tax reduction effect brought about by the GWDS has a more significant effect on non-state-owned firms. Second, we explore the heterogeneous results for different-sized firms. The median of firm assets is used to differentiate firm size.

Table 17. Heterogeneous regression results of firm size.

	TFP
	(1)	(2)	(3)
Panel A: Large firms (fixed assets above the median)
GWDS	0.144 ***	0.150 ***	0.111 ***
	(0.041)	(0.042)	(0.041)
Observations	1241	1241	1241
Panel B: Small firms (fixed assets below the median)
GWDS	0.244 **	0.241 **	0.223 **
	(0.111)	(0.105)	(0.109)
Observations	1159	1159	1159
Kernel	Triangle	Epanech.	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at the county level. *** and** indicate significance at the 1% and 5% levels, respectively.

presents the heterogeneous results for firm size, showing that the GWDS has a greater impact on smaller firms than on larger firms. This may stem from the fact that smaller firms have a weaker loan capacity and are subject to greater financing constraints. Therefore, some preferential measures of the GWDS may have a more significant effect on smaller firms. Last, we test the heterogeneous results for different industry types. As can be seen in

Table 18. Heterogeneous regression results of industry.

	TFP
	(1)	(2)	(3)
Panel A: Capital-intensive industries
GWDS	0.111 ***	0.088 **	0.094 **
	(0.038)	(0.041)	(0.041)
Observations	1715	1715	1715
Panel B: Labor-intensive industries
GWDS	0.276 **	0.237 *	0.196 **
	(0.134)	(0.123)	(0.084)
Observations	685	685	685
Kernel	Triangle	Epanech.	Uniform

Note. TFP estimated using the method of Levinsohn and Petrin [24]. Standard errors are clustered at the county level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

, the GWDS has a greater impact on labor-intensive industries than on capital-intensive industries. The possible cause could be that the GWDS boosts wages in firms. Consequently, more skilled workers are employed, increasing the efficiency of labor-intensive firms.

6. Conclusions and Policy Implications

Like many other countries, China is faced with obvious regional development gaps, which impedes balanced economic development. As an effective tool for regional development, place-based policies have been widely used in China. Existing studies have confirmed the role of place-based policies in promoting employment growth and regional economic growth, but there is a lack of firm-level research on the effect of place-based policies on firm TFP. Hence, this research provides an empirical study of the impact of place-based policies on firms in China, the largest developing country in the world.

Using detailed firm-level data on Chinese industrial firms together with geographic data, this study analyzed the effects of the GWDS. Employing a geographic regression discontinuity design, the study used firms within 10 km on either side of the GWDS boundary as research samples to identify the causal effects of the GWDS on the TFP of Chinese manufacturing firms. The results reveal that the GWDS increased firm TFP on the western side of the boundary in the range of 11.2% to 13.7%. A series of tests confirmed the robustness of these results. The result of this paper is consistent with the results of Kline and Moretti [7] and Lu et al. [3], who also found that place-based policies can promote the firm TFP in the affected areas.

The study further explored the mechanisms of this effect. The results indicate that although the GWDS does not increase the labor and capital inputs of firms, it does raise the wages of firms and increase the employment of high-quality human capital, thereby boosting labor and capital factor productivity. Unlike most studies such as those by Hansen et al. [19] and Shenoy [20], our study found that the GWDS has no significant promoting effect on local employment. However, our result is similar to that of Neumark and Kolko [39], who also did not find that place-based policies increase employment. Moreover, the results of our study are in line with the findings of Kline and Moretti [7] and Chaurey [18], who found that place-based policies can increase employee wages in affected areas. Furthermore, the preferential tax policy of the GWDS reduces corporate income tax and alleviates the financing constraints facing firms. The promotion effect of tax incentives on firm TFP has been confirmed by much of the literature. For example, Zhang [28] found that China’s export tax rebate policy alleviated the financing constraints of firms and thus increased their TFP. Jacob [29] pointed out that the dividend tax exemption in Switzerland could promote the input of labor and capital and thus improve firm productivity. Liu et al. [30] found that a policy of accelerating depreciation of fixed assets in China promoted the increase of firm productivity. Together, these mechanisms improve firm TFP. However, the results also reveal that the GWDS does not improve firm innovation or boost firm exports. The most relevant study is the one conducted by Falck et al. [40], who showed that the place-based innovation policies in Germany only promoted an increase in firm R&D activities in the short term, but that the effect was not sustainable. The heterogeneous effects of different types of firms were further explored. It was found that private firms, small firms, and labor-intensive firms are more affected.

Based on the above analysis results, we propose the following recommendations: Firstly, place-based preferential policies should be enhanced with a specific focus on promoting firm innovation. As China faces increasing pressure to achieve balanced economic development amid the “new normal status,” it is crucial to continue implementing place-based policies to reduce firms’ operating costs and improve their TFP. Although the GWDS effectively reduces firms’ financing constraints, it has limited impact on promoting firm innovation. Therefore, the government should prioritize policies that reduce research and development costs or provide additional tax incentives to promote firm innovation.

Secondly, the business environment in the western region should be improved. In response to increasing economic pressure, stabilizing employment and ensuring people’s livelihoods should be prioritized. This study found that although the GWDS promotes the development of firms, it does not have a significant impact on employment. This may be due to the relatively poor local business environment. Hence, the government should introduce a series of preferential policies that place greater emphasis on promoting local employment.

Thirdly, incentives for capital-intensive industries in the western region should be increased, and greater efforts should be made to transform and upgrade the regional industrial structure. Although the development of capital-intensive industries in the western region has achieved initial targets, the industrial structure there remains inferior to that in the eastern region. Therefore, the government should continue to maintain an appropriate amount of fixed asset investment in the western region and increase investment in capital-intensive industries to stabilize economic growth and facilitate the transformation and upgrading of the regional industrial structure.

The findings of this paper provide an empirical study of the impact of place-based policies on firm productivity in developing countries. However, there are still some limitations in this study, which also provide opportunities for future research. One limitation is that this paper only focuses on the impact of the GWDS on firm productivity. In fact, the effect of place-based policies is multifaceted. For example, Chaurey [18] found that a location-based tax incentive scheme in India not only increased the firm output, fixed capital, and the number of firms, but also improved the welfare of residents. In the future, we will further explore the impact of the GWDS on employment, business creation, and household consumption and income. Another limitation is that this paper only evaluates the effect of the first round of the GWDS due to the limitations of the data. The main limitations of the data are the absence of key variables and data anomalies in the ASIF database after 2007. In the follow-up studies, we will integrate additional and updated firm-level databases when available to carry out a more in-depth study on the effects of the current GWDS. In addition, research on the impact of government policies on environment has been increasing dramatically in recent years [41,42,43]. In future investigations, we can further explore the impact of the GWDS on carbon emission and pollution.