Robot Application and Adjustment of Export Product Scope: Can We Have Both Efficiency and Quality?
-
Jianhong Qi
Abstract
Based on the matching data of China Industrial Enterprise Database and China Customs Database from 2000 to 2015, this paper examines the influence of robot application (RA) on the export product scope (EPS) of multi-product enterprises (MPEs), the influence channels, and the export competition strategies of these enterprises. The research results show that RA has promoted the expansion of EPS of MPEs, and this promotion effect is prominent especially for the export of labor-intensive products, resource-intensive products and for non-state-owned enterprises (SOEs). The adjustment of EPS by RA not only enhances efficiency through productivity improvement effect and cost-saving effect, but also improves quality through emission-reduction effect and product quality effect, which is very obvious in long-quality-ladder enterprises. The RA-expanded product scope covers both old products and new products manufactured in the same industry. Facing fierce competition in the same industry, robot-using enterprises prefer quality competition strategy, which further promotes the expansion of EPS.
1 Introduction and Literature Review
According to the Guiding Opinions of the Central Committee of the Communist Party of China and the State Council on Promoting High-Quality Development of Trade (HQDT) issued in November 2019, promoting HQDT has become a major strategy for China’s economic development. At the same time, “promoting the integration of Internet, Internet of Things, big data, AI, blockchain and trade”,[1] especially the integration of AI and trade, has also become an important driving force of China’s HQDT.
The application of AI in production is mainly realized by industrial robots (hereinafterreferred to as robots). Different from the early robots, the new-generation robots can not only replace human labor in some monotonous, complicated and long-hour jobs, but also have the advantages of automatic control, re-programmability and the ability of performing multi-objective tasks. If robots can achieve “machine learning”, they will be very likely to replace the demographic dividend in a larger scope and to a greater extent, and alleviate the adverse impact of the aging population and rising labor costs on HQDT. Furthermore, since robots are more stable and accurate in standardized, mechanical and repetitive labor, they can help alleviate the problem of “low quality and low price” of China’s export products. As Bessen (2018) pointed out, the significance of AI is not only to replace labor with machines, but also to reduce the price yet improve the quality of products in a competitive market. It can be seen that AI, as one of the driving forces of HQDT, may change the decision-making of export enterprises.
MPEs are the main force of HQDT in China, contributing to most of the country’s exports. Facing the opportunities created by AI for HQDT, MPEs will inevitably consider different product portfolios, that is, the adjustment of EPS, when making export decisions, whether the adjustment of the scale, structure or the quality of export products. After the application of robots, MPEs will inevitably adjust EPS to redistribute their factor resources among different varieties of products in order to improve productivity and maximize profits. Therefore, the core topic that this paper focuses on is: How and why does MPEs adjust EPS in the face of all kinds of benefits brought by AI?
According to the existing literature, the research on AI’s influence on international trade has emerged from scratch in recent years. Goldfarb and Trefler (2018) were the first to point out that economies of scale, knowledge creation and knowledge diffusion emphasized by traditional trade theories will be given new connotations in the AI era, thus affecting a country’s trade model. This view was echoed by Korinek and Stiglitz (2021). On this basis, Artuc et al. (2018) investigated the influence of robots on the trade pattern, wage level and welfare of different countries, finding that for those countries where robots are more widely used, not only their imports from underdeveloped countries will increase in the same industry, but also their own exports will further increase. In terms of export volume, Brynjolfsson et al. (2019) found that machine translation software, as one of AI technologies, can significantly increase the export volume; Żukrowska (2021), based on the trade relationship between the EU and external partners, found that the wide use of AI can increase the scale of trade and reduce the negative impact of COVID-19 on trade. In terms of export quality, Destefano and Timmis (2021) were the first to focus on the effect of AI on export quality, believing that RA can reduce the error probability in the production process and improve the quality of export products; Hong et al. (2022) further found that RA has a U-shaped relationship with product quality.
The marginal contributions of this paper are as follows. Firstly, at the reseach perspective level of research and product, this paper examines the impact of AI on MPEs’ resource allocation, thus enriching and expanding the research on the relationship between AI and trade. Secondly, from the perspective of influencing mechanism, it puts forward that robots, as a human-created technology, not only improve productivity, reduce marginal cost, but also reduce pollutant emissions, improve product quality, and optimize both efficiency and quality. When investigating the channels for RA to influence product quality, the paper further discusses the types of enterprises that benefit from robots from the perspective product quality ladder, which makes up, to some extent, for the defects of ignoring RA’s influence on product quality and underestimating its influence on productivity in the existing literature. Thirdly, from the perspective of content expansion, the paper not only considers the structural adjustment of EPS for old and new products and different industries, but also discusses the competition strategies of enterprises with different RA levels and the influence of these strategies on EPS, once again highlighting the importance of AI to HQDT.
2 Theoretical Hypotheses
It has been found in the existing literature that the introduction of AI is conducive to the improvement of enterprises’ productivity, which is an important way to expand EPS. Specifically, according to the heterogeneous enterprise theory, only enterprises with high productivity will enter the export market. The research of Acemoglu and Restrepo (2018) found that AI can increase productivity by replacing low-skilled labor, upgrading production technology and acting on a specific production link, thus increasing the output and export volume of enterprises. There are three reasons for this. First, the improvement of productivity will prompt enterprises to charge relatively higher prices in the markets that feature a wide variety of products (Bernard et al., 2019), so enterprises have bigger incentives to expand EPS out of the pursuit of high profits. Second, enterprises with higher productivity tend to have more efficient distribution networks (Bernard et al., 2019), which will increase their desire for higher market shares, motivate them to produce and sell more kinds of products, and encourage them to constantly explore new markets (Feenstra and Ma, 2007). Third, the improvement of productivity enables enterprises to balance resource allocation, transfer the excess capacity of core products to the production non-core products, turn some loss-making non-core products into profitable products, and maintain the original market share. With the increase of new products entering the market and the decrease of original products exiting the market, EPS is expanded. Accordingly, we put forward Hypothesis 1.
Hypothesis 1: Export enterprises using AI can expand EPS by improving productivity.
The application of AI can reduce the production cost of enterprises by improving productivity. Moreover, it may directly increase or decrease the production cost, which will affect the profit of enterprises and prompt them to make adjustments to EPS. Although the procurement of robots will directly increase fixed costs, it can encourage enterprises to dynamically adjust production decisions, improve the efficiency of supply chain management, and reduce the production and operating costs of enterprises (Agrawal et al., 2018). In comparison, the impact of AI on labor costs is complex and inconclusive. Acemoglu and Restrepo (2020) believed that robots can do repetitive and low-tech jobs, resulting in unemployment of low-skilled laborers and lowering the actual wage level of enterprises. However, some other scholars hold the opposite view that although AI can replace human capital-intensive jobs and produce a job substitution effect, it can also bring a job creation effect, create new jobs that are difficult to be replaced by robots, increase the employment of high-tech workers, and improve the average wage level and labor cost of enterprises. The reduction in labor cost will lead to an increase in the profit of all products and expand the EPS of MPEs; on the contrary, when the labor cost rises, enterprises will inevitably raise the product price to transfer on the increased cost burden to customers, and the rise of product price will weaken the competitiveness of enterprises and affect their EPS. Therefore, we put forward Hypothesis 2.
Hypothesis 2: Enterprises using AI can adjust EPS through an increase (decrease) in production cost.
According to the Guiding Opinions of the Central Committee of the Communist Party of China and the State Council on Promoting High-Quality Development of Trade, export enterprises must consider the coordination between trade development and environmental protection. As a human-created technology, AI can not only improve productivity and reduce pollutant emissions (Huang et al., 2022), but also directly change the production mode of enterprises and promote green development (Liu et al., 2021). This is because, on the one hand, robots can supervise the pollutant discharge of enterprises in real time, improve the efficiency of pollution control and reduce the pollution caused by incorrect manual operation;[1] on the other hand, robots mostly use clean energy as the main energy source, which reduces pollutant emissions from the source. The green and emission-reduction effect of robots can promote the expansion of EPS of enterprises, and its mechanism can be explained from two perspectives. The first is “induced” EPS expansion, that is, under the condition that the production of enterprises is constrained by the pollutant discharge limit, the reduction in pollutant discharge can be understood as an increase in the pollutant discharge limit to a certain extent, which means that enterprises can increase their output and expand their production scale, thus providing the possibility for EPS. The second is “associated” EPS expansion, that is, RA can reduce or even avoid the pollution caused by technical restrictions, and turn pollutants into by-products, so as to increase the variety of products of enterprises.[2] From this, we put forward Hypothesis 3.
Hypothesis 3: Enterprises using AI can expand EPS by promoting emission reduction.
Using robots in production can realize the integration of production links, reduce product loss, and ensure the quality and accuracy of production. On the one hand, the role of robots in improving product quality is mainly reflected in a series of repetitive and high-precision tasks (Destefano and Timmis, 2021). Compared with human workers, robots can accurately identify and screen out unqualified products. After the product standard is entered into the robot program, the robot can accurately judge whether a product is qualified or not through automatic scanning and other functions, thus reducing the existence of unqualified products in the market. On the other hand, as a precision machine, robots require production inputs to have higher quality. High-quality inputs combined with stable production links are conducive to the ultimate improvement of product quality. Although there are only a handful of researches on AI’s influence on product quality, the influence of product quality on EPS has been recognized by some scholars. Especially in today’s context of rising labor costs, export enterprises wish to increase their profits through “process innovation effect” and “quality improvement effect”, which will force MPEs to make innovations and expand EPS. Therefore, we put forward Hypothesis 4.
Hypothesis 4: AI helps to improve the quality of export products and expand the EPS of MPEs.
As mentioned above, robots expand the EPS of enterprises by enhancing productivity, saving costs, reducing emissions and improving product quality, but the expansion of EPS may further aggravate “erosive competition”. In other words, while expanding EPS, robots will also lead to fierce competition in the same industry, resulting in a “zero-sum game” that damages industrial competitors’ market share and employment rate (Acemoglu et al., 2020; Bonfiglioli et al., 2020). For example, Kugler et al. (2020) found that the use of US robots reduced the employment and income of Colombian workers (especially female workers) in the local labor market, and RA was likely to turn benign competition in the same industry into erosive competition. Different from previous technologies, robots not only enhance productivity, reduce production costs, but also improve product quality. When the export markets overlap with each other, the lowest quotations that different enterprises can offer are similar, so in this case only by improving product quality can they gain a competitive advantage. In the context of increasingly erosive competition, therefore, export enterprises using robots are more likely to adopt a quality competition strategy to gain strategic advantages. Accordingly, we put forward Hypothesis 5.
Hypothesis 5: MPEs using robots are more likely to adopt a quality competition strategy to gain competitive advantage.
3 Data and model design
3.1 Data Source
Because the application of AI in production is mainly realized by robots, but the number of robots used by enterprises is not available, this paper refers to the practice of Acemoglu et al. (2020) and Fan et al. (2021), taking the number of imported robots as the number of robots actually used by enterprises, and the data comes from China Customs Database. At present, there are two kinds of product coding systems for industrial robots: broad-sense coding and narrow-sense coding. In this paper, the broad-sense HS6 coding system[1] is used in benchmark regression, and the narrow-sense HS8 coding system is used in the robustness test. If the products imported by an enterprise include the above-mentioned coded products, it is deemed that robots are used by the enterprise.
The relevant data of MPEs come from China Industrial Enterprise Database and China Customs Database from 2000 to 2015. After matching the data from the two databases, 682863 observations at the enterprise-year level are finally obtained. Specifically, 9461 MPEs have robots in use, covering 17644 observations. On this basis, we exclude the samples with missing main variables and less than 10 employees, and winsorize the main variables at 1% and 99% quantiles. In addition, this paper uses the fixed assets investment price indices and the producer price indices published in the China Statistical Yearbook to deflate the prices of total assets and total industrial output value, respectively.
3.2 MPEs and RA
Figure 1 shows the development trend of RA in Chinese enterprises from 2000 to 2015. It can be seen that the number of enterprises, and the flow and stock data of robots calculated in accordance with the HS6 coding system are slightly higher than those calculated in accordance with the HS8 coding system, but the difference is not big, and they are all increasing. The left axis reflects the number of robots imported under the two coding systems of HS6 and HS8 and the installation flow of robots in China according to the statistics of the International Federation of Robotics (IFR). It can be seen that these three quantity lines have the same trend, and the total number ofrobots imported is close to the authoritative installation flow statistics published by IFR no matter whether the HS6 or HS8 coding system is used. This finding is consistent with that of Fan et al. (2021), which once again shows the rationality of measuring the degree of RA by the number of robots imported. The period of 2000–2008 was the initial stage for Chinese enterprises to import robots, with a small import volume, and the import flow and stock data varied greatly among enterprises. After 2009, however, Chinese enterprises began to import robots in large quantities, with both flow and stock achieving rapid growth. Therefore, using the period of 2000–2015 as the sample period for this paper can reflect the evolution of RA in Chinese export enterprises, while the influence of RA in other development stages will be discussed later on in this paper.
Change in the Number of Robots Imported by China from 2000 to 2015
In this paper, enterprises exporting more than one variety of products (HS6 coding system) in the same year are identified as MPEs. From 2000 to 2015, MPEs always had an absolute advantage in terms of quantity among China’s export enterprises, with the largest number of enterprises exporting 2~10 varieties of products. After 2004, the export volume of MPEs increased year by year, and the gap with single-product enterprises gradually widened. After 2013, the gap became bigger and bigger. Since MPEs have become the mainstay of China’s export, taking them as the research object is of great significance for optimizing resource allocation.
Based on the above facts, in order to preliminarily verify the influence of robots on the EPS of MPEs, we have also drawn the kernel density function curves of enterprises with and without RA (Figure 2). Figure 2a shows the kernel density function of EPS. It can be seen that most enterprises that do not use robots have smaller EPS, while the EPS of enterprises that use robots is obviously larger than that of the enterprises that do not use robots. Figure 2b shows the kernel density function of core product concentration, from which it can be seen that the average product concentration of enterprises using robots is obviously lower than that of those not using robots, and most of the former are in the range of low product concentration, indicating that robots not only promote the expansion of EPS, but also reduce the concentration of core products. Of course, whether this causal relationship is established needs more rigorous empirical test later on in this paper.
EPS and Kernel Density of Product Concentration
3.3 Model Setting
To test the influence of RA on the EPS of enterprises, this paper chooses the fixed effect model for regression, according to the Hausman test result, and the specific model is set as follows:
In Equation (1), the subscript i denotes an enterprise and t represents a year. The dependent variable Scope selects two indicators that reflect the change in EPS: one is EPV (Variety), which is a direct indicator measured by the logarithm of the number of HS6 product varieties exported by enterprise i in year t; the second is product concentration (Core), which is an indirect indicator measured by the proportion of core product exports to total exports. When EPS expands, the number of non-core products exported by enterprises will increase, and the product concentration will decrease accordingly, which indirectly reflects the expansion of EPS. The core independent variable, Robot, represents the logarithm of the number of robots imported according to HS6 coding statistics. The set of controlled variables, Z, selected in this paper mainly includes: enterprise size (Size), which is expressed by the logarithm of the total assets of the enterprise; enterprise age (Age), which is expressed by the logarithm of enterprise age; financing constraint (FC), which is measured by the ratio of interest expense to net fixed assets; export scale (Exp), which is expressed by the logarithm of export delivery value; corporate capital intensity (KL), which is measured by the ratio of net fixed assets to the number of employees.[1] λi and δt represent the enterprise-fixed effect and year-fixed effect respectively, and εit is a random perturbation term.
4 Regression Results and Analysis
4.1 Benchmark Regression Result
From Table 1, it can be seen that whether the fixed effect is controlled or not, the influence coefficient of independent variable Robot, which is the focus of this paper, is always significantly positive on Variety and significantly negative on Core, indicating that the application of robots in MPEs can significantly expand EPS and reduce the concentration of core export products. After other variables and fixed effects are controlled, every 1% increase in the number of robots used by enterprises will expand EPS by 3.3% and reduce product concentration by 0.9%. The possible reason is that robots can improve the unit output and productivity of enterprises, and the improvement of productivity makes enterprises more motivated to explore new markets and gain higher market share, which will ultimately expand their EPS. At the same time, for some non-core products with negative profits, thanks to the productivity improvement effect of robots, enterprises can maintain their existing market share, and upgrade these products in quality, which will increase the export volume of non-core products, thus reducing the concentration of core products.
Benchmark Regression Results
| Variety |
Core |
|||||
|---|---|---|---|---|---|---|
| (1) | (2) | (3) | (4) | (5) | (6) | |
| Robot | 0.049*** | 0.043*** | 0.033*** | −0.003*** | −0.018*** | −0.009*** |
| (15.455) | (9.340) | (7.299) | (−3.403) | (−11.384) | (−5.420) | |
| Size | 0.053*** | 0.012*** | 0.036*** | −0.001*** | −0.002*** | 0.015*** |
| (42.862) | (8.584) | (20.799) | (−3.651) | (−3.560) | (27.744) | |
| Age | 0.005* | 0.028*** | 0.021*** | −0.015*** | −0.042*** | −0.010*** |
| (1.844) | (8.950) | (5.415) | (−17.498) | (−35.503) | (−8.071) | |
| FC | 0.000** | −0.002** | −0.002*** | −0.000*** | −0.000 | 0.000** |
| (2.337) | (−2.419) | (−2.654) | (−3.826) | (−0.338) | (2.028) | |
| Exp | 0.026*** | 0.009*** | 0.010*** | −0.001*** | 0.000 | 0.000* |
| (70.858) | (26.076) | (27.553) | (−8.621) | (1.546) | (1.763) | |
| KL | −0.782*** | −0.255*** | −0.221*** | 0.052*** | 0.006* | 0.035*** |
| (−77.346) | (−24.162) | (−20.168) | (17.582) | (1.788) | (10.132) | |
| Enterprise-fixed effect | Not controlled | Controlled | Controlled | Not controlled | Controlled | Controlled |
| Year-fixed effect | controlled Not | controlled Not | Controlled | controlled Not | controlled Not | Controlled |
| Sample size | 520 251 | 387 071 | 387 071 | 520 251 | 387 071 | 387 071 |
| R2 | 0.060 | 0.746 | 0.750 | 0.004 | 0.700 | 0.715 |
Note: The values in brackets are t values. Enterprise-level robust standard error of clustering is used. *, ** and *** represent the significance at 10%, 5% and 1% levels respectively, the same below. Unless otherwise specified, enterprise-controlled variables, enterprise-fixed effect and year-fixed effect are all controlled in the next table.
As mentioned above, it is not that the bigger the EPS, the better it is. Here, the prerequisite is that the marginal spillover effect of EPS expansion is positive. Therefore, this paper takes EPS, product concentration and its square term as the core independent variables, and takes the logarithmic value of the total export of enterprises (Export) and the logarithmic value of export delivery value (Delivery) as dependent variables for empirical test. The estimation results show that the primary term of EPS has a significant positive effect on export volume and export delivery value, while itssquare term has a significant negative effect, which confirms the inverted U-shaped relationship between EPS and export earnings.[1] At the same time, according to the benchmark regression results, the promotion effect of RA on EPS expansion (0.033) is still on the left side of the symmetrical axis (1.074/(2×0.069)), which indicates that the effect is still positive during the sample period of this paper, thus meeting the prerequisite for the validity of benchmark regression results.
4.2 Robustness Test
Endogenous problems. Firstly, it is necessary to consider the possible reverse causal relationship between the number of robots imported by enterprises and their EPS. In order to solve the endogenous problem, this paper draws on the practice of Kugler etal. (2020) and Bonfiglioli et al. (2020), and selects the logarithmic value of the number of robots used in an industry in the United States (Robot_US) and the robot suitability index in the same industry (Suitability) as instrumental variables. Secondly, in order to ensure that the instrumental variables meet the exogenous conditions, this paper uses the heteroscedasticity instrumental variable method proposed by Lewbel (2012) to make the test again. Thirdly, the paper examines the dynamic relationship between variables through intra-group differences, so as to eliminate the endogenous problems caused by some omitted variables that do not change with time.
Sample selection bias. First of all, because the core independent variable Robot in this paper has a large number of zero values, the estimation results may be biased. To reduce the bias, the paper limits the sample to MPEs that import at least one robot. Secondly, the robots observed in the sample period may be imported by Chinese robot manufacturers for the purpose of intermediate input or R&D, not for the production of final export products. To solve the above problem, the paper refers to the practice of Fan et al. (2021), and identifies robot manufacturers by looking at whether the company names contain “robot”; if yes, they will be identified as robot manufacturers. After this part of samples is removed, the benchmark model is regressed again. Finally, RA may cause enterprises to switch between single-product manufacturing and multi-product manufacturing, and may also cause them to frequently enter and exit the export market, which may lead to biased estimation results. Therefore, this paper excludes the enterprises that have just entered or exited the export market from the sample, and only retains the MPEs that continuously exist in the export market for regression.
Core variable substitution. Because the number of in-use robots at the enterprise level is not directly available, there will be some errors in the selection of proxy variables. The paper adopts two methods for core variable substitution. The first is to construct the core independent variable by using the narrow-sense HS8 coding system of robot products; the second is to construct enterprise-level robot penetration indicator (AI) by using the number of robots installed in China published by IFR.
Replacement of estimation method. On the one hand, in order to alleviate the estimation bias in traditional OLS regression, this paper redefines the core independent variable as whether the enterprise uses robots (Robot_dummy), adopts the treatment effect model to solve the potential endogenous problems, and continues to use Robot_US as the instrumental variable of Robot_dummy. On the other hand, considering that the sample in this paper only contains export enterprises, excluding non-export enterprises, and there is a certain sample selection bias, we replace the sample with all industrial enterprises and use Heckman two-step method to test the robustness.
Changes in sample range. Firstly, industries importing robots differ greatly in China, and some industries import much more robots than other industries, which may dominate the final empirical regression results. To avoid the dominant influence of this situation on the regression results, we make re-estimation by excluding the Top 1 industry and Top 3 industries in terms of the number of imported robots,[1] respectively. Secondly, in order to avoid the overestimation of the benchmark regression result due to the frequent adjustment of EPS due to too few varieties of products, the samples used in this paper for robustness test are MPEs with at least five varieties of export products. Thirdly, considering the fact that an enterprise’s export volume of certain products is too small and will be, in theory, included in statistics, but in reality it does not represent real diversification of products, in order to avoid overestimation of the regression results, this paper excludes the products whose export volume accounts for less than 1% of an enterprise’s total export volume, and reconstructs EPS (Variety_cut) and product concentration index (Core_cut). Fourthly, the previous typical facts show that between the two periods of 2000–2008 and 2009–2015, the scale of RA is obviously different, which may affect the conclusion of benchmark regression. Therefore, we divide the sample period into two stages, 2000–2008 and 2009–2015, and make regression respectively.
Eliminating the influence of other factors. Firstly, in order to eliminate the influence of trade liberalization and foreign investment liberalization on EPS, this paper takes industry-specific import tariffs (Importtariff) and export tariffs (Exporttariff) and the ratio of foreign-funded enterprises’ output to the total output of the industry (Foreignratio) as controlled variables, and includes them in the benchmark regression equation for estimation. Secondly, in order to control the influence of wide use of ICT on enterprises’ RA and EPS, this paper introduces controlled variables such as each province’s density of long-distance optical cables (Cable), Internet penetration (Internet), and number of mobile base stations (Mobile) based on the benchmark model. Thirdly, considering the fact that there are many factors affecting EPS and it is impossible to enumerate and control them one by one, this paper, on the basis of enterprise-fixed effect, also controls the interaction items between province-fixed and year-fixed effects and between industry-fixed and year-fixed effects.
The results of the above robustness test[2] are consistent with the results of benchmark regression, which confirms the robustness of the conclusions of this paper.
5 Investigation of Transmission Channels
The foregoing text shows that AI represented by robots significantly expands the EPS of MPEs and reduces the concentration of core products, but the channels through which this expansion effect is transmitted remain to be further investigated. This paper adopts Liu and Lu’s endogenous mediation effect test model (2015), and continues to take the logarithmic value of the number of robots installed in the same industryin the United States (Robot_US) and the robot suitability index in the same industry (Suitability) as instrumental variables of the endogenous mediation model,[1] so as to examine more rigorously the various transmission channels through which robots expand the EPS of enterprises.
5.1 Productivity Channel
In this paper, productivity (TFP) is selected as the channel variable, and the total factor productivity is measured using the LP method. Column (1) of Table 2A reports the two-stage regression results of the relationship between the core independent variable and TFP. It can be seen that RA has significantly promoted the productivity of enterprises, and it is significant at the level of 1%. At the same time, the test condition (16.38) that Kleibergen-Paap rk F values are all greater than 10% shows that there is no problem of weak instrumental variables. The results listed in Columns (2) and (3) further show that after TFP is included in the model, the increase in productivity significantly expands the EPS of MPEs and reduces the concentration of core products, thus verifying Hypothesis 1 of this paper.
Transmission Channels
| Productivity effect |
Cost effect |
|||||
|---|---|---|---|---|---|---|
| A Efficiency channel improvement |
TFP |
Variety |
Core |
MC |
Variety |
Core |
| (1) | (2) | (3) | (4) | (5) | (6) | |
| Robot | 0.296*** | −1.073*** | ||||
| (5.457) | (−5.278) | |||||
| TFP | 1.038*** | −0.160** | 2.493*** | −0.391*** | ||
| (4.343) | (−2.438) | (6.849) | (−5.723) | |||
| MC | −1.299*** | 0.204*** | ||||
| (−3.692) | (3.132) | |||||
| Kleibergen-Paap rk F value | 300.713 | 31.449 | 95.618 | 19.307 | ||
| Sample size | 386974 | 386974 | 382111 | 382535 | 382438 | 382438 |
| Emission reduction effect | Product quality effect | |||||
| B Quality channel improvement | Emit | Variety | Core | TQ | Variety | Core |
| (7) | (8) | (9) | (10) | (11) | (12) | |
| Robot | −2.649*** | 0.053* | ||||
| (−7.067) | (1.753) | |||||
| Emit | −0.449*** | 0.069*** | ||||
| (−3.073) | (2.616) | |||||
| TQ | 0.125*** | −0.016*** | ||||
| (5.848) | (−2.647) | |||||
| TFP | 2.862*** | −0.402*** | 1.016*** | −0.157** | ||
| (16.384) | (−12.897) | (4.242) | (−2.365) | |||
| Kleibergen-Paap rk F value | 103.652 | 64.298 | 299.922 | 15.292 | ||
| Sample size | 382060 | 381965 | 381965 | 387064 | 386967 | 386967 |
5.2 Production Cost Channel
In this paper, the labor cost of RA and the purchase, maintenance and repair costs of robots are taken into account, all of which are expressed by the concept of marginal cost (MC); the cost-plus rate is calculated using the method of De Loecker and Warzynski (2012), and the marginal cost is obtained by dividing the cost-plus rate by the product price. The results in Column (4) of Table 2A show that the influence of Robot on MC is significantly negative, indicating that RA significantly reduces the marginal cost of enterprises. Columns (5) and (6) show the influence of marginal cost on EPS. In order to get the direct impact of robots on production costs and exclude the indirect impact of productivity improvement, this paper takes total factor productivity into the model as an endogenous variable. The estimation results show that marginal cost has a significant negative impact on EPS and a significant positive impact on product concentration, which indicates that reducing marginal cost will expand the EPS of MPEs and reduce the concentration of core products, thus verifying Hypothesis2 of this paper. Combining Hypotheses 1 and 2, we believe that RA expands EPS by improving productivity and reducing production costs.
5.3 Emission Reduction Channel
Based on the practice of Fan et al. (2021b), this paper uses the logarithmic value (Emit) of the ratio of industrial waste emissions to total industrial output value as the proxy variable of the emission reduction channel. The smaller the value is, the less pollutant emissions per unit output value, and the greener and more environmentally friendly the production and operation of enterprises are. The estimation results in Column (7) of Table 2B show that the influence of Robot on Emit is significantly negative, indicating that RA significantly reduces the emission of pollutants and promotes the green upgrading of enterprises. The results in Columns (8) and (9) further show that the influence of emission reduction is significantly negative on EPS but significantly positive on product concentration, indicating that pollution reduction can expand the EPS of MPEs, thus verifying Hypothesis 3 of this paper.
5.4 Quality Improvement Channel
As mentioned above, RA may also improve product quality and lead to changes in the EPS of MPEs. Therefore, product quality is brought into channel analysis for the first time in this paper. We use the methods of Baldwin and Harrigan (2011) and Khandelwal et al.(2013) to estimate the quality of export products at the enterprise-product level, and use the method of Shi (2014) for normalization, and then add product quality to the enterprise level, so as to calculate the export product quality (TQ) at the enterprise level. The estimation results in Column (10) of Table 2B show that the influence of RA on the quality of export products is significantly positive, indicating that robots significantly improve the quality of export products of MPEs. The estimation results in Columns (11) and (12) of Table 2B show that the improvement of product quality is beneficial to the expansion of EPS and the reduction of product concentration, which verifies Hypothesis 4 of this paper. Based on Hypotheses 3 and 4, this paper holds that RA promotes the expansion of EPS by reducing pollutant emissions and improving product quality. Therefore, RA can not only improve efficiency, but also upgrade quality, thus making products more cost-effective.
Although robots improve the product quality of enterprises on a whole, not all enterprises or products benefit from RA. On the one hand, in order to further explore the types of enterprises benefiting from RA, this paper uses the method of Khandelwal (2010) method to construct a product quality ladder (Ladder) based on the difference between the highest product quality and the lowest product quality of enterprises, divides them into long-quality-ladder and short-quality-ladder enterprises according to the median of the product quality ladder, and respectively investigates the influence of RA on the product quality of these two types of enterprises. As shown in Columns (1) and (2) of Table 3, the impact of RA is significantly positive on the product quality of long-quality-ladder enterprises, but not significant on that of short-quality-ladder enterprises. On the other hand, in order to explore the influence of RA on the quality of different categories of products, we focus on the product level and make regression of the quality of core and non-core products, respectively. As shown in Columns (3) and (4) of Table 3, the influence of Robot on the quality of core and non-core products is significantly positive, that is, RA can significantly improve the quality of both core products and non-core products. As an input in the production process, robots replace the low-skilled labor force of enterprises, not for a particular type of products but for all types of products, so their quality improvement effect not only acts on core products, but also acts on non-core products.
Types of Enterprises and Products Benefiting from the Quality Improvement Effect of RA
| Enterprise level |
Product level |
|||
|---|---|---|---|---|
| Long quality ladder |
Short quality ladder |
Core products |
Non-core products |
|
| (1) | (2) | (3) | (4) | |
| Robot | 0.015*** | 0.002 | 0.031*** | 0.008** |
| (5.387) | (0.505) | (6.694) | (2.332) | |
| Product-fixed effect | Not controlled | Not controlled | Controlled | Controlled |
| Sample size | 177820 | 151090 | 386248 | 3634446 |
6 Further Analysis
6.1 Heterogeneity Investigation
The above results show that RA significantly expands the EPS of MPEs, but this conclusion is based on the mean regression model, which may cover up the heterogeneity effect of enterprises. Therefore, this paper introduces enterprise heterogeneity in different dimensions to further observe the influence of robots on the EPS of MPEs.[1]
Product heterogeneity. According to the method of broad economic classification (BEC) internationally traded goods, export products can be divided into intermediate goods, consumer goods and capital goods. This paper defines capital goods as capital-intensive products, and intermediate goods and consumer goods as labor-intensive products. The results show that RA mainly promotes the expansion of EPS and the reduction of product concentration of labor-intensive products, but has no significant impact on capital-intensive products. The reason is that labor-intensive products are mainly produced by low-skilled labor that robots can replace. On the one hand, robots reduce labor input and production costs to realize their cost-reduction effect; on the other hand, they reduce the probability of errors in production and improve product quality to realize their quality-improvement effect, thus expanding EPS and reducingproduct concentration.
Heterogeneity of resource allocation. This paper constructs an enterprise-level Theil index based on the export value of different products of enterprises, and takes the median of Theil index as the standard for enterprise classification. If the Theil index of an enterprise is higher than the median, it is defined as a resource-intensive enterprise; if not, it is regarded as a resource-dispersed enterprise. The results show that for resource-intensive enterprises, RA promotes the expansion of EPS and the reduction of product concentration; for resource-dispersed enterprises, RA’s influence on EPS is still significantly positive, but the coefficient and significance of influence are smaller and its influence on product concentration is not significant. After RA, resource-intensive enterprises can improve production efficiency, reduce production costs, and allocate the resources originally concentrated on core products to non-core products, thus expanding EPS, reducing product concentration and maximizing resource allocation efficiency.
Heterogeneity of ownership. SOEs and non-SOEs differ greatly in production technology and resource allocation efficiency, and this difference is ultimately reflected in the gap in production efficiency, so RA’s impact on the EPS of these two types of enterprises may be different. The test results show that RA can significantly promote the adjustment of EPS in non-SOEs, but this effect is not significant in SOEs. The possible reasons are that SOEs’ incentive system is imperfect, managers pay insufficient attention to enterprise performance, production efficiency is low, and SOEs’ strict employment management system makes it difficult for labor to be replaced. In contrast, private enterprises in non-SOEs have stronger financing constraints and higher fixed costs of export, and therefore are more motivated to adopt robots to improve productivity and reduce costs; foreign-funded enterprises benefit from the diffusion of foreign technology, have higher productivity and more flexibility in production and employment, and can flexibly adjust their EPS according to their own needs.
6.2 Sources of EPS Expansion
The source of EPS expansion. new products or old products. In this paper, the products that have been in production for a long time are defined as old products, and those that have recently been put into production are defined as new products. According to this classification, we respectively calculate the EPS of new and old products of enterprises, and define the EPS and concentration of new products as Variety_New and Core_New, and the EPS and concentration of old products as Variety_Old and Core_Old. In order to make the EPS of new products comparable to that of old products, this paper uses the SUR method for judgment. The estimation results in Table 4A show that the influence of Robot on Variety_New and Variety_Old is significantly positive at the level of 1%, but its influence on Core_New and Core_Old is significantly negative at the level of 1%. At the same time, the χ2 statistic of SUR test of the EPS of new and old products is 27.90, and the χ2 statistic of the concentration of new and old products is 5.53, indicating that EPS and concentration of new and old products are comparable. Through comparison, it is found that RA significantly increases the variety of both new and old products, but this effect is more obvious for old products. In other words, RA not only provides impetus for the innovation and introduction of new products, but also helps to support the economies of scale of old products and realize the optimization and upgrading of export product mix.
Sources of EPS Expansion
| A New product or old product |
EPS |
Product concentration |
||
|---|---|---|---|---|
| Variety_New |
Variety_Old |
Core_New |
Core_Old |
|
| (1) | (2) | (3) | (4) | |
| Robot | 0.376*** | 1.058*** | -0.009*** | -0.013*** |
| (6.660) | (24.148) | (−6.695) | (−4.575) | |
| χ2 | 27.90*** | 5.53** | ||
| Sample size | 222596 | 300839 | 222595 | 300839 |
| R2 | 0.063 | 0.095 | 0.010 | 0.047 |
| B Different industries or the same industry | EPS | Product concentration | ||
| Variety_diff | Variety_same | Core_diff | Core_same | |
| (5) | (6) | (7) | (8) | |
| Robot | 0.079* | 0.595*** | −0.009* | −0.023*** |
| (1.837) | (6.193) | (−1.823) | (−4.243) | |
| χ2 | 9.87** | 29.88** | ||
| Sample size | 140411 | 82145 | 140411 | 82145 |
| R2 | 0.122 | 0.042 | 0.015 | 0.009 |
The source of EPS expansion: different industries or the same industry. As mentioned above, RA increases the entry of new products into product scope, so are the new products from the same industry or different industries? To answer this question, according to the classification of HS2 coding system, this paper divides new products into products of the same industry (with the same HS2 code) and products of different industries (with different HS2 codes), calculates the EPS (Variety_same and Variety_diff) and the product concentration (Core_same and Core_diff) of the same industry and different industries, and takes them as dependent variables to perform SUR tests again. The estimation results in Table 4B show that RA has a significant impact on the EPS and product concentration of both the same industry and different industries, and the χ2 statistics are greater than the significance level of 5%. Further comparison shows that robots have a greater impact on the EPS and product concentration of the same industry, indicating that RA mainly promotes the increase of product varieties in the same industry. The reason is that AI application can improve the R&D efficiency and innovativeness of enterprises and enable them to produce more kinds of new products. At the same time, because it is easier to obtain economies of scale by producing similar products in the same industry, enterprises will gradually expand the EPS of the same industry with their core products as the axis by virtue of the advantages of robots in reducing production costs, realizing emission reduction and improving product quality.
6.3 Competition Strategy and EPS Expansion
In order to explore the respective competition strategies of leading enterprises and following enterprises in the context of “erosive competition”, this paper uses the method of Eckel et al. (2015) and constructs the following model to obtain the competition strategies adopted by each of the two types of MPEs:
wherein, j represents product category, lnPrice represents the logarithmic value of the export product price, and Rank is a variable denoting the ranking of each product of MPEs in terms of export volume from small to large. The higher the ranking of a product, the closer it is to the core product with the largest export volume. According to the meaning of competition strategy, when δ1 is greater than 0, the enterprise adopts a quality competition strategy; otherwise, it adopts a cost competition strategy. In this paper, a competition strategy variable Comp is introduced into the benchmark model. If the enterprise adopts the quality competition strategy, a value of “1” is assigned to Comp, and if the cost competition strategy is adopted, a value of “0” is assigned to it. On this basis, Comp and Robot together form an interaction item which is then included in the benchmark model to further verify the influence of different competition strategies on EPS adjustment. The results in Table 5 show that the influence of Robot×Comp on leading enterprises and following enterprises is significantly positive, which verifies Hypothesis 5 of this paper. This shows that after RA, both leading and following enterprises actively adopt quality competition strategies to expand EPS in the fierce competition in the same industry, and RA’s coefficient of influence is obviously bigger on following enterprises than on leading enterprises.
Late-Comer Advantages of Robots in Adjusting EPS and the Choice of Competition Strategies
| Product concentration |
||||
|---|---|---|---|---|
| Leading enterprise | Following enterprise | Leading enterprise | Following enterprise | |
| (1) | (2) | (3) | (4) | |
| Robot×Comp | 0.035*** | 0.048*** | −0.003 | −0.006*** |
| (4.368) | (7.707) | (−1.576) | (−3.618) | |
| Robot | 0.072*** | 0.085*** | −0.007*** | −0.008*** |
| (8.618) | (12.282) | (−2.699) | (−3.920) | |
| Comp | −0.092*** | −0.121*** | 0.020*** | 0.028*** |
| (−28.010) | (−40.181) | (17.930) | (26.226) | |
| Sample size | 155766 | 218441 | 155766 | 218441 |
| R2 | 0.854 | 0.803 | 0.770 | 0.721 |
7 Conclusions and Insights
After putting forward the hypotheses about how robots affect the EPS of MPEs, this paper empirically tests these hypotheses by using the matching data of China Industrial Enterprise Database and China Customs Database from 2000 to 2015. Four conclusions are drawn from the research. Firstly, RA can promote the expansion of EPS of MPEs and reduce the concentration of core products. After a series of robustness tests, this conclusion still holds. Secondly, robots not only improve productivity, reduce marginal costs, but also expand EPS by reducing pollutant emissions and improving product quality. In other words, robots expand the EPS of enterprises by ensuring both efficiency and quality, and their quality improvement effect is particularly prominent in long-quality ladder enterprises. Thirdly, the role of robots in expanding EPS is more significant in labor-intensive, resource-intensive and non-state-owned enterprises. Fourthly, RA expands EPS, but the expanded EPS includes both old products and new products of the same industry, which may lead to fierce competition within the same industry. Therefore, according to the chronological order of RA in different enterprises, this paper finds that, after RA, the tendency to choose a quality competition strategy is obviously enhanced in both leading and following enterprises. In particular, following enterprises rely on their late-comer advantages to expand EPS more actively and reduce product concentration by adopting the quality competition strategy in order to cope with the fierce competition in the same industry.
The conclusions of this paper are of obvious significance for cultivating China’s new advantages in international trade competition and promoting HQDT. First of all, robots can effectively improve productivity and product quality while reducing marginal costs. Although leading export enterprises have first-mover advantages in this area, following export enterprises can also gain late-comer advantages by using robots, so as to fully grasp the new opportunities in the AI era and become much more intelligent. Secondly, based on the heterogeneity of enterprises affected by robots, it is suggested that export enterprises, whether labor-intensive, resource-dispersed or non-state-owned, should consider starting RA as soon as possible so as to give full play to the role of robots in expanding EPS, increasing Chinese exports and achieving more steady and sustained export growth. Finally, RA helps to shorten the product quality ladder of export enterprises and encourage them to adopt the quality competition strategy, which will not only alleviate the outstanding problem of low quality and low price of Chinese export products, but also prove that AI is an indispensable and important driving force of and breakthrough point in China’s pursuit of HQDT.
References
Acemoglu, D., & Restrepo, P. (2018). Artificial Intelligence, Automation and Work. NBER Working Papers, 24196.10.3386/w24196Search in Google Scholar
Acemoglu, D., & Restrepo, P. (2020). Robots and Jobs: Evidence from US Labor Markets. Journal of Political Economy, 128(6), 2188–2244.10.1086/705716Search in Google Scholar
Acemoglu, D., Lelarge, C., & Restrepo, P. (2020). Competing with Robots: Firm-Level Evidence from France. AEA Papers and Proceedings, 110, 383-388.10.1257/pandp.20201003Search in Google Scholar
Agrawal, A., McHale, J., & Oettl, A. (2018). Finding Needles in Haystacks: Artificial Intelligence and Recombinant Growth. NBER Working Papers, 24541.10.3386/w24541Search in Google Scholar
Artuc, E., Bastos, P., & Rijkers, B. (2018). Robots, Tasks, and Trade. World Bank Policy Research Working Papers, 8674.10.1596/1813-9450-8674Search in Google Scholar
Baldwin, R., & Harrigan, J. (2011). Zeros, Quality, and Space: Trade Theory and Trade Evidence. American Economic Journal: Microeconomics, 3(2), 60–88.10.1257/mic.3.2.60Search in Google Scholar
Bernard, A., Blanchard, E., Van Beveren, I., & Vandenbussche, H. (2019). Carry-Along Trade. Review of Economic Studies, 86(2), 526–563.10.1093/restud/rdy006Search in Google Scholar
Bernard, A., Redding, S., & Schott, P. (2010). Multiple-Product Firms and Product Switching. The American Economic Review, 100(1), 70–97.10.1257/aer.100.1.70Search in Google Scholar
Bes sen, J. E. (2018). AI and Jobs: The Role of Demand. NBER Working Papers, 24235.10.3386/w24235Search in Google Scholar
Bonfiglioli, A., Crinò, R., Fadinger, H., & Gancia, G.(2020). Robot Imports and Firm-Level Outcomes. CESifo Working Paper Series, No.8741.10.2139/ssrn.3744604Search in Google Scholar
Brynjolfsson, E., Hui, X., & Liu, M.(2019). Does Machine Translation Affect International Trade? Evidence from a Large Digital Platform. Management Science, 65(12), 5449–5460.10.1287/mnsc.2019.3388Search in Google Scholar
De Loecker, J., & Warzynski, F.(2012). Markups and Firm-Level Export Status. The American Economic Review, 102(6), 2437–2471.10.1257/aer.102.6.2437Search in Google Scholar
Destefano, T., & Timmis, J. D.(2021). Robots and Export Quality. Policy Research Working Paper Series.10.1596/1813-9450-9678Search in Google Scholar
Eckel, C., Iacovone, L., Javorcik, B., & Neary, J. (2015). Multi-Product Firms at Home and Away: Cost-versus Quality-Based Competence. Journal of International Economics, 95(2), 216–232.10.1016/j.jinteco.2014.12.012Search in Google Scholar
Fan, H., Hu, Y., & Tang, L. (2021). Labor Costs and the Adoption of Robots in China. Journal of Economic Behavior and Organization, 179(11), 16–23.10.1016/j.jebo.2020.11.024Search in Google Scholar
Feenstra, R., & Ma, H. (2007). Optimal Choice of Product Scope for Multiproduct Firms under Monopolistic Competition. NBER Working Papers, 13703.10.3386/w13703Search in Google Scholar
Goldfarb, A., & Trefler, D. (2018). AI and International Trade. NBER Working Papers, 24254.10.3386/w24254Search in Google Scholar
Hong, L., Liu, X., & Zhan, H. (2022). Use of Industrial Robots and Chinese Enterprises’ Export Quality Upgrading: Evidence from China. Journal of International Trade & Economic Development, 31(6), 860–875.10.1080/09638199.2021.2018021Search in Google Scholar
Huang, G., He, L., & Lin, X. (2022). Robot Adoption and Energy Performance: Evidence from Chinese Industrial Firms. Energy Economics, 107, 105837.10.1016/j.eneco.2022.105837Search in Google Scholar
Khandelwal, A. (2010). The Long and Short of Quality Ladders. Review of Economic Studies, 4, 1450–1476.10.1111/j.1467-937X.2010.00602.xSearch in Google Scholar
Khandelwal, A., Schott, P., & Wei, S. (2013). Trade Liberalization and Embedded Institutional Reform: Evidence from Chinese Exporters. The American Economic Review, 103(6), 2169–2195.10.1257/aer.103.6.2169Search in Google Scholar
Korinek, A., & Stiglitz, J. (2021). Artificial Intelligence, Globalization, and Strategies for Economic Development. CEPR Discussion Papers, 15772.10.3386/w28453Search in Google Scholar
Kugler, A., Kugler, M., Ripani, L., & Rodrigo, R. (2020). U.S. Robots and Their Impacts in the Tropics: Evidence from Colombian Labor Markets. NBER Working Papers, 28034.10.3386/w28034Search in Google Scholar
Lewbel, A. (2012). Using Heteroscedasticity to Identify and Estimate Mismeasured and Endogenous Regressor Models. Journal of Business & Economic Statistics, 30(1), 67–80.10.1080/07350015.2012.643126Search in Google Scholar
Liu, L., Yang, K., Fujii, H., & Liu, J. (2021). Artificial Intelligence and Energy Intensity in China’s Industrial Sector: Effect and Transmission Channel. Economic Analysis and Policy, 70, 276–293.10.1016/j.eap.2021.03.002Search in Google Scholar
Liu, Q., & Lu, Y. (2015). Firm Investment and Exporting: Evidence from China’s Value-Added Tax Reform. Journal of International Economics, 97(2), 392–403.10.1016/j.jinteco.2015.07.003Search in Google Scholar
Żukrowska, K. (2021). Artificial Intelligence (AI) and International Trade. Artificial Intelligence and Its Contexts, 225–240.10.1007/978-3-030-88972-2_14Search in Google Scholar
© 2023 Jianhong Qi, Zhitong Zhang, Published by DeGruyter
This work is licensed under the Creative Commons Attribution 4.0 International License.
Articles in the same Issue
- Frontmater
- Structural Evolution of RMB Exchange Rate Reform: Historical Review, Experience and Prospect
- A Study of Coordinating China’s Two-Pillar Regulatory Policy under the Shock of the Fed’s Interest Rate Hike—From the Perspective of “Stable Growth” and “Risk Prevention”
- The “Stagflation” Risk and Policy Control: Causes, Governance and Inspirations
- Robot Application and Adjustment of Export Product Scope: Can We Have Both Efficiency and Quality?
- The Effect of “Pressure-Type” Fiscal Incentives on Industrial Restructuring in China
- Convergence Analysis of Village Collective Economy Based on a Long-Term (1978–2018) Observation of 40 Villages in the Suburbs of Beijing
Articles in the same Issue
- Frontmater
- Structural Evolution of RMB Exchange Rate Reform: Historical Review, Experience and Prospect
- A Study of Coordinating China’s Two-Pillar Regulatory Policy under the Shock of the Fed’s Interest Rate Hike—From the Perspective of “Stable Growth” and “Risk Prevention”
- The “Stagflation” Risk and Policy Control: Causes, Governance and Inspirations
- Robot Application and Adjustment of Export Product Scope: Can We Have Both Efficiency and Quality?
- The Effect of “Pressure-Type” Fiscal Incentives on Industrial Restructuring in China
- Convergence Analysis of Village Collective Economy Based on a Long-Term (1978–2018) Observation of 40 Villages in the Suburbs of Beijing
