GVC Participation, Productivity, and Energy Efficiency: A Firm-Level Evidence

(1)

MASTER THESIS

GVC Participation, Productivity, and Energy

Efficiency: A Firm-Level Evidence

Name

: Aryo A. Sunaryo

Student number : S2842084

Email address :

a.a.s.sunaryo@student.rug.nl

Supervisor

: Prof. Dr. Catrinus J. Jepma

Co-Assessor

: Prof. Inmaculada Martínez-Zarzoso, Ph.D.

Program

:

MSc. International Economics and Business (Double Degree

with Georg-August-Universität Göttingen, Germany)

Department Global Economics and Management (GEM)

Faculty of Economics and Business

(2)

(3)

3

ABSTRACT

As firms may not equally be involved in the global value chain (GVCs), and hence equally affected by trade policies, we believe that a firm-level analysis is needed to complement the study of country- and industry-level analysis. We use Propensity Score Matching combined with Difference-in-Difference (PSM-DiD) estimation to analyze the effects of firms’ participation into GVCs on their economic and energy performance in three South East Asian (SEA) countries over the year 2009 and 2015. We find that firm participation into GVCs leads to higher productivity and energy efficiency of the participated firms, and owning internationally-recognized quality certification can even provide stronger learning effects for them. These results are confirming the economic and environmental benefits of global production network participation for GVC-firms in emerging markets.

(4)

4

I. INTRODUCTION

The rise of global value chains (GVCs) has led to a (vertical) specialization of the country in specific activities and stages within value chains, rather than in the entire industries. OECD reported that in 2014 over 70% of global trade was in the form of intermediate goods and capital goods. Hence countries have been more interdependence than before, requiring the needs of new view to formulate trade policy. Accordingly, the analysis of GVCs now becomes one of the most discussed topics in the international trade literature.

GVCs is the range of activities required in the production and delivery of a good or service to the final consumer. These include design, procurement, manufacturing, assembly, distribution, marketing, and after-sales service (Buckley and Strange, 2015). GVCs activities have formed a highly complex production networks because the parts and the components produced in different countries can come from several firms, can be integrated (offshore) or not (outsource) within a company’s border, and can be assembled through various steps along the value chain or in a final location (Prete and Rungi, 2015).

Against this background, the international trade pattern should be analyzed beyond the country-level framework. Besides some growing studies in the industry-country-level analysis using Input-Output table (e.g. Hummels, Ishii, and Yi, 2001; Weber, 2009; Timmer et al., 2013), a firm-level analysis has become a common trend in international trade literature thanks to the work of Melitz (2003). In a global production network setting, firms do not necessarily need to develop the capacity to perform all production steps since they can focus on specific tasks and support the value chain as suppliers of intermediate inputs or subcontractors (Humphrey and Schmitz, 2002). Focusing on specific activities also implies that firms may not equally be involved in GVCs, even though they work within the same industry. Some firms may participate in high value-added activates, thus generate higher income, while others participate in low value-added activities.1

In the firm-level analysis of international trade, most of the empirical discussion is surrounding the test for Melitz’s model of firm heterogeneity (i.e. whether more productive firms self-select into the export market) and the effects of international trade exposures to firms’ productivity. Driven by this spirit, we would like to analyze whether more productive firms self-select to join GVCs and to assess the effects of joining GVC to firms’ performances. The latter becomes relevant as “firm participation in a global supply chain and cooperation within a network of

(6)

6 upstream and downstream partners can enhance a firm’s information flows and learning possibilities, introduce new business practices and more advanced technology, which in turn enhancing productivity” (Prete, Giovannetti and Marvasi, 2016).

Besides its impact on productivity, another concern over international trade and increasing global market integration is its impact on the environment. While there is a common ground on the positive impacts of trade on productivity, an intense debate underlies the trade-environment nexus. On the one hand, the international trade supporters argue that international trade is good for the environment since it can enhance knowledge, environmentally friendly practice and clean-technology transfer from developed to developing countries. On the other hand, pollution haven hypothesis (PHH) suggest that weaker environmental standards in developing countries have caused the emergence of environmental concern (e.g. air pollution) associated with irresponsible GVCs activities. These two opposing arguments have risen more concern among researchers and policymakers regarding the impact of international trade, or more specifically of GVCs activities on the environment.

Following Baldwin and Yan (2016) we define firm participation in GVC when a firm involves in two-way trading activities (i.e. both import and export). In addition, Prete, Giovanetti, and Marvasi (2016) argued that in order to join in international supply chains, internationally-recognized certification is often required since it can guarantee and signal the ability of the firm to meet the international standards. We also argued that GVC-firms that own international certification may have a closer trading relationship with advanced economies. As a result, these firms may have more opportunity to learn modern technology and business process, which in turn may increase the benefits of GVC participation even more. In this regards, we will use a stronger definition of GVC participation, which is two-way traders with international certification (i.e. certified two-way traders), to examine the impact of international standardization in shaping the learning effects of joining GVC.

Furthermore, we utilize the data of total factor productivity provided by the World Bank (2017) as a proxy for firm-level productivity. Additionally, we also use several other measurements to capture labor productivity, such as sales per workers and wages. With regards to environmental performance variables, since the pollution data at the firm level is hard to collect, we use energy intensities as a proxy for environmental performance, i.e. energy use divided by total sales.2 Finally, the discussions bring us to the following research questions: What are the effects of

(7)

7 firm participation in the global value chain to productivity and energy efficiency? Does international certification matters in driving those effects?

To investigate such causal relationship between firm’s participation in GVC on productivity and energy efficiency, we utilize data from World Bank’s Enterprise Survey (WBES) for three Southeast Asia (SEA) countries, namely Indonesia, Vietnam and the Philippines for the year of 2009 and 2015. The choice of using these countries is relevant for a couple of reasons. First, SEA region is one of the dominant participants in GVCs activities with 31.5% of their export contains foreign value added (FVA). Second, SEA countries has exported 98 billion dollars of value added to satisfy final demands of Europe, making them the third largest trading partner for SEA countries. It is worth noting because about 90% of EU citizens are aware of environmental product and quality (Roy and Yasar, 2015). Third, most existing studies are discussing the effects of international offshoring and outsourcing activities on productivity of firms active in developed countries (e.g. Görg, Hanley, and Strobl, 2008; Ito, Tomiura, and Wakasugi, 2011), but it is also important to see these effects on emerging markets standpoint. As is known, firms in developing countries are the recipients of those outsourcing and offshoring activities. With regards to the period of study, since WBES is not conducted every year, we cannot utilize a long time horizon for the dataset. The two periods (2009 and 2015) are the only period where the panel data identification for firms are provided.

Using Propensity Score Matching combined with Difference-in-Difference (PSM-DiD) estimation, we find evidence that the involvement of global production networks leads to higher productivity and energy efficiency. Additionally, we also found that GVC-firms that own international certification can experience stronger learning effects from their GVC activities. These finding suggests that trade liberalization and GVCs participation may generate substantial economic, and environmental benefits for firms in emerging markets.

(8)

8 studies had estimated total factor productivity with the method proposed by Olley and Pakes (1996) or Levinsohn and Petrin (2003), we are one of the first studies that utilize firm-level total factor productivity dataset provided by the World Bank (2017).

(9)

II. LITERATURE REVIEW

2.1. GVCs Activities of SEA Countries

Back in the 1980s, managing an international business was a lot more complicated, phone calls were expensive, shipping was slow, and early computer technology could not give many advantages for business. Nowadays, phone calls and video conference are virtually free, shipping is cheap and easy, and modern computing and the Internet technology have made people more connected than ever. These revolutions have been transforming how companies make products and distribute them worldwide. Take the automobile industry, the Japanese car manufacturer, Toyota work on design and upstream activities of its car production in Japan, but the transmission gear are produced in Toyota’s production plant in Philippines and India, engine control unit (ECU) in Malaysia, gasoline engine in Indonesia, and diesel engine and final assembly in Thailand.3 In the past, this could not be possible because manufacturers were not sure about the quality of parts they would receive. Thanks to the information and communication technology (ICT) revolution, it has never been easier to outsource and to coordinate complex activities from long-distance like today (Baldwin and Evenett, 2015). Current technological advances in ICT and lower transportation cost and trade barriers have allowed the production of goods and services to become more disaggregated into several stages and be conducted in different locations worldwide. The wage differences between advanced and developing countries are the drivers that made the disaggregation profitable for companies (Baldwin and Evenett, 2015). Some of the production stages that require labor-intensive activities are typically located in emerging economies, while other capital- or knowledge-intensive activities stages are located in more advanced economies (Buckley and Strange, 2015). Baldwin and Evenett (2015) called this new international trade pattern as the second leap of globalization unbundling, where “production stages previously performed in close proximity are now dispersed geographically.”

Some countries in Asia and Eastern Europe have been able to gauge these opportunities and participate in global production networks. For China, the benefits are even higher, since it started to upgrade its position from only specializing in final assembly into leading supplier of intermediate goods (Lu et al., 2015). For some countries in Southeast Asia (SEA) region, however, this has not been the case. Although they heavily involved in GVCs, they still work on low value-added activities.

(10)

10 Figure 2.1 shows the backward (measured by foreign value added embodied in exports or FVX) and forward integration (measured by domestic value added embodied in foreign country exports or DVX) of SEA countries in GVCs. As shown by the figures, the share of FVX is much higher than DVX, implying that low value-added activities still dominate SEA’s participation in GVCs. These are opposite with the figure for EU28 and NAFTA countries, where high value-added dominated their GVCs activities.

Figure 2.1 GVC participation of SEA countries

Note: CHN: China; IND: India; IDN: Indonesia; MYS: Malaysia; PHL: The Philippines; SGP: Singapore; THA: Thailand; VNM: Vietnam. Source: Trade in Value Added Database, OECD

Interestingly, despite being located far away from the western economies, North America and European countries are the main trading partner of SEA countries. By looking into domestic (SEA) value added embodied in foreign final demand (DVA), Figure 2.2 shows that SEA countries had traded 118 billion and 97 billion dollars of value added to satisfy NAFTA’s and EU28’s final demands in 2011, respectively. In terms of foreign value added embodied in SEA’s final demand (FVA), EU28 has been the largest value-added contributors, followed by China and Japan. Given its close trading relationship with advanced economies, especially the Europe, positive spillover effects through technological transfer which then leads to higher productivity and energy efficiency are likely to occur.

0 10 20 30 40 50 60 2005 2011 2005 2011 2005 2011 2005 2011 2005 2011

ASEAN EU28 NAFTA CHN IND

% o f G ro ss G DP FVX DVX 0 10 20 30 40 50 60 70 2005 2011 2005 2011 2005 2011 2005 2011 2005 2011 2005 2011

IDN MYS PHL SGP THA VNM

(11)

11 Figure 2.2. Trading partners of SEAs

Source: Trade in Value Added Database, OECD

2.2. GVCs Activities of Firms

The above mentioned macro-level facts reveal that the increasing fragmentation of global production process has made countries and firms more interconnected than before. Consequently, the use of gross export to measure country competitiveness and comparative advantage become less informative (Timmer et al., 2013; Brakman and Van Marrewijk, 2016). Hence, the firm-level analysis of global production network is also needed to complement the country and sectoral analysis in order to set appropriate international trade policies (Prete, Giovannetti and Marvasi, 2016). For instance, the increasing tariff of intermediate goods might harm firms that use a significant amount of foreign intermediate goods but might benefit other domestic enterprises that supply intermediate goods, even if they involve within the same industry.

Ideally, a firm participation in GVC should be measured by how much value-added that a firm can contribute in the production process of a product. However, such data are difficult to collect. Thus, we follow Baldwin and Yan (2016) to define a firm’s GVC-status, “manufacturers that both import intermediate inputs and export intermediate or finished products” (or two-way traders). Although one-way trading firms that only export or only import might, in some cases, also be considered as GVC participants, our criteria highlights a stronger definition of firms’ involvement in global production network (Baldwin and Yan, 2016). It also means that this definition should exclude some firms that are indirectly integrated into GVCs, such as those that use other domestic intermediate products, or supply local firms that in turn

(12)

12 export (indirect export). Later, in addition to this two-way trader’s definition, we will follow Prete, Giovannetti, and Marvasi (2016) who define firms’ involvement in GVC when such firms are two-way traders and possess internationally-recognized quality certification.

Before we go further, it is worth to mention that since the data of pollution on firm-level detail is difficult to collect, we will proxy environmental performance of a firm by its energy usage. This proxy is commonly used by the existing studies since Eskeland and Harrison (2002) found that energy use, on average, is positively associated with firm’s emission. Thus, for the sake of consistency with our variables, we will always use energy efficiency to construct our hypothesis, although the literature discussion may sometimes refer to sother indicators of environmental performances.

2.3. GVC Participation and Self-Selection Hypothesis

Since the work of Melitz (2003), the “new” new trade theory (NNTT) has become the underlying framework to conduct firm-level analysis in international trade. In a nutshell, Melitz (2003) formalized the nature of firm’s heterogeneity to explain the pattern of international trade. He argued that only more productive firms self-select into exporting because only those firms that are able to overcome the costs of entering export markets and compete in internationally competitive markets (“self-selection hypothesis”). Following this theoretical framework, a plenty of studies had been conducted to compare various characteristics of exporters and non-exporters, such as productivity, size, and wage. Most studies agreed that exporters are typically larger, more productive, more skill and capital intensive, and paying higher wages (e.g., Bernard et al., 2007).4

While the study for exporting firms has been a mainstream, the study that compares characteristics of GVCs firms and non-GVCs firms is surprisingly few.Nevertheless, the study from Prete, Giovannetti, and Marvasi (2016) and Baldwin and Yan (2016) found that GVCs firms are also larger, more productive and have higher trade shares. They argue that to become a GVC firms, a firm must incur fixed costs such as direct transportation and tariffs costs, developing a logistics network, communicating product specifications, satisfying international

(13)

13 standards, adopting new technology, and coordinating with international co-workers. In heterogeneous firm models of international trade, only the most productive firms that can afford all of those costs and then can join GVC (Baldwin and Yan, 2016). Thus, the self-selection hypothesis might also hold for self-selection into GVC -- more productive firms are more likely to join GVC.

In terms of environmental performance, Ray and Yasar (2015) states that “the concerns over reverse causation are not only relevant in the context of productivity but also in energy efficiency.” In other words, the positive association between GVCs participation and firm’s energy efficiency might be because more energy efficient firms are self-select to be part of GVCs, not the other way around. To this, we can assume that upon involving in GVCs, firms may adopt newer, more energy-efficient technologies before they involve in international activities (Batrakova and Davies, 2012). This assumption seems reasonable because adopting clean-technology may reduce waste of production, total cost of energy, pollution abatement cost and increase the willingness to pay of environmentally conscious consumers, which in turn generate more profits. As we also want to know whether high productivity and energy efficiencies emanates prior (ex-ante) firm’s participation in GVCs, the following hypothesis need to be tested:

Hypothesis 1. More productive and more energy efficient firms are more likely to self-select

into global value chains.

2.4. GVC Participation and Learning Hypothesis

Besides self-selection hypothesis, the literature suggests other mechanisms that can explain a positive relation between international activities of a firm and its productivity. Based on our definition of GVC status (i.e. two-way traders), there are at least three mechanisms that are relevant to explain the effects of firms’ participation into GVC. They are an export effect, an import effect and a combined effect of the two. Lastly, we will also discuss how the learning effects may be affected when GVC-firms own an internationally-recognized certification.

2.4.1. Export Effects

(14)

14 via exporting. De Loecker (2007) argue that exporting firms may gain some insights, knowledge, and expertise from their international buyers which allow them to improve their efficiency level. To survive in tougher international competition is another reason why exporting firms tend to adopt new technology after they join international trades, which in turn improve their efficiency (Sjoholm, 1999). Additionally, exporting firms also have an opportunity to exploit scale economies as they now expose to a larger market, so that their production costs per output are declining (Baldwin and Yan, 2016).

Unlike the self-selection hypothesis which has been confirmed by most trade economists, the empirical evidence on the learning-by-exporting hypothesis is not clear. Nevertheless, some profound results have been found. Using micro data from Slovenian manufacturing firms, De Loecker (2007) found that exporting firms become more productive once they start exporting, and the productivity gap between exporters and their domestic counterparts diverse over time. By using firm-level UK manufacturing data, Crespi, Criscuolo, and Haskel (2008) found that exporters have more opportunities to learn from clients which then induce faster productivity growth. Girma, Greenaway, and Kneller (2004) found supports for both self-selection and learning-by-exporting hypothesis for British manufactures. That is, exporters are more productive ex-ante and they do self-select, but at the same time their ex-post productivity are also increasing after they start exporting.

In the trade-environment nexus, Batrakova and Davies (2012) had adopted the Melitz’s (2003) model of heterogeneous firms to develop the theoretical model that can explain the effects of exporting on firm-level environmental performance. He explained that when a firm starts exporting, its output tends to rise, increasing its demand for energy and emits the pollution it is responsible for. However, the scale economy of export markets will eventually be able to cover the cost of adopting clean and more energy-efficient technologies.

(15)

15 Shimamoto (2005) and Albornoz, Cole, and Elliott (2009) found exporting and foreign ownership encourage the adoption of environmental management systems among Japanese and Argentinean firms, respectively. More recently, Forslid, Okubo, and Ultveit-Moe (2015) analyzed firm-level data from Sweden and found that exporters have 10-30 percent lower emission intensity of carbon dioxide, sulfur dioxide and nitrogen oxides. Using Indonesian establishment data, Roy and Yasar (2015) found that exporting status can discourage the use of ‘dirty’ source of energy for production (i.e. fuel) relative to electricity.

2.4.2. Import Effects

Importing intermediate inputs can also enhance firm productivity by providing access to foreign inputs and technologies that are unavailable or more expensive to obtain domestically (Grossman and Rossi-hansberg, 2008). Using Indonesian firm-level data, Amiti and Konings (2007) found that lower tariffs on final goods can increase productivity by inducing stronger import competition, whereas cheaper imported inputs raise productivity by learning, variety, and quality effects. This is confirmed by Kasahara and Rodrigue (2008) who found evidence that becoming an importer of foreign intermediates improves the productivity of Chilean manufacturers.

The empirical studies on the effect of imported intermediate inputs on environmental performance at the detailed firm level, however, are relatively scarce. Martin (2011) found that reductions in tariffs on intermediate inputs led to 23% improvement in fuel efficiency of Indian manufacturers, with the entire effects are coming from within-firm improvement rather than between-firm (i.e. industry) improvement. Imbruno and Ketterer (2016) extended the analysis by developing a formal theoretical framework. Their empirical analysis showed that firms that enter intermediate input markets can improve their performance, and reduce their energy intensity, relative to non-importing firms. Through importing, they said, “firms can have access to a larger range of differentiated intermediate inputs, which entails not only a more efficient usage of intermediates itself but also a more efficient usage of energy, implying beneficial effects for the environment.”

2.4.3. Joint Effects (Export and Imports)

(16)

16 importing, vice versa. This idea underlies the definition of GVC-firms proposed by Baldwin and Yan (2016). They argued that when firms involve in GVCs, they might actually benefit from the combined effect of being both an importer and an exporter.

Meanwhile, if any, the studies regarding GVCs activities of a firm are mostly taking a view from advanced countries (home countries) standpoint. In this regards, those scholars are trying to find whether outsourcing activities of parent companies in developed countries’ can affect firms’ productivity. However, firms that active in developing countries, as a recipient (or host) of outsourcing activities, have not yet received proper attention.

Nevertheless, some scholars such as Prete, Giovanetti, and Marvasi (2016) have pioneered the analysis of GVC firms in developing countries. They suggested that since production processes can be sliced into single tasks, firms can specialize in a particular stage of the production and can active in international trade, despite being small. Using propensity score matching and difference-in-difference techniques, they found that GVCs participation can enhance productivity of North African firms. This finding is confirmed by Agostino et al. (2015) who argued that the productivity, technical capabilities and competitiveness of small and medium enterprises are encouraged by joining global production network.

Besides having higher opportunity to utilize the technological transfers from both their upstream and downstream partners, being involved in both importing and exporting can also benefit firms in terms of cost-saving. Using plant-level data for Chilean manufacturing industries, Kasahara and Lapham (2013) found that both exporting and importing entail large start-up costs, but they estimated that a firm can save between 7 and 26 percent of fixed costs and sunk costs associated with trade by simultaneously engaging in both export and import activities.

(17)

17 2.4.4. International Certification and Learning Effects

Prete, Giovanetti, and Marvasi (2016) argued that internationally-recognized certification, such as ISO 9000 and 14000, is an essential requirement to enter international supply chains, as they can guarantee and signal the ability of the firm to meet the international standards. Such certifications are even more relevant for firms that active in developing countries. In their study, they found that participating in GVC as certified two-way traders can enhance productivity by 40%-50%, on average, which is much higher than the average productivity effects found in the existing studies. For example, Baldwin and Yan (2016) found that participating in GVC as two-way traders (without certification) can “only” enhance productivity by 5-15%, on average.

Some international business scholars confirm PGM finding. Starke et al. (2012) found that ISO 9000 certification is associated with an increase in sales revenues, decrease in cost of goods sold and increase in the asset turnover ratios of certified companies in Brazil. They argued that international certification can reduce the asymmetric information between the buyers and the sellers since it can guarantee the quality of products and hence firms can charge a premium price. They also argued that the main reason to adopt international certification is to improve the efficiency of the production processes, which in turn can generate competitive advantages and more profits. Cost-saving is another reason to adopt international standard. In this regards, once firms adopt the international standard, the cost associated with failure, appraisal and prevention costs may decrease (Starke et al., 2012). Finally, we argue that GVC-firms that satisfy international standard may have a closer trading relationship with advanced economies. As a result, these firms have more opportunity to learn modern technology and business process both from the upstream and downstream of the supply chain. More importantly, international certification may also indicate that such firms have a set of qualities that can absorb technological transfer more easily so that the learning effects can be generated more effectively and faster.

(18)

18 as they are audited regularly (Russo, 2009). Based on those discussions, we argue that once the two-way traders have internationally-recognized quality certifications, they will benefit even more from participating in global production networks. Taking all together, the following hypothesis emanates from the previous discussion:

Hypothesis 2. Firms’ participation in GVCs will improve the average productivity and energy

efficiency of the participated firms, and such improvements will even be stronger if the participated firms own internationally-recognized certification.

2.5. GVC Participation and Pollution Haven Hypothesis

While most scholars agree that international activities of firms can enhance economic performance (i.e. productivity), the effects of international activities on firm’s environmental performance have not been reaching a common ground. A cornerstone of the debate on the globalization-environment nexus is the pollution haven hypothesis (PHH). This hypothesis states that globalization may allow pollution-intensive industries to shift their production facilities to countries with weaker environmental regulations, resulting in a race to the bottom in overall environmental standards and increased pollution levels (Pethig, 1976; McGuire, 1982). In the context of GVC, PHH might imply that in addition to the motive of seeking the cheapest production factors, industries from advanced countries tend to relocate (by offshoring or outsourcing) their polluting activities in developing countries that have lower environmental standard and weaker enforcement. By doing so, these manufacturers can avoid expensive costs associated with satisfying the environmental requirements in their home countries. For the host countries, however, job creation and wages rate enhancement have been built at the expense of the environment. Thus it is predicted that the reduction of trade barriers may also imply an expansion of ‘dirty’ production activities because it makes countries with low emission standards to become pollution havens (Forslid, Okubo and Ulltveit-Moe, 2015).

(19)

19 location choice for equity joint ventures (EJVs) in China, Dean, Lovely and Wang (2009) found thathighly-polluting industries are attracted by weak environmental standards. On the contrary, He (2006) argued that the rise in environmental regulation stringency only has a modest deterrent effect on FDI inflow to China because FDI enterprises generally produce with high pollution efficiency. This result is confirmed by Eskeland and Harrison (2002) who found US-owned plants are significantly more energy efficient and use cleaner types of energy, and hence benefiting from “pollution havens” is not the intention of US multinationals.

In SEA context, the study regarding PHH is quite a few. Nonetheless, Mukhopadhyay (2006) found that Thailand was indeed a pollution haven country. The more comprehensive findings by Merican et al. (2007) suggest that FDI adds to pollution in Malaysia, Thailand, and the Philippines but not in Indonesia where FDI is inversely related to pollution, and Singapore where it is proved insignificant. Those all opposing results possibly occur because it is difficult to measure the cost of production associated with environmental regulations in a given jurisdiction relative to others, as Levinson and Taylor (2008) stated, “the compliance costs stemming from these regulations could come in the form of environmental taxes, regulatory delays, the threat (or execution) of lawsuits, product redesign, or emissions limits”.

The inconclusive results of previous studies are opening more opportunity to conduct research concerning PHH. Using a firm-level analysis, we may assume that the weak environmental regulation and enforcement in SEA regions has attracted foreign companies to shift their polluting activities to SEA countries, either through arm-length (outsourcing) or offshoring. Therefore, it can be expected that firms that involve in global production network may actually be those firms that are chosen or are established to perform pollution-intensive activities since the parent firms are not allowed (or too expensive) to do that in their home countries. In the context of energy usage, GVC firms may be those firms that use dirty sources of energy (e.g., coal) or use dirty technologies that are not energy-efficient. The following hypothesis has arisen from the previous discussion:

Hypothesis 3. Firms’ participation in GVCs will decrease the average environmental

(20)

20

III. DATA AND METHODOLOGY

3.1. Data Specification

The recent literature on international trade have stressed the increasing importance of micro-level analysis. Driven by this spirit, our empirical analysis will exploit data on the individual firm-level. We use the World Bank Enterprise Survey (WBES) database, which is an establishment-level survey in manufacturing and services sectors conducted by the World Bank in cooperation with local partners.5 Through interviews (telephone and/or face-to-face meetings) with firms in the manufacturing and services sectors, the purpose of this survey is to create business environment indicators that are comparable across countries. This survey provides information on the characteristics of firms across various dimensions, including infrastructure and services, sales and supplies, the degree of competition, capacity, land and permits, crime, finance, business-government relations, labor, the business environment, and performance. The stratified random sampling method was used to select the sample of firms.6 For empirical analysis, we analyze a balanced panel of 948 establishments in three South East Asia countries, namely Indonesia, the Philippines, and Vietnam, that active in the years 2009 and 2015.7 Ideally, one would analyze all ten members of Association of South East Asia Nation (ASEAN), but due to the data limitation, we are only able to analyze these three countries. For instance, the data for Thailand is only available for the year 2016. Other potential drawbacks from this data are the six years gap between the two periods. Thus, if the productivity and energy efficiency gains are diminishing over time, we will not be able to observe such causality. On the other hand, if such diminishing is not robust, we will find a high economic magnitude as the gains may already be accumulated.

To prepare the data prior to analysis, the data need to be cleaned. For instance, some negative values (that logically should be positive) of the variable interests, such as energy usage, export share, import share, employment, sales and capital stock, are treated as missing values. Since we are only using two periods of years with a significant gap, it is not possible to replace the missing values with the values from the subsequent years. One may also think that those

5_{Typically, previous studies use Annual Manufacturing Survey dataset to analyze the behavior of international} trade on firm-level. However, due to difficulty to find the data (and if available, it is very expensive), we decide to use WBES database. Furthermore, it should be noted that most enterprises in the survey are single-plant firms. It means that the term establishment, plant, enterprise and firm refer to the same meaning, and thus sometimes would be used interchangeably.

6_{A stratified random sample is one obtained by separating the population elements into non-overlapping groups,} called strata, and then selecting a simple random sample from each stratum (World Bank, 2009). For further detail on survey method see: http://www.enterprisesurveys.org

(21)

21 observations with missing values should be dropped. However, it might not be wise to do that in our baseline estimation because it will cut down a half of observations. Furthermore, we also remove some outliers that we considered illogical, for example, the data on energy cost per total cost should not exceed 100%.

Among other variables, for each year and firm, the survey includes most important characteristics of establishment that will be used to conduct the empirical analysis. Those are export share, import share, total factor productivity, total sales, total employment, energy cost, foreign ownership, capital stock, labor cost, the cost of raw material and cost of intermediate inputs. Moreover, industries covered by this survey are classified by 4-digit of ISIC rev. 3.1. As we utilize data for three different countries, we need to express the variables with local currency unit into the same common currency unit (i.e. USD). In doing so, we divide the local currency unit with the respective exchange rate to USD. Next, we deflated this nominal values into the same constant value (i.e. 2009 price) by using respective GDP deflator. Both of the data for exchange rate and GDP deflator are collected from the World Development Indicator database, the World Bank. Finally, as suggested by the World Bank (2017), it is better to not treat each establishment equally for regression purpose. Instead, one is suggested to use the certain sampling weight provided in the survey to weight the observations. In this case, we use median-assumption for the sampling weight.8

As we want to observe the economic and environmental impact of GVC participation, we need to use several dependent variables to capture such effects. In general, we distinguish two types of productivity. The first one is the firm’s production efficiency, measured by total factor productivity, which is basically the output residual after we take into account all inputs, such as capital, labor, energy, and raw materials. The data for revenue-based TFP is already estimated by the World Bank (2017), although we also estimate another measure of TFP which will be used interchangeably in the estimation.9_{The World Bank’s (2017) method for} estimating revenue-based TFP is explained in Appendix A.

Following Prete, Giovannetti, and Marvasi (2016) we used labor productivity as the second indicator of firms productivity. This is measured by sales, value-added, and profit that are generated by each worker. Once firms become more productive, we expect that such firms can sell more products, gain more profits, and added more value to their products, given the same or less number of worker used. Additionally, we also use average wages per worker as the

8_{Weight with median assumption is low if the eligible establishments are those that rejected the screener} questionnaire, and answering telephone or fax was the only response (World Bank, 2009).

(22)

22 measure of labor productivity since microeconomic theory suggests that the wage a worker earns equals to the amount of output an incremental worker would produce (marginal product of labor), cateris paribus. Besides, average wages can also indicate the skill-intensity within a firm (Imbruno and Ketterer, 2016), where higher wages usually associated with high-skill workers.

Unlike the productivity indicators, the direct measurement of environmental performance (e.g. pollution intensity) are not straightforward because there are no data available on pollution at the firm level. Some scholars such as Eskeland and Harrison (2003), Cole et al. (2008), and Batrakova and Davies (2012) used energy efficiency as the proxy of firm’s environmental performances. In this regards, the first variable that we use to capture the degree of energy efficiency is energy consumption per sales (i.e. energy intensity). Another possible indicator for energy efficiency is the cost of energy shares, which is the total cost of energy divided by total cost of production (Roy and Yasar, 2015). In the WBES survey, energy cost includes firm’s purchases of fuels and electricity.

Furthermore, GVC-status is our main variable of interest. Following Baldwin and Yan (2016), we treat a firm as global value chain participant if such firm is both import intermediate (or raw) inputs from abroad and export intermediate (or final) products to foreign countries (two-way traders).10 Thus, this variable is equal to 1 if a firm export and import in year t and is 0 otherwise. With regards to certified two-ways traders’ definition, we treat a firm as a GVC participant, or taking value of 1, if such firm is a two-way traders and have internationally-recognized certification, and taking value of 0 otherwise.11

To control for other aspects of firms’ characteristics, we include firm’s capital intensity, with the idea that firms using a high amount of capital may require generate higher productivity and require more energy. Moreover, due to the nature of low value-added task of GVCs in developing countries, such capital-intensive firms may be less likely to participate in GVCs. The capital intensity is measured by the total cost of machinery, vehicles, and equipment (World Bank, 2017), divided by a total number of employment. Also, earlier studies suggest that foreign ownership is associated with higher productivity and energy efficiency because the parent companies may provide the subsidiary with better technology (Eskeland and Harrison, 2003; Batrakova and Davies, 2012). Thus, we include an ownership variable equal to one if a

10_{A firm is called exporter if its export per sales (export share) is more than 10%, similarly a firm is called importer} if its imported input over total input (import share) is more than 10%.

(23)

23 firm is foreign owned (with more than 51% of shares) and 0 otherwise. We also include the size of a firm (measured as a total number of employees) to control the labor intensity of the firm, since global value chain activities in developing countries are often associated with labor-intensive activities. Moreover, firms with larger size is often associated with more profitable, more productive and more energy-efficient (Bernard et al., 2007; Batrakova and Davies, 2012). Finally, we include 2-digit industry classification dummies, year dummies, and country dummies in order to control for industry (e.g. chemical industry is widely known to be pollution intensive), year (e.g. energy price shock) and country-specific variation (e.g. environmental regulation). Table B.1, B.2, B.3 in Appendix B displays a list of variables used, industry classification and the summary statistics, respectively.

3.2. Empirical Strategy

In the preliminary analysis, we would like to know whether there is heterogeneity in firm’s characteristics of GVC-firms and those of non-GVC firms. To do so, we will compare the mean differences between those two groups of firms by using a simple t-test.

(24)

24 that always participate in GVC during the period (Always GVC). However, the latter two are not the focus of our analysis.

Second, Blundell and Dias (2000) argued that when a randomized experimental design cannot be done, matching is a way of re-creating the similar circumstances. In an ideal experimental setting, outcomes for firms that change GVC status would be compared with outcomes they would have experienced had they not changed their GVC status or “counterfactual” (Prete, Giovannetti and Marvasi, 2016). Since the latter is unobservable, Rosenbaum and Rubin (1983) suggest that we can use propensity-score matching to create such counterfactual from the control group.

In a nutshell, matching techniques utilize the non-experimental data by assuming that selection into treatment, in our case GVC participation, is fully determined by observed firms’ characteristics (i.e. conditional independence assumption or CIA). Thus we can use these observed characteristics to match the observations, resulting in two similar groups, one is the treated group and the other one is the “matched” control group (or counterfactual). Notice that given the similar characteristics of firms in the treated and the “matched” control group, the selection to GVC status (i.e. receiving treatment) is now assumed to be random rather than driven by certain firms’ characteristics. In other words, this matching technique may mitigate the firm self-selection bias into GVC.

Conditioning on too many covariates, however, can cause a dimensionality problem, and hence matching observations will become very difficult (Batrakova and Davies, 2012). Rosenbaum and Rubin (1983) proposed a solution to this issue, which is to calculate propensity score that measures the probability of receiving a treatment given the observed firm’s characteristics. By doing so, we can match the treated and the control group on single value of probability rather than the whole range of covariates because the propensity score already summarized the covariates information.

In this step, we select a number of observable variables that predict the probability of participating in GVC, which later will be used to estimate the “propensity scores”. Similar to previous studies, we utilize probit estimation that predicts the probability of entering GVC (GVC-entry) in year t, conditional to the firm’s characteristics at year t-1:

𝑃𝑟(𝐺𝑉𝐶_𝑖𝑡 = 1|𝑋_𝑖𝑡−1) = Φ(𝛾_𝑐 + 𝛾_𝑠+ 𝛾_𝑡+ 𝛽𝑋_𝑖𝑡−1) (2)

(25)

25 intensity, productivity, employment, capital intensity, wages and foreign ownership. Additionally, in order to take into account all possible differences in institutional, industrials and time characteristics, we control for country (𝛾𝑐), industry classification by technology (𝛾𝑠) and year fixed effects (𝛾_𝑡). Before doing so, however, we conduct correlation analysis to check a possible multicollinearity among variables that will be used in this estimation. It can be seen that according to Table B.4 in Appendix B, all of the correlation values are less than 0.50 which implies the multicollinearity problem are less likely to occur.

Third, we matched all firm i that become GVC (treated group) with a similar non-GVC firm j based on their calculated propensity scores (“matched” control group). Due to its nature of continues variable, it is impossible, however, to match the exact propensity score. We then use Nearest Neighbor Matching (NNM) approach to match the observations.12 In essence, NNM chooses a match for each observation if a minimum distance in the propensity scores is smaller than a pre-specified value (i.e. caliper). To do so, we also employ a common support technique to exclude those firms for whom a match could not be found because the propensity scores are too far apart from the other.

When matching is performed, a careful consideration needs to be established between the CIA and the common support. That is because, selecting a large number of covariates might cause difficulty in finding the common support, while selecting a minimal number of the explanatory variable may violate CIA assumption. Hence, we ensure the validity of the matching exercise by examining whether the pre-entry firm characteristics of the treated and the control group are similar. In doing so, we conduct a balancing property test, which is a simple significance test (i.e. t-test) that compare the statistical differences between the treated and the “matched” control group in their observed characteristics before the treatment take place (i.e. year t-1). Finally, after we successfully match the observation, the appropriate average treatment effects can be observed. Previous studies use Difference-in-Difference (DiD) estimation to estimate the average effects of the treatment (i.e. changing GVC status) in the treated group.13 Essentially, this approach is comparing the average outcomes of productivity and energy intensity of the treated and the control group, before and after the firms participate in GVC. The DiD estimators are thus as follows:

𝐸(𝑌_1𝑡− 𝑌_0𝑡|𝑋, 𝐷 = 𝐺𝑉𝐶) − 𝐸(𝑌1𝑡− 𝑌0𝑡|𝑋, 𝐷 = 𝑁𝑜𝑛𝐺𝑉𝐶) = 𝛿 (3)

12_{We will also use other matching method, namely Kernel matching method for robustness check.}

(26)

26 where 𝑌0𝑡 represents the outcome variables one period before a firm entering GVC, and 𝑌_1𝑡 represents the outcomes variables after a firm entering GVC. As is mentioned before, the outcome variables here are productivity and energy efficiency. The differences between these two periods eventually generate within observations’ differences (∆𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑣𝑖𝑡𝑦_𝑖 and ∆𝐸𝑛𝑒𝑟𝑔𝑦𝑖). Next, the differences between observations (i.e. treated vs control group) is calculated in order to estimate the average causal or treatment effects, or difference-in-differences estimator (𝛿). Figure 3.1 illustrates how DiD technique is applied.

Figure 3.1 Illustration of DiD estimation

Source: Hill, Griffiths, and Lim (2011)

To sum up, the procedures to observe the treatment effects are: (i) Deciding who will be in the treated group and the control group, (ii) Estimating the probability of receiving treatment (i.e. changing GVC status) based on observable covariates and use that to calculate propensity score, (iii) Creating counterfactual by matching firms in the treated and the control group with the most similar characteristics reflected in their propensity scores, and (iv) Using DiD approach to examine the average treatment effects of GVC participation on productivity and energy efficiency.

(27)

27

IV. EMPIRICAL RESULTS

4.1. Preliminary Analysis

In this section, we present some preliminary analysis that will help us to better understand the dataset before we can examine the causal effects in the next section. First of all, we provide distribution of firm based on their trading status in Table 4.1. We can see that around a half (45%) of SEA’s firms were international traders, either an importer, an importer or both (GVC). It can be seen that importing is the most frequent international activity of firms in this region (20.25%). Interestingly, based on our definition of GVC firms, the share of two-way traders (16.61%) is larger than that of exporter-only (8.7%). The majority of firms (55%), however, are still serving the domestic market.14

Table 4.1. Firms’ participation in international trade, 2009-2015 Trading Status Frequency (No of

observations) Percent (%) Domestic-only 1,032 54.43 Exporters-only 165 8.7 Importers-only 384 20.25 Two-way traders (GVC) 315 16.61 Total 1,896 100

In Figure 4.1, we show that on average the GVC firms are exporting 60% of their sales and importing 60% of their material inputs. With regards to the ownership of GVC firms, on average, foreigners own 40% of the shares. All of these figures are considerably higher than those of non-GVC firms. Additionally, we also show the fact that while export shares are dominated by low-tech GVC-firms (70%), most of import shares (also 70%) are contributed by high-tech GVC-firms.15 These simple facts suggest that GVC-firms are indeed playing significant roles in driving the economy of SEA countries.

14_{Exporters are defined as firms that make more than 10% of their total sales abroad (export share>10%), while} importers are defined as firms that purchase more than 10% of their material inputs and supplies abroad (import share>10%).

(28)

28 Figure 4.1. International activities of GVC Firms, 2009-2015

Furthermore, in Figure 4.2 we show that GVC firms also differ from non-GVC firms in other characteristics besides their trading activities. It can be seen that in terms of economic performance, GVC-firms are clearly superior in compare to non-GVC firms. For instance, while the average sales per worker of GVC-firms is $200,000 per year, those of non-GVC firms is only a half of it (i.e. around $100,000). The energy performance of GVC and non-GVC firms, however, are almost indistinguishable.

Figure 4.2. GVC firms’ performances, 2009-2015

To complement previous graphical analysis, we conduct a simple t-test that shows the mean differences in some firms’ characteristics among the trading firms. In Table 4.2, we show that GVC firms generally have better economic performance than all non-GVC firms (i.e. exporters, importers or domestic firms). GVC firms are significantly more productive, more capital intensive, employing more people, and paying a higher wage than all non-GVC firms.

0 20 40 60 % non-gvc gvc

export share (%) import share (%) foreign ownership (%) 0 20 40 60 80 %

low-tech medium-low-tech medium-high-tech high-tech export share(%) import share(%)

0 2 4 6 8 % non-gvc gvc

total energy intensity (%) electricity intensity (%) fuel intensity (%) 0 .0 5 .1 .1 5 .2 mil li o n U SD non-gvc gvc

(29)

29 However, we cannot find any evidence that GVC firms have significantly lower energy intensity than non-GVC firms, which corroborates the previous graphical analysis. These findings also not hold if we compare the performance of GVC firms with pure exporters and importers. It shows that GVC firms neither have superior economic nor energy performances in compare to pure exporters or importers. The only characteristics of GVC-firms that remain significantly different from exporter and importer are employment and foreign ownership. This is possible because GVC-firms are often characterized by its involvement in multinational enterprise’s (MNE) production networks and labor-intensive activities. More importantly, these results are confirming the hypothesis that firms are indeed heterogeneous in terms of their characteristics (Melitz, 2003; Antras and Helpman, 2004).

Table 4.2. Average differences in characteristics of GVC and non-GVC firms

Firm Characteristics GVC firms versus All non-GVC firms Exporters Importers Export share 54.967*** 1.987 Import share 45.769*** 3.686 TFP 0.212** -0.007 -0.017

Sales per worker 0.878* 0.027 0.027

Profit per worker 0.058* _0.096 _0.005

VA per worker 0.05 0.059 0.004 Energy intensity -0.253 0.946 0.415 ln(Employment) 1.782*** _0.855*** _1.223*** ln(Capital Intensity) 0.445*** 0.447 0.452* ln(Wage) 0.387*** 0.172 0.041 Foreign Ownership 35.319*** _27.038*** _29.379*** Note: s*** p < 0.01, ** p < 0.05, * p < 0.1

4.2. GVC Participation and Firms’ Performances

(30)

30 As is mentioned in section 3, we first specify two groups of firms: (i) GVC-entrants (treated group) is a firm that did not join GVC in year t-1 but enter global production networks afterward, while (ii) never-GVC is firms that never join global production network (control group). Between our periods of analysis (2009-2015), we found that 45 firms enter into global production network (treated), 634 firms never join GVC (control), 75 firms stop to join GVC (quitters), and 91 firms always participate in GVC (always) (Table 4.3).

Table 4.3. Distribution of treated and control group, 2009-2015 GVC Groups Frequency (No of firms) Percent (%)

Never (control) 634 75.03

Enter (treated) 45 5.33

Quitters 75 8.88

Always 91 10.77

Total 84516 _100%

To test self-selection hypothesis, we run a probit estimation to estimate the probability of a firm joining GVC in year t based on its characteristics in t - 1, as illustrated in equation (2). In Table 4.4, we do not find any evidence that more productive and energy-efficient firms self-select to join GVC. However, we do find that foreign-owned firms are more likely to participate in GVC. It might be because foreign-owned firms can relatively easy to buy inputs and sell products through their multinational network (Imbruno and Ketterer, 2016). We also find that firms that employ more people have a higher probability of joining GVC. There are at least two explanations for this. First, if the number of workers represent the size of firms, it might be the case that only these big firms that can afford fixed cost of entering GVC. Second, as we have discussed in section 2, firms in developing countries mostly participate in low value-added tasks, which is often characterized by labor-intensive activities. That is probably why, firms with more workers, on average, have a higher probability to enter GVC.

(31)

31 Table 4.4. Selection into GVC as Two-Ways Traders

Dependent variable is taking value 1 if a firm enter into GVC, and 0 otherwise. Independent variables are covariates in year t-1. Standard errors in parenthesis. The regression include industry and country dummies. Inference: *** p < 0.01, ** p < 0.05, * p < 0.1.

As we found that initial level of productivity and energy efficiency do not affect firms’ decision to enter GVC, the self-selection hypothesis (hypothesis 1) cannot be supported. Although this is consistent with the finding of Batrakova and Davies (2012), it contradicts with the common findings in international trade literature (e.g. Bernard and Jensen, 1999) that typically conclude that more productive firms self-select to involve in international trade activities. One possible argument to explain these contradictive results is the underlying motives of firms in South East Asia countries to join GVC. Instead of the more productive firm have the initiative to self-select into GVC, it can be the case that firms in developing countries are joining GVC because they are chosen as a partner by big firms in advanced countries (through offshoring or outsourcing), regardless their initial productivity and energy-efficiency.

Next, we utilize the same probit estimation to estimate the firm-level propensity scores, which then is used to match the observations in the treated and control group. That is being done because, in order to observe the actual effects of entering GVC on firm’s performance, we need to compare the productivity and energy intensity of treated and “matched” control group (i.e. counterfactual), after and before they join GVC.

Dependent variable Two-Ways Entryt

(32)

32 As is mentioned in section 3, we use Nearest Neighbor Matching (NNM) approach to match the treated and the untreated group. Afterwards, we conduct the balancing property test to evaluate the quality of this matching. We confirm that the “matched” control group has been identified with similar pre-treatment (i.e. pre-GVC) firm characteristics with the treated group, showed by the insignificant mean differences of the pre-treatment variables between two groups (See Table B.5 in Appendix B).

Finally, Table 4.5 shows the effects of entering GVC on firm’s performance. We found that almost all of the coefficients of treatment effects (i.e., the effects of joining GVC) are significant and are having the expected sign, which hence support our second hypothesis (hypothesis 2).17_{That is, firms’ participation in GVC as two-way traders (export and import)} can enhance the average productivity and energy efficiencies of the participated firms, implying the benefits of a global production network participation on the economy and the environment. These findings also confirm that the pollution haven hypothesis (hypothesis 3) cannot be supported.

Table 4.5. The effects of entering GVC as Two-Ways Traders

Productivities Energy Efficiencies

Ln(TFP) Sales (million) Profit (million) VA (million) Ln(Wage) Energy intensity (%) Energy cost (%) DiD 0.247* _0.665*** _0.355*** _0.281*** _0.263 _-3.482** _-5.425*** (0.144) (0.191) (0.104) (0.101) (0.181) (1.383) (1.781) N 434 585 585 544 576 558 559

All regressions include industry and country dummies. Matching is conducted by NNM method. N is the number of matched pairs. DiD is the coefficient of treatment effects. Standard error in parenthesis. Inference: *** p < 0.01, ** p < 0.05, * p < 0.1. The treatment is taking the value of 1 if the firm was not two-ways traders in year t-1 but switch into two-ways traders in year t (Two-ways entry). TFP, Sales, Profit, VA, Wage, Energy intensity, and Energy cost are log of total factor productivity, sales per worker (mil.USD), profit per worker (mil.USD), value-added per worker (mil.USD), log of wage per worker, energy per sales (%) and energy cost per total cost (%), respectively.

With regards to the magnitude of the learning effects, our results show that after joining GVC, the participated firms are experiencing productivity gain by 25%, on average, indicating the technological transfers from GVC activities. This learning effects on productivity are comparable with existing findings. For instance, De Loecker (2007) found that exporting significantly causes productivity gains by around 8-20% depending on to which country the exporting firms is exporting to. Imbruno and Ketterer (2016) found that firms that import intermediate inputs have increased their productivity by 23%, on average. With regards to GVC

(33)

33 activity, Prete, Giovannetti, and Marvasi (2016) found that being involved in GVC significantly increase the participated firms’ productivities by 30%, on average.

Besides its impact on firm’s productivity, we also show that firms that join GVC are experiencing some improvements in their labor productivities. For instance, after joining GVC, these firms can increase their sales, profit, and value-added by $665000, $335000, and $281000 per worker per year, respectively. The increasing in sales per workers may confirm the argument that in addition to technological effects captured in TFP, firms that join GVC can also exploit scale economies as they now exposed to a larger market.

Furthermore, we also show that firms can decrease their energy needed per sales (energy intensity) by 3.48% after they join GVCs. This is confirmed by the energy cost shares’ coefficient, which shows that after firms join GVCs, their energy cost per total cost is decreasing by 5.45%. Although this is a good sign, in compare to the existing studies, however, these magnitude effects are considerable smaller. For instance, Batrakova and Davies (2012) found that exporting can increase the energy efficiency of Irish firms by 25%, whereas Imbruno and Ketterer (2016) found that importing intermediate inputs can decrease Indonesian firms’ energy intensity by 11%. Those scholars, however, also found that the learning effects of exporting and importing on energy intensity are no longer present after firms active in international trade for three years. This diminishing effects of learning is one possible explanation why we found such smaller magnitude. As the year gap in our dataset is quite large (i.e. 6 years), it might be the case that firms already joined GVC for several years (i.e. 1-5 years) by the time we estimate their outcomes (i.e. year 2015), so that the average treatment effects of GVC participation on energy efficiency might already be diminished.

4.2.1. Does International Certification Matters?

Prete, Giovanetti, and Marvasi (2016) argued that internationally-recognized certification, such as ISO 9000 and 14000, is an essential requirement to enter international supply chains, as they can guarantee and signal the ability of the firm to meet the international standards. This is even more relevant as we analyze firms in developing countries. By adopting international standards, we expect that a firm can have more opportunity to learn from advanced economies and can absorb technological transfer more effectively. In other words, we believe that the learning effects of GVC participation are partly driven by whether or not the participated firms own international certification.

(34)

34 international certification in year t, but not in year t-1 (the treated group). The control group is then consisting of a firm that never satisfies any of those conditions in both years (i.e. never be a certified two-ways traders).

We observe that there are 181 two-ways traders that have international certification, of which 30 of them enter into GVC according to previous definition (treated group), while 699 firms were never involved in GVC based on this definition (control group). Same as before, we conduct probit estimation to test the self-selection hypothesis and to estimate the propensity score. Consistent with the previous findings, we also find that self-selection hypothesis (hypothesis 1) cannot be supported, as the productivity and energy efficiency coefficients do not significantly affect the decision of a firm to become certified two-ways traders. In this case, we only find that foreign owned, larger and more capital intensive firms are more likely to become certified two-way traders (See Table B.6 in Appendix B).

Table 4.6 shows the learning effects of participating in GVC as certified two-ways traders. It can be seen that the treatment effects are strongly significant and giving the expected signs for both productivities and energy performances, which again corroborates our second hypothesis (hypothesis 2). More importantly, it shows that the effects of joining GVC as certified two-ways traders are giving stronger effects in compare to the effects of simple two-two-ways traders in Table 4.5. For instance, the effects of being certified two-way traders on TFP is 35%, which is 10% higher than that in the previous estimation. This magnitude is consistent with Prete, Giovannetti, and Marvasi (2016) who found that joining GVC as certified two-ways trades can boost productivity by 30%-50%.

In a similar vein, sales, profit and value-added are also increased by a considerable amount, i.e. $882000, $535000, and $609000 per worker per year, which is around $200000 higher than that of simple two-way traders’ estimation. Notice that, by participating as a certified two-ways traders, the average wages can increase by 55%, while we showed previously that simple GVC participation (without certification) do not have any causal effects on wages. That is probably because as GVC-firms start to have international certification, they need to employ more high-skilled workers, which then increase the average wages within these firms.

(35)

35 Table 4.6. The effects of entering GVC as Certified Two-Ways Traders

Productivities Energy Efficiencies

Ln(TFP) Sales (million) Profit (million) VA (million) Ln(Wage) Energy intensity (%) Energy cost (%) DiD 0.351*** 0.882*** 0.535*** 0.609*** 0.551*** -12.061*** -11.386*** (0.124) (0.168) (0.101) (0.108) (0.184) (1.525) (1.824) N 544 658 658 611 646 624 625

All regressions include industry and country dummies. Matching is conducted by NNM method. N is the number of matched pairs. DiD is the coefficient of treatment effects. Standard error in parenthesis. Inference: *** p < 0.01, ** p < 0.05, * p < 0.1. The treatment is taking the value of 1 if the firm was not certified two-ways traders in year t-1 but switch into certified two-ways traders in year t (Certified two-ways entry). TFP, Sales, Profit, VA, Wage, Energy intensity, and Energy cost are log of total factor productivity, sales per worker (mil.USD), profit per worker (mil.USD), value-added per worker (mil.USD), log of wage per worker, energy per sales (%) and energy cost per total cost (%), respectively.

Those high productivity and energy efficiency premium entail the importance of international standardization for GVC-firms. It is widely believed that adopting international standards can improve firm’s efficiency and competitiveness. Once it is combined with international trade activities, a firm will benefit even more because it can help the firm to connect with buyers and suppliers from advanced countries which increase the learning opportunity of the firm. Moreover, as a company with international certifications need to be audited periodically, they are likely to maintain their existing standards (Russo, 2009). As a result, the diminishing effects of learning may be alleviated and hence the accumulated benefits from international trade participation would be more persistent and higher. Those mechanisms may explain why joining GVC as certified two-ways traders are able to give much higher learning effects (i.e. magnitudes) in compare to simple two-way traders. To sum up, we confirm that international certification does matter in driving the learning effects, which again supports our second hypothesis.

GVC Participation, Productivity, and Energy Efficiency: A Firm-Level Evidence

MASTER THESIS