The Impact of Industrial Robots on the Sustainable Development of Zombie Firms in China

Abstract

Utilizing panel data from listed Chinese manufacturing companies spanning the period from 2011 to 2018, this paper investigates the impact of industrial-robot applications on the resurgence of zombie firms. The fixed-effects-estimation results reveal that a 1% increase in industrial-robot utilization corresponds to a 0.49% rise in the revival rate of financially distressed entities. The paper underscores the instrumental role of industrial robots in fostering the sustainable development of zombie firms, emphasizing improvements in capital returns, labor productivity, and overall factor productivity. Heterogeneity analysis further highlights the varied influences of geographical location, capital intensity, and pollution emissions on this relationship. This paper contributes to a comprehensive understanding of the crucial role played by industrial automation in revitalizing struggling enterprises within the Chinese manufacturing landscape.

1. Introduction

In recent years, the phenomenon of “robots replacing humans” has rapidly gained momentum across various industries. In 2022, China’s industrial-robot consumption reached 297,000 units, maintaining its global lead for ten consecutive years. Zombie firms denote companies that are facing operational challenges and low profitability but are still surviving due to financial aid or lenient credit terms. Therefore, can the use of industrial robots help zombie firms overcome challenges? On one hand, increasing the number of robots used will significantly raise the purchasing and maintenance costs for firms. Limited availability of skilled professionals may hinder the effective utilization of robots, and the rapidly escalating costs are unfavorable for the revival of zombie firms. However, on the other hand, increasing the use of robots can reduce the likelihood of human errors in production, contribute to improving product quality, and enhance overall enterprise efficiency. Simultaneously, the use of robots contributes to increasing operational efficiency, expanding production scale, lowering production costs, and generating economies of scale, thereby improving overall enterprise performance and promoting the revival of zombie firms.

In order to investigate the impact of robot usage on the revival of zombie enterprises and the relevant mechanisms, we utilize data from Chinese A-share manufacturing companies from 2011 to 2018. China’s A-share manufacturing companies denote manufacturing firms with shares listed and traded on the Shanghai and Shenzhen stock exchanges in China. We analyze the relationship between industrial-robot usage and the revival of zombie firms, study the channel mechanisms through which robot usage affects the revival of zombie firms, and analyze the heterogeneous effects of robot usage on different types of zombie firms.

Our research is closely related to two categories of the literature: one is about the economic effects of robot applications and their impacts on employment, and the other is about zombie firms. Regarding the economic effects of robot applications, some scholars have analyzed the impacts of robot applications on the micro-performance of enterprises. For instance, Graetz and Michaels (2018), Koch et al. (2021) found that increasing the use of robots can significantly improve the labor productivity of companies [1,2]. Ghobakhloo and Fathi (2021) argue that industrial robots, as a form of general technological advancement, can drive the rational allocation of production factors, thereby optimizing the production process [3]. Another group of scholars has analyzed the impacts of robot applications on macroeconomic development. For example, Vinuesa et al. (2020) believe that the integration of technologies like robots with economic activities is a crucial engine driving the transformation and upgrading of the manufacturing industry [4]. Kromann et al. (2020) analyzed the impact of industrial-robot applications on economic growth and income inequality among residents [5].

Regarding the impacts of robot applications on employment, the focus is mainly on aspects such as the total employment and the structure of employment. Concerning the impact on total employment, some scholars argue that robot applications generate a substitution effect on labor. For instance, Bidwell (2013) suggests that in situations where machines can replace human labor, employers are more inclined to adopt flexible and elastic employment relationships to adapt to constantly changing market demands and withstand external economic shocks [6]. Frey and Osborne (2017), using a Bayesian-classifier method and analyzing 702 occupations in the United States, found that 47% of positions are at high risk of being highly automated [7]. Dauth et al. (2017) found that the application of industrial robots can significantly reduce labor costs for companies, with each additional industrial robot replacing an average of 1.6 full-time employees [8]. Acemoglu and Restrepo (2018) analyzed U.S. data and discovered a negative correlation between robot applications and employment in the United States [9]. However, some scholars believe that robot applications can increase employment opportunities. For example, Autor (2015), using data from the U.S. manufacturing sector, found that the application of robots led to the creation of 230,000 new jobs from 2010 to 2015 [10]. Alexopoulos and Cohen (2016) argue that technological progress increases production efficiency, promotes the expansion of production scale, and increases demand for labor [11]. Dauth et al. (2021) found that robot applications reduce employment in the manufacturing sector, but these effects are completely offset by new jobs in the service industry, resulting in an overall increase in labor demand [12]. Autor et al. (2022) found that automation innovation significantly increases new tasks and new occupations, thereby increasing the potential for workers’ job transitions [13].

Regarding the impact on employment structure, Benzel et al. (2019) found that under the application of new technologies such as robots, when the skills of workers do not match the skill levels required for the job positions, it can lead to structural unemployment [14]. When workers anticipate that their current jobs are about to be replaced by intelligent devices, robots, and algorithms, their willingness to undergo job transitions became stronger (Brougham and Haar, 2018) [15]. Aghion et al. (2020) found that the negative impact of robots on employment was greater for an uneducated group of workers compared to an educated group [16]. When analyzing the heterogeneity of robot impact on employment, Acemoglu and Restrepo (2020) found that the substitution effect of robot applications is more pronounced for blue-collar occupations involved in routine and physical labor [17]. Xie et al. (2021) pointed out in their analysis that enterprises introducing artificial intelligence will reduce the demand for low-skilled labor while continuously increasing the demand for high-skilled labor [18].

Research on zombie firms primarily focuses on two aspects. First, regarding the causes of zombie firms, scholars believe that factors such as the explicitization of non-performing loans by banks (Peek and Rosengren, 2005), unreasonable bankruptcy systems (McGowan et al., 2017), and government subsidies (Hoshi and Kshyap, 2010; Chang et al., 2021) are the main triggers for the formation of zombie firms [19,20,21,22]. Second, concerning the economic effects of zombie firms, Kwon et al. (2015) and Tan et al. (2016) argue that zombie firms occupy a significant amount of resources, suppressing the production efficiency of non-zombie firms and hindering the rational allocation of resources [23,24]. The negative effects of zombie firms are also evident in areas such as financing costs (Lin et al., 2015; Imai, 2016), the ratio of new entrants (Hoshi and Kim, 2012), and the deterioration of excess production capacity (Shen and Chen, 2017) [25,26,27,28]. Furthermore, scholars have analyzed the impact of measures such as subsidy preferential policies (Bruche and Llobet, 2014) and credit allocation efficiency (Dai et al., 2021) on the exit of zombie firms from the market [29,30].

While a considerable body of literature has explored the influence of robot applications on macroeconomic development and employment, there is a dearth of research on their effects on micro-enterprise development, particularly concerning zombie firms. In addressing the developmental challenges of zombie firms, our focus is on facilitating their transformation into normal firms. Nevertheless, the existing literature has not delved into the impacts of robot applications on the revival of zombie firms. Thus, relying on the current literature, it proves challenging to ascertain whether robot applications can facilitate the revival of zombie firms, not to mention the channel mechanisms through which robot applications affect the revival of zombie firms. Additionally, it remains unclear whether robot applications will exhibit heterogeneous effects on the revival of various types of zombie firms.

Through an exploration of the transmission mechanism and the heterogeneous effects of robot applications on the revival of zombie firms, we can conduct a thorough analysis of the varied impacts of robot applications on different types of zombie firms, thereby aligning with the research objective of understanding the impacts of robot applications on the revival of zombie firms. In comparison to the existing literature, this paper makes the following marginal contributions. First, from the perspective of the profound integration of robots into the real economy, we investigated the impacts of robot applications on the revival of zombie firms, thereby enriching the relevant research on robot applications and zombie firms. Second, we analyze the mechanisms through which robot applications impact the revival of zombie firms. We discover that enhancing capital returns, labor productivity, and TFP serve as effective channels for fostering the revival of zombie firms. Third, relying on regional characteristics, capital intensity, and pollution emissions of enterprises, we analyze the heterogeneous effects of robot applications on the revival of various types of zombie firms. Our research will offer valuable insights for the transformation and development of zombie firms.

The remainder of this paper is organized as follows. Section 2 analyzes the characteristics of the main variables and theoretically examines the mechanisms through which robot applications affect the revival of zombie firms. Section 3 introduces the model settings and variable selection. Section 4 empirically tests the impacts of robot applications on the revival of zombie firms and their channel mechanisms, as well as the heterogeneous effects on the revival of different types of zombie firms. Section 5 conducts robustness tests. Section 6 presents the main research conclusions.

2. Empirical Facts and Theoretical Model

2.1. Empirical Facts

Table 1. Density of robot usage by industry.

Industry	Robot-Usage Density (Units/10,000 People)	Robot-Usage Density (Units/10,000 People)	Increase in Robot-Usage Density (Units/10,000 People)
	2011 Year	2018 Year	2011–2018
Automotive manufacturing	62.95	600.95	538.01
Electrical and electronic	4.77	116.12	111.35
Metal products	6.06	67.13	61.07
Other manufacturing industries	2.08	53.83	51.75
Plastics and chemicals	8.71	33.01	24.30
Industrial machinery and equipment manufacturing	0.66	27.85	27.19
Railway, shipbuilding, aerospace, and other transportation equipment manufacturing industries	1.59	18.20	16.60
Food and beverage	1.04	16.09	15.05
Non-Metallic mineral products	0.55	8.39	7.84
Basic metals	0.63	7.42	6.79
Wood and furniture	0.00	4.70	4.70
Paper	0.24	2.73	2.49
Textile	0.01	0.65	0.64

Notes: This table summarizes descriptive statistical information on the use of robots in different manufacturing industries. Data Source: Calculated by the authors.

shows the density of robot usage by industry. Looking at the density of robot usage by industry, the automotive manufacturing industry, electrical and electronic industry, metal products industry, plastics and chemicals industry, and the industrial machinery and equipment manufacturing industry, as well as other manufacturing industries, have higher robot-usage density. In contrast, the food and beverage industry, non-metallic mineral products industry, wood and furniture industry, paper industry, and textile industry have lower robot-usage density. Looking at the growth in the density of robot usage by industry, the increase is still higher in high-end manufacturing industries such as automotive manufacturing and electrical and electronic industries. In contrast, traditional industries such as wood and furniture, paper, and textile industries have lower growth rates.

2.2. Theoretical Model

Drawing on the research approach of Hsieh and Klenow (2009), we establish an economic model to analyze the mechanism through which robot applications impact zombie firms [31]. Assuming the total output $Y_s$ of industry s is in the CES-aggregation form of industry outputs,

$$Y_s={\left(\sum {Y_{si}}^{\frac{\sigma -1}{\sigma }}\right)}^{\frac{\sigma }{\sigma -1}}$$

(1)

where σ represents the degree of differentiation among products from different firms, and $Y_{si}$ represents the output of representative firm i in the industry, given by

$$Y_{si}=Y_s{\left(\frac{P_s}{P_{si}}\right)}^{\sigma }$$

(2)

where $P_{si}$ and $P_s$ represent the product prices of the representative firm i and the industry s, respectively. Additionally, we assume that the production function $Y_{si}$ for the representative zombie firm i satisfies the constant returns to scale in the Cobb-Douglas function form

$$Y_{si}=A_{si}{K}_{si}^{{\alpha }_s}{L}_{si}^{{1-\alpha }_s}$$

(3)

where $K_{si}$ and $L_{si}$ represent the capital and labor inputs for firm i, respectively, and $A_{si}$ is the productivity of firm i. The parameter ${\alpha }_s$ represents the elasticity coefficient for industry s, reflecting the contribution of the capital input to output, while ${1-\alpha }_s$ reflects the contribution of the labor input to output.

We denote the deviation degree of robot application on capital yield as ${\tau }_{ksi}$ and the deviation degree on labor productivity as ${\tau }_{lsi}$.

The profit function for the representative zombie firm i can be expressed as

$${\mathsf{\pi }}_{si}=P_{si}{Y}_{si}-\left(1+{\tau }_{ksi}\right)r{K}_{si}-\left(1+{\tau }_{lsi}\right)\omega L_{si}$$

(4)

where r and ω represent the prices of capital and labor inputs, respectively. Firms make decisions based on the profit-maximization condition to determine the product price $P_{si}$, as well as the allocation of capital input $K_{si}$ and labor input $L_{si}$. From the first-order conditions of ${\mathsf{\pi }}_{si}$ we obtain the following:

$${\alpha }_s\frac{\sigma -1}{\sigma }{P}_{si}{A}_{si}{\left(\frac{K_{si}}{L_{si}}\right)}^{{\alpha }_s-1}=r(1+{\tau }_{ksi})$$

(5)

$$\left({1-\alpha }_s\right)\frac{\sigma -1}{\sigma }{P}_{si}{A}_{si}{\left(\frac{K_{si}}{L_{si}}\right)}^{{\alpha }_s}=\omega (1+{\tau }_{lsi})$$

(6)

It can be observed that the existence of ${\tau }_{ksi}$ and ${\tau }_{lsi}$ due to robot usage causes the deviation of marginal returns from marginal costs for firm i. Subsequently, we can calculate the prices and output for the enterprise as follows:

$$P_{si}=\frac{\sigma }{\sigma -1}\frac{1}{A_{si}}{\left(\frac{\omega \left(1+{\tau }_{lsi}\right)}{1-{\alpha }_s}\right)}^{{1-\alpha }_s}{\left(\frac{r\left(1+{\tau }_{ksi}\right)}{{\alpha }_s}\right)}^{{\alpha }_s}$$

(7)

$$Y_{si}={\left(\frac{\sigma -1}{\sigma }\right)}^{\sigma }{A_{si}}^{\sigma }{\left(\frac{1-{\alpha }_s}{\omega \left(1+{\tau }_{lsi}\right)}\right)}^{\sigma {(1-\alpha }_s)}{\left(\frac{{\alpha }_s}{r\left(1+{\tau }_{ksi}\right)}\right)}^{{\sigma \alpha }_s}{P_s}^{\sigma }{Y}_s$$

(8)

We measure enterprise performance using corporate profits, and the performance of zombie firm i can be expressed as follows:

$$E_{si}=P_{si}{Y}_{si}={\left(\frac{\sigma -1}{\sigma }\right)}^{\sigma -1}{A_{si}}^{\sigma -1}{\left(\frac{1-{\alpha }_s}{\omega \left(1+{\tau }_{lsi}\right)}\right)}^{(\sigma -1){(1-\alpha }_s)}{\left(\frac{{\alpha }_s}{r\left(1+{\tau }_{ksi}\right)}\right)}^{{(\sigma -1)\alpha }_s}{P_s}^{\sigma }{Y}_s$$

(9)

where $D_s={\left(\frac{\sigma -1}{\sigma }\right)}^{\sigma -1}{\left(\frac{1-{\alpha }_s}{\omega }\right)}^{(\sigma -1){(1-\alpha }_s)}{\left(\frac{{\alpha }_s}{r}\right)}^{{(\sigma -1)\alpha }_s}{P_s}^{\sigma }{Y}_s$. The performance function for zombie firm i can be expressed as follows:

$$E_{si}=\frac{1}{{\left(1+{\tau }_{ksi}\right)}^{({\sigma -1)\alpha }_s}}{\frac{1}{{\left(1+{\tau }_{lsi}\right)}^{(\sigma -1){(1-\alpha }_s)}}{A}_{si}}^{\sigma -1}{D}_s$$

(10)

Assuming that industry characteristics ($D_s$) are constant, the performance of zombie firms depends on factors such as the deviation of capital yield (${\tau }_{ksi}$), the deviation of labor productivity (${\tau }_{lsi}$), and enterprise productivity ($A_{si}$). Specifically, the performance of zombie firms is inversely proportional to the deviation of capital yield and labor productivity but directly proportional to enterprise productivity. This indicates that when the application of robots leads to a decrease in the deviation of capital yield and labor productivity, it will improve firm performance, thereby increasing the probability of zombie firms’ revival. When the application of robots results in an increase in enterprise productivity, it enhances firm performance, thus activating existing zombie firms. Therefore, we believe that increasing the use of robots can promote the revival of zombie firms by enhancing capital yield, labor productivity, and total-factor productivity.

3. Empirical Specification

3.1. Data Source

The robot data we used is sourced from the International Federation of Robotics (IFR), labor-structure data is from the Wind database, and other data such as enterprise financial indicators are sourced from the CSMAR database. Since the IFR database only contains robot data up to 2018, and the Wind database has labor-structure data from 2011 onwards, we have chosen data from Chinese A-share listed manufacturing companies from 2011 to 2018 to address our research questions. We have selected data from listed manufacturing companies based on two considerations: First, IFR provides global data on industries such as agriculture, forestry, animal husbandry, and fishing, as well as segmented data on robots in the manufacturing industry. Second, the global manufacturing industry data can be matched one-to-one with China’s manufacturing industry two-digit-code data. Third, the CSMAR database and Wind database can provide more comprehensive and recent labor-structure data for listed companies. To ensure the rigor and reliability of empirical analysis, we excluded samples with significant missing financial data.

3.2. Model Specification

Building on the mechanism analysis in Section 2, we conduct a two-step study on the impact of robot application on the revival of zombie firms and its underlying mechanisms. In the first step, to examine the impact of robot application on the revival of zombie firms in the manufacturing industry, we construct the following baseline model:

$${ZF}_{it}={\alpha }_0+\beta {Rd\_Ch}_{it}+{\mathrm{X}}_{it}^{\prime }\gamma +{\mu }_i+{\mu }_t+{\epsilon }_{it}$$

(11)

where i and t represent the enterprise and the year, and the dependent variable ${ZF}_{it}$ is a dummy variable indicating whether a zombie firm i is revived in year t, taking the value of 1 if the enterprise revives and 0 otherwise. The key explanatory variable ${Rd\_Ch}_{it}$ represents the density of robot usage at the enterprise level. ${\mathrm{X}}_{it}^{\prime }$ includes other control variables, specifically controlling for the influence of factors at the enterprise, industry, and provincial levels on the revival of zombie firms. ${\mu }_i$ and ${\mu }_t$ represent individual fixed effects and time fixed effects, respectively, and ${\epsilon }_{it}$ is the disturbance term. The estimated coefficient β reflects the impact of robot-usage density on the revival of zombie firms. Specifically, if β > 0, it indicates that increasing robot usage has a positive promoting effect on the revival of zombie firms; if β < 0, it implies that increasing robot usage has a negative inhibitory effect on the revival of zombie firms.

In the second step, to analyze the channels through which robot application affects the revival of zombie firms, we introduce three intermediary variables: capital yield rate (Cap), labor productivity (Lab_pro), and total-factor productivity (TFP), for use in constructing a mediation effects model.

3.3. Variable Definitions

Regarding explanatory variables, we draw on the method of Acemoglu and Restrepo (2020) and use robot-usage density Rd_Ch to measure the application of robots in Chinese manufacturing enterprises [17]. Firstly, using data from the “China Industrial Statistics Yearbook” and the China-specific data from the IFR, we match data for 13 manufacturing industries, including food and beverage, textile, wood and furniture, paper, plastic and chemical products, non-metallic mineral products, basic metals, metal products, industrial machinery and equipment manufacturing, electrical and electronic, automotive manufacturing, railways, ships, aerospace, and other transport equipment manufacturing. Next, selecting the year 2010 as the base year, we calculate robot-usage density using the Bartik instrumental variable method

$${Rd\_Ch}_{it}=\frac{P_{ijt=2011}}{{MP}_{t=2011}}\times \frac{R_{jt}}{L_{j,t=2010}}$$

(12)

where $P_{ijt=2011}$ is the personnel ratio of enterprise i in industry j in the year 2011, ${MP}_{t=2011}$ is the median personnel ratio of the manufacturing industry in the year 2011, $R_{jt}$ is the stock of robots in industry j in year t, and $L_{j,t=2010}$ is the employment quantity in industry j at the base year 2010.

Regarding the dependent variable, we draw on the methods defined by Caballero et al. (2008), Fukuda and Nakamura (2011) to define the revival of zombie firms [32,33]. In the first step, we calculate the minimum interest payment of enterprise i in year t, denoted as $R_{i,t}^{\ast }$

$$R_{i,t}^{\ast }=r_{s,t-1}\times b_{s,t-1}+\frac{1}{5}\sum _{j=1}^5{r}_{l,t-j}\times b_{l,t-1}+{rcb}_{min}\times {bonds}_{i,t-1}$$

(13)

where s represents the short term, l represents the long term, $r_{s,t-1}$ and $r_{l,t-j}$ are the benchmark loan interest rates for one year and the five-year average, respectively, and ${rcb}_{min}$ is the minimum coupon rate for convertible bonds within five years. $b_{s,t-1}$, $b_{l,t-1}$ and ${bonds}_{i,t-1}$ represent the short-term loans, long-term loans, and outstanding bond balance of enterprise i, respectively.

In the second step, using the actual interest payment $R_{i,t}$ of enterprise i in year t, we calculate the interest differential of the enterprise, denoted as $x_{i,t}$

$$x_{i,t}=\frac{R_{i,t}-R_{i,t}^{\ast }}{b_{s,t-1}+b_{l,t-1}+{bonds}_{i,t-1}+{cp}_{i,t-1}}$$

(14)

where ${cp}_{i,t-1}$ represents the balance of interest-bearing notes payable for the enterprise. If the interest differential of the enterprise is negative, the enterprise is identified as a zombie firm.

In the third step, we utilize the profit information of the enterprise to reduce the likelihood of misidentifying the enterprise as a zombie firm.

$${FN}_{i,t}={EBIT}_{i,t}-R_{i,t}^{\ast }$$

(15)

If the post-subsidy earnings before interest and taxes (${EBIT}_{i,t}$) of an enterprise, after deducting government subsidies, are higher than the minimum interest payment, then the enterprise is identified as a normal company. If a certain enterprise simultaneously satisfies having post-subsidy earnings before interest and taxes less than the minimum interest payment, a leverage ratio (${LEV}_{i,t}$) greater than 0.5, and if the total debt (${DEBT}_{i,t}$) in period t compared to the total debt in period t − 1 (${DEBT}_{i,t-1}$) has increased, then the enterprise is identified as a zombie firm once again.

In the fourth step, to avoid the existence of one-time zombie firms, we stipulate that if an enterprise is identified as a zombie firm in both the previous period and the current period, it is considered a zombie firm in the current period. In the fifth step, when an enterprise is a normal company in the current period and a zombie firm in the previous period, it is considered that the zombie firm has revived in the current period.

Regarding control variables affecting zombie firm revival at the enterprise level, we select variables such as enterprise age (Age), cash flow ratio (Cfo), leverage ratio (Lev), working capital (Loc), operating income (Reve), and financial asset allocation (Allo). For factors affecting zombie firm revival at the industry and provincial levels, we select variables such as the proportion of zombie firms in the industry (Zb_s) and per capita GDP at the provincial level (Pergdp). Descriptive statistics for the main variables are provided in

Table 2. Variable definitions and summary statistics.

Variable	Definitions	Obs.	Mean	Minimum	Maximum
Zombie	A dummy variable that equals one if the firm revives	2032	0.098	0	1
Rd_Ch	The density of robot usage	2032	2.094	0	6.684
Age	The enterprise age	2032	2.929	1.386	3.689
Cfo	The cash flow ratio	2032	0.586	0	8.820
Lev	The leverage ratio	2032	0.626	0.016	63.971
Loc	The working capital	2032	0.025	−12.493	0.912
Reve	The operating income	2032	66.935	0	1050.203
Allo	The financial asset allocation	2032	0.205	0	0.966
Zb_s	The proportion of zombie firms in the industry	2032	0.179	0	1
Pergdp	The per capita GDP at the provincial level	2032	10.912	9.706	11.939

Notes: This table summarizes the descriptive statistical data of different variables.

.

4. Empirical Results and Analysis

We begin by analyzing the impacts of robot applications on the revival of zombie firms. Subsequently, we examine the channels through which robot applications affect the revival of zombie firms, focusing on capital returns, labor productivity, and total-factor productivity. Finally, we explore the heterogeneous effects of robot applications on the revival of different types of zombie firms.

4.1. Baseline Results

Table 3. Impacts of robot applications on the revival of zombie firms.

Variable	(1)	(2)	(3)	(4)
Rd_Ch	0.4038 *	0.3933 *	0.4436 *	0.4877 **
	(1.76)	(1.70)	(1.90)	(2.05)
Age		2.0406	2.0535	2.4921
		(0.81)	(0.82)	(0.98)
Cfo		−0.2645	−0.2766	−0.3009
		(−0.93)	(−0.94)	(−0.99)
Lev		−0.4501	−0.4556	−0.4456
		(−0.88)	(−0.91)	(−0.87)
Loc		−0.5744	−0.5868	−0.5829
		(−1.09)	(−1.13)	(−1.11)
Reve		−0.0030 *	−0.0031 *	−0.0030 *
		(−1.65)	(−1.70)	(−1.66)
Allo		0.2550	0.1978	0.2468
		(0.32)	(0.25)	(0.31)
Zb_s			−1.0758	−1.0754
			(−1.47)	(−1.47)
Pergdp				−2.0669 *
				(−1.92)
Firm FE	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes
Observations	1474	1474	1474	1474

Notes: * and ** represent statistical significance at the 10% and 5%, levels, respectively. z-statistics are presented in parentheses.

presents the regression results of the impacts of robot applications on the revival of zombie firms. In column (1), the results include the core explanatory variable, robot-usage density, and the regression results with individual and time fixed effects. It can be observed that the estimated coefficient of robot-usage density (Rd_Ch) is significantly positive, suggesting a preliminary indication that robot applications can promote the revival of zombie firms. In column (2), with the addition of enterprise-level control variables such as enterprise age, the estimated coefficient of robot-usage density remains significantly positive, indicating that a higher robot-usage density is associated with a higher likelihood of zombie firm revival. Column (3) presents the regression results further controlling for the industry-level variable, which is the proportion of zombie firms in the industry. The estimated coefficient of robot-usage density remains significantly positive and passes the significance test at the 10% level. Column (4) provides the complete regression results controlling for the provincial-level variable, which is per capita GDP. It can be observed that the estimated coefficient of robot-usage density is significantly positive at the 5% level, with an estimated coefficient of approximately 0.49. The results suggest that for every 1 percentage point increase in an enterprise’s robot-usage density the zombie firm revival rate increases by 0.49 percentage points. One possible reason for this is that increasing the usage of robots in enterprises improves operational performance, which is beneficial for promoting the revival of zombie firms.

4.2. Mechanism

The research described above indicates that the use of industrial robots can effectively assist zombie firms in overcoming challenges. Combining the theoretical mechanisms analyzed earlier, in this section we will analyze the mechanisms through which robot applications impact the revival of zombie firms, specifically focusing on capital returns, labor productivity, and total-factor productivity. Regarding the intermediary variables, we calculate the capital return rate (Cap), measure labor productivity (Lab_pro) using the per capita operating income of an enterprise. Additionally, we adopt the method proposed by Ackerberg et al. (2015) to compute total-factor productivity (TFP) [34]. The regression results are presented in

Table 4. Analysis of mechanisms through which robot applications impact the revival of zombie firms.

Variable	(1)	(2)	(3)	(4)	(5)	(6)
Rd_Ch	0.0502 ***	0.5156 **	0.1493 ***	0.4401 *	0.1541 ***	0.4347 *
	(2.72)	(2.13)	(4.31)	(1.83)	(4.28)	(1.81)
Cap		3.6262 ***
		(3.70)
Lab_pro				0.3227 **
				(1.97)
TFP						0.3288 **
						(2.03)
Age	0.4162 **	2.3364	0.1090	2.6533	0.1767	2.6603
	(2.23)	(0.91)	(0.31)	(1.04)	(0.48)	(1.04)
Cfo	−0.1387 ***	−0.3345	0.5766 ***	−0.5524	0.4986 ***	−0.5391
	(−7.71)	(−0.99)	(17.09)	(−1.54)	(14.21)	(−1.51)
Lev	−0.7766 ***	0.1499	−0.0283 ***	−0.4677	−0.0322 ***	−0.4117
	(−159.82)	(0.20)	(−3.10)	(−0.72)	(−3.39)	(−0.63)
Loc	−0.2162 ***	−0.6397	0.5159 ***	−0.7019	0.5616 ***	−0.6533
	(−14.36)	(−1.00)	(18.26)	(−1.12)	(19.12)	(−1.05)
Reve	0.0004 ***	−0.0036 *	0.0015 ***	−0.0035 *	0.0019 ***	−0.0037 *
	(3.28)	(−1.92)	(5.87)	(−1.87)	(7.15)	(−1.94)
Allo	−0.2311 ***	−0.1275	−0.3312 ***	0.2330	−0.5574 ***	0.2897
	(−3.47)	(−0.16)	(−2.65)	(0.29)	(−4.29)	(0.36)
Zb_s	0.0197	−0.9627	−0.3638 ***	−0.8623	−0.3493 ***	−0.8540
	(0.35)	(−1.31)	(−3.49)	(−1.17)	(−3.22)	(−1.16)
Pergdp	−0.1369	−1.7421	−0.1932	−1.9735 *	−0.2146	−1.9866 *
	(−1.54)	(−1.60)	(−1.16)	(−1.82)	(−1.24)	(−1.83)
Sobel Test	Z = 2.1917		Z = 1.7925		Z = 1.8313
Firm FE	Yes	Yes	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	2032	1474	2032	1474	2032	1474

Notes: Columns (1) and (2) report the regression results of Cap as a mediator variable. Columns (3) and (4) report the regression results of Lab_pro as a mediator variable. Columns (5) and (6) report the regression results of TFP as a mediator variable. *, **, *** represent statistical significance at the 10%, 5%, and 1% levels, respectively. z-statistics are presented in parentheses.

.

4.2.1. Capital Yield Rate

In Table 4, column (1) presents the regression results of robot-usage density (Rd_Ch) on the capital yield rate. The estimated coefficient of robot-usage density is significantly positive and passes the significance test at the 1% level. This indicates that as robot-usage density increases, the capital yield rate also rises. Column (2) shows the regression results with the capital yield rate (Cap) as the mediating variable. The estimated coefficient of the capital yield rate is positive and passes the significance test at the 1% level. This suggests that increasing the capital yield rate can effectively promote the revival of zombie firms. The regression results also pass the Sobel test, indicating that enhancing the capital yield rate is an effective channel through which robot applications significantly promote the revival of zombie firms. This implies that when the production advantage of robots surpasses that of labor, companies reduce their demand for labor, leading to a substitution of machines for human effect. This can enhance capital efficiency, particularly in sectors where the repetitive nature of labor is prominent.

4.2.2. Labor Productivity

In Table 4, column (3) presents the regression results of robot-usage density on labor productivity. The estimated coefficient of robot-usage density is positive and passes the significance test at the 1% level, indicating that higher robot-usage density is associated with higher labor productivity. Column (4) shows the regression results with labor productivity (Lab_pro) as the mediating variable. The estimated coefficient of labor productivity is positive and passes the significance test at the 5% level. This suggests that increasing labor productivity significantly promotes the revival of zombie firms. The regression results also pass the Sobel test, indicating that enhanced labor productivity is the second important channel through which robot applications influence the revival of zombie firms. This may be explained by the fact that robot applications lead to more intelligent production processes, motivating employees to acquire advanced skills, and ultimately contributing to the revival of zombie firms.

4.2.3. TFP

In Table 4, column (5) presents the regression results of robot-usage density on total-factor productivity. The estimated coefficient of robot-usage density (Rd_Ch) is positive and significantly passes the 1% significance level, indicating that higher robot-usage density is associated with higher TFP. Column (6) shows the regression results with TFP as the mediating variable. The estimated coefficient of TFP is positive and significantly passes the 5% significance level, suggesting that increasing TFP significantly promotes the revival of zombie firms. The regression results also pass the Sobel test, indicating that TFP is another channel through which robot applications influence the revival of zombie firms. Robot applications can enhance the technical capabilities of firms, improve overall productivity, and increase the probability of the revival of zombie firms. This may be explained by the fact that robot applications contribute to the better alignment of production factors, improving overall productivity and the efficiency of intelligent manufacturing and resource allocation, thus promoting the upgrading of firms and increasing the likelihood of the revival of zombie firms.

4.3. Heterogeneous Responses

4.3.1. Different Regions

We have previously analyzed the impacts of industrial-robot applications on the revival of zombie firms, finding that the use of robots can effectively help zombie firms to overcome challenges. However, the preceding analysis primarily reflects the overall impacts of robot applications on the revival of zombie firms, without revealing the heterogeneous effects on different types of zombie firms. Given that different zombie firms are located in diverse regions, have varying levels of capital intensity, and exhibit different degrees of pollution emissions, the impacts of robot use on the revival rates of different types of zombie firms should logically differ. To conduct a more detailed analysis of the impact of robot use on the revival of zombie firms, we categorize firms based on their geographical location, capital intensity, and pollution-emission levels, distinguishing between southern and northern enterprises, capital-intensive and labor-intensive enterprises, and also high-pollution and low-pollution enterprises. This allows us to investigate the heterogeneous effects of robot use on the revival of different types of zombie firms.

Heterogeneity analysis was conducted based on different regions, with the dummy variable assigned a value of 1 for the southern region and a value of 0 for the other regions. Interaction terms between dummy variables (D_r) representing different regions and robot applications (Rd_Ch) were incorporated into the benchmark model. This analysis aimed to assess whether the impacts of robot applications on the revival of zombie firms exhibit significant variations across different regions. The heterogeneity analysis results for different regions are presented in column (1) of

Table 5. Heterogeneous responses.

Variable	(1)	(2)	(3)
Rd_Ch	0.1124	0.2217	−0.2003
	(0.39)	(0.78)	(−0.45)
D_r×Rd_Ch	0.4147 **
	(2.25)
D_c×Rd_Ch		0.3896 *
		(1.67)
D_p×Rd_Ch			0.5922 *
			(1.81)
Controls	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes
Time FE	Yes	Yes	Yes
Observations	1474	1474	1474

Notes: * and ** represent statistical significance at the 10% and 5%, levels, respectively. z-statistics are presented in parentheses.

. The coefficient of D_r×Rd_Ch is significantly positive at the 5% level, suggesting that robot applications exert a more pronounced impact on the revival of zombie firms in the southern region.

4.3.2. Different Capital-Intensity Levels

Heterogeneity analysis is performed based on the degree of capital intensity, with a value of 0 assigned to capital-intensive firms and a value of 1 assigned to labor-intensive firms. Interaction terms between dummy variables representing different levels of capital intensity (D_c) and robot applications (Rd_Ch) were incorporated into the benchmark model. This analysis aimed to assess whether the impacts of robot applications on the revival of zombie firms exhibit significant differences based on varying levels of capital intensity. The heterogeneity analysis results for different levels of capital intensity are presented in column (2) of Table 5. The coefficient of D_c×Rd_Ch is significantly positive at the 10% level, suggesting that robot applications exert a more pronounced impact on the revival of labor-intensive zombie firms.

4.3.3. Different Pollution Levels

Heterogeneity analysis is performed based on the degree of pollution, assigning a value of 0 to high-polluting firms and a value of 1 to low-polluting firms. Interaction terms between dummy variables representing different levels of pollution (D_p) and robot applications (Rd_Ch) were incorporated into the benchmark model. This analysis aimed to assess whether the impacts of robot applications on the revival of zombie firms exhibit significant differences based on varying levels of pollution. The heterogeneity analysis results for different levels of pollution are presented in column (3) of Table 5. The coefficient of D_p×Rd_Ch is significantly positive at the 10% level, suggesting that robot applications exert a more pronounced impact on the revival of low-pollution zombie firms.

5. Robustness Checks

In order to further validate the reliability of the conclusion that robot application can promote the revival of zombie firms, we conducted the following robustness tests. Firstly, we redefined the revival of zombie firms. Secondly, we excluded data from the automotive industry. Thirdly, to minimize potential endogeneity issues, we employed instrumental variable methods for empirical testing. Fourthly, in order to minimize the impact of omitted variables, we added control variables and fixed effects for empirical testing.

5.1. Replacement of the Dependent Variable

In the baseline regression, if an enterprise was a zombie enterprise in the previous period and a non-zombie enterprise in the current period, we considered the enterprise revived in the current period. This method mainly emphasizes the “temporary” revival of zombie enterprises and does not consider the sustainability of zombie enterprise revival. To test the robustness of the results, we redefine the revival of zombie enterprises. We define a zombie enterprise’s revival if it meets either of the following conditions: first, it was a zombie enterprise in the previous period and a non-zombie enterprise in the current and subsequent periods; second, it was a zombie enterprise in the previous period and a non-zombie enterprise in the current and remaining years. Repeating the empirical analysis, we obtained the results in columns (1) and (2) of

Table 6. Robustness test.

Variable	(1)	(2)	(3)	(4)	(5)	(6)
Rd_Ch	0.4877 **	0.7728 ***	0.4768 **		0.1341 **	0.5103 **
	(2.05)	(2.66)	(2.01)		(2.26)	(2.11)
Age	2.4921	1.4568	2.7233	0.3610	0.1327	2.3090
	(0.98)	(0.49)	(1.06)	(1.59)	(0.62)	(0.90)
Cfo	−0.3009	−0.2735	−0.3330	0.0111	−0.0252	−0.3341
	(−0.99)	(−0.95)	(−1.06)	(0.51)	(−1.23)	(−1.00)
Lev	−0.4456	−1.4267 *	−0.4449	0.0007	−0.0040	0.1625
	(−0.87)	(−1.89)	(−0.86)	(0.11)	(−0.72)	(0.22)
Loc	−0.5829	−1.8144 **	−0.5817	−0.0261	−0.0167	−0.6284
	(−1.11)	(−2.54)	(−1.10)	(−1.44)	(−0.97)	(−0.98)
Reve	−0.0030 *	−0.0032 *	−0.0029	−0.0002	−0.0003 *	−0.0036 *
	(−1.66)	(−1.69)	(−1.62)	(−1.30)	(−1.87)	(−1.94)
Allo	0.2468	1.5886 *	0.2413	−0.1277	0.0232	−0.1020
	(0.31)	(1.68)	(0.30)	(−1.58)	(0.30)	(−0.12)
Zb_s	−1.0754	−0.3844	−1.2170	0.2180 ***	−0.1287 *	−0.9542
	(−1.47)	(−0.47)	(−1.63)	(3.24)	(−1.94)	(−1.29)
Pergdp	−2.0669 *	−1.7827	−2.0422 *	0.3362 ***	−0.2185 **	−1.8968 *
	(−1.92)	(−1.55)	(−1.82)	(3.13)	(−2.10)	(−1.65)
Roa						3.6132 ***
						(3.69)
Pro_indus						0.1874
						(0.57)
Sec_gdp						0.0222
						(0.42)
Instrument				0.0218 ***
				(15.90)
Non-identifiability Test (LM Statistic)				253.010 ***
Weak Instrument Test (F-Statistic)				293.626 ***
Industry FE	No	No	No	No	No	Yes
Firm FE	Yes	Yes	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	1474	1331	1434	2032	2032	1474

Notes: Columns (1) and (2) report the robustness test results after redefining the resurrection of zombie enterprises. Column (3) reports the robustness test results after excluding data from the automotive industry. Columns (4) and (5) report the results of the robustness test using instrumental variable methods. Column (4) includes the main regressors and controls, while column (5) adds the instrumental variable. The instrument variable is the average robot density in major importing countries of Chinese robots. The non-identifiability test (LM Statistic) and weak instrument test (F-Statistic) assess the validity of the instrumental variable. Column (6) reports the results of adding control variables and fixed effects. *, **, *** represent statistical significance at the 10%, 5%, and 1% levels, respectively. z-statistics are presented in parentheses.

. The results show that, using two different methods to define the revival of zombie enterprises, the regression coefficient of robot-usage density remains significantly positive under the inclusion of control variables, individual fixed effects, and time fixed effects. This is consistent with the previous empirical results, indicating that our baseline regression results are robust.

5.2. Exclusion of the Automotive Industry Sample

In the above baseline regression, we used data from the entire manufacturing industry to analyze the impact of robot application on the revival of zombie enterprises. However, from the analysis of characteristic facts, it can be seen that the robot-usage density in the automotive industry is significantly higher than in other manufacturing industries, which may affect the accuracy of the regression results. To verify the robustness of our baseline regression results, we follow the method of Acemoglu and Restrepo (2020) to exclude the automotive industry sample [17]. Column (3) of Table 6 presents the regression results after removal of the automotive industry sample. After including control variables and controlling for individual fixed effects and time fixed effects, the estimated coefficient of robot-usage density (Rd_Ch) remains significantly positive and passes the significance test at the 5% level. This is consistent with the previous empirical results, indicating that the increased use of robots significantly increases the probability of zombie enterprise revival. Thus, our baseline regression results are robust.

5.3. Endogeneity Treatment

Considering that there may be omitted variables or bidirectional causality between robot usage and zombie enterprise revival in the baseline regression, we use instrumental variable methods to address potential endogeneity issues. We use the average robot density in major importing countries of Chinese robots, such as the United States, Japan, South Korea, Germany, and Sweden, as an instrument. Although there is a strong correlation between robot application in China and the robot production in these countries, there is no evidence to suggest that robot application in these countries would affect the Chinese manufacturing industry’s zombie enterprises. Therefore, the instrument variable meets the exogeneity condition. Columns (4) and (5) of Table 6 present the regression results using the instrumental variable. It can be seen that, after including control variables and controlling for individual fixed effects and time fixed effects, the estimated coefficient of robot-usage density remains significantly positive and passes the significance test at the 5% level. Moreover, the non-identifiability test shows that the regression results are significant at the 1% level, rejecting the hypothesis of non-identifiability of the instrumental variable. The weak instrumental variable test shows that the F-statistic is greater than the critical value at the 10% level, rejecting the hypothesis that the instrumental variable is weak. This implies that, even after employing instrumental variables, the estimated coefficient of robot-usage density retains its significant positive value. In comparison with the benchmark regression results, there is no significant alteration in the sign and significance level of the estimated coefficient for the core variable of robot-usage density. A noteworthy observation is that augmenting the utilization of robots substantially promotes the revival of zombie firms, signifying the robustness of our benchmark regression results.

5.4. Add Control Variables

Acknowledging the potential presence of omitted variables between robot usage and the revival of zombie firms, we have introduced additional control variables with the aim of mitigating the influence of other factors on the revival of zombie firms. These encompass the firm-level return on assets (Roa), industry-level average operating profit margin (Pro_indus), and the regional-level proportion of the manufacturing industry in GDP (Sec_gdp). Furthermore, fixed effects for the firm, industry, and year have been incorporated. In column (6) of Table 6, the regression results are presented, with the estimated coefficient of robot-usage density being positive and passing the significance test at the 5% level. The estimates of the core explanatory variable show no significant changes in terms of their signs and significance levels with the introduction of control variables at various levels and the inclusion of fixed effects for the firm, industry, and year. This implies that the regression results in this paper demonstrate robustness and are not contingent on the choice of control variables.

6. Conclusions

We utilized data from Chinese-manufacturing-listed companies from 2011 to 2018 to investigate the impact of industrial-robot applications on the revival of zombie firms. Our results indicate that the use of industrial robots can effectively help zombie firms to overcome challenges and achieve sustainable development. Empirical results indicate that for each 1% increase in the density of industrial-robot usage the revival rate of zombie enterprises rises by 0.49 percentage points.

Examining the channels of influence, the use of robots can promote the revival of zombie enterprises by increasing capital returns, labor productivity, and total-factor productivity. Heterogeneity analysis indicates that the impact of industrial robots on the revival of zombie enterprises is constrained by various factors, such as the region where the enterprise is located, the intensity of capital, and pollution-emission levels. Specifically, industrial-robot applications increase the revival rates of zombie enterprises in the southern region, labor-intensive zombie enterprises, and low-pollution zombie enterprises. The impacts on zombie enterprises in the northern region, capital-intensive zombie enterprises, and high-pollution zombie enterprises are not significant.

Lastly, constrained by data availability, there exists an inevitable limitation in the selection of indicators for gauging the utilization of robots. Hence, for future research in this domain, we intend to explore other pertinent variables to undertake a more exhaustive analysis of the influence of robot applications on zombie firms. In subsequent research endeavors, we will delve deeper into analyzing the differentiated impacts of robot applications on zombie firms across diverse markets. Additionally, within the framework of low-carbon transformation and development, we will scrutinize the synergistic effects and implementation pathways of robot applications in zombie firms to contribute to pollution reduction and carbon reduction.