WPS7491 Policy Research Working Paper 7491 Domestic Value Added in Exports Theory and Firm Evidence from China Hiau Looi Kee Heiwai Tang Development Research Group Trade and International Integration Team November 2015 Policy Research Working Paper 7491 Abstract China has defied the declining trend in domestic content individual processing exporters caused China’s domestic in exports in many countries. This paper studies China’s content in exports to increase from 65 to 70 percent in rising domestic content in exports using firm- and customs 2000–2007. Such substitution was induced by the coun- transaction-level data. The approach embraces firm hetero- try’s trade and investment liberalization, which deepened its geneity and hence reduces aggregation bias. The study finds engagement in global value chains and led to a greater vari- that the substitution of domestic for imported materials by ety of domestic materials becoming available at lower prices. This paper is a product of the Trade and International Integration Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at hlkee@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Domestic Value Added in Exports: Theory and Firm Evidence from China Hiau Looi Keey Heiwai Tangz World Bank Johns Hopkins University Key Words: Firm heterogeneity, Domestic value added, Value added trade, China, Global value chains, Trade liberalization, FDI liberalization JEL Classi…cation Numbers: F10, F14 We thank Richard Baldwin, Andrew Bernard, Thibault Fally, Gordon Hanson, Russell Hillberry, David Hummels, Robert Johnson, Wolfgang Keller, Pravin Krishna, Aaditya Mattoo, Marcelo Olarreaga, Nina Pavcnik, James Riedel, Georg Schaur, Peter Schott, Zhi Wang, and Shang-Jin Wei for discussions and comments. We are also grateful for conference/seminar participants at the AEA, Barcelona GSE Summer Forum, Brandeis, CESIfo, Dartmouth, FREIT (UCSC), Georgetown, Graduate Institute - Geneva, GTDW, IMF/WB/WTO Workshop, Nottingham, Penn State-Tsinghua Conference, Colorado Boulder, USITC, and the World Bank for feedbacks. Research for this paper has in part been supported by the World Bank’ s Multidonor Trust Fund for Trade and Development and the Strategic Research Program on Economic De- velopment. The results and opinions present in this paper are our own, and do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. y Development Research Group, The World Bank, Washington, DC 20433, USA; Tel.: (202) 473-4155; Fax: (202)522-1159; e-mail: hlkee@worldbank.org. z Johns Hopkins University, School of Advanced International Studies, 1717 Massachusetts Ave NW, Washington, DC 20036, USA. Tel.: (202) 663-5679; email: hwtang@jhu.edu. “Production processes are more and more fragmented ... The nature of trade has changed, but our trade data have not ... Many goods are assembled in China, but their commercial value comes from the numerous countries ... We want to know the value added by each country in the production process of …nal goods.” –Pascal Lamy, Director-General of WTO, “Made in the World”Initiative, 2011 1 Introduction Over the past two decades, increasing global production fragmentation has allowed exporting …rms to rely less on domestic inputs for production. Indeed, research …nds that domestic content in exports has been declining in most countries. China is an intriguing exception.1 What caused China to defy the declining trend in domestic content in exports in most countries, despite its deep engagement in global value chains? There are several possible answers to this question with con‡icting implications. The rising domestic content could re‡ect the changing composition of Chinese exports, suggesting that China has shifted its comparative advantage towards the industries with high domestic content. It could also be a result of its increasing domestic production costs, which would imply that the country has become less competitive. Yet another possible answer is that it could be due to the gradual substitution of domestic for imported materials by its exporters. This would imply that China has become more competitive, particularly in the intermediate input sectors. s rising domestic content in exports can provide Understanding the determinants of China’ important development policy insights for other countries. This paper uses customs transaction-level data merged with …rm survey data to measure 1 Koopman, Wang and Wei (2012) …nd that China’ s DVAR rose between 2002 and 2007. Johnson and Noguera (2014), using the GTAP IO tables, show that from 1970 and 2009, the DVAR of all countries in their sample are declining, except for the Republic of Korea and Indonesia. 1 s rising domestic content in exports, or the ratio of domestic value added and analyze China’ in exports to gross exports (DVAR). Our transaction-level data cover the universe of Chi- nese exporters during the period of 2000-2007, allowing us to construct …rm, industry and aggregate DVARs over time to study their evolution. The recent burgeoning literature on measuring industry and aggregate DVARs relies on input-output (IO) tables. While using IO tables has the advantage of capturing IO linkages within and across countries, the presence of …rm heterogeneity may result in signi…cant aggregation biases in the estimates of the DVAR. Our ground-up approach embraces …rm heterogeneity by measuring industry and aggregate DVARs as the weighted averages of the underlying …rms’DVARs.2 This unique methodol- ogy further allows us to compute bootstrapped standard errors for our aggregate estimates, which are then used to perform statistical tests on the rising trend. Finally, we use the customs data merged with manufacturing …rm survey data to examine whether changes in s export composition, …rms’production costs and material shares are responsible for China’ rising DVAR. s DVAR has been rising, but are Our DVAR estimates con…rm existing studies that China’ s aggregate higher than previous estimates. Speci…cally, we …nd that the DVAR of China’ exports increased from 65% to 70% between 2000 and 2007, with similar magnitudes of s bilateral exports to its major trading partners. The increase increases in the country’ in the aggregate DVAR is statistically signi…cant and con…rms the upward trend found in Koopman, Wang, and Wei (2012) (KWW12 hereafter), which adopts an IO table-based approach. However, our DVAR estimate for processing exports is signi…cantly higher than those of KWW12. The …nding that our DVAR estimates are higher than those of KWW12 exempli…es 2 Ahmad et al. (2014) also allow …rm heterogeneity to a¤ect the estimates of a country’s aggregate DVAR. They use …rm-level data to generate indicators by exporter status, which are then used to re…ne IO-table based estimates of domestic value added in exports from Turkey. 2 that ignoring …rm heterogeneity may lead to downward aggregation bias in the IO table- based approach. Samples that are used to construct IO tables often consist mainly of large …rms.3 Given that large …rms tend to have a lower DVAR due to their high import-to-sales ratios, over-sampling large …rms in the construction of IO tables can lead to lower estimates of the aggregate DVAR.4 To illustrate this point, we conduct a decomposition exercise. In particular, we show how our DVAR estimate can be lowered to a level that is not statistically di¤erent from that of KWW12, just by using a sample that includes only those large …rms that satisfy the sample selection criteria behind the construction of the Chinese IO tables. This suggests that aggregation bias driven by …rm heterogeneity alone is su¢ cient to explain the wedge between our estimates. s aggregate DVAR? Our …rm-level regressions reveal What has caused the rise in China’ that it is mainly driven by individual processing exporters substituting domestic for imported materials, both in terms of volume and varieties. Other factors, such as rising production costs due to higher wages, changing composition of Chinese exports towards the high-DVAR industries, or churning of …rms with di¤erent DVARs, cannot explain the upward trend during the sample period. We also …nd that the substitution of domestic for imported materials was induced by s trade and FDI liberalization since the early 2000s. To guide our empirical the country’ analysis, we build a model featuring a translog cost function, which permits an estimation of the time-varying elasticity of substitution between domestic and foreign input varieties to s DVAR. We …nd that for China, study how various government policies may a¤ect a country’ increasing FDI and declining input tari¤s have led to a greater variety of domestic materials becoming available at lower prices during the sample period. For the entire processing 3 See the United Nations Handbook of Input-Output Table Compilation and Analysis (1999). 4 Recent research by Amiti, Itskhoki and Konings (2014) and Blaum, Lelarge and Peters (2014) shows that larger …rms tend to have a higher import-to-sales ratio. 3 sector and for most industries within that sector, imported and domestic materials are gross substitutes, with the estimated elasticity of substitution ranging between 1.9 and 6.6. These large elasticities explain why lower prices of domestic materials can result in such signi…cant increases in the DVAR at the …rm and thus the aggregate level in China. Despite its simplicity, our methodology can be applied widely. In its basic form, our methodology can be used directly to measure the DVAR of processing exporters that operate in many export-oriented industries deeply embedded in global value chains.5 Furthermore, with additional assumptions, our methodology can be applied to constructing the DVAR for countries that have little dependence on processing exports.6 Our paper is related to several strands of literature. It relates to the literature on measur- ing value added trade.7 In particular, our DVAR estimates complement the IO table-based estimates, by incorporating …rm heterogeneity and thus minimizing aggregation bias. This paper is also related to the literature on the e¤ects of trade and FDI liberalization on do- mestic product varieties.8 Our results con…rm existing …ndings that the reduction in input tari¤s and increased presence of FDI in downstream sectors could lead to an expansion of domestic product variety. Finally, our paper contributes to the literature on international production sharing and global value chains,9 as well as the studies on China’ s increasing 5 The industries of garment, shoes, and toys in Bangladesh, Cambodia, Dominican Republic, and Mauri- tius, as well as the electronics industry in Malaysia, Thailand, and Vietnam are some of the examples. 6 In research in progress, we apply our current methodology to contruct the DVAR for a wide range of countries, where we obtain matched importer-exporter customs transaction data from the World Bank’ s Exporter Dynamics Database (Cebeci, et al., 2012). Preliminary results, based on Bangladesh, Guatemala, Madagascar and Morocco, show an upward trend in the DVAR of these countries’aggregate exports. 7 This literature starts with Hummels, Ishii and Yi (2001) to use industry input-output (IO) tables to calculate the value added to exports ratios for many countries. Recent related work includes Antràs, Chor, Fally, and Hillberry (2012), Johnson and Noguera (2012 and 2014), Koopman, Wang and Wei (2012, 2014), Antràs and Chor (2013), De la Cruz, Koopman, Wang and Wei (2013), and Johnson (2014). 8 Goldberg et. at. (2008) studies the impact of trade liberalization of India on its export variety. Kee (2015) shows that the increased presence of FDI in the garment sector of Bangladesh caused a greater variety of domestic materials to become available which led to product scope expansion and productivity gains in domestic garment …rms. 9 See Feenstra (1998) for a review of the early literatre on foreign outsourcing. More recent work includes, among others, Baldwin (2012) which postulates how participating in a global supply chain should be viewed as a new strategy of industrialization; and Timmer et al. (2014) which summarizes the main …ndings in the 4 engagement in the global economy.10 Our results speak to both bodies of work by showing s rising DVAR is due to the substitution of domestic for imported materials. that China’ Such substitution indicates that the country is relying less on imports and becoming more competitive in intermediate input sectors. This suggests that China has been moving up the value chains, and thus may have signi…cant implications for world trade and the global economy, given its sheer size.11 The paper proceeds as follows. Section 2 de…nes our measures of …rm DVAR. Section 3 shows how we use …rms’DVAR to compute industry and aggregate DVARs, and analyze their patterns. We also discuss the associated aggregation biases in the standard IO table- based approach, extend our methodology to include the non-processing sector, and calculate s aggregate exports in this section. Section 4 presents the pattern of the DVAR of China’ …rm DVAR. Section 5 develops a simple model to theoretically and quantitatively study the determinants of …rm DVAR. Section 6 concludes. In the Appendix, we describe our data sets and the construction of the main variables, such as the number of upstream varieties, import varieties, and industry exchange rates. A theoretical model that features a Cobb-Douglas production function is also presented there. literature on global value chains. 10 Johnson and Noguera (2012) show that the US-China trade imbalance in 2004 is 30-40 percent smaller when trade is measured in value added. Autor, Dorn, and Hanson (2013) show that increasing Chinese imports cause signi…cantly suppressed job creation, lower wages, lower labor market participation, and higher unemployment in the U.S. Pierce and Schott (2015) …nd that U.S. industries with the larger decline in tari¤s against imports from China experienced the slower employment growth, lower job creation, and higher job destruction. 11 These …ndings are consistent with a recent paper by Constantinescu, Mattoo and Ruta (2015) who suggest that China’ s structural transformation may be an important reason for the recent global trade slowdown, as China is relying less on foreign materials, thanks to its increasingly competitive domestic intermediate input industries. 5 2 De…ning Firm-Level Domestic Value Added We use two micro data sets in this paper: Chinese customs transaction-level trade data from 2000 to 2007, and the Annual Surveys of Industrial Firms from the National Bureau of Statistics of China over the same period. Readers are referred to the appendix for details.12 For the ease of exposition, we …rst focus on processing exporters, which are required by law to sell all their outputs abroad and may import materials free of duties.13 In Section 3.3, we will extend our methodology to study non-processing exports and thus aggregate exports of China. Let us …rst de…ne the main variable of interest – domestic value added in exports (DVA), s total revenue. A …rm’ starting from the accounting identity of a …rm’ s (i) total revenue (P Yi ), by de…nition, consists of the following components: pro…ts, ( i ) ; wages (wLi ) ; cost of capital (rKi ) ; cost of domestic materials P D MiD , and cost of imported materials P I MiI : P Yi i + wLi + rKi + P D MiD + P I MiI : (1) Some domestic materials may embody foreign content, while some imported materials may embody domestic content. Let us denote the foreign content in domestic materials and F D domestic content in imported materials by i and i , respectively. Then P D MiD can be 12 We employ the procedures commonly used to organize these data. We remove trade intermediaries, identi…ed by the methods proposed by Ahn et al. (2011), in the customs data. We also remove import and export transactions with China itself. As pointed out by Liu (2013), China’ s re-imports from itself accounted for about 9% of its total imports. These abnormal trade ‡ ows could arise from tax and transport cost saving incentives. 13 China’s Customs regulates processing trade under several regimes, with pure assembly (PA) and import and assembly (IA) being the two main types. The main di¤erence between these two regimes lies in the allocation of control rights of the imported inputs. In the PA regime, a foreign …rm supplies components to a Chinese assembly plant and retains ownership and control over the imported inputs throughout the production process. In the IA regime, a Chinese assembly plant imports components of its own accord and retains control over their use. Readers are referred to Feenstra and Hanson (2005) for a more detailed description of the two regimes. While this distinction will not a¤ect our DVAR estimates, it may a¤ect the way one should model …rm sourcing decisions. See Fernandes and Tang (2012) which exploits these regulatory di¤erences to study the organizational form of o¤shoring. Later on we will report regression results separately for the two types of processing. 6 F D written as the sum of i and a part that constitutes purely domestic content, qi : Likewise, D P I MiI can be written as the sum of i and a part that constitutes purely foreign content, F qi : F D P D MiD i D + qi ; and P I MiI i F + qi : s gross domestic product, we de…ne the DVA of a …rm Similar to the concept of a country’ s output. In other as the total value of domestic goods and services embodied in the …rm’ s DVA equals the sum of its pro…ts, wages, rental costs of capital, and both words, a …rm’ direct or indirect domestic materials purchased:14 D D DV Ai i + wLi + rKi + qi + i : (2) For a processing …rm that exports all its output and imports some of its intermediate inputs and capital equipment, its export (EXPi ) equals its revenue, while its import (IM Pi ) equals K the costs of imported materials, P I MiI , and imported capital, i . Thus, (1) implies D F K EXPi = DV Ai + IM Pi i + i i ) (3) D F K DV Ai = (EXPi IM Pi ) + i i + i . Equation (3) shows that we may use EXPi s DVA IM Pi to measure a processing …rm’ D F K after adjusting for i ; i and i . For China, KWW12 and Wang, Wei, and Zhu (2014) D …nd that i is very close to 0 for processing exports.15 Moreover, in our current data set, 14 Note that while some …rms may have foreign capital in its ownership, the returns to this foreign capital as well as its pro…ts are still included in its DVA in exports. This is because the service of these capital is rendered within the country’ s borders. Moreover, a …rm’ s DVA contains domestic materials produced by other …rms and is therefore larger than its own value added by de…nition. 15 Based on the GTAP Multi-Country IO tables, Koopman, Wang, and Wei (2014) estimate that the domestic content embedded in imported materials accounted for 0.7% of China’ s ordinary exports in 2004, and essentially 0 for its processing exports. Using the IO tables from the World Input-Output Database (WIOD), Wang, Wei and Zhu (2014) update these estimates and show that the domestic content embedded 7 processing …rms’imports of capital are recorded separately from material imports, implying K i = 0.16 Thus, the only necessary adjustment here is to remove foreign content in domestic F materials, i , which causes EXPi IM Pi to overestimate DV Ai in exports. From (3), …rm s ratio of domestic value added in exports to gross exports (DVAR) depends only on the i’ F share of imported materials in total revenue (P I MiI =P Yi ) with adjustments for i =EXPi : F DV Ai P I MiI i DV ARi =1 (4) EXPi P Yi EXPi F P M Mi P I MiI i = 1 M ; (5) P Yi P Mi EXPi where P M Mi = P D MiD + P I MiI : F Without …rm-level information on i =EXPi , we refer to KWW12 for the industry es- timates for 2007 and impute the estimates backward for each industry-year between 2000 and 2007, using the weighted average of the growth rates of the number of ordinary (non- processing) importers across upstream industries.17 These industry estimates range from 0.4 F to 5.7 percent, which we use to proxy for i =EXPi s DVAR.18 in (4) to construct a …rm’ D in imports used by Chinese exporters ( EXP i i ) increased from 0.1% in 1995 to 1.3% in 2007. They also show that these estimates can vary widely across sectors, ranging from 2.5% for the Chemical Products sector to 0.2% for the Leather and Footwear sector. Unfortunately, such estimates are not available separately for processing and ordinary exports. Nevertheless, given the low estimated domestic content in imported materials at the aggregate level, adjusting for it is unlikely to have a signi…cant e¤ect on both the levels and the trends of our aggregate DVAR estimates. Our approach therefore may underestimate the DVAR for sectors that use imported material with high domestic content. Given that our DVAR estimates for most sectors are already higher than the existing estimates based on IO tables, accounting for returned domestic content in imports will only strengthen our point that the existing estimates are subject to a downward aggregation bias. 16 The Chinese customs data record material and capital imports separately from a …rm’ s total imports, in a category called “Equipment for Processing Trade” (code number = 20). We thank a referee for pointing this out. 17 The rationale is that the net entry of ordinary importers, stimulated by China’ s continuous trade liberal- ization, may increase the supply of intermediate inputs that embody more foreign content. This assumption is grounded on the …ndings in Brandt et al. (2015), which shows that while the cost share of imports in total materials has been stable, the aggregate import share has increased substantially due to a large entry of new importers since China’ s accession to the WTO in late 2001. 18 Table A7 in the appendix reports the estimates of F i =EXPi by industry-year. Notice that our approach does not double count DVA as long as we exclude indirect trade between processing …rms and focus on mesauring DVA of the processing trade regime. We need additional assumptions to deal with the double- 8 Equation (5) shows that, once we control for the share of materials in total revenue P M Mi =P Yi , factors that do not a¤ect the share of imported materials in total materials s DVAR. This is an accounting identity, independent of the choice of will not a¤ect a …rm’ s DVAR, one should production functions. It highlights that in order to understand a …rm’ focus on the determinants of the share of imported materials in total materials. In Section 5, we will develop a simple but general model that features a translog cost function to formally study these determinants.19 3 From Firm DVAR to Industry and Aggregate DVAR Inferring the DVAR of an industry or aggregate exports from …rms’DVAR is straightforward. If …rms only engage in direct trading (i.e. do not import or export for other …rms) and only produce in one industry, then we can compute the DVAR of industry j as follows: X IM Pi i2 j X EXPi EXPi IM Pi X EXPi DV ARj = 1 X = X = X DV ARi ; (6) EXPi i2 EXPi EXPi i2 j EXPi j i2 j i2 j i2 j where j is the set of …rms in industry j . Industries are de…ned according to the industry classi…cation by the United Nations.20 By construction, the DVAR of industry j is a weighted average of the DVAR of all …rms in industry j with weights equal to the export shares of the …rms. Likewise, we can sum up all industry imports and exports …rst and then compute counting issue when we measure DVA for non-processing and aggregate exports. 19 To show that our main theoretical results are not speci…c to the functional form choice, we also solve for a model that features a Cobb-Douglas production function in the Appendix. 20 See http://unstats.un.org/unsd/tradekb/Knowledgebase/HS-Classi…cation-by-Section for the UN indus- try classi…cation. There are originally 20 sectors in the UN list. Sectors 1-3, which are agricultural sectors, are excluded since we cannot match most of the transactions to the manufacturing survey data. Sector 5 - Mining and Sector 19 - Arms and Ammunition are excluded for the same reason. Examples of a sector include Chemical Products (HS2 = 28-38), Textiles (HS2 = 50-63), Footwear and Headgear, etc. (HS2 = 64-67), and Machinery, Mechanical, Electrical Equipment (HS2 = 84-85). 9 the DVAR of aggregate exports as follows: XX IM Pi j i2 j XX EXPi DV AR = 1 XX = XX DV ARi : (7) EXPi j i2 j EXPi j i2 j j i2 j s DVAR, the aggregate DVAR constructed based on (7) is a weighted Similar to an industry’ average of the DVAR of all …rms, with weights re‡ecting the export shares of the …rms.21 While our ground-up approach is appropriate for inferring the aggregate DVAR, there are two caveats. The …rst caveat is about multi-industry exporters, for whom the allocation of imported materials (IM Pij ) to the production of output in di¤erent industries (EXPij ) s DVAR based on is generally unobservable in the data, making the inference of an industry’ (6) impossible. Thus, we only use the subset of single-industry exporters to infer industry DVARs.22 The second caveat relates to processing exporters importing indirectly through other …rms in China. Under the current customs regulations in China, processing …rms can legally sell imported materials to other …rms and bene…ted from tari¤ exemption, as long as the buyers are also registered processing …rms. Complicating this problem is that such transactions are not con…ned within the same industry or geographic location.23 The transactions of imported materials between two processing …rms in the domestic economy appear to be widespread according to our data. This practice of indirect importing certainly impacts the way we construct the …rm-level 21 In reporting the aggregate DVAR, we …rst aggregate …rm DVARs to the industry level. To make sure that the industry-level analysis, particularly the between-and-within analysis, is not driven by potential noises due to merging the customs data with the …rm data, we use industry weights based on the export value of single-industry exporters in the customs data set. 22 Nevertheless, since the construction of the …rm-level DVAR is not restricted by the multi-industry concerns, we will also include multi-industry exporters in the …rm-level regressions below. 23 See Regulations Concerning Customs Supervision and Control over the Inward Processing and Assembling Operation by China’ s Ministry of Commerce. For example, a shoe processing exporter may import leather and sell it to a handbag processing exporter. 10 and industry-level DVAR. In particular, for those …rms that import more than their needs, which we call excessive importers, using (4) may underestimate their DVARs and in the extreme case result in negative DVARs.24 On the other hand, for those …rms that buy imported materials from other processing …rms locally, which we call excessive exporters, using (4) may overestimate their DVARs, and in the extreme case bias the DVARs towards 1. To address the issue of indirect importing, we …rst use balance-sheet data to identify both the excessive importers and exporters. We de…ne excessive importers as those …rms that import more than their total material costs as recorded in the NBS Annual Survey of Industrial Firms (2000-2007), given that total material costs should equal to the sum of imported materials and domestic materials.25 These excessive importers import more than their total materials and are dropped from our sample. To identify excessive exporters, we …rst identify all registered ordinary (non- processing) exporters that only export in a single industry. Unlike processing exporters, s Customs to sell all outputs abroad. They can ordinary exporters are not required by China’ use imported materials to produce for both domestic and foreign sales. In addition, ordinary exporters need to pay import tari¤s and thus should have less incentive to import materials. The DVAR of ordinary exporters should be on average higher than that of processing ex- porters in the same industry. Thus, we use the 25th percentile of ordinary exporters’DVARs as an upper bound for processing exporters’ DVAR, and identify all processing …rms that have a DVAR higher than this cuto¤ as excessive exporters. Our …rm-level regression results 24 In the raw data, about 10 percent of the single-industry …rms have negative net exports. 25 Without a common …rm identi…er shared by the two data sets, we use …rm names to merge the customs transaction data with the NBS Annual Surveys of Industrial Firms. For rare cases that have duplicate …rm names, we use the …rm’ s address to improve the merging. See Ma, Tang, and Zhang (2014) for details about the merging procedures. Tables A2 and A3 in the Appendix present the representation of the merged and …ltered samples, relative to the original customs sample. In terms of the number of exporters, about 39% of the single-industry processing exporters from the customs data sets can be merged with the NBS data, and about 22% survive our …lters that remove excessive importers and exporters. In terms of export value, our …nal sample covers over 46% of exports based on the original customs data. 11 below are robust to using higher percentiles of ordinary …rms’DVAR as …lters. In summary, we focus on a subset of single-industry processing exporters that have their IM P EXP bounded between the two cuto¤s: OT IM P IM P P DM D + P I M I ; (8) EXP (25) EXP EXP OT IM P OT where DV AR(25) =1 EXP (25) is the 25 percentile of the DVAR of ordinary exporters in the same industry.26 Table 1 summarizes the main issues, assumptions and solutions of our approach to constructing the DVAR at the …rm, industry and aggregate levels.27 3.1 Movement of the Industry and Aggregate DVAR of Processing Exports The …nal data set is an unbalanced panel of 17,903 observations for 8,459 single-industry processing exporters over 8 years (2000-2007).28 Our sample covers a balanced panel of 15 industries throughout the sample period. An advantage of using the micro approach is that we can construct random samples drawn from the …rm sample and compute bootstrapped standard errors for our estimates of the aggregate DVAR. Figure 1 shows our benchmark 26 IM P OT Table A8 in the appendix reports EXP (25) by industry-year. We will check the sensitivity of our regression results by including both excessive importers and exporters in the sample below. 27 Sometimes, …rms have incentives to stock up imported materials when the international prices of com- modities are low, particularly in those industries that use a lot of commodities, such as iron, copper and crude oil, as inputs. Thus, imports may not be fully used to produce goods in the same period. For these …rms, the calculation of the DVAR based on (4) may not be accurate. However, there is no easy way to resolve the issue of inventory management. As we will show in the next section, all …rm observations with negative DVA are no longer negative once we use (8) to restrict our sample. This suggests that inventory management does not appear to drive our results. 28 Our sample covers both types of processing trade in China – pure assembly (PA) and import-and- assembly (IA). While we will check the robustness of our regression results below by repeating the analysis separately for the two regimes, it is important to point out that IA accounts for a much larger share of processing, in terms of the volume as well as the number of exporters, compared to PA. In our regression sample, over 90% of the observations belong to IA, in which exporters take control and hold ownership over the imported materials. We show in Figure A5 in the Appendix that even at its peak in 2000, PA never accounts for more than 30% of total processing exports, and continuously declined to less than 20% by 2007. 12 estimates of the DVAR of Chinese processing exports, along with the 95-percent con…dence intervals based on 100 randomly drawn samples with replacement. Chinese processing ex- ports’ DVAR has been increasing from 0.46 in 2000 to 0.55 percent in 2007. Depending on the year, the 95-percent con…dence interval is between 5 to 11 percentage-point wide, with an average of 7 percentage points over the 8 years in our sample. Most importantly, based on the bootstrapped standard errors, the di¤erence between the DVAR of Chinese aggregate exports in 2007 and that of 2000 is statistically signi…cant, lending strong support for KWW12, who also …nd an upward trend of similar magnitude based on IO tables and aggregate trade data. Figures A4 and A6 in the Appendix show similar trends, despite using samples with di¤erent cuto¤s from (8), and a sample that includes multi-industry …rms. Figure 2 plots the DVAR of processing exports across time for di¤erent industries, to- gether with the 95-percent con…dence intervals based on 100 random samples drawn with replacement. The DVAR increased for all industries besides two (wood and articles; and base metals). For the industries that exhibit an upward DVAR trend, the tight con…dence intervals convincingly reject the null hypothesis that the DVAR estimates are the same be- tween 2000 and 2007 (see Table A9 in the Appendix for details). For wood and articles, and base metals industries, their DVAR are not statistically di¤erent between 2000 and 2007. Overall, none of the industries exhibits a declining trend in the DVAR that is statistically signi…cant during the sample period. The micro data also permit a decomposition of the aggregate trend into between- and within-industry changes. Speci…cally, the change in the aggregate DVAR (from year t 1 to t) can be decomposed according to the following identity: X X 4DV ARt = wjt (4DV ARjt ) + DV ARjt (4wjt ); j j | {z } | {z } within between 13 1 EXPjt EXPjt 1 where wjt = 2 EXPt + EXPt 1 is the average share of industry j in total exports over year 1 t 1 and t, while DV ARjt = 2 (DV ARjt + DV ARjt 1 ) is the simple average of industry s DVAR over year t j’ 1 and t. Figure 3 shows that the increase in the aggregate DVAR over the sample period is all driven by within-industry increases in the DVAR rather than a between-industry reallocation of resources from the low-DVAR industries to the high-DVAR industries. s With these estimates, we further construct the bilateral DVAR with respect to China’ major trading partners. For each country-year, we compute the weighted average of the s share in total exports to the DVAR across industries, with weights equal to each industry’ destination. Figure 4 shows that in all top 5 trading partners (i.e., the U.S., Hong Kong SAR, China, Japan, the Republic of Korea, and Germany), there is a clear upward trend in the bilateral DVAR. In particular, the DVAR of Chinese processing exports to the US has increased from 0.47 to 0.55 between 2000 and 2007. 3.2 Firm Heterogeneity and Aggregation Bias How may …rm heterogeneity a¤ect the aggregate DVAR estimates? In a nutshell, …rm heterogeneity may lead to aggregation bias when the underlying sample used to construct the aggregate DVAR is not representative. This could happen if the following two conditions hold: (i) …rm size is used as the sample selection criteria and (ii) there is a systematic relationship between …rm size and import intensity. The above two conditions may hold in the samples used to construct IO tables in general, and speci…cally for China. According to the United Nations Handbook of Input-Output Table Compilation and Analysis (1999, section V: Compilation of Production Accounts of Industries), the intermediate input consumption and input structure of large establishments could be applied to small establishments, given that they are often not covered by industry 14 statistics (p. 110). This suggests that small and medium size …rms are routinely omitted from the IO table samples, and that the industry import intensities inferred are often based on data of mostly large …rms. For China, according to the National Input-Output Survey s Methods of China (2007) published by the National Bureau of Statistics of the People’ Republic of China, the sample used to construct the IO tables consists of all large …rms that have at least 300 million yuan in revenue (about 38 million USD during the sample period), along with some small- and medium-sized …rms sampled with unknown proportions (see item 5 on p. 3 about sample selection and item 4 on p. 27 about size cuto¤s). In other words, the sampling method behind the construction of Chinese IO tables is heavily biased towards the very large …rms. Second, recent research by Amiti, Itskhoki and Konings (2014) and Blaum, Lelarge and Peters (2014) shows that large …rms tend to have a higher import-to-sales ratio. This is also s import intensity on con…rmed by our sample of Chinese …rms. When we regress a …rm’ …rm size (measured by log(sales)), controlling for industry-year …xed e¤ects, we …nd that s import doubling …rm sales is associated with a 0.5 percentage-point increase in the …rm’ intensity (signi…cant at the 1% level).29 Given that …rms’import intensity and DVAR are negatively correlated, by omitting the smaller …rms, IO tables by construction tend to include …rms with a lower DVAR. Such sample selection criteria could cause the aggregate DVAR estimates to be signi…cantly biased downward. To demonstrate the signi…cant aggregation bias driven by sample selection when the underlying population of …rms are heterogeneous in size and import intensity, we conduct s 2004 …rm census data, which covers the following decomposition exercise relying on China’ the universe of all manufacturing …rms and is therefore much larger than our original …rm survey data set which only includes manufacturing …rms with a minimum 5 million RMB 29 Results are available upon request. 15 revenue.30 The …rst row of Table 3 shows the aggregate DVAR estimates based on the total popula- tion of …rms from the manufacturing census in 2004. The estimated DVAR is 0.479. In the next row, we restrict the census sample to include only …rms that overlap with our original manufacturing survey. The estimated DVAR dropped slightly to 0.478, which is not statisti- cally di¤erent from the previous row. This con…rms that sample selection bias is not an issue in our manufacturing survey sample, despite the exclusion of …rms with less than 5 million RMB revenue in our sample.31 Listed in Row (3) is the IO table-based DVAR estimate of 0.408 from KWW12. Consistent with our results in previous section, the IO table-based DVAR estimate is statistically lower than the DVAR estimates in Rows (1) and (2), based on their respective bootstrapped standard errors. In Row (4), we further restrict the census sample, according to the sample selection criteria speci…ed in the Chinese IO table manual –…rms with over 300 million RMB revenue. The aggregate DVAR estimate drops to 0.453. Not only is the resulting DVAR estimate based on this large …rm sample lower than the estimates in Rows (1) and (2), it is also not statistically di¤erent from the IO table-based estimate by KWW12 in Row (3), based on a standard error of 0.034 from bootstrapping with 100 repetitions. This exercise con…rms that DVAR decreases when samples that only include larger …rms are used, due to …rms’heterogenous input sourcing. Thus, the result in Table 3 nicely shows that ignoring …rm heterogeneity may lead to downward aggregation bias in the IO table-based approach. While there can be many reasons why our …rm-based estimates and the IO table-based estimates of KWW12 are di¤erent, such 30 Unlike our industrial survey dataset, the census dataset does not provide direct information on …rms’ costs of materials. We follow the guideline of the user manual of the census dataset to compute a …rm’s cost of materials by substracting its total sales by its value added. As such, our DVAR estimates based on the census dataset is not directly comparable to the estimates in the previous sections based on the industrial surveys, which provide direct information on …rms’costs of materials. 31 Recall that our sample consists of …rms from the Annual Surveys of Industrial Firms, which has sales cuto¤ of 5 million RMB (about 600,000 USD) and above, while the 2004 census covers all industrial …rms. 16 as di¤erences in methodology or estimation errors, this decomposition exercise focuses solely on the role of …rm heterogeneity in explaining the wedge. Firm heterogeneity matters because …rms of di¤erent sizes have di¤erent import intensities. By restricting the census sample to large …rms according to the IO cuto¤ criteria, we are able to account for the di¤erence between our aggregate DVAR and that of the IO table-based estimate of KWW12. 3.3 Extension to Non-Processing and Aggregate Exports The methodology we have developed above is suitable for pure exporters who export all their products, and that the products are produced by using up all the materials they have imported. It requires the condition that no …nal products or imported materials may leak to the domestic economy. A lot of exporters that engage in global value chains should satisfy this condition, in the form of processing trade, such as garment producers in Bangladesh, Guatemala, and other emerging economies. However, many exporters are not processing exporters. Unlike processing exporters, non- processing exporters do not export all their outputs. In addition, they often use some of their imported materials to produce goods for domestic sales. Thus, the condition that no …nal output or imported materials leak to the domestic economy is not met. How …rms split their imported inputs between production for domestic sales and exports is generally unknown. To extend our methodology to measure the DVAR of the non-processing exporters, we s need to make one proportionality assumption at the …rm level: the allocation of the …rm’ inputs to the production for exports is proportional to the share of exports in total sales, which we may infer from our industrial survey data. This assumption is equivalent to as- suming that the DVAR is the same between exports and domestic sales of the …rms. Our proportionality assumption will likely be non-binding if …rms produce the same products for 17 both the domestic and export markets. In addition, it is considerably less restrictive than the industry-level proportionality assumption commonly made by existing studies, as we still allow …rms to be heterogeneous in terms of their shares of exports in total sales. Thus, the DVA and DVAR of a non-processing exporter are: K F EXPi DV AO i = EXPi IM Pi i + i ; (9) P Yi K F O DV Ai IM Pi i + i DV ARi = =1 ; (10) EXPi P Yi O’stands for ordinary exports. Similar to processing exports, there where the superscript ‘ are transactions between non-processing exporters and the rest of the economy. After the adjustment based on the proportionality assumption, we follow the same procedures as outlined in Table 1 to adjust the estimates of the DVAR, similar to what we did for processing F exporters. We …rst obtain imputed based on the estimates from KWW12. Then we identify imported capital based on the United Nations Broad Economic Categories (BEC) K list of capital goods, and adjust for i . Finally, we drop excessive importers. However, unlike what we can do for processing exporters that export excessively, there is no corresponding …lter we can use to drop the excessive ordinary exporters. Including them in the sample will result in an overestimation of the DVAR of ordinary exports. With this caveat in mind, our approach is transparent and general enough to be applied to estimate the DVAR of di¤erent types of exporting …rms and thus countries with varying prevalence of processing trade. We use the ground-up approach to measure the DVAR of Chinese aggregate exports, by taking the weighted average over the DVARs of processing and ordinary exports, with weights equal to the corresponding export shares.32 As shown in Table 2, the average DVAR 32 Here we measure the DVAR for single-industry exporters only. As we have done for processing exports, we can also do it for multiple-industry …rms as well. The drawback is that excessive processing importers are identi…ed as those that have import-export less the 25th percentile of the DVAR of ordinary exporters in the same year, but not the same sector-year. These numbers are available upon request. 18 of ordinary exports during the sample period is around 0.9, substantially higher than that of processing exports but consistent with similar …ndings by KWW12. Moreover, the DVAR of ordinary exports has declined slightly between 2000 and 2007, from 0.92 to 0.90. However, given the small decline compared to the much larger increase in the DVAR of processing s total exports exports, coupled with the fact that the share of processing exports in China’ has been stabilized at around 55% throughout the sample period, the DVAR of Chinese aggregate exports increased from 0.65 to 0.70 between 2000 and 2007 (see Figure 5, Table s DVAR has increased signi…cantly in 2 and Figure A7 in the appendix). In short, China’ recent years, almost entirely driven by the rise in the DVAR in the processing export sector. 4 Time-series Trend of Firm DVAR In this section, we provide reduced-form evidence of the time-series changes in …rms’DVAR s rising DVAR and other related variables. A formal analysis of the determinants of China’ will be presented in the next section. Given the …nding in the previous section that the entire increase in the DVAR is caused by processing exports instead of ordinary exports, we will focus on providing …rm-level evidence based on processing exporters only from this section and on. We start o¤ by estimating the following speci…cation at the …rm level: DV ARit = i + t + X Xit + it ; (11) where i stands for …rm, t represents year, and it is the regression residual. The …rm and year …xed e¤ects are i and t respectively, with the year e¤ect for 2000 dropped to avoid 0 the dummy variable trap. Thus, positive and rising ts (i.e. 0 < t < t+1 ; 8t > 2000) will imply a within-…rm increase in the DVAR over time. PMM s material-to-sales ratio, Control variables in Xit include a …rm’ PY and its labor it 19 PMM s cost (total wages or the ratio of wages to total sales). The inclusion of a …rm’ PY is it to examine whether the …rm substitutes between domestic and imported materials, keeping the total material cost share constant, according to (5). Labor cost is included to verify the s rising DVAR in popular claim that increasing labor costs are a main reason behind China’ PMM 0 0 exports. Controlling for PY ; if ts are positive, signi…cant and rising, while Xs are it not positive or insigni…cant, then it implies that the DVAR is rising within …rms, due to a substitution of domestic materials for imported materials. Table 4 presents our baseline results. Bootstrapped standard errors, clustered at the industry level, are used for all the regressions reported in this section. Column (1) shows positive, signi…cant, and increasing year …xed e¤ects, suggesting that …rms’DVAR is rising during the sample period. On average, …rm DVAR increases by 15 percentage points between 2000 and 2007. This within-…rm increase is larger than the 9 percentage-point increase at the aggregate level (see column 4 in Table 2), implying that exiting …rms have a higher DVAR than new entrants on average. In other words, the upward trend of the aggregate DVAR of Chinese exports is entirely driven by the rising DVAR among the surviving exporters, not PMM due to the exit of low-DVA …rms.33 Furthermore, by controlling for the …rm’ s PY , it we con…rm that the rising DVAR is due to …rms’ substitution of domestic for imported materials. wL s wage-to-sales ratio In column (2), we add the …rm’ P Y it as a control. The insigni…cant wL PMM coe¢ cient on P Y it supports the prediction based on (4) that once PY is controlled it s DVAR. Columns (3) to (5) show for, labor costs should not have any direct impact on a …rm’ the same upward trend for three di¤erent samples –domestic exporters only, foreign-invested exporters only, and multi-industry exporters included. In column (6), we repeat the same 33 According to Table A10 in the appendix, the exiting …rms tend to be smaller in terms of sales and exports. Given that …rm size and DVAR are negatively correlated, it is not surprising to see that the exiters have higher DVAR as shown in the table. Furthermore, to the extent that …rm size proxies for …rm productivity, it is also not surprising that these smaller and thus high-DVAR …rms are more likely to exit. 20 analysis using an un…ltered sample that includes both excessive importers and exporters. The magnitudes of the estimated year …xed e¤ects are very close to those in column (2) when the …ltered sample is used, suggesting that our …ndings are not driven by the removal of excessive importers and exporters. In summary, we …nd that the within-…rm increase in the DVAR is widespread and it is not driven by sample selection. The within-…rm increase in the DVAR over time should arise from exporters’substituting domestic for imported materials, at both the intensive and extensive margins. To examine this claim, we estimate the following speci…cations: P IMI = i + t + X Xit + it ; (12) PMM it ln(import_varietyit ) = i + t + X Xit + ! it ; (13) P IMI where PMM is the share of imported materials in total material cost for …rm i in year it t, while ln(import_varietyit ) stands for the (log) number of import variety, measured by the number of imported HS6-country pairs.34 Firm …xed e¤ects are denoted by i and i in the respective speci…cations, while t and t are the year …xed e¤ects, with the year e¤ects for 2000 omitted to avoid the dummy variable trap. Control variables in Xit include wL K s wage-to-sales ratio, …rm’ P Y it , (log) capital-labor ratio, ln L it , and material-to-sales PMM ratio, PY . We include these controls to capture the e¤ects of changing labor costs and it capital deepening of the …rm on imports. The residuals for each of the speci…cations are it and ! it ; respectively. If …rms are using more domestic materials for imported materials, the year …xed e¤ects are expected to be negative, signi…cant and declining (i.e., t < t 1 <0 and t < t 1 < 0; 8t > 2000). Column (1) in Table 5 shows that the share of imported materials is gradually declining 34 The HS classi…cation has changed twice (2002 and 2007) during our sample period. We use the concor- dance …le created by Cebeci et al. (2012) to de…ne a consistent set of varieties over time. 21 P IMI s within …rms over time. In particular, …rm’ PMM dropped by about 17 percentage it points on average in 2007 compared to 2000. This result supports our …nding that Chinese processing exporters are substituting more domestic materials for imported materials over time. Firm wage-sales ratio and capital-labor ratio do not appear to be related to its import share. The results remain robust when we split the sample into the domestic private and foreign …rm samples (columns (2)-(3)) or include multi-industry …rms (column (4)). Consistent with the …ndings that …rms decrease their imports, Table 6 shows negative, signi…cant and declining year …xed e¤ects, suggesting that on average, processing …rms also import fewer input varieties over time. At the sample mean, the number of import varieties decreased by 0.35 log points in 2007 relative to 2000.35 Other …rm-level controls are insigni…cant. Columns (2) and (3) show that the decline mostly happens for foreign …rms but not domestic private …rms. The results remain robust to including multi-industry …rms in the sample (column (4)). Along with the results from the previous tables, we …nd that …rms’ average DVAR is rising through substitution of domestic inputs for foreign inputs, at both the intensive (the cost share of imported materials) and extensive margins (import variety).36 For processing …rms to substitute domestic for imported input varieties, an increased availability of the latter is expected. Unfortunately, data on domestic input variety in China are not available. To examine the phenomenon, we rely on the number of varieties exported by ordinary (non-processing) …rms as proxies instead. Note that unlike processing exporters, ordinary exporters consist mainly of the indigenous Chinese …rms that also sell in the do- 35 In unreported results, we …nd that most of the decline is due to …rms importing fewer products (HS6) instead of importing from fewer countries. 36 It is interesting to note that many of the dropped import varieties are parts and components from the neighboring countries, such as parts of refrigerators, computers, and electric conductors from Singapre and Japan, pick-up cartridges from Hong Kong SAR, China, iron and steels products from the Republic of Korea, and television cameras from Taiwan, China. Other varieties also include parts of electrical machines from Italy, and cathode-ray tubes from Germany. These observations are consistent with our hypothesis that processing exporters are substituting domestic for imported materials. 22 mestic market. Some of these local …rms become big and start exporting. By tracking the number of varieties exported by ordinary …rms, we are picking up the tip of the iceberg as some of these domestic varieties may not make it to the foreign markets.37 Nevertheless, the following evidence is insightful. Table A12 in the appendix lists 67 products that were imported by processing exporters and were not exported by ordinary exporters in 2000, but were exported by ordinary exporters in 2007. Some of them are important inputs used by large exporters across many industries, accounting for an import value of close to US$392 million. By 2007, not only were these products no longer imported by processing …rms, ordinary exporters have started exporting them with a total value of over US$1.55 billion. These results suggest that processing exporters’demand for these imported products is now being met by local suppliers.38 To verify that the decline in import variety is not due to exporters’specialization in their core competencies, we estimate the following speci…cation: ln(export_varietyit ) = i + t + X Xit + uit , (14) wL PMM where Xit includes P Y it and PY as in (11). Dependent variable, export_varietyit is it s number of exported HS6-country pairs.39 Firm …xed e¤ects ( i ), year measured by …rm i’ 37 We use products produced by ordinary (non-processing) exporters to proxy for domestic variety, in the belief that a …rm’ s export product scope is a subset of its domestic product scope. There could be export varieties that were not sold domestically or vice versa. There could also be domestic varieties produced by non-exporters that were not exported. In these regards, our proxy should be considered as a lower bound of domestic variety. 38 In the last column of Table A12 in the appendix, we also report the share of exports by foreign …rms for each product in 2007. Out of the 67 products listed in the table, 15 products have over 20% of exports by foreign …rms in 2007, and 5 products were exported solely by them. These results suggest that foreign …rms may have moved into some of the intermediate good sectors in China. These results are also consistent with the assertions of recent studies, such as Autor et al. (2013) and Pierce and Schott (2015), that changes in policies in the U.S. and China may have encouraged foreign …rms to o¤shore production to China, potentially contributing to China’ s growing competitiveness. That said, the majority of these new export products are actually produced by indigenous domestic Chinese …rms. We thank David Hummels for suggesting this exercise. 39 We also repeat the same analysis using the number of HS6 (without the country dimension) to measure export variety. The results remain robust. 23 …xed e¤ects ( t ), and other …rm controls are included as before. As Table 7 shows, despite the declining cost share of imported materials and decreasing variety, processing …rms’export variety is rising over time, particularly after 2002, one year after China joined the WTO. In summary, our results suggest that the domestic content in Chinese processing exports is rising over time. The rise is mainly driven by …rms actively substituting domestic for imported materials, but not rising production costs. Chinese exporters have been expanding their product scope while reducing imports, both at the intensive and extensive margins.40 5 Determinants of Firm DVAR s trade and FDI liberaliza- In the rest of the paper, we will focus on studying whether China’ tion since 2000 could explain its rising DVAR. We …rst develop a simple model to guide our empirical exploration of the determinants of the rising …rm DVAR. This model focuses on the time-series movement of …rms’ DVAR and thus the aggregate DVAR, and deliberately abstains from explaining the cross-sectional di¤erences in the DVAR.41 5.1 A Simple Model Recall the accounting identity (4): I PtI Mit I PtM Mit PtI Mit DV ARit = 1 + 'it = 1 + 'it ; Pit Yit Pit Yit PtM Mit 40 There can be concerns that the regression results are di¤erent between the two processing trade regimes, as described in Section 3. To this end, we repeat all four regression analysis using the sample of import- and-assembly (IA) and pure-assembly (PA) …rms, respectively. As reported in Table A11 in the appendix, results remain robust and quatitatively identical to the results reported so far. This is not surprising given that 90% of the observations in our sample belong to the IA regime. It is assuring to see that …rm DVAR is also increasing within PA exporters. The magnitude of the coe¢ cients on the year …xed e¤ects are similar. Similar trends are also found using this sample for other dependent variables of interest, though the statistical signi…cance may sometimes be smaller due to the much smaller sample of PA …rms. 41 In the Appendix, we derive a model that features a Cobb-Douglas production function, and show how …rm heterogeneity in price-cost margins may lead to a cross-sectional variation in …rm DVAR. 24 where 'it is a well-behaved classical regression error term that captures the unobservable F it EXPit s DVAR depends only on the share of imported materials in total . Thus, a …rm’ PtIMI MM Pt it materials, M it Pt Mit ; once we control for the share of materials in total revenue Pit Yit : Without loss of generality, assuming that the unit material cost function, P M PtI ; PtD , is a translog function of the prices of imported and domestic materials, which is symmetric, homogenous of degree one and can provide a second-order approximation to any functional form of price aggregates: ln P M PtI ; PtD = i + 0I ln PtI + 0D ln PtD (15) 1 2 1 2 + II ln PtI + ID ln PtI ln PtD + DD ln PtD : 2 2 The assumptions of symmetry and homogeneous of degree one imply the following restrictions on the translog parameters: II < 0; DD < 0; 0I + 0D = 1; II + ID = DD + ID = 0; and II = DD = ID <0) ID > 0: (16) Let mI D it and mit be the requirement of imported and domestic materials for producing one unit of total materials Mit : k Mit mk it = ; k = I; D: Mit s Lemma, the share of imported or domestic materials is the elasticity of the By Shephard’ 25 unit material cost function with respect to the price of imported or domestic materials: @P M PtI ; PtD = mk I i Pt ; Pt D , for k = I; D @Ptk @P M PtI ; PtD Ptk Ptk k I D Ptk Mik PtI ; PtD = m P ; P = : @Ptk P M (PtI ; PtD ) P M (PtI ; PtD ) i t t P M (PtI ; PtD ) Mit Thus, when the unit cost function is translog, the share of imported materials in total materials is a log-linear function of the relative input prices: PtI Mit I @ ln P M PitI D ; Pit = (17) PtM Mit I @ ln Pit I D = 0I + II ln Pit + ID ln Pit PtI = 0I ID ln ; PtD PtI where D Pt is the ratio of the price index of imported input varieties to that of domestic input PtMM it varieties. From (4), once we control for the share of materials in total sales, Pit Yit , …rm PtI DVAR depends only on D Pt positively (given that ID > 0): PtM Mit PtI DV ARit = 1 + 0I + ID ln + 'it ; 8i; t: (18) Pit Yit PtD Thus, by assuming a translog cost function, we show that the only factor that a¤ects PtI s DVAR is a …rm’ D, Pt after controlling for the share of total material cost in total sales, PtMM it 42 Pit Yit . Other factors, such as wages, productivity and other costs of production do not directly enter (18), as long as the share of total materials in total sales is controlled for. We explore three obvious factors that can a¤ect …rm DVAR, namely import tari¤s facing upstream suppliers, foreign direct investment (FDI), and exchange rates in the next section. 42 Note that it is the …rm’s total sales in the denominator, not output. Thus, a …rm’s mark-up, which we do not aim to estimate, is already embedded in the formula. 26 In addition to its ‡exibility of providing a second order approximation to any cost func- tion, the translog cost function (15) has the advantage of not restricting the elasticity of substitution between domestic and imported materials to be a constant.43 This modeling ‡exibility is particularly important since a rising …rm DVAR could be driven by a rising elas- ticity of substitution between imported and domestic input varieties. By using a translog speci…cation, we let the data reveal whether and how the elasticity was changing over the sample period. Speci…cally, let t be the elasticity of substitution between domestic and imported ma- terials in year t. According to Blackorby and Russell (1989), the elasticity of substitution between the two variables equals the cross-price elasticity "ID t minus the own price elas- ticity "DD t : t = "ID t "DD t : (19) In this case, using (15), we can express both "ID t and "DD t as functions of ID and sD t : 44 @ ln MtD DD ID "DD t = D + sDt 1= + sD t 1; @ ln PtD st sD t @ ln MtD ID "ID t I = I + sD t ; @ ln Pt st which according to (19) gives ID t = + 1 > 1; (20) sD t (1 sD t ) since ID > 0. We will be able to test these restrictions when we estimate ID based on (17) and construct t from (20). Note that t could change over time (and across industries) due 43 This property is in contrast with the case of a constant-elasticity-of-substitution (CES) production function. Readers are referred to the Appendix for a derivation of a …rm’ s DVAR when the production function is Cobb-Douglas. 44 See Kee, Nicita and Olarreaga (2008) for the derivation. 27 to changing sD t . Before discussing our estimation of t in detail later, let us return to the PtI discussion about the determinants of D Pt s DVAR. and thus a …rm’ 5.1.1 Exchange Rates One obvious factor that could cause …rm DVAR to increase is the exchange rate. De…ne the exchange rate, Et , as the foreign-currency price of a Chinese yuan. The price of imported materials in yuan is equal to the world price of foreign materials, PtI , divided by Et , i.e., I Pt PtI = Et . A yuan appreciation (a higher Et ) decreases the yuan price of imported materials, possibly lowering …rm DVAR according to (18): @ PtI =PtD @DV ARit @DV ARit @ PtI =PtD <0) = < 0: (21) @Et @Et @ (PtI =PtD ) @Et 5.1.2 Input Tari¤s Facing Domestic Input Suppliers The relative price of materials could change due to the varying supply of input varieties. We assume that sector-level materials are CES aggregates of di¤erent varieties of domestic and imported inputs as follows: 2 D 3 1 2 I 3 1 Vt Vt X 1 X 1 D Mit = 4 mD vi 5 I ; Mit =4 mI vi 5 ; > 1; v =1 vi =1 where VtD and VtI are the numbers of domestic and foreign input varieties available to the …rm. Let us assume that the elasticities of substitution, , between any two varieties of imported materials, as well as between any two varieties of domestic materials, are constant. The average price of imported and domestic materials can then be expressed as PtD = 2 D 311 2 I 311 Vt Vt X X 4 D 1 5 I 1 Pvt and PtI = 4 Pvt 5 , where Pvt D I and Pvt represent the price of a v =1 v =1 domestic and a foreign input variety, respectively. An increase in domestic material varieties 28 will raise the relative price of imported materials, which in turn raise …rm DVAR: @PtD @ PtI =PtD @DV ARit @DV ARit @ PtI =PtD < 0 ) > 0 ) = > 0: (22) @VtD @VtD @VtD @ (PtI =PtD ) @VtD The intuition is similar to the positive e¤ects of an increase in import varieties on aggregate productivity and welfare (e.g., Broda and Weinstein, 2006 and Feenstra and Kee, 2008). What caused an increase in domestic and imported material varieties? We explore two s gradual trade lib- factors previously explored in the literature. The …rst factor is China’ eralization. Goldberg et al. (2010) show that in India, input tari¤ liberalization results in domestic …rms’ expansion of product scope. The main reason is that after trade liberal- ization, domestic …rms have access to cheaper and new imported input varieties. Over our sample period (2000-2007), China experienced a continuous decline in import tari¤s and s accession to the WTO in other trade restrictions, which was accelerated after the country’ December 2001. It is worth noting that such liberalization does not directly a¤ect process- ing …rms, which have always been exempted from tari¤s on imported inputs. That said, tari¤ reduction could have a signi…cant impact on those non-processing …rms that supply materials to the downstream processing exporters.45 With access to new, cheaper, or better imported materials after tari¤ liberalization, non-processing …rms experience lower produc- tion costs and may produce more varieties. Processing exporters in downstream sectors can now purchase these varieties domestically, replacing previously imported input varieties. This substitution at the extensive margin, as we will show below, plays an important role in driving the DVAR of the downstream processing exporters. More formally, let t denote the (average) input tari¤ of the upstream industries. Tari¤ reduction may increase domestic input varieties, which in turn raise the relative price of imported materials and thus the 45 As long as the imported materials stay inside the processing regime, domestic transactions are still exempted from tari¤s. 29 DVAR of downstream exporters. These relationships can be expressed as: @VtD @DV ARit @DV ARit @ PtI =PtD @VtD <0) = < 0: (23) @ t @ t @ (PtI =PtD ) @VtD @ t 5.1.3 Foreign Direct Investment The last factor is related to the rising FDI in the processing sector, as China increased its engagement in global value chains, due to its FDI liberalization since 2000.46 Participating in global value chains has been proposed to be a new and e¤ective way of industrialization (Baldwin, 2012). In particular, Rodriguez-Clare (1996) and Kee (2015) show that more own-industry FDI can increase the demand for domestic materials, raising the supply and quality of domestic material varieties from the upstream industries.47 Given > 1 in our model, a higher demand by downstream exporters will lower the price of domestic materials, which in turn increase the DVAR for all exporters. More formally, we have @VtD @DV ARit @DV ARit @ PtI =PtD @VtD >0) = > 0: (24) @F DIt @F DIt @ (PtI =PtD ) @VtD @F DIt The following section will empirically examine how the three factors discussed in this section shape the movement of …rm DVAR. 5.2 Three-Stage Least Squares Regressions Our model shows that factors such as exchange rates, FDI and upstream input tari¤s may raise …rms’ DVAR, through a¤ecting domestic input varieties and hence the relative price 46 With China’ s accession to the WTO in December 2001, the government has committed to a deeper and more comprehensive liberalization to FDI, though revising the Law on Foreign Capital Enterprises in October 2000. In particular, the revised law lifted the requirement for foreign enterprises to export the majority of their output. 47 For example, FDI in the garment industry may increase the demand for domestic textile products and cause the domestic textile industry to increase their product varieties. 30 of imported materials. We …rst empirically establish these channels without imposing the translog cost structure and let the data show the relationship between these variables. In the next section, we will formally estimate the translog parameters to assess how well our highly stylized model may explain …rm DVAR. We …rst isolate the part of the within-…rm changes in the DVAR that is common across I Pjt all …rms within an industry, given that D Pjt is industry-speci…c. To this end, we estimate the average within-…rm change in the DVAR by industry according to (11) and allow year …xed e¤ects to be industry-speci…c: DV ARit = i + jt + X Xit + it : The estimated jt , ^ jt , captures the average within-…rm change in DVAR of each industry j in each year relative to 2000. We then estimate the following system of three equations using 3SLS: ! I Pjt ^ jt = !1 + !1 4 ln + 1 (25) j p D jt ; Pjt ! I Pjt 4 ln D = !2 2 2 D j + ! E 4 ln Ejt + ! v 4 ln Vjt + 2 jt ; (26) Pjt D U 4 ln Vjt = !3 3 3 3 j + ! T 4 ekt + ! F 4 ln F DIjt + ! E 4 ln Ejt + 3 jt ; (27) where ! 1 2 3 j , ! j , and ! j stand for industry …xed e¤ects in three di¤erent equations, and 1 2 jt , jt , 3 and jt are the corresponding error terms. The …rst equation uses the change in the price of imported materials relative to domestic I Pjt materials, 4 ln D Pjt ; to explain the within-…rm change in the DVAR that is common across I Pjt all …rms within an industry. The second equation explains how 4 ln D Pjt can be caused by the change in the exchange rate, 4 ln Ejt , de…ned as the increase in the foreign price of the 31 D yuan, and the change in domestic upstream variety, 4 ln Vjt . The last equation explains how D 4 ln Vjt can be caused by the change in own-industry FDI, 4 ln F DIjt ; the change in the average input tari¤s facing …rms in the upstream industry, 4eU kt ; and 4 ln Ejt . We include the exchange rate in (27) to test the hypothesis that a stronger yuan, in addition to a¤ecting import prices directly as speci…ed by (26), may also decrease the demand for domestic inputs as …rms may choose to increase imported inputs. The ways that we measure imported input prices, domestic input prices, exchange rates, and domestic upstream variety are discussed in detail the Appendix. Our model predicts that ! 1 2 2 p > 0 in (25); ! E < 0 and ! v > 0 in (26); !3 3 3 T < 0, ! F > 0, and ! E < 0 in (27). Table 8 reports the results. Since ^ jt are estimated with errors, bootstrapped standard errors (with 500 repetitions) are used in all equations. Column (1) shows a positive and I Pjt signi…cant correlation between the relative price index of imported materials, D Pjt , and the average within-…rm change in the DVAR in the same industry. Column (2) presents the results of (26) ; which shows that controlling for industry …xed e¤ects, upstream variety has a strong and positive in‡uence on the relative price of materials. On the other hand, the estimated coe¢ cient on exchange rate has a wrong sign, but is only marginally signi…cant with a t-stat of 1.66. At any rate, given that the average annual change in Ejt is close to zero during the sample period, the exchange rate is economically insigni…cant in a¤ecting the relative price of imported materials. This result suggests that empirically most of the changes in the relative price of materials during the sample period were driven by the expansion of domestic upstream variety and not necessarily due to exchange rate changes. Column (3) reports the estimates of (27). The result shows that all three factors (own-industry FDI, upstream input tari¤ liberalization, and the exchange rate) are statistically signi…cant in explaining the expansion of upstream domestic variety. In particular, the result that input tari¤ liberalization in the upstream industry is associated with an expansion of the variety 32 of upstream materials is consistent with the …ndings by Goldberg et al. (2010). Over our sample period, Chinese ordinary exporters experienced a continuous decline in input tari¤s, s accession to the WTO in 2002. From 2000 to 2007, which was accelerated by the country’ the average input tari¤ facing suppliers in the upstream sectors declined by about 55%. The coe¢ cient of -0.012 implies that the reduction in tari¤s is associated with a 0.7% increase in domestic input varieties, about one-…fth of the average increase across sectors from 2000 to 2007. It is worth noting again that processing …rms are exempted from tari¤s for imported materials, so tari¤ reduction will not a¤ect their production costs directly but only indirectly through other general equilibrium e¤ects in the domestic economy. Tari¤ reduction leads to an increased supply of input varieties, which in turn lowers the average domestic material price and contribute to the rise in the DVAR of processing exporters. Likewise, the presence of own-industry FDI has a positive impact on the variety of upstream materials, supporting the …ndings of Rodriguez-Clare (1996) and Kee (2015). Speci…cally, given that the average FDI stock in an industry is about 1.16 log-point higher in 2007 compared to 2000, the coe¢ cient of 0.017 implies that the increase in FDI in the downstream sectors is associated with a 2% increase in domestic input varieties. Finally, the negative sign on 4 ln Ejt is consistent with the hypothesis that a stronger Chinese yuan will lead to more imported variety and thus less domestic variety due to import competition. However, during the sample period, the average annual change in Ejt is close to zero, implying that the exchange rate plays an economically insigni…cant role in the expansion of the domestic input market. Overall, the results presented in Table 8 is consistent with our model, highlighting that the change in …rm DVAR is driven by the changes in the relative prices of imported and domestic materials, due to the underlying expansion of domestic input variety, in response to the upstream input tari¤ liberalization and the increased presence of own-industry FDI 33 in downstream industries.48 5.3 Quantitative Analysis In this section, we estimate our model structurally in order to assess how much of the change in the DVARs at the …rm and aggregate levels can be explained by our model. We need to …rst estimate the translog parameter, ID : s DVAR depends on According to (18), a …rm’ PtMM it the share of materials in total sales, Pit Yit ; and the translog parameter, ID , as follows: PtM Mit PtI DV ARit = 1 + 0I + ID ln + 'it : Pit Yit PtD PtI The partial impact of a change in ln D Pt on …rm DVAR is @DV ARit PtM Mit I = ID : @ ln Pt Pit Yit D Pt PtMM it With the estimate of ID and the actual data on Pit Yit , we can calculate how much of the change in …rm and industry DVAR is due to the change in the relative price as predicted 48 s request, we have also checked whether FDI into upstream sectors could a¤ect the DVAR Per a referee’ of downstream industries. To this end, we regress the change in the upstream input variety of an industry on the change in the weighted average of FDI across upstream industries, in addition to all the right-hand side variables included in column (3) of Table 8 (i.e., changes in own-industry FDI, upstream input tari¤s, and exchange rates). We …nd that upstream FDI does not explain the increase in upstream input variety, while all the other variables remain signi…cant and have the same sign as those reported in Table 8. In particular, the coe¢ cient on upstream FDI presence is -.0038 and is not statistically signi…cant. This …nding is consistent with our previous results that most of the new intermediate inputs were produced by indigenous Chinese …rms and not foreign …rms (see Table A12). Interpreting this result through the lens of our model would suggest that upstream FDI does not a¤ect the relative price of imported materials and hence industry DVAR. However, it is plausible that upstream FDI may have an independent e¤ect on industry DVAR, not through a¤ecting domestic input variety, but that is beyond the scope of our model and paper. 34 by our model: PtM Mit PtI DV ARit = b ID ln ; (28) Pit Yit PtD 0 1 X EXPi B X EXP P M M C I Pjt B i t it C DV ARjt = X DV ARit = B X C b ID ln D ;(29) i2 EXPi @i2 EXP i Pit Yit A Pjt j j i2 j i2 j where the change in industry j 0 s DVAR equals the weighted average of the changes in the DVAR of all …rms in industry j (i 2 j ), derived from (6) ; and j subscript is added to the relative price for clarity. In addition, with the estimate of ID ; we can also construct the elasticity of substitution between imported and domestic materials, jt , for each industry j and year according to (20). Such estimates allow us to assess the time-series variation in jt and examine whether the rise in …rm DVAR is driven by an increasing jt or not. To estimate ID , we estimate the following econometric counterpart of (17) : I PtI Mit PtI = ai ID ln + it ; (30) PtM Mit PtD where ai is the …rm …xed e¤ect that subsumes 0I in (17) and it is the residual. In other words, ID is estimated from the within-…rm variation in the relative price between im- ported and domestic materials. Since the dependent variables are measured with errors, we bootstrap the standard errors (based on 500 randomly drawn samples). Moreover, we use the exchange rate indices, (log) FDI and (log) upstream input tari¤s of the sector as the PI instrumental variables for ln P D t . t Table 9 reports the estimated b ID , …rm average share of imported materials in total material cost, and the implied bjt for 15 industries and both 2000 and 2007. Estimated b ID 0 for all industries and years are positive and the resulting s are all greater than 1, satisfying 35 the theoretical restrictions speci…ed in (16) and (20). In addition, when the entire sample of …rms is used, the F-statistics for the …rst stage is highly signi…cant with p-value of 0 and thus passing the weak instrument test of Stock and Yogo (2005) by a wide margin.49 . Likewise, across all industries, most of the F-statistics are larger than 100 with the minimum …rst stage F-statistics is 44. Of the 15 industries, the IV estimates of ID are signi…cant for 12 industries at the 1% signi…cance level. The estimated t for the whole sample is 2.68 for 2000 and 2.83 for 2007. Both of them are statistically signi…cant at the 1% level. Of the 12 industries for which b2007 is signi…cantly di¤erent from 0, b2007 ranges from 1.90 for “plastic & rubber (HS2 = 39-40)” to 6.56 for “beverages and spirit (HS2 = 16-24)”. Even for the industries for which the estimates are imprecise, the coe¢ cients are positive, implying that the implied is larger than 1. In other words, foreign and domestic input varieties are gross substitutes for processing exports in all industries in China. Based on the estimates of jt 0 for both 2000 and 2007, we perform simple t-tests and con…rm that jt s are statistically constant within the sample period and for each industry.50 Using these estimates, we can do the following back-of-the-envelope calculations. In 2007, PtI the average (across industries) D Pt is about 0.419 log-points higher that that of 2000. The estimated average b ID for the pooled sample, based on the instrumental variables estimation, is 0.376; while the average (across …rms) share of material cost in total sales in 2007 is 0.786. Using (28), the predicted increase in DV ARit is 0:376 0:786 0:419 12%, which is not statistically di¤erent from 14:7%, the estimated within-…rm increase in the DVAR from 2000 49 According to Table 1 of Stock and Yogo (2005), the critical value of the …rst stage F-statistics for the weak instruments test for three instrumental variables used for one endogeneous variable is 13.91, if the bias of the IV estimator is restricted to be no more than 5 percent of the OLS bias. 50 To assess the time series movement of ; we test H0 : 2007 2000 = 0: T-tests are performed based on the following variance for : var ( ) var ( ) = 2; 2 s (1 s) which is derived from (20). We construct the standard errors based on data from both 2000 and 2007. None of the t-statistics is statistically signi…cant. These test statistics are available upon request. 36 to 2007 as reported in Table 4.51 Likewise, the predicted change in industry DVAR is also about 13%, according to (29), which explains fully the average increase in the industry DVAR during the sample period. This suggests that our simple translog model of using the relative price of materials (driven by upstream input tari¤, FDI, and exchange rates) to explain the s and aggregate DVAR …ts the data very well. …rm’ Let us summarize the main …ndings of the paper. First, the DVAR of processing …rms is increasing within …rms across time in our sample and has led to an upward trend in both the industry and aggregate DVAR. Firm entry and exit do not explain the upward trend. Second, such an increase is mainly driven by …rms substituting domestic for imported mate- rials. Third, such a substitution is a response to the declining relative prices of domestic to imported materials caused by the expansion of domestic input variety. Fourth, the expansion of domestic input variety is induced by an increasing presence of foreign …rms in processing exports and decreasing input tari¤s facing upstream suppliers. Fifth, based on the decrease in the relative price of domestic to imported materials, our model explains nearly all of the s and aggregate DVAR from 2000 to 2007. increase in the …rm’ 6 Concluding Remarks s rising ratio of domestic value added in This paper provides micro-level evidence of China’ s customs transaction data over the 2000- exports to gross exports (DVAR). We use China’ s DVAR and show how the increase in …rm DVAR might 2007 period to measure a …rm’ explain the aggregate trend. We …nd that the drastic increase in the DVAR of Chinese processing exports is observed across all industries and trading partners, and accounts for s aggregate exports during the period. almost the entire rise in the DVAR of the country’ 51 The 95% con…dence interval of DV \ ARit is (11.2%, 13.6%), which overlaps with (11.3%, 18.0%), the 95% con…dence interval of the 2007 …xed e¤ect in Table 4. 37 These …ndings resonate well with the existing IO table-based studies, such as Koopman, Wang, and Wei (2012). We exploit our …rm-level data to con…rm that the increase in the DVAR not only exists within industries, but also within …rms. Neither reallocation of resources across industries nor …rm entry and exit contributes to the increase in the DVAR of aggregate exports. Firm- level regressions show that the rising DVAR is due to an active substitution of domestic for imported materials by individual processing exporters. Such substitution is revealed at both the intensive margin, represented by a lower cost share of imported materials, and the extensive margin, manifested by decreasing import varieties. Behind this substitution is a continuous decline in the relative prices of domestic to imported input varieties. s DVAR and We build a simple model to analyze the time-series determinants of a …rm’ show that during the sample period, the continuous tari¤ reduction facing upstream …rms and the rising FDI since 2000 have contributed signi…cantly to the increase in domestic input varieties and thus the decline in their prices. These micro-level …ndings provide comprehen- sive explanations about how Chinese exporters have expanded their activities along global value chains away from the …nal stages of production. They also highlight that trade and s DVAR, through input-output linkages and FDI liberalization may actually raise a country’ spillovers that go beyond the targeted industries. While it is beyond the scope of the current paper, our approach is general enough to examine the micro-foundation and mechanism of a host of interesting economic issues. It can be used to study the relationship between …rm DVAR and productivity, and shed light on the desirability for a developing nation to promote high value-added exports as a growth strategy. It can also be used to assess the validity of the proposal for emerging markets to “move up the value chains”or to raise the DVAR. Regarding the last remark, we have started a new project on measuring the DVAR for a 38 wide range of countries, based on the matched importer-exporter customs transaction data s Exporter Dynamic Database. Preliminary results show an upward from the World Bank’ trend in the DVAR for countries such as Bangladesh, Guatemala, Madagascar and Morocco. One common trait of these countries is their conducive trade and FDI policies that allow their participation in global value chains, similar to the case of China. This is clearly a promising avenue for future research. References [1] Ahmad, Nadim, Sonia Araujo, Alessia Lo Turco, and Daniela Maggioni (2013). “Using Trade Microdata to Improve Trade in Value-Added Measures: Proof of Concept Using Turkish Data.” in Trade in Value Added: Developing New Measures of Cross-Border Trade, Edited by Mattoo, Aaditya, Zhi Wang, and Shang-Jin Wei, World Bank. [2] Amiti, Mary, Oleg Itskhoki and Jozef Konings (2014). “Importers, Exporters and Ex- change Rate Disconnect,” American Economic Review, 104(7): 1942-1978. [3] Antràs, Pol, Davin Chor, Thibault Fally, and Russell Hillberry (2012). “Measuring the Upstreamness of Production and Trade Flows.”American Economic Review Papers and Proceedings 102, no. 3: 412-416. [4] Antràs, Pol, and Davin Chor (2013). “Organizing the Global Value Chain.” Economet- rica 81, no. 6: 2127–2204. [5] Autor, David, David Dorn, and Gordon Hanson (2013). “The China Syndrome: Local Labor Market E¤ects of Import Competition in the United States,”American Economic Review 103, no. 6: 2121-2168. 39 s Second Un- [6] Baldwin, Richard (2012) “Trade and Industrialization after Globalization’ bundling: How Building and Joining a Supply Chain are Di¤erent and Why it Matters.” in Globalization in an Age of Crisis: Multilateral Economic Cooperation in the Twenty- First Century. University of Chicago Press. [7] Blackorby, Charles and Robert Russell (1989). “Will the Real Elasticity of Substitution Please Stand Up? (A Comparison of the Allen/Uzawa and Morishima Elasticities),” American Economic Review 79(4), 882-888. [8] Blaum, Joaquin, Claire Lelarge, and Michael Peters (2014). "Estimating the Produc- tivity Gains from Importing." Yale University Working Paper. [9] Brandt, Loren, Johannes Van Biesebroeck, Luhang Wang, and Yifan Zhang (2015). “WTO Accession and Performance of Chinese Manufacturing Firms,” Working Paper, University of Toronto. [10] Broda, Christian and David Weinstein (2006). “Globalization and the Gains from Va- riety,” Quarterly Journal of Economics, 121(2), 541-585. [11] Cebeci, Tolga, Fernandes, Ana, Freund, Caroline. and Martha Pierola (2012). “Exporter Dynamics Database,”World Bank Policy Research Working Paper 6229. [12] Constantinescu, C, Aaditya Mattoo and Michele Ruta (2015) “The Global Trade Slow- down: Cyclical or Structural?”World Bank Policy Research Working Paper, WPS 7158. [13] De La Cruz, Justino, Robert B. Koopman, Zhi Wang, and Shang-Jin Wei. (2011). “Es- s Manufacturing Exports,” US International timating Foreign Value-added in Mexico’ Trade Commission Working Paper 2011-04A. 40 [14] Fernandes, Ana and Heiwai Tang (2012). “Determinants of Vertical Integration in Ex- port Processing: Theory and Evidence from China,” Journal of Development Eco- nomics, 99(2), 396-414. [15] Feenstra, Robert (1998) “Integration of Trade and Disintegration of Production in the Global Economy,” Journal of Economic Perspectives, 12(4): 31-50. [16] Feenstra, Robert and Hiau Looi Kee (2008). “Export Variety and Country Productiv- ity: Estimating the Monopolistic Competition Model with Endogenous Productivity,” Journal of International Economics, 74(2), 500–518. [17] Goldberg, Pinelopi, Amit Khandelwal, Nina Pavcnik, and Petia Topalova (2010). “Im- ported Intermediate Inputs and Domestic Product Growth: Evidence from India,” Quarterly Journal of Economics, 125 (4), 1727-1767. [18] Hummels, David, Jun Ishii and Kei-mu Yi (2001). “The Nature and Growth of Vertical Specialization in World Trade,” Journal of International Economics, 54:75-96. [19] Johnson, Robert (2014). Five Facts about Value-Added Exports and Implications for Macroeconomics and Trade Research, Journal of Economics Perspectives, 28 (2). [20] Johnson, Robert and Guillermo Noguera (2012). “Accounting for Intermediates: Pro- duction Sharing and Trade in Value Added,” Journal of International Economics, 86 (2), 224-236. [21] Johnson, Robert and Guillermo Noguera (2014). “A Portrait of Trade in Value Added over Four Decades.”Dartmouth College Working Paper. [22] Kee, Hiau Looi (2015). “Local intermediate inputs and the shared supplier spillovers of foreign direct investment.” Journal of Development Economics, 112, 56-71. 41 [23] Kee, Hiau Looi, Nicita, Alessandro, & Olarreaga, Marcelo (2008). “Import demand elasticities and trade distortions.” Review of Economics and Statistics, 90(4), 666-682. [24] Koopman, Robert, Zhi Wang, and Shang-Jin Wei (2012). “Estimating Domestic Content in Exports When Processing Trade Is Pervasive,” Journal of Development Economics, 99:1, pp.178-89. [25] Koopman, Robert, Zhi Wang, and Shang-Jin Wei (2014). “Tracing Value-Added and Double Counting in Gross Exports.” American Economic Review, 104(2): 459-94. [26] Liu, Xuepeng (2013). “Tax avoidance through re-imports: The case of redundant trade,” Journal of Development Economics, 104: 152-164. [27] Ma, Yue, Heiwai Tang, and Yifan Zhang (2014). “Factor Intensity, Product Switching, and Productivity: Evidence from Chinese Exporters,” Journal of International Eco- nomics, 92 (2), 349-362. [28] Ministry of Commerce of China Regulations Concerning Customs Supervision and Control over the Inward Processing and Assembling Operation URL: eng- lish.mofcom.gov.cn/aarticle/lawsdata/chineselaw/200211/20021100053665.html s Republic of China (2007). National Input- [29] National Bureau of Statistics of the People’ Output Survey Methods of China. [30] Pierce, Justin and Peter Schott (2015). “The Surprisingly Swift Decline of U.S. Manu- facturing Employment,”Working Paper, Yale University. [31] Rodriguez-Clare, Andres (1996). “Multinationals, Linkages, and Economic Develop- ment,” American Economic Review, 86(4), 852-873. 42 [32] Stock, James H and Motohiro Yogo (2005). “Testing for Weak Instruments in Linear IV Regression,” In: Andrews DWK Identi…cation and Inference for Econometric Models. New York: Cambridge University Press, 80-108. [33] Timmer, Marcel P., Abdul Azeez Erumban, Bart Los, Robert Stehrer, and Gaaitzen J. de Vries. (2014) “Slicing up global value chains.” Journal of Economic Perspectives, 28(2), 99-118. [34] United Nations (1999). Handbook of Input-Output Table Compilation and Analysis. [35] Wang, Zhi, Shang-Jin Wei, and Kunfu Zhu (2014). “Quantifying International Produc- tion Sharing at the Bilateral and Sector Level,”NBER Working Paper, 19677. 43 Figure 1: DVAR of Processing Exports (2000-2007), with 95% (Bootstrapped) Con…dence Intervals .6 .55 DVAR .5 .45 .4 2000 2001 2002 2003 2004 2005 2006 2007 Year Measured DVAR 95 c.i. (upper bound) 95 c.i. (lower bound) Table 1: Issues and Assumptions or Solutions Issues Assumptions or Solutions 1 Domestic content in imported materials. Negligible according to KWW (2012). 2 Imported content in domestic materials. Lower DVAR by 1.5% - 5.7%. 3 Firms import capital equipment. Remove equipment from …rm imports. 4 Firms buy imported materials from …rms. Drop excessive exporters. 5 Firms sell imported materials to other …rms. Drop excessive importers. 6 Multi-industry …rms hinder the calculation Restrict the sample to single-industry of industry DVAR. …rms. Table 2: Domestic Value Added Ratio Processing (P) Ordinary (O) Aggregate (A) Year DVAR (Filter 1) DVAR (Filter 2) DVAR (Filter 3) DVAR (Filter 1) DVAR (Filter 3) 2000 0.487 0.475 0.459 0.924 0.650 2001 0.495 0.488 0.468 0.915 0.652 2002 0.517 0.505 0.488 0.918 0.668 2003 0.502 0.494 0.478 0.914 0.661 2004 0.539 0.531 0.507 0.900 0.674 2005 0.579 0.571 0.544 0.893 0.695 2006 0.565 0.558 0.520 0.904 0.697 2007 0.599 0.587 0.548 0.900 0.701 Notes: Filter 1: Include exporters that have material > imports, exports >= imports. Filter 2: Include exporters that satisfy Filter 1 and DVAR < 50th Pct(DVARO ). Filter 3: Include exporters that satisfy Filter 1 and DVAR < 25th Pct(DVARO ). DVARA = Processing_Shr DVARP +(1-Processing_Shr) DVARO 44 Figure 2: DVAR Trend (2000-2007) by Industry with 95% (Bootstrapped) Con…dence Inter- vals 04:beverages & spirit 06:chemical products 07:plastics & rubber 08:raw hides & skins 1 .5 0 09:wood & articles * 10:pulp of wood 11:textiles 12:footwear & headgear, etc. 1 .5 0 13:stone, plaster, cement, etc. 14:precious metals 15:base metals * 16:machinery, mechnical & elec eqmt 1 .5 0 2000 2002 2004 2006 17:vehicles & aircrafts 18:optical, photographic, etc. 20:misc manufacturing 1 .5 0 2000 2002 2004 2006 2000 2002 2004 2006 2000 2002 2004 2006 year Dashed lines = 95% confidence interval. * = industries with average DVAR lower in 2007 than 2000. Figure 3: Decomposing the DVAR Growth into Within- and Between-industry Growth .1 .08 .06 .04 .02 0 -.02 2000 2001 2002 2003 2004 2005 2006 2007 year within between total change 45 s Exports to its Top 5 Trading Partners Figure 4: DVAR of China’ Germany Hong Kong, SAR Japan .55 .5 .45 DVAR of Exports 2000 2002 2004 2006 Korea United States .55 .5 .45 2000 2002 2004 2006 2000 2002 2004 2006 year Graphs by two-digit iso code s Aggregate (Processing + Ordinary) Exports Figure 5: DVAR of China’ .75 .7 .65 .6 .55 .5 .45 2000 2001 2002 2003 2004 2005 2006 2007 year DVAR (Processing + Ordinary Exp) DVAR (Processing Exp) All firms with materials < imp & exp < imp, and processing firms with DVAR>DVAR (25 percentile of Ord Exporters) are excluded. 46 Table 3: Decomposition Exercise: Firm Heterogeneity and Aggregation Bias DVAR of Total Exports Number of …rms in the sample (1) Census 0.479 (0.021) 3419 (2) Original Sample 0.478 (0.023) 2623 (3) KWW (2012) Estimates 0.408 N/A (4) Large Firms Only 0.453 (0.034) 123 Notes: With the exception of (3), all numbers are calculated by the authors based on di¤erent samples. (1) refers to the 2004 Census of Manufacturing Plants; (2) restricts the sample in (1) to the original survey dataset. (3) is the IO table-based estimate from KWW (2012); (4) restricts the sample in (2) to only …rms with total exports larger than 300 million RMB. Bootstrapped standard errors are reported in parentheses. 47 Table 4: Dependent Variable: The Ratio of Domestic Value Added in Exports to Gross Exports (DVAR) (1) (2) (3) (4) (5) (6) Sample All All Dom private Foreign Multiple Ind Un…ltered 2001 0.0301*** 0.0299*** 0.0764 0.0327*** 0.0256*** 0.0268*** (0.007) (0.006) (0.080) (0.006) (0.005) (0.005) 2002 0.0490*** 0.0493*** 0.0810 0.0492*** 0.0466*** 0.0493*** (0.004) (0.004) (0.106) (0.004) (0.006) (0.004) 2003 0.0657*** 0.0663*** 0.190** 0.0656*** 0.0709*** 0.0681*** (0.008) (0.008) (0.078) (0.008) (0.005) (0.005) 2004 0.0669*** 0.0674*** 0.140 0.0677*** 0.0749*** 0.0715*** (0.008) (0.011) (0.127) (0.011) (0.005) (0.010) 2005 0.0962*** 0.0969*** 0.198 0.0978*** 0.117*** 0.101*** (0.007) (0.009) (0.124) (0.008) (0.005) (0.010) 2006 0.135*** 0.136*** 0.257* 0.136*** 0.146*** 0.133*** (0.010) (0.011) (0.133) (0.012) (0.005) (0.010) 2007 0.147*** 0.147*** 0.300** 0.146*** 0.161*** 0.150*** (0.013) (0.017) (0.140) (0.016) (0.006) (0.014) P D M D +P I M I PY -0.0236*** -0.0234*** 0.0190 -0.0230** -0.0207*** -0.0108*** it (0.007) (0.008) (0.060) (0.010) (0.006) (0.004) wL P Y it -0.0010 0.0522 -0.0010 -0.0040 -0.0032 (0.016) (0.155) (0.017) (0.009) (0.006) N 17903 17871 858 16726 28925 31965 R-sq .0729 .0733 .104 .074 .0955 .0597 Notes: Firm and year …xed e¤ects are always included. Data set: merged NBS-customs data. Columns (1) and (2) use the whole sample; columns (3) and (4) include only domestic private and foreign-invested …rms, respectively. Column (5) includes …rms that operate in multiple industries as well. Column (6) includes single-industry …rms that do not satisfy our rules to …lter …rms that engage in indirect trade. Bootstrapped standard errors, clustered at the industry level, are reported in parentheses. * p<0.10; ** p<0.05; *** p<0.01. 48 Table 5: Dependent Variable: Share of imports in total materials Sample All Dom private Foreign Multiple Ind 2001 -0.0237** 0.0538 -0.0241** -0.0200*** (0.010) (0.047) (0.011) (0.006) 2002 -0.0278*** 0.137** -0.0293*** -0.0223*** (0.006) (0.062) (0.007) (0.007) 2003 -0.0674*** 0.0761 -0.0695*** -0.0678*** (0.007) (0.067) (0.007) (0.007) 2004 -0.0837*** 0.0813 -0.0852*** -0.0830*** (0.008) (0.061) (0.008) (0.006) 2005 -0.114*** 0.0483 -0.115*** -0.116*** (0.010) (0.066) (0.009) (0.006) 2006 -0.155*** 0.0312 -0.157*** -0.144*** (0.011) (0.081) (0.009) (0.007) 2007 -0.170*** -0.00236 -0.171*** -0.154*** (0.017) (0.086) (0.013) (0.007) wL P Y it 0.0336 0.417** 0.0328 0.0481 (0.042) (0.190) (0.039) (0.042) ln (K=L)it -0.0035 -0.0252 -0.0040 -0.0040 (0.003) (0.030) (0.003) (0.003) N 17831 858 16688 28875 R-sq .0898 .104 .0918 .0896 Note: Firm and year …xed e¤ects are always included. Data set: merged NBS and customs data. Column (1) uses the whole sample; columns (2) and (3) include only domestic private and foreign-invested …rms, respectively. Column (4) includes …rms that operate in multiple industries as well. Bootstrapped standard errors, clustered at the industry level, are reported in parentheses. * p<0.10; ** p<0.05; *** p<0.01. 49 Table 6: Dependent Variable: ln(number of import varieties) Sample All Dom private Foreign Multiple Ind 2001 -0.114*** -0.208* -0.106*** -0.134*** (0.016) (0.124) (0.018) (0.013) 2002 -0.110*** 0.216 -0.0990*** -0.128*** (0.016) (0.284) (0.016) (0.016) 2003 -0.217*** -0.0606 -0.208*** -0.240*** (0.029) (0.419) (0.026) (0.016) 2004 -0.274*** 0.186 -0.267*** -0.279*** (0.039) (0.352) (0.035) (0.015) 2005 -0.342*** 0.0535 -0.335*** -0.360*** (0.046) (0.367) (0.045) (0.016) 2006 -0.197*** 0.122 -0.183*** -0.215*** (0.054) (0.336) (0.054) (0.019) 2007 -0.351*** 0.131 -0.344*** -0.356*** (0.090) (0.345) (0.081) (0.020) P D M D +P I M I PY 0.0144 -0.106 0.0171 0.0104 it (0.025) (0.332) (0.019) (0.020) wL P Y it -0.0327 0.899 -0.0374 -0.0608 (0.038) (1.033) (0.054) (0.059) N 17871 858 16726 28925 R-sq .0571 .0609 .0589 .0565 Note: Firm and year …xed e¤ects are always included. Data set: merged NBS and customs data. Column (1) uses the whole sample; columns (2) and (3) include only domestic private and foreign-invested …rms, respectively. Column (4) includes …rms that operate in multiple industries as well. Bootstrapped standard errors, clustered at the industry level, are reported in parentheses. * p<0.10; ** p<0.05; *** p<0.01. 50 Table 7: Dependent Variable: ln(number of export varieties) Sample All Dom private Foreign Multiple Ind 2001 -0.0280 0.138 -0.0233* -0.0272 (0.022) (0.223) (0.012) (0.021) 2002 0.0599 0.318 0.0712** 0.0729*** (0.042) (0.221) (0.029) (0.020) 2003 0.103** 0.479 0.107*** 0.130*** (0.049) (0.318) (0.035) (0.018) 2004 0.124** 0.598** 0.126*** 0.161*** (0.056) (0.267) (0.039) (0.016) 2005 0.210*** 0.821*** 0.210*** 0.236*** (0.040) (0.310) (0.029) (0.019) 2006 0.286*** 0.945*** 0.283*** 0.316*** (0.050) (0.316) (0.033) (0.017) 2007 0.275*** 1.086*** 0.267*** 0.306*** (0.046) (0.338) (0.030) (0.022) P D M D +P I M I PY 0.0130 0.182 0.0110 0.0017 it (0.018) (0.266) (0.017) (0.018) wL P Y it -0.0474 -0.126 -0.0517* -0.0606* (0.059) (0.726) (0.028) (0.031) N 17871 858 16726 28925 R-sq .0399 .121 .0388 .0486 Note: Firm and year …xed e¤ects are always included. Data set: merged NBS and customs data. Column (1) uses the whole sample; columns (2) and (3) include only domestic private and foreign-invested …rms, respectively. Column (4) includes …rms that operate in multiple industries as well. Bootstrapped standard errors, clustered at the industry level, are reported in parentheses. * p<0.10; ** p<0.05; *** p<0.01. 51 Table 8: Determinants of the Within-…rm Increase in the DVAR (1) (2) (3) Dep. Var D 4t;00 DV ARjt 4t;00 ln(P I =P D )jt 4t;00 ln Vjt 4t;00 ln P I =P D jt 0.269*** (0.026) 4t;00 ln(Ejt ) (RMB appreciation) 1.479* -0.089*** (0.891) (0.031) D 4t;00 ln Vjt 17.108*** (3.177) U 4t;00 ln ejt -0.012* (0.007) 4t;00 ln (F DIjt ) 0.017*** (0.002) Industry Fixed E¤ects N 105 105 105 R-sq 0.030 0.106 0.006 4t;00 is the operator that subtracts the variable of interest from its corresponding value in 2000. Bootstrapped standard errors (with 500 repetitions) are reported in parentheses. Coe¢ cients are estimated using 3SLS. Columns (1), (2), and (3) are third, second, and …rst stages, respectively. * p<0.10; ** p<0.05; *** p<0.01. 52 Table 9: Estimated Elasticity of Substitution between Domestic and Foreign Input Varieties IV Industry sD2000 sD 2007 ID s.e. ID s.e. 2000 2007 whole sample 0.661 0.710 0.376*** (0.019) 0.351*** (0.017) 2.678 2.826 2.566 2.705 04: beverages & spirit (16-24) 0.921 0.885 0.566*** (0.211) 0.553*** (0.170) 8.779 6.561 8.600 6.434 bIV bIV b2000 b2007 06: chemical products (28-38) 0.770 0.722 0.309*** (0.072) 0.296*** (0.066) 2.745 2.539 2.671 2.475 07: plastics & rubber (39-40) 0.734 0.732 0.175*** (0.058) 0.176** (0.069) 1.896 1.892 1.901 1.897 08: raw hides & skins (41-43) 0.603 0.717 0.315*** (0.112) 0.403*** (0.110) 2.316 2.552 2.683 2.986 09: wood & articles (44-46) 0.620 0.742 0.568 (0.529) 0.283 (0.434) 3.411 3.967 2.201 2.478 10: pulp of wood (47-49) 0.769 0.793 0.506*** (0.180) 0.525*** (0.183) 3.848 4.083 3.955 4.198 11: textiles (50-63) 0.690 0.771 0.938*** (0.066) 0.891*** (0.066) 5.385 6.313 5.165 6.046 12: footwear & headgear, etc. (64-67) 0.770 0.771 0.427*** (0.059) 0.426*** (0.062) 3.411 3.418 3.405 3.413 53 13: stone, plaster, cement, etc. (68-70) 0.802 0.694 0.121 (0.121) 0.142 (0.154) 1.762 1.570 1.894 1.669 14: precious metals (71) 0.664 0.730 0.238 (0.155) 0.202 (0.166) 2.067 2.208 1.905 2.025 15: base metals (72-83) 0.882 0.768 0.292*** (0.091) 0.287** (0.118) 3.806 2.639 3.758 2.611 16: machinery, mechanical electrical & equipmt (84-85) 0.571 0.644 0.278*** (0.024) 0.273*** (0.027) 2.135 2.213 2.114 2.191 17: vehicles & aircraft (86-89) 0.580 0.852 0.405*** (0.069) 0.396*** (0.076) 2.663 4.212 2.626 4.140 18: optical, photographic, etc. (90-92) 0.713 0.728 0.284*** (0.048) 0.286*** (0.050) 2.388 2.434 2.398 2.444 20: misc manufacturing (94-96) 0.691 0.765 0.291*** (0.057) 0.335*** (0.061) 2.363 2.619 2.569 2.863 Bootstrapped standard errors (with 500 repetitions) are reported in parentheses. Firm …xed e¤ects are always included when estimating ID : ID is estimated using OLS, while IV ID is estimated using 2SLS with instruments including import-weighted exchange rates, upstream input tari¤s, and the (log) level of FDI in the same industry. See the Appendix for the details of these instruments. * p<0.10; ** p<0.05; *** p<0.01. 1 Appendix (Not for Publication) 1.1 Data Description The main data set for this paper covers the universe of Chinese import and export trans- s actions in each month between 2000 and 2007. It reports values (in US dollars) of a …rm’ exports (and imports) at the HS 8-digit level (over 7000 products) to each destination (from each source) country. We drop trading companies (intermediaries) in our sample, using the methods proposed by Ahn et al. (2013) to identify them. This level of disaggregation is the …nest for empirical studies in international trade –i.e., transactions at the …rm-product- country-month level. s export growth. Processing trade has been playing a signi…cant role in driving China’ From 2000 to 2007, processing exports have increased by over four folds from 138 billion USD to 680 billion USD with the share of processing exports in total exports held steadily around 55 percent, as shown in Figure A1. In addition, Table A1 shows that, the U.S. consistently ranked as the top destination, accounting for about 25 percent of Chinese total processing exports. Following the U.S. is Hong Kong SAR, China, which accounted for slightly over 20 percent of the total. Japan has been the third largest market for Chinese processing exports, but its prominence has declined from 18 percent in 2000 to 10 percent s top 10 export destinations, as in 2007. Processing exports are widespread among China’ seen in Figure A2. It accounted for 63 percent of Chinese exports to the U.S. in 2007 and 81 percent for Hong Kong SAR, China, the highest share among the top 10 destinations. We present in Figure ?? the share of processing exports in 2007 by industry sector, according to the United Nations groupings of HS2 categories. There exists a substantial heterogeneity in the prevalence of processing exports across industries. The share is about 20 percent for the “wood & articles” sector (HS2 = 6 -14) and is over 80 percent for the 1 “machinery, mechanical, and electrical equipment”sector (HS2 = 84-85). The advantage of focusing on processing exporters is that we need not worry about their imports for …nal consumption, as by de…nition, all imports in processing trade have to be used as intermediate inputs.52 However, not all processing exporters import for their own use. Some of them import for other processing …rms, which also implies that some processing …rms must export more than what their imported materials can support. We develop sys- tematic rules to identify processing …rms that potentially import from and export for other …rms. To this end, we merge the customs transaction data with the …rm-level data from s National Bureau of Statistics the Annual Surveys of Industrial Firms conducted by China’ (NBS hereafter). The surveys cover all state-owned enterprises (SOEs) and non-state-owned …rms that have sales above 5 million yuan in a given year.53 The NBS data contain detailed information for most of the standard balance sheet information, such as …rm ownership, output, value added, industry code (480 categories), exports, employment, original value of …xed asset, and intermediate inputs. Tables A2 and A3 present the percentages of …rms and s median of …rm sales that are covered by the merged data. Table A4 presents the industry’ materials-to-sales ratios. 1.1.1 Transforming Chinese I/O Tables to One Based on UN Industry Code s National Bureau of Statistics to match multiple IO 1. Use the concordance from China’ codes with multiple HS 6-digit codes (revision 2002). 2. Match multiple HS6 codes to multiple UN industry sector codes (20 of them). 52 Manova and Yu (2013) examine how …nancial constraints a¤ect exporters positions in global supply chains in China and thus their pro…ts. In this paper, we simply take advantage of the special features of the processing regime without getting into the details about …rms’transition from one regime to another. 53 The industry section in the o¢ cial statistical yearbooks of China is constructed based on the same data source. The unit of analysis is a …rm, and not the plant, but other information in the survey suggests that more than 95% of all observations in our sample are single-plant …rms. 5 million yuan is roughly exchanged to 600,000 US dollars during the sample period. 2 3. For each IO code, pick the UN code that has the largest number of HS6 shared. This will guarantee that all IO codes will be covered. 4. For UN codes that are matched with multiple IO codes, manually choose a unique UN code for the match. It happens in only one case. 5. Then add up the values of intermediate inputs for each pair of upstream-downstream relationship. A matrix of 20 groups by 20 groups will be built. 6. Recompute the IO coe¢ cients based on the UN industry sector classi…cation. 1.1.2 Computing Domestic Upstream Variety To compute domestic upstream variety, we use the weighted average of the number of HS6 products exported by non-processing …rms across all upstream industries as a proxy for domestic upstream varieties, since data on domestic varieties are not available. The belief is s export product scope is a subset of its domestic product scope.54 Speci…cally, that a …rm’ P we compute the weighted average of the number of upstream varieties by Vjt = I i=1 sij Vit , where sij is the share of industry i0 s goods used in total input costs of industry j , according to the Chinese input-output table for 2002. Vit is the number of HS6 products exported by non-processing …rms in industry i in year t. Since the HS classi…cations have changed twice (in 2002 and 2007, respectively) during our sample period, we use the concordance …le created by Cebeci et al. (2012) to de…ne a consistent set of varieties over time. As reported in Table A5, the number of varieties available to the downstream processing exporters is increasing over time for most industries. Some industries have systematically higher input varieties (e.g. machinery, mechanical, and electrical equipment). This industry-speci…c feature is already 54 There could be export varieties that were not sold domestically or vice versa. There could also be domestic varieties produced by non-exporters that were not exported. In these regards, our proxy should be considered as a lower bound of the number of domestic varieties. 3 controlled for by industry …xed e¤ects in the regressions. 1.1.3 Computing Upstream Input Tari¤s s upstream tari¤s involves two steps. For each upstream industry, Computing an industry’ input tari¤s are measured as a weighted average of tari¤s facing all input suppliers to that s inputs in total material cost of industry. Speci…cally, we obtain the share of industry i’ industry j , sij , from the Chinese IO table for 2002. Then for each industry j , we compute P the weighted average of input tari¤s as ejt = I i=1 sij it , where it is the average tari¤ rate for industry i in year t and I is the total number of industries. Finally, for each downstream industry k , we use the IO coe¢ cients again to compute the weighted average of upstream PI input tari¤s eUkt = j =1 sjk ejt . The idea to use the IO tables twice is that we need the measure of tari¤s facing domestic input suppliers, not downstream exporters. For example, a garment …rm uses fabrics, zippers and buttons. Fabrics …rms use cotton yarns, zipper …rms use steel, and button …rms use plastics. Thus, the upstream input tari¤ for a garment …rm is a weighted average tari¤ rates on cotton yarns, steel and plastics. 1.1.4 Computing Industry-speci…c Exchange Rate Indices We use the Tornqvist method to construct an industry-speci…c time-varying exchange rate. For each industry j , let Ijt be the set of common countries …rms in industry j import from in two consecutive years, t and t s currency price of a yuan in year 1: Denote country c’ t and t s shares in industry j 0 s total imports 1 by Ect and Ect 1 ; and denote country c’ in year t and t 1 by scjt and scj;t 1 . The industry-speci…c rate of yuan appreciation with respect to the countries from which industry j imports in year t is de…ned as X1 4 ln Ejt = (scjt + scj;t 1 ) (ln Ect ln Ec;t 1 ) : c2 I 2 jt 4 Using this weighted average of appreciation rates, we de…ne the industry-speci…c exchange rate for imports as Ejt = Ej;t 1 exp (4 ln Ejt ) ; with Ejt normalized to 1 in the base year (i.e., 2000) or any starting year for each industry. 1.1.5 Computing Industry-speci…c Domestic Input Price Indices Computing the input price indices involves two steps. First, we use the Tornqvist method to construct an industry-speci…c time-varying domestic input price indices. For each industry j (15 of them), let Ijt be the set of common sub-industries in two consecutive years, t and t s output price index in year t and t 1: Denote sub-industry s’ 1 by Pst and Ps;t 1 ; and s sales in industry j 0 s total sales in year t and t denote the share of sub-industry s’ 1 by s ! sjt and ! sj;t 1 . Data on output price indices at the 4-digit sector level (based on China’ NBS classi…cation) are obtained from Brandt et al. (2012).55 The industry-speci…c rate of output price in‡ation in year t is de…ned as X1 ejt = 4 ln P (! sjt + ! sj;t 1 ) (ln Pst ln Ps;t 1 ) : s2I 2 jt Using this weighted average of in‡ation rates, the sector-speci…c output price level is de…ned as ej;t ejt = P P 1 ejt ; exp 4 ln P ejt normalized to 1 in 2000. with P ejt , with weights equal to the The second step is to compute the weighted average of P coe¢ cients from the Chinese IO table for 2002. The goal is to compute the average domestic prices facing processing …rms in industry j . Speci…cally, for each industry j , the weighted 55 http://www.econ.kuleuven.be/public/N07057/CHINA/appendix/ 5 D PJ ekt , where ak is the share of industry k goods in average of input prices is Pjt = k=1 akj P total material costs for production of a unit of industry j goods and J is the number of D industries. Notice that Pjt ejt , since akj is varies across time purely due to the variation in P …xed throughout the sample. 1.1.6 Computing Industry-speci…c Imported Input Price Indices To compute the imported input indices, we use the Tornqvist method to construct an industry-speci…c time-varying import price indices based on …rm-level imports from the customs transaction data. For each industry j (15 of them), let Ijt be the set of common product (at the HS 8-digit level) in two consecutive years, t and t s 1: Denote product s’ import prices in year t and t 1 by pI I s imports st and ps;t 1 ; and denote the share of product s’ in industry j 0 s total imports in year t and t 1 by $sjt and $sj;t 1 . Product-level import prices (by processing …rms only) are computed as total import value divided by total quan- tity of import at the HS8 level, using customs transaction-level data. Then sector-speci…c rate of import price in‡ation in year t is de…ned as X1 ejt 4 ln P I = ($sjt + $sj;t 1 ) ln pI st ln pI s;t 1 : j 2I 2 jt Using this weighted average of in‡ation rates, the sector-speci…c import price level is de…ned as ejt P I ej;t =P I 1 ejt exp 4 ln P I ; ejt with P I normalized to 1 in 2000. Table A6 reports the ratio of the imported material price index to the domestic material price index across industry-years. 6 1.2 Theoretical Derivation of Firm DV AR (the Cobb-Douglas Case) In the main text, we derive the theoretical expression of …rm DV AR based on a translog production function. In this section, we use a more convenient form of production function –the Cobb-Douglas production function, as the basis to derive …rm DV AR. D For each year t; consider …rm i with productivity, i, which uses both domestic Mit I and imported materials Mit ; alongside capital (Kit ) and labor (Lit ) to produce output Yi , according to the following production production: Yit = i Kit K LitL Mit M ; (31) 1 1 D I 1 Mit = Mit + Mit ; (32) K + L + M = 1 and > 1: Each …rm faces input prices rt ; wt ; PtD ; PtI for capital, labor, domestic materials, and im- ported materials. Given (32) it can be shown that the price index of total materials is a constant-elasticity-of-substitution (CES) function over PtD and PtI : 1 1 1 1 PtM = PtD + PtI Firms’cost minimization implies the following total cost of producing Yit units of output: Yit rt K wt L PtM M Cit rt ; wt ; PtD ; PtI ; Yit = ; with (33) i K L M PtM Mit = M: Cit 7 Thus, the marginal cost (cit ) of producing Yit units of …nal goods is @Cit 1 rt K wt L PtM M cit = = ; (34) @Yit i K L M which is constant over output. Note that while input prices and input elasticities are common across all …rms within an industry-year, …rms have di¤erent productivity, i; which results in di¤erent marginal cost, cit ; across …rms. Then we can express the share of imported materials in total revenue as: I PtI Mit I PtI Mit PtM Mit Cit = Pit Yit PtM Mit Cit Pit Yit I PtI Mit cit = M M Pt Mit Pit I PtI Mit = M (1 it ) ; PtM Mit Pit cit where i = Pit 2 [0; 1] is the price-cost margin of the …rm.56 Finally, the share of imported materials in total cost of materials can be obtained by the following minimization problem: I min PtI Mit D + PtD Mit 1 1 D I 1 s:t: Mit = Mit + Mit : 56 Note that price-cost margin, i s markup, which is usually de…ned as is closely related to …rm’ Pit 1 i = = : cit 1 i If price equals marginal cost, as it is in the case of perfect competition, i equals 0 and i = 1: When i > 1; then i > 0: 8 Solving it gives the following ratio of imported material cost to total material cost: I PtI Mit 1 = 1: (35) PtM Mit PtI 1+ D Pt s DV AR in period t, based on (4), as We can then express …rm i’ 1 DV ARit = 1 M (1 it ) I 1: (36) Pt 1+ D Pt s DV AR can be analyzed as follows: According to (36), the determinants of a …rm’ 1. Cross-sectional distribution of the DV AR within an industry-year Given input prices and elasticities, the cross sectional distribution of DV AR within an s price-cost margin, industry-year depends on the distribution of …rm’ i, given that DV AR is an a¢ ne transformation of i. Thus, within an industry-year, a …rm with a higher i will have a higher DV AR. Factors that a¤ect the price-cost margin will therefore a¤ect …rm DV AR: Perfect Competition If the industry is perfectly competitive, it = 0; 8i; t; the cross-sectional distribution of DV AR degenerates to the following constant that does not vary across …rms: 1 DV ARit = 1 M I 1; 8i; t: Pt 1+ D Pt Monopolistic Competition with CES preferences Under monopolistic competition with CES preferences, it = ; 8i; since markup is con- stant across all …rms, the cross-sectional distribution of DV AR degenerates to the following 9 constant that also does not vary across …rms within the same industry: 1 DV ARit = 1 M (1 ) I 1; 8i; t: Pt 1+ D Pt Note that the cross-sectional distribution of DV AR does not depend on the distribution of …rm productivity under CES preferences, as long as markup is constant across …rms. Empirically, if we observe varying DV AR across …rms within the same industry-year, it indicates that the CES preference assumption is not supported and that the industry is likely not perfectly competitive. 2. Time-series movement of DV AR within …rms Eq. (36) shows that the time-series movement of DV AR is determined by the price of PtI imported inputs to domestic inputs, D Pt , which is common across …rms within the same PtI industry-year. Factors that a¤ect D Pt s DV AR over time. It is worth will a¤ect a …rm’ PtI emphasizing that factors that do not a¤ect D Pt s wages (w) or directly, such as the …rm’ productivity ( i ), do not directly a¤ect the time-series movement of DV AR within …rms.57 References [1] Ahn, J., A. Khandelwal, and S.J. Wei (2011) “The Role of Intermediaries in Facilitating Trade,” Journal of International Economics, vol 84, 73–85. [2] Cebeci, Tolga, Fernandes, Ana, Freund, Caroline. and Martha Pierola (2012). “Exporter Dynamics Database,”World Bank Policy Research Working Paper 6229. 57 Domestic wages can still indirectly a¤ect …rm DV AR through a¤ecting the price of domestic materials. In the regression analysis below, controlling for the relative price of materials, we should expect no impact from wages on …rm DV AR. 10 [3] Koopman, Robert, Zhi Wang, and Shang-Jin Wei (2012). “Estimating Domestic Content in Exports When Processing Trade Is Pervasive,” Journal of Development Economics, 99:1, pp.178-89. [4] Manova, Kalina and Zhihong Yu (2013). “Firms and Credit Constraints along the Global Value Chain: Processing Trade in China,”NBER Working Paper 18561. 11 s Processing Exports, 2000-2007 Figure A1: Share of China’ 1200 1000800 Billion USD .56 600 .53 .55 400 .55 .55 200 .55 .55 .55 2000 2001 2002 2003 2004 2005 2006 2007 year Total Exports Processing Exports Share of Processing Exports 12 s Top 10 Export Destinations (2000 & Figure A2: Shares of Processing Exports in China’ 2007) .8 Share of Processing in Exports .2 .4 0 .6 US HK JP KR DE NL GB SG TW IT 2000 2007 Figure A3: Shares of Processing Exports by Industry Sector (2007) wood & articles (44-46) stone & plaster (68-70) chemical products (28-38) base metals (72-83) beverages & spirit (16-24) textiles (50-63) raw hides & skins (41-43) misc manu. (94-96) footwear & headgear (64-67) vehicles & aircrafts (86-89) precious metals (71) pulp of wood (47-49) plastics & rubber (39-40) machinery/ mechanical/ elec equip (84-85) optical, photographic (90-92) 0 .2 .4 .6 .8 Share of Processing in Exports Figure A4: DVAR of Processing Exports - Di¤erent Filtered Samples (2000-2007) .6 .55 .5 .45 2000 2001 2002 2003 2004 2005 2006 2007 year Filter: Exp >= Imp & Material >= Imp Filter + DVAR < 25% DVAR_Ord Filter + DVAR < Med DVAR_Ord 14 Figure A5: Export Share of the Two Types Processing (2000-2007) .8 .6 exp_shr .4 .2 2000 2002 2004 2006 year Import and Assembly Pure Assembly Figure A6: DVAR of Processing Exports (Multi-industry Firms, 2000-2007) .6 .55 DVAR .5 .45 .4 2000 2001 2002 2003 2004 2005 2006 2007 Year Measured DVAR 95 c.i. (upper bound) 95 c.i. (lower bound) 15 Figure A7: DVAR of Aggregate Exports (Single-industry Firms, 2000-2007) .75 .7 DVAR .65 .6 2000 2001 2002 2003 2004 2005 2006 2007 Year Measured DVAR (Aggregate Exp) 95 c.i. (upper bound) 95 c.i. (lower bound) s Processing Exports Table A1: Top 10 Destinations of China’ 2000 2007 Rank USD (Bil) USD (Bil) 1 United States 35.17 United States 152.51 2 Hong Kong SAR, China 31.02 Hong Kong SAR, China 150.00 3 Japan 23.17 Japan 60.25 4 Germany 5.62 Netherlands 29.08 5 Korea, Republic of 5.34 Germany 29.00 6 Netherlands 3.90 Korea, Republic of 26.70 7 United Kingdom 3.90 Singapore 19.04 8 Singapore 3.62 United Kingdom 17.41 9 Taiwan, China 2.92 Taiwan, China 13.22 10 France 2.10 France 11.81 Source: China’s Customs Trade Data. 17 Table A2: Representation of Di¤erent Subsamples by Numbers of Exporters Industry Number of Firm-year Observations customs merged w/ NBS % of customs …ltered % of customs 04:beverages & spirit (16-24) 830 356 42.89 257 30.96 06:chemical products (28-38) 2278 920 40.39 410 18.00 07:plastics & rubber (39-40) 7139 2656 37.20 1190 16.67 08:raw hides & skins (41-43) 3472 1242 35.77 678 19.53 09:wood & articles (44-46) 637 169 26.53 77 12.09 10:pulp of wood (47-49) 2570 1204 46.85 337 13.11 11:textiles (50-63) 20054 7619 37.99 4806 23.97 18 12:footwear & headgear, etc. (64-67) 4776 2158 45.18 1329 27.83 13:stone, plaster, cement, etc. (68-70) 993 401 40.38 226 22.76 14:precious metals (71) 1826 446 24.42 219 11.99 15:base metals (72-83) 4278 1725 40.32 786 18.37 16:machinery, mech, elect eqmt (84-85) 22574 9420 41.73 4986 22.09 17:vehicles & aircraft (86-89) 1281 627 48.95 405 31.62 18:optical, photographic, etc. (90-92) 3498 1211 34.62 810 23.16 20:misc manufacturing (94-96) 5376 1954 36.35 1391 25.87 Total 81582 32108 39.36 17907 21.95 Source: China’ s Customs Trade Data and National Bureau of Statistics (NBS) Manufacturing Survey. Sections 1, 2, 3, 5, and 19 are non-manufacturing sectors and are excluded from the analysis. Sample pooled across 2000-2007. Table A3: Representation of Di¤erent Subsamples By Export Values Industry Sales (million usd) customs (mil usd) merged % of customs …ltered % of customs 04:beverages & spirit (16-24) 1447 1042 72.02 822 56.78 06:chemical products (28-38) 4401 2584 58.71 1308 29.72 07:plastics & rubber (39-40) 14156 9535 67.36 6331 44.72 08:raw hides & skins (41-43) 6639 4199 63.25 1843 27.77 09:wood & articles (44-46) 718 434 60.48 217 30.17 10:pulp of wood (47-49) 2760 1923 69.66 1130 40.93 11:textiles (50-63) 42272 29606 70.04 20168 47.71 19 12:footwear & headgear, etc. (64-67) 18123 13333 73.57 10567 58.31 13:stone, plaster, cement, etc. (68-70) 1575 1133 71.92 706 44.82 14:precious metals (71) 13299 9838 73.97 1616 12.15 15:base metals (72-83) 12562 6439 51.25 4166 33.16 16:machinery, mech, elect eqmt (84-85) 223527 151238 67.66 102399 45.81 17:vehicles & aircraft (86-89) 25232 19782 78.40 17525 69.45 18:optical, photographic, etc. (90-92) 10041 8039 80.06 4155 41.38 20:misc manufacturing (94-96) 13514 9050 66.97 6690 49.50 Total 390268 268173 68.72 179641 46.03 Source: China’ s Customs Trade Data and National Bureau of Statistics (NBS) Manufacturing Survey. Sections 1, 2, 3, 5, and 19 are non-manufacturing sectors and are excluded from the analysis. Sample pooled across 2000-2007. Table A4: Median of Materials to Sales Ratio by Industry and Year Industry Sector Year 2000 2001 2002 2003 2004 2005 2006 2007 04:beverages & spirit (16-24) 0.785 0.774 0.779 0.724 0.833 0.784 0.797 0.774 06:chemical products (28-38) 0.813 0.824 0.777 0.790 0.814 0.771 0.787 0.772 07:plastics & rubber (39-40) 0.806 0.791 0.791 0.799 0.830 0.806 0.798 0.798 08:raw hides & skins (41-43) 0.806 0.810 0.788 0.766 0.772 0.792 0.763 0.741 09:wood & articles (44-46) 0.801 0.788 0.769 0.741 0.776 0.801 0.796 0.815 10:pulp of wood (47-49) 0.800 0.796 0.778 0.785 0.818 0.799 0.769 0.771 11:textiles (50-63) 0.791 0.782 0.770 0.771 0.769 0.758 0.753 0.736 12:footwear & headgear, etc. (64-67) 0.795 0.778 0.754 0.770 0.763 0.745 0.749 0.720 13:stone, plaster, cement, etc. (68-70) 0.795 0.768 0.735 0.777 0.750 0.777 0.739 0.753 14:precious metals (71) 0.780 0.754 0.739 0.749 0.744 0.711 0.724 0.762 15:base metals (72-83) 0.826 0.817 0.797 0.782 0.812 0.791 0.787 0.810 16:machinery, mech, elect & eqmt (84-85) 0.800 0.803 0.773 0.773 0.804 0.796 0.780 0.780 17:vehicles & aircraft (86-89) 0.811 0.829 0.800 0.776 0.811 0.787 0.809 0.788 18:optical, photographic, etc. (90-92) 0.806 0.785 0.750 0.759 0.773 0.753 0.753 0.727 20:misc manufacturing (94-96) 0.796 0.776 0.757 0.764 0.783 0.755 0.758 0.761 Source: China’s Customs Trade Data and National Bureau of Statistics Manufacturing Survey. 20 Table A5: Upstream Variety Counts Industry Sector Year 2000 2001 2002 2003 2004 2005 2006 2007 01:live animals (1-5) 287.7 288.2 292.1 289.9 291.2 293.5 291.5 293.4 02:vegetables (6-14) 333.4 335.0 340.4 339.2 340.9 344.1 342.3 342.8 03:animal or vegetable oil (15) 294.2 294.5 297.9 295.6 296.4 299.2 297.3 298.0 04:beverages & spirit (16-24) 307.3 308.4 313.3 311.7 313.2 316.2 314.3 315.3 05:mineral products (25-27) 253.5 256.0 258.9 261.2 262.6 265.2 266.5 265.4 06:chemical products (28-38) 304.5 307.4 312.4 313.5 315.5 318.6 319.8 316.9 07:plastics & rubber (39-40) 263.6 263.6 268.4 268.1 270.9 273.2 273.6 272.1 08:raw hides & skins (41-43) 308.1 309.1 312.8 310.8 312.1 314.5 314.1 312.2 09:wood & articles (44-46) 186.2 188.2 192.3 192.0 194.1 195.2 193.6 193.2 10:pulp of wood (47-49) 202.6 205.3 207.3 209.4 209.6 213.3 210.8 209.8 11:textiles (50-63) 445.7 447.2 452.0 449.6 452.3 454.4 453.1 451.8 12:footwear & headgear, etc. (64-67) 374.5 374.6 378.6 376.5 379.5 381.1 380.4 378.6 13:stone, plaster, cement, etc. (68-70) 282.2 284.2 288.9 289.9 292.3 294.6 295.5 293.6 14:precious metals (71) 310.3 313.5 319.3 320.1 323.8 326.3 326.9 324.4 15:base metals (72-83) 348.7 352.7 359.5 361.0 366.4 369.0 370.4 367.4 16:machinery, mech, elect eqmt (84-85) 447.6 450.9 456.3 457.6 461.6 463.6 464.3 462.9 17:vehicles & aircraft (86-89) 296.4 297.3 302.6 304.7 308.1 309.4 310.9 311.0 18:optical, photographic, etc. (90-92) 421.6 424.6 430.7 430.9 435.7 437.6 438.3 435.8 20:misc manufacturing (94-96) 326.8 328.5 333.5 333.0 336.6 338.4 338.5 336.4 Source: China’ s Customs Trade Data and National Bureau of Statistics Manufacturing Survey. Each variety is de…ned as a HS-6 digit product. 21 Table A6: Price Index of Imported Materials/ Price Index of Domestic Materials Industry Sector Year 2000 2001 2002 2003 2004 2005 2006 2007 04:beverages & spirit (16-24) 1 0.980 0.975 1.075 1.067 1.092 1.187 1.220 06:chemical products (28-38) 1 0.981 1.028 1.145 1.219 1.385 1.564 1.657 07:plastics & rubber (39-40) 1 0.997 1.053 1.139 1.183 1.288 1.418 1.526 08:raw hides & skins (41-43) 1 1.000 0.997 1.098 1.125 1.192 1.279 1.355 09:wood & articles (44-46) 1 0.960 0.991 1.077 1.112 1.162 1.233 1.262 10:pulp of wood (47-49) 1 0.998 1.024 1.116 1.168 1.241 1.332 1.486 11:textiles (50-63) 1 0.995 1.004 1.087 1.108 1.153 1.228 1.253 12:footwear & headgear, etc. (64-67) 1 0.994 1.019 1.101 1.150 1.234 1.328 1.396 13:stone, plaster, cement, etc. (68-70) 1 0.996 1.007 1.095 1.197 1.356 1.510 1.659 14:precious metals (71) 1 0.985 0.960 1.048 1.094 1.208 1.316 1.403 15:base metals (72-83) 1 0.978 0.991 1.043 1.112 1.256 1.403 1.488 16:machinery, mech, elect eqmt (84-85) 1 1.021 1.115 1.237 1.305 1.431 1.572 1.890 17:vehicles & aircraft (86-89) 1 1.044 1.053 1.136 1.245 1.390 1.547 1.890 18:optical, photographic, etc. (90-92) 1 1.015 1.120 1.299 1.416 1.541 1.672 2.022 20:misc manufacturing (94-96) 1 0.992 1.009 1.105 1.175 1.286 1.413 1.563 Source: China’s Customs Trade Data and National Bureau of Statistics Manufacturing Survey. Both prices are normalized to 1 for year 2000. Table A7: Percentage of Foreign Content in Domestic Materials Industry Sector Year 2000 2001 2002 2003 2004 2005 2006 2007 04:beverages & spirit (16-24) 0.727 0.795 0.960 1.176 1.560 2.032 2.029 2.084 06:chemical products (28-38) 0.670 0.744 0.921 1.151 1.595 2.134 2.183 2.318 07:plastics & rubber (39-40) 0.386 0.433 0.544 0.691 0.975 1.312 1.374 1.466 08:raw hides & skins (41-43) 0.718 0.788 0.972 1.210 1.652 2.169 2.210 2.291 09:wood & articles (44-46) 1.110 1.209 1.465 1.826 2.518 3.353 3.352 3.493 10:pulp of wood (47-49) 0.892 1.012 1.286 1.680 2.389 3.211 3.374 3.549 11:textiles (50-63) 1.058 1.163 1.443 1.800 2.436 3.226 3.288 3.426 12:footwear & headgear, etc. (64-67) 0.927 1.027 1.290 1.631 2.263 3.023 3.133 3.293 13:stone, plaster, cement, etc. (68-70) 1.204 1.338 1.662 2.094 2.944 3.967 4.103 4.381 14:precious metals (71) 0.918 1.024 1.276 1.607 2.249 3.053 3.188 3.450 15:base metals (72-83) 1.146 1.282 1.602 2.026 2.857 3.907 4.122 4.511 16:machinery, mech, elect eqmt (84-85) 1.089 1.230 1.544 1.974 2.737 3.689 3.939 4.375 17:vehicles & aircraft (86-89) 1.414 1.586 1.981 2.528 3.564 4.855 5.134 5.657 18:optical, photographic, etc. (90-92) 0.730 0.820 1.028 1.311 1.829 2.466 2.617 2.877 20:misc manufacturing (94-96) 1.015 1.129 1.412 1.787 2.502 3.366 3.513 3.759 Source: From Koopman, Wang, and Wei (2012) and authors’imputation based on the growth rate of the number of ordinary importers 22 Table A8: 25th-percentile of Ordinary Exporters’DVAR by Industry and Year Industry Sector Year 2000 2001 2002 2003 2004 2005 2006 2007 04:beverages & spirit (16-24) 0.909 0.928 0.897 0.884 0.876 0.922 0.911 0.931 06:chemical products (28-38) 0.880 0.906 0.895 0.942 0.880 0.914 0.904 0.915 07:plastics & rubber (39-40) 0.811 0.862 0.853 0.838 0.795 0.845 0.849 0.848 08:raw hides & skins (41-43) 0.792 0.846 0.876 0.894 0.870 0.792 0.803 0.777 09:wood & articles (44-46) 0.820 0.848 0.855 0.878 0.859 0.898 0.870 0.901 10:pulp of wood (47-49) 0.804 0.850 0.826 0.873 0.775 0.946 0.893 0.895 11:textiles (50-63) 0.802 0.852 0.855 0.873 0.858 0.890 0.893 0.891 12:footwear & headgear, etc. (64-67) 0.756 0.789 0.792 0.855 0.804 0.870 0.823 0.888 13:stone, plaster, cement, etc. (68-70) 0.942 0.889 0.912 0.907 0.861 0.876 0.877 0.892 14:precious metals (71) 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 15:base metals (72-83) 0.851 0.861 0.896 0.916 0.876 0.917 0.926 0.953 16:machinery, mech, elect eqmt (84-85) 0.830 0.833 0.841 0.893 0.836 0.900 0.910 0.915 17:vehicles & aircraft (86-89) 0.944 0.971 0.978 0.967 0.943 0.980 0.982 0.989 18:optical, photographic, etc. (90-92) 0.808 0.867 0.843 0.882 0.897 0.901 0.915 0.915 20:misc manufacturing (94-96) 0.730 0.804 0.892 0.901 0.899 0.912 0.932 0.923 Source: China’s Customs Trade Data and National Bureau of Statistics Manufacturing Survey. Table A9: DVAR by Industry and Year Industry Sector Year 2000 2001 2002 2003 2004 2005 2006 2007 04:beverages & spirit (16-24) 0.650 0.685 0.699 0.694 0.725 0.680 0.732 0.750 06:chemical products (28-38) 0.386 0.463 0.500 0.481 0.384 0.452 0.564 0.443 07:plastics & rubber (39-40) 0.418 0.458 0.364 0.403 0.357 0.507 0.417 0.443 08:raw hides & skins (41-43) 0.426 0.343 0.410 0.418 0.504 0.525 0.531 0.573 09:wood & articles (44-46) 0.438 0.604 0.445 0.289 0.552 0.594 0.347 0.390 10:pulp of wood (47-49) 0.304 0.401 0.395 0.393 0.452 0.547 0.562 0.515 11:textiles (50-63) 0.495 0.464 0.525 0.546 0.558 0.599 0.620 0.561 12:footwear & headgear, etc. (64-67) 0.590 0.571 0.613 0.663 0.628 0.657 0.686 0.693 13:stone, plaster, cement, etc. (68-70) 0.550 0.517 0.538 0.617 0.587 0.504 0.530 0.554 14:precious metals (71) 0.248 0.262 0.094 0.306 0.531 0.291 0.504 0.528 15:base metals (72-83) 0.525 0.468 0.545 0.477 0.556 0.356 0.426 0.491 16:machinery, mech, elect eqmt (84-85) 0.402 0.428 0.467 0.436 0.489 0.540 0.479 0.529 17:vehicles & aircraft (86-89) 0.501 0.657 0.507 0.628 0.554 0.617 0.721 0.767 18:optical, photographic, etc. (90-92) 0.469 0.530 0.509 0.529 0.463 0.574 0.641 0.558 20:misc manufacturing (94-96) 0.617 0.572 0.599 0.606 0.620 0.584 0.663 0.650 Source: China’s Customs Trade Data and National Bureau of Statistics Manufacturing Survey DVAR is computed using single-industry …rm sample and Filter 2 stated in Table 2. 23 Table A10: Characteristics of Exiting Exporters (1) (2) (3) (4) (5) (6) (7) (8) Dep Var Exitt statet 1 DV ARt 1 ln(salest 1 ) ln(expt 1 ) statet 1 0.0680* 0.0606* 0.0686* 0.0617 (0.039) (0.035) (0.039) (0.038) DV ARt 1 0.101*** 0.108*** (0.015) (0.014) ln(salest 1 ) -0.0035* -0.0039 (0.002) (0.003) 24 ln(expt 1 ) -0.0088*** -0.0073*** (0.002) (0.002) Exitt 0.0035** 0.0374*** -0.0461* -0.151*** (0.002) (0.005) (0.026) (0.028) Controls Industry-Year Fixed E¤ects N 15271 15274 15271 15274 15274 15304 15299 15304 R2 .0737 .075 .0702 .0711 .0148 .0828 .0944 .0753 Note: Industry-year …xed e¤ects are always included. Data set: merged NBS and customs data. Columns (1)-(4) examine the relation between the (lagged) …rm characteristics and the probability of exits. Columns (5) and (8) examine the characteristics of exiting …rms. Bootstrapped standard errors are in parentheses. * p<0.10; ** p<0.05; *** p<0.01. Table A11: Import and Assembly versus Pure Assembly (1) (2) (3) (4) (5) (6) (7) (8) Dep. Var. DVAR Imp/Material ln(Exp Variety) ln(Imp variety) Sample: IA PA IA PA IA PA IA PA Year Dummies: 2001 0.0298*** 0.0237 -0.0232*** 0.00386 -0.123*** -0.0524 -0.0366* -0.0264 (0.007) (0.032) (0.007) (0.033) (0.019) (0.101) (0.022) (0.088) 2002 0.0494*** 0.0422 -0.0295*** 0.0359 -0.114*** -0.0604 0.0601** 0.0335 (0.008) (0.034) (0.007) (0.035) (0.020) (0.106) (0.024) (0.093) 2003 0.0682*** 0.0618* -0.0700*** 0.00539 -0.224*** -0.0959 0.101*** 0.119 (0.007) (0.034) (0.007) (0.038) (0.021) (0.107) (0.023) (0.093) 2004 0.0706*** 0.0486 -0.0917*** 0.0271 -0.286*** -0.133 0.118*** 0.217** (0.008) (0.032) (0.008) (0.044) (0.022) (0.106) (0.024) (0.096) 2005 0.0980*** 0.100*** -0.118*** -0.0290 -0.349*** -0.221** 0.203*** 0.228** (0.008) (0.034) (0.009) (0.047) (0.024) (0.107) (0.025) (0.105) 2006 0.140*** 0.132*** -0.161*** -0.0467 -0.202*** -0.136 0.283*** 0.285*** (0.008) (0.038) (0.010) (0.045) (0.025) (0.106) (0.029) (0.102) wL P Y it -0.0044 0.0009 0.0270 0.251* -0.0343 -0.231 -0.0417 0.0059 (0.016) (0.065) (0.052) (0.136) (0.055) (0.226) (0.037) (0.241) P D M D +P I M I PY -0.0247*** 0.0073 0.0143 -0.0867 0.0097 -0.123 it (0.009) (0.058) (0.025) (0.164) (0.026) (0.224) ln(K=L)it -0.0037 -0.0071 (0.005) (0.011) N 13062 1744 13040 1733 13062 1744 13062 1744 R2 .0686 .0459 .0867 .0579 .0647 .0208 .0419 .0372 Note: Firm and year …xed e¤ects are always included. Data set: merged NBS-customs data. IA and PA stand for import and assembly and pure assembly, respectively. Columns (1) and (2) use …rm DVAR as the dependent variable; columns (3) and (4) use …rm imports-to-materials ratio as the dependent variable; columns (5) and (6) use log of the …rm’ s export varieity as the dependent variable; columns (7) and (8) use log of the …rm’s export varieity as the dependent variable. Bootstrapped standard errors are in parentheses. * p<0.10; ** p<0.05; *** p<0.01. 25 Table A.12: Products that used to be imported by processing exporters but not exported by ordinary exporters in 2000 Rank HS6 (96) Description Imp00 Exp07 % Exp07 by FIE 1 740200 Unre…ned copper; copper anodes 94775.05 1.785 1.5 2 530121 Broken and scutched 69219.71 73.338 0.0 3 740311 Re…ned copper - Cathododes 52945.12 115.669 0.0 4 510130 Carbonised 47167.51 4099.934 19.2 5 291733 Aromatic polycarboxylic acids 22195.56 71.764 63.5 6 740321 Copper alloys - Copper-zinc base alloys 13405.72 21.957 5.0 7 710610 Powder 10303.45 6269.82 47.3 8 291412 Acyclic ketones without oxygen function 9354.077 20100.525 13.8 9 740329 Other copper alloys 8589.997 250.009 1.2 10 410122 Other hides and skins of bovine animals 7923.013 409.437 91.7 11 30375 Other …sh, excluding livers and roes 7108.482 403.583 18.1 12 470720 Other paper or paperboard 5220.848 57.024 0.0 13 750712 Tubes and pipes - of nickel 4757.735 1073.887 1.5 14 750511 Bars, rods and pro…les, of nickel 4255.77 87.14 0.0 15 721113 Not further worked than hot-rolled 3560.055 1737.362 0.0 16 400260 Isoprene rubber (IR) 3206.528 2492.855 0.6 17 870423 Other, with compression-ignition 2527.633 796856.69 8.4 18 481031 Kraft paper and paperboar 2410.466 2424.858 2.1 19 370120 Instant print …lm 2332.919 351.927 0.0 20 370256 Other …lm, for colour photography 2135.713 55.455 0.0 21 722530 Other, not further worked 2130.281 69535.009 10.3 22 40110 Of a fat content 2022.768 0.023 100.0 23 40410 Whey and modi…ed whey 1992.98 0.71 0.0 24 721020 Plated or coated with lead 1506.084 2511.163 0.9 25 540342 Other yarn, multiple or cabled 1413.818 80.048 7.3 26 530129 Flax, broken, scutched, hackled - other 1163.462 135.442 49.7 27 370510 For o¤set reproduction 1067.683 91.158 10.4 28 740312 Re…ned copper - Wire-bars 1028.783 0.455 100.0 29 370231 Other …lm, without perforations 888.111 38.389 0.0 30 480240 Wallpaper base 772.938 6382.673 28.0 31 80221 Hazelnuts or …lberts 617.869 5.9 0.0 32 50710 Ivory; ivory powder and waste 540.557 20.158 0.0 33 151329 Palm kernel or babassu oil 445.65 24.453 99.4 34 80211 Almonds - In shell 376.58 3.5 0.0 35 890392 Motorboats, other than outboard 360 607.729 0.0 36 841013 Hydraulic turbines and water wheels 300 2133.552 0.0 t o be continued to the next page 26 Rank HS6 (96) Description Imp00 Exp07 % Exp07 by FIE 37 293211 Compounds containing unfused furan ring 298.517 3480.953 62.8 38 30541 Smoked …sh, including …lletsi 268.626 51.527 15.2 39 290121 Unsaturated - Ethylene 228.697 53980.444 62.4 40 720450 Remelting scrap ingots 213.786 0.15 0.0 41 320120 Wattle extract 186.009 4.052 61.1 42 330112 Essential oils of citrus fruit 182.584 216.775 14.5 43 180320 Wholly or partly defatted 132.859 3.155 100.0 44 220860 Vodka 70.474 110.711 83.5 45 382313 Industrial monocarboxylic fatty acids 60.583 58.399 0.0 46 151229 Cotton-seed oil and its fractions 51.215 1788.796 55.8 47 520625 Single yarn, of combed …bres 50.501 721.513 1.0 48 470319 Unbleached - Non-coniferous 40.203 97.423 0.0 49 271129 In gaseous state - Other 39.653 14.256 18.4 50 722720 Of silico-manganese steel 37.912 48480.139 17.5 51 180310 Not defatted 37.019 1449.275 51.3 52 550520 Of arti…cial …bres 33.626 195.591 7.0 53 150300 Lard stearin, lard oil, oleostearin 32.134 1.57 100.0 54 20319 Fresh or chilled - Other 28.441 25052.286 0.0 55 292213 Amino-alcohols, their ethers and esters 25.68 58.781 0.0 56 711510 Catalysts in the form of wire cloth 18.672 0.432 0.0 57 151000 Other oils and their fractions 14.377 0.035 0.0 58 151521 Maize (corn) oil and its fractions 11.338 20758.875 22.8 59 151110 Crude oil 9.91 0.137 0.0 60 262011 Containing mainly zinc 7.8 226.859 0.0 61 180400 Cocoa butter, fat and oil 6.861 27570.497 45.3 62 270730 Xylole 6.047 41.119 0.0 63 630631 Sails - Of synthetic …bres 5 1073.53 0.0 64 722592 Otherwise plated or coated w/ zinc 1.681 1002.997 0.0 65 252230 Hydraulic lime 1.344 11.135 0.0 66 310229 Ammonium sulphate; double salts 0.992 155.239 0.0 67 854340 Electric fence energisers 0.54 441628.86 25.1 Total 392,126 1,546,760 16.63 Imp00 is the value of imports by processing exporters in 2000, in thousands USD. Exp07 is the value of exports by non-processing exporters in 2007, in thousands USD. 27