NBER WORKING PAPER SERIES HOW MUCH OF CHINESE EXPORTS IS REALLY MADE IN CHINA? ASSESSING DOMESTIC VALUE-ADDED WHEN PROCESSING TRADE IS PERVASIVE

NBER WORKING PAPER SERIES HOW MUCH OF CHINESE EXPORTS IS REALLY MADE IN CHINA? ASSESSING DOMESTIC VALUE-ADDED WHEN PROCESSING TRADE IS PERVASIVE Robert Kooman Zhi Wang Shang-Jin Wei Working Paer 14109 htt://www.nber.org/aers/w14109 NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA 02138 June 2008 The views exressed in this aer are those of the authors alone. They do not necessarily reflect the views of the US International Trade Commission, any of its individual Commissioners, any other organization that the authors are or have been affiliated with, nor the National Bureau of Economic Research. NBER working aers are circulated for discussion and comment uroses. They have not been eerreviewed or been subect to the review by the NBER Board of ors that accomanies official NBER ublications. 2008 by Robert Kooman, Zhi Wang, and Shang-Jin Wei. All rights reserved. Short sections of text, not to exceed two aragrahs, may be quoted without exlicit ermission rovided that full credit, including notice, is given to the source.

How Much of Chinese Exorts is Really Made In China? Assessing Domestic Value-Added When Processing Trade is Pervasive Robert Kooman, Zhi Wang, and Shang-Jin Wei NBER Working Paer No. 14109 June 2008, Revised December 2011 JEL No. F1,O1,O53 ABSTRACT The rise of China in world trade has brought both benefits and anxiety to other economies. For many olicy questions, it is crucial to know the extent of domestic value added (DVA in exorts, but the comutation is more comlicated when rocessing trade is ervasive. We roose a method for comuting domestic and foreign contents that allows for rocessing trade. By our estimation, the share of domestic content in exorts by the PRC was about 50% before China s WTO membershi, and has risen to over 60% since then. There are also interesting variations across sectors. Those sectors that are likely labeled as relatively sohisticated such as electronic devices have articularly low domestic content (about 30% or less. Robert Kooman Research Division Office of Economics US International Trade Commission 500 E Street SW Washington, DC 20436 Robert.Kooman@usitc.gov Zhi Wang Research Division Office of Economics US International Trade Commission 500 E Street SW Washington, DC 20436 zhi.wang@usitc.gov Shang-Jin Wei Graduate School of Business Columbia University Uris Hall 619 3022 Broadway New York, NY 10027-6902 and NBER shangin.wei@columbia.edu

1. Introduction Made in China is one of the most common labels one encounters in a shoing mall in the United States and Euroe. Increasingly, many roducts that are suosed to be technically sohisticated and therefore likely to be associated with exorts from highincome countries, such as digital cameras and comuters, also carry that label. Since the most salient characteristic of the factor endowment in China is a vast suly of unskilled labor relative to either hysical or human caital, is the country s actual exort structure inconsistent with the redictions from the international trade theory based on its endowment? A ossible resolution to the uzzle is that China is simly the last section of a long global roduction chain that ends u assembling comonents from various countries into a final roduct before it is exorted to the US and EU market. Indeed, a MacBook comuter carries a label at its back (in small tye that reads Designed by Ale in California; Assembled in China. This label is likely to be oversimlified already, as it reorts only the head and the tail of a global roduction chain, but skis many other countries that suly other comonents that go into the roduct. China is the archetye of a national economy that is well integrated into a global roduction chain. It imorts raw material, equiment, and intermediate inuts, and then exorts a big fraction of its outut (on the order of 37% of GDP in 2006 to the world market. The PRC is not the only country whose roduction and exorts are a art of a global chain; Jaan, Korea, Singaore, and Malaysia are some other examles of countries that articiate actively in the international divisions of labor. However, the PRC is noteworthy due to its sheer size. In addition, its exort/gdp ratio, at 35% or higher in recent years, is extraordinarily high for a large economy, when comared with 8% for the US and 13% for India. With a reutation as a world factory, China is a to sulier of manufacturing outsourcing for many global comanies. For many olicy issues, it is imortant to assess the extent of domestic content in exorts. For examle, what is the effect of a currency areciation on a country s exorts? The answer deends crucially on the share of domestic content in the exorts. Other things being equal, the lower the share of domestic content in the exorts, the smaller the effect on trade volume a given exchange rate areciation would have. As 1

another examle, what is the effect of trading with the PRC on US income inequality? The answer deends in art on whether the PRC simly exorts roducts that are intensive in low-skilled labor or whether its exorts are more sohisticated. Rodrik (2006 notes that the er caita income tyically associated with the kind of goods bundle that the PRC exorts is much higher than the country s actual income. He interrets this as evidence that the skill content of its exorts is likely to be much higher than its endowment may imly. Schott (2008 documents an aarent raid increase in the similarity between the PRC s exort structure and that of high-income countries, and interrets it as evidence of a rise in the level of sohistication embedded in the country s exorts. Wang and Wei (2008a use disaggregated regional data to investigate the determinants of the rise in exort sohistication. Indeed, many other observers have exressed fear that the PRC is increasingly roducing and exorting sohisticated roducts and may be roviding wage cometition for mid- to high-skilled workers in the US and Euroe. However, the calculations by Rodrik (2006 and Schott (2008 do not take into account the imorted content in the country s exorts. If the domestic content in exorts from the PRC is low, esecially in sectors that would have been considered sohisticated or high-skilled in the US, then imorts from the PRC may still generate a large downward ressure on the wage of the low-skilled Americans after all (as ointed out by Krugman, 2008. These are imortant olicy questions and have imlications for both develoing and develoed countries. A good understanding of the nature and extent of global suly chains can rovide imortant insights for economists and olicy makers. How would one assess foreign versus domestic content in a country s exorts? Hummels et al. (2001 (HIY in subsequent discussion roose a method to decomose a country s exorts into domestic and foreign value added share based on a country s inut-outut (IO table. They make a key assumtion that the intensity in the use of imorted inuts is the same between roduction for exorts and roduction for domestic sales. This assumtion is violated in the resence of rocessing exorts. Processing exorts are characterized by imorts for exorts with favorable tariff treatment: firms imort arts and other intermediate materials from abroad, with tariff exemtions on the imorted inuts and other tax references from local or central governments, and, after rocessing or assembling, exort the finished roducts. The olicy references for 2

rocessing exorts usually lead to a significant difference in the intensity of imorted intermediate inuts in the roduction of rocessing exorts and that in other demand sources (for domestic final sales and normal exorts. Since rocessing exorts have accounted for more than 50% of China's exorts every year at least since 1996, the HIY formula is likely to lead to a significant under-estimation of the share of foreign value added in its exorts. In fact, most economies offer tariff reductions or exemtions on imorted intermediate inuts used in roduction for exorts. Ignoring rocessing exorts is likely to lead to estimation errors, esecially for economies that engage in a massive amount of tariff/tax-favored rocessing trade, such as the China, Mexico and Viet Nam. In this aer, we aim to make two contributions to the literature. First, we resent a formula for comuting shares of foreign and domestic value added in a country s exorts when rocessing exorts are ervasive. We develo this formula because the roduction technology and inut sourcing differs for goods roduced for domestic consumtion and normal exorts comared to those roduced under exort rocessing regimes. The HIY formula is a secial case of this general formula. Second, we aly our methodology to China using data for 1997, 2002, and 2007. We estimate that the share of foreign value added in China s manufactured exorts was about 50% during 1997 to 2002 before China s WTO membershi, almost twice as high as that imlied by the HIY formula, and has risen to over 60% in 2007 after 5 years of its WTO membershi. There are also interesting variations across sectors. Those sectors that are likely labeled as relatively sohisticated such as comuters, telecommunication equiments, and electronic devices have articularly low domestic content (about 30% or less. By design, this is an accounting exercise, and does not intend to examine the determinants, driving forces and the consequences of changes in domestic contents in China s gross exorts thoroughly. However, a solid methodology to estimate and account domestic and foreign value added in develoing countries exorts is the necessary first ste toward a better understanding of these issues. Besides the aers on vertical secialization in the international trade literature, this aer is also related to the I/O literature. In articular, Chen et al. (2004 and Lau et al (2007 are the first to develo a non-cometitive tye I/O model for China (i.e., one 3

in which imorted and domestically roduced inuts are accounted for searately and to incororate rocessing exorts exlicitly. However, these aers do not describe a systematic way to infer searate inut-outut coefficients for roduction of rocessing exorts versus those for other final demands. It is therefore difficult for others to relicate their estimates or aly their methodology to other countries. In addition, they use an aggregated version of China s 1995 and 2002 inut-outut tables, resectively, to erform their analysis, with 20 some goods roducing industries. We rovide a more uto-date and more disaggregated assessment of foreign and domestic values added in Chinese exorts with 83 goods roducing industries. Finally, they imose an assumtion in estimating the imort use matrix from the cometitive tye I/O table ublished by China s National Statistical Bureau: within each industry, the mix of the imorted and domestic inuts is the same in caital formation, intermediate inuts, and final consumtion. We relax this assumtion by refining a method roosed in Dean, Fung, and Wang (2007 that combines China s rocessing imorts statistics with United Nations Broad Economic Categories (UNBEC classification. The rest of the aer is organized as follows. Section 2 resents a concetual framework for estimating shares of domestic and foreign value added in a country s exorts when rocessing exorts are ervasive. It also describes a mathematic rogramming rocedure to systematically infer a set of I/O coefficients called for by the new formula but not tyically available from a conventional I/O table. Section 3 resents the estimation results for Chinese exorts. Section 4 evaluate the accuracy of the estimation results by using an IO table with searate rocessing trade account comlied by Mexico national statistical agency based 2003 economic census. Section 5 concludes. 2. Concetual Framework and Estimation Method 2.1 When secial features of rocessing exorts are not taken into account We first discuss how domestic and foreign contents in a country s exorts can be comuted when it does not engage in any rocessing trade. The discussion follows the inut-outut literature, and is the aroach adoted (imlicitly by Hummels et al. (2001 4

and Yi (2003. Along the way, we will oint out a clear connection between the domestic content concet and the concet of vertical secialization. 2 When imorted and domestically roduced intermediate inuts are accounted searately, a value-based inut-outut table can be secified as follows: 3 A A D M ua D D X + Y = X (1 M X + Y = M (2 M + ua + A = u (3 v where A D = [a D i] is an nn x matrix of direct inut coefficients of domestic roducts; A M = [a M i] is an nn x matrix of direct inuts of imorted goods; Y D is an n x1 vector of final demands for domestically roduced roducts, including usage in gross caital formation, rivate and ublic final consumtion, and gross exorts; Y M is an n 1 vector of final demands for imorted roducts, including usages in gross caital formation, rivate and ublic final consumtion; X is a n 1 vector of gross outut; M is a n 1 vector of imorts; A v = [a v ] is a 1 n vector of each sector s ratio of value added to gross outut, and u is a 1 n unity vector. Subscrits i and indicate sectors, and suerscrits D and M reresent domestically roduced and imorted roducts, resectively. Equations (1 and (2 define two horizontal balance conditions for domestically roduced and imorted roducts, resectively. A tyical row k in Equation (1 secifies that total domestic roduction of roduct k should be equal to the sum of the sales of roduct k to all intermediate and final users in the economy (the final sales include domestic consumtion and caital formation, lus exorts of roduct k. A tyical row h in Equation (2 secifies that the total imorts of roduct h should be equal to the sum of the sales of roduct h to all users in the economy, including intermediate inuts for all sectors, lus final domestic consumtion and caital formation. Equation (3 is both a vertical balance condition, and an adding-u constraint for the inut-outut coefficients. 2 We use the terms domestic value added and domestic content interchangeably. Similarly, we use the terms foreign value added, foreign content, and vertical secialization to mean the same thing. 3 Such a model is called a non-cometitive model in the IO literature. HIY (2001 do not secify this system exlicitly but go straight to the imlied Leontief inverse while Chen et al. (2004 secify only the first two equations. A fully secified model facilitates better understanding of the connection between vertical secialization and domestic content, and a comarison with the model in the next sub-section that features rocessing exorts. 5

It imlies that the total outut (X in any sector k has to be equal to the sum of direct value added in sector k, and the cost of intermediate inuts from all domestically roduced and imorted roducts. ( I A D From Equation (1 we have X 1 D 1 D = ( I A Y (4 4 is the well-known Leontief Inverse, a matrix of coefficients for the total domestic intermediate roduct requirement. Define a vector of share of domestic content, or domestic value added, in a unit of domestically roduced roducts, DVS = {dvs }, a 1 n vector, as the additional domestic value added generated by one additional unit of final demand of domestic roducts ( Y D = u : DVS = A v X / Y D = A ( I A v D 1 = A ( I A v D 1 where A an nn x diagonal matrix with a ˆV v as its diagonal elements. Equation (5 indicates that the domestic content for an IO industry is the corresonding column sum of the coefficient matrix for total domestic intermediate goods requirement, weighted by the direct value-added coefficient of each industry. Because standard model assumes that exorts and domestic sales are roduced by the same technology, the share of domestic content in final demand and the share of domestic content in total exorts are the same. So Equation (5 is also the formula for the share of domestic content in total exorts for each industry. Define a vector of share of foreign content (or foreign value added in final demand for domestically roduced roducts by FVS = u DVS. By making use of Equation (3, it can be verified that FVS = u A v ( I A D 1 = (5 M D 1 ua ( I A (6 For each industry, this is the column sum of the coefficient matrix for total intermediate imort requirement. This turns out to be the same formula used to comute vertical secialization by Hummels et al. (2001. In other words, the concets of vertical secialization and of foreign content are identical. I A has to be full rank. 4 D ( 6

2.2 Domestic Content in Exorts When Processing Trade is Prevalent We now turn to the case in which tariff-favored rocessing exorts are revalent; these exorts have a different intensity in the use of imorted inuts than do domestic final sales (and normal exorts. Concetually, we wish to kee track searately of the IO coefficients of the rocessing exorts and those of domestic final sales and normal exorts. For now, we ignore the fact that these IO coefficients may not be directly available, and will discuss a formal aroach to estimate them in the next subsection. The exanded I/O table with a searate account for rocessing exorts is reresented by Figure 1. Figure 1: Inut-outut table with searate roduction account for rocessing trade Domestic Intermediate Inuts Production for domestic use & normal exorts (D Processing Exorts (P Intermediate Inuts from Imorts Intermediate use Production for Production of Final use domestic use & normal exorts rocessing exorts (C+I+G+E DIM 1,2,, N 1,2,, N 1 1 1. DD. DP D P Z Z Y E. N 1... N Value-added 1 Gross outut 1 1... N 0 0 P E Z V MD D P X E Z V MP P P E Gross Outut or Imorts P X E P E M Y M We use suerscrit P and D, resectively, to reresent rocessing exorts on one hand, and domestic sales and normal exorts on the other. This exanded IO model can be formally described by the following system of equations: I A 0 DD A I DP X E P E P Y = E D P E P (7 A MD P MP P M ( X E + A E + Y = M (8 ua DD + ua MD + A D v = u (9 7

ua DP + ua MP + A P v = u This is a generalization of the model discussed in the revious subsection. Equations (7 and (8 are a generalization of Equations (1 (2, and Equations (9 (10 are a generalization of Equation (3, with a searate account for rocessing exorts. Equations (9 and (10 are also the new adding-u constraint for the IO coefficients. The analytical solution of the system is X E P E P = I A 0 DD A I DP 1 Y E The generalized Leontief inverse for this exanded model can be comuted as follows: B = I A 0 DD A I DP 1 B = B DD PD B B DP PP D P E P ( I A = 0 DD 1 Substituting Equation (12 into Equation (11, we have: X E P = ( I A DD 1 ( Y D E P + (1 A DD 1 A DP ( I A E DD 1 Substituting Equation (13 into Equation (8, the total demand for imorted intermediate inuts is I A DP (10 (11 (12 (13 M MD DD 1 D P MD DD 1 DP P MP M Y = A ( I A ( Y E + A (1 A A E + A E (14 It has three comonents: the first term is total imorted content in final domestic sale and normal exorts, and the second and the third terms are indirect and direct imorted content in rocessing exorts, resectively. We can comute vertical secialization (VS or foreign content share in rocessing and normal exorts in each industry searately: VSS VSS D T P MD = MD DD 1 ua ua (1 A ( I A DD A DP 1 + ua MP T (15 The total foreign content share in a articular industry is the sum of the two weighted by the share of rocessing and non-rocessing exorts s and u s, where both s and u are a 1 by n vector: D VSS = (16 P VSS P P VSS ( u s, s The foreign content (or foreign value-added share in a country s total exorts is: 8

TVSS = ua MD ( I A DD 1 E E te P + u( A MD (1 A DD 1 A DP + A MP P E te where te is a scalar, the country s total exorts. Equation (16 is a generalization of Equation (7, the formula to comute industry-level share of vertical secialization. Equation (17 is a generalization of the formula for country-level share of vertical secialization roosed by Hummels et al. (2001, age 80. In articular, either when DD A = DP A and formula for VS. MD A = A MP (17, or when E P /te = 0, Equation (18 reduces to the HIY Similarly, the domestic content share for rocessing and normal exorts at the industry level can be comuted searately: DVS DVS D T P = A B = ( A = A v D v A D v ( I A D v ( I A DD ( I A A v 0 1 DD A 1 DP + A P v DD T 1 ( I A DD I 1 A DP (18 The total domestic content share in a articular industry is a weighted sum of the two: D DVS = (19 P DVS P P DVS ( u s, s The domestic content share in a country s total exorts is: Either when P D DD 1 E E D DD 1 DP P TDVS = AV ( I A + ( AV (1 A A + AV te DD A = DP A and P E te (20 D A v = A P v, or when E P /te = 0, Equation (20 reduces to the HIY formula in Equation (5. Note we can easily verify that for both rocessing and normal exorts, the sum of domestic and foreign content shares is unity. 2.3 Estimation Issues Equations (18-20 allows us to comute the shares of domestic content in rocessing and normal exorts for each industry as well as in a country s total exorts. However, statistical agencies tyically only reort a traditional I/O matrix, A D, and sometimes A M, but not A DP, A DD, A MP and A MD searately. Therefore, a method to estimate these matrices, based on available information, has to be develoed. In this sub-section, 9

we roose to do this via a quadratic rogramming model by combining information from trade statistics and conventional I/O tables. The basic idea of this model is to use information from the standard I/O table to determine sector-level total imorts/exorts, and information from trade statistics to determine the relative roortion of rocessing and normal exorts within each sector, thus use u all available data to slit the national economy into rocessing and nonrocessing blocks, each with its own IO structure. Using the data from the I/O table to determine sector-level total imorts/exorts hels to ensure that the balance conditions in the official I/O account are always satisfied, and that the I/O table with searate rocessing and non-rocessing accounts estimated from the model always sums to the ublished official table. Such a method is a formalization of the calibration methods widely used in macroeconomics and CGE modeling when the number of endogenous variables larger than the number of equations. The following data are observable from a standard I/O table: x i = Gross outut of sector i; z i = Goods i used as intermediate inuts in sector ; v = Value-added in sector ; m i = imorts of sector i goods; and y = final demand excet for exorts of goods i. i ` We combine those observed data from the I/O table and rocessing trade shares 5 observed from trade statistics to determine the values for: m i = Imorts of sector i good used as intermediate inuts to roduce rocessing exorts; d m i = Imorts of sector i goods used as intermediate inuts for domestic roduction and normal exorts; e = Normal exorts of sector i ; and n i e i = Processing exorts of sector i. The artition of imorts into intermediate and final use is based on a combination of China custom imort statistics and UN BEC classification, as described in Dean, Fang 5 Processing trade are defined by China Customs, which include trade regime Process & assembling "(14 and Process with imorted materials" (15 in China Customs statistics. These statistics are relatively accurate because they involve duty exemtion and value-added tax rebates which under intensive Customs monitoring. 10

and Wang (2011. The results of such artition and the actual numbers used in our emirical estimation are reorted and discussed in the data source subsection later. Parameters on domestic and imorted final demand can be inferred from the observed data discussed above: m y i. m y i = Final demand of goods i from imorts (residuals of mi - m i - d m i ; d y i = Final demand of goods i rovided by domestic roduction (residual of yi - All those data based on official statistics are entered our estimation model as constants. Define dd z i = Domestically roduced intermediate good i used by sector for domestic sales and normal exorts; d z i = Domestically roduced intermediate good i used by sector for rocessing exorts; md z i = Imorted intermediate good i used by sector for domestic sales and normal exorts; m z i = Imorted intermediate good i used by sector for rocessing exorts ; d v = value added by domestic and normal exort roduction in industry ; v = value added by rocessing exort roduction in industry. Then the IO coefficients for the exanded IO model can be written as: A DD dd md d zi z dd i v MD md D vd = [ ai ] = [ ], A = [ ai ] = [ ], Av = [ a ] = [ ], x e x e x e d m zi zi v DP d MP m P v A = [ ai ] = [ ], A = [ ai ] = [ ], Av = [ a ] = [ ], e e e To obtain these unobservable IO coefficients, we need estimate within-industry transactions [z dd i ], [z d i ], [z md i ], and [z m i ], as well as sector-level value added [v d ], and [v ], subect flowing I/O account identities and statistical adding u constraints: K dd d ( z zi = x i = 1 + i ei e n i y d i (21 K = 1 ( z md i + z m i = m i y m i (22 11

K dd md ( z + z + v i i = 1 d = x e (23 K i= 1 K =1 K =1 d m ( z i + z + v z md i z m i = = i m d i m i = e (24 (25 (26 K = 1 z v ( z dd i d dd i K d d + zi = zi ( mi + mi z d i + z = 1 md i + z m i = z i (27 + (28 + v v (29 = The economic meanings of these 9 grous of constraints are straightforward. Equations (21 and (22 are row sum identities for the exanded I/O account. They state that total gross outut of sector i has to equal to the sum of domestic intermediaries, final demand and exorts (both rocessing and normal exorts in that sector. Similarly, total imorts have to equal imorted intermediate inuts lus imorts delivered to final users. Equations (23 and (24 are column sum identities for the exanded I/O account. They define the value of rocessing exorts in sector as the sum of domestic and imorted intermediate inuts as well as rimary factors used in roducing rocessing exorts; these four grous of constraints corresond to equations (7-(10 in the extended I/O model resectively. Equations (25 to (29 are a set of adding u constraints to ensure that the solution from the model is consistent with official statistics on sector-level trade and within-industry transactions. Suose there are K sectors, then there will be 4K 2 + 2K unknowns and only 7K+K 2 indeendent constraints, so we have to cast the calibration roblem as a constrained otimization rocedure to minimize following obective functions: 12

Min S = + K K i= 1 K i= 1 K i= 1 = 1 ( z m i ( z dd i z0 z0 z0 z0 m 2 i m i dd i dd i + K 2 =1 + ( v d K K i=1 = 1 v0 v0 d d ( z 2 + K = 1 z0 d d 2 i i d z0i ( v + v0 v0 K K i= 1 = 1 2 ( z md i z0 z md i md i 2 (30 Where z s and v s are variables to be estimated, those variables with a 0 in the suffix denote initial values. These initial values are guesses about the values of variables based on official statistics using roortional assumtions (will discuss below. However, because all arameters in the 9 grous of linear constraints (right hand side of equations (21 to (29 were directly or indirectly obtained from observable official statistical sources, model solutions thus are restricted into a convex set and will be relatively stable resect to variations in these initial values as long as all the arameters in these linear constraints ket as constants. The initial value of z i md and z i md, are generated by allocating roortion to inut i s usage in sector as equation (31: k ik k k 13 k ik k k k d m i and m i in zi ( e / x z md i ( x m e / x d z0 i = m N i z0i = m N i (31 z ( e / x z ( x e / x The slit of total inter-sector intermediate inuts flow from sector i to sector between normal and rocessing use are based on their roortion in gross outut. The residuals of the total intermediate inuts and the imorted intermediate inuts estimated from equation (31 are taken the initial values for domestically roduced intermediate inuts as equations (32 and (33: ( x e dd md z0i = zi z0i (32 x e d m z0i = zi z0i (33 x The initial values for direct value added in the roduction for rocessing exorts in sector ( v0, are generally set to be the residuals imlied by Equation (24. However, we set a minimum value at the sum of labor comensation and dereciation in

a sector multilied by the share of rocessing exorts in that sector s total outut. In other words, the initial value v0 is set to equal the greater of the residuals from Equation (24 or the minimum value. The initial value for direct value added in the roduction for d domestic sales and normal exorts ( v0 is set as the difference between v (from the I/O table and v0. We conduct some sensitivity checks using alternative initial values to emirically verify whether the model solutions are sensitive to these initial values. It turns out that these alternative initial values do not materially alter our basic conclusions. We imlement this quadratic rogramming model in GAMS (Brooke et al, 2005, related comuter rograms and data files will be available at the USITC website for downloading. 3. Estimation Results After describing the data sources, we reort and discuss the estimation results for shares of domestic and foreign content in Chinese exorts at the aggregate level, and by sector, firm ownershi and maor destination countries. 3.1 Data Inter-industry transaction and (direct value-added data are from China s 1997, 2002 and 2007 benchmark I/O tables ublished by the National Bureau of Statistics of China (NBS. We use detailed exorts and imorts data of 1997, 2002, and 2007 from the General Customs Administration of China to hel differentiate the rocessing and normal trade in each sector. The trade statistics are first aggregated from the 8-digit HS level to China s I/O industries, and then used to comute the share of rocessing exorts in each I/O industry. Modifying a method from Dean, Fung and Wang (2009, we artition all imorts in a given commodity classification into three arts based on the distinction between rocessing and normal imorts in the trade statistics, and on the UN BEC classification scheme: (a intermediate inuts in roducing rocessing exorts; (b intermediate inuts for normal exorts and other domestic final sales; and (c those used in gross caital formation and final consumtion. A summary of these trade statistics as a ercentage of China s total imorts along with share of rocessing exorts during 1996-14

2008 is reorted in Table 1, which shows a downward trend for the use of imorted inuts in roducing rocessing exorts, and an uward trend in their use in roducing normal trade and domestic final sales. Processing exorts as a share of China s total merchandise exorts also gradually decline in recent years. Such trend seems artially reflects the consequence of a series of olicy measures to change the references to rocessing trade and foreign invested enterrises has adoted by Chinese government since the end of 2006. Detailed trade share arameters for each I/O industry in the three benchmark year (1997, 2002, and 2007 are listed in Aendix tables A-C. These data comuted directly from detailed Chinese official trade statistics are imortant to understand our estimates of domestic and imorted content in Chinese gross exorts, esecially their change trends over time. (Inset Table 1 here 3.2 Domestic and foreign contents in total exorts Table 2 resents the results for the decomosition of aggregate foreign and domestic value-added shares in 1997, 2002 and 2007. For comarison, the results from the HIY method that ignores rocessing trade are also reorted. The estimated aggregate domestic value added share in China s merchandise exorts was 54% in 1997, and 60.6% in 2007. For manufacturing roducts, these estimated shares are slightly lower in levels but trending uward significantly at 50% in 1997 and 59.7% in 2007, resectively. In general, the estimated direct domestic value-added shares are less than half of the total domestic value-added shares. However, the estimated indirect foreign value-added share was relatively small; most of the foreign content comes from directly imorted foreign inuts, esecially in 1997 and 2002. The indirect foreign value-added increase over time, and reach about a quarter of China s directly imorted foreign inuts in 2007, indicating the share of simle rocessing and assembling of foreign arts is declining, while more imorted intermediates are being used in the roduction of other intermediate inuts that are then used in the roduction rocess. (Insert table 2 here 15

Relative to the estimates from the HIY method, our rocedure roduces estimates of a much higher share of foreign value added in Chinese gross exorts and with a different trend over time. To be more recise, estimates from the HIY method show that the foreign content share (total VS share increased steadily from 17.6% in 1997 to 28.7% in 2007 for all merchandise exorts, and from 19.0% to 27.1% for manufacturing only during the same eriod. In contrast, our estimates suggest a trend in the oosite direction, with the share of foreign value added in all merchandise exorts falling from 46% in 1997 to 39.4% in 2007, and a similar decline for the share in manufacturing exorts, it fell from 50% in 1997 to 40.3% in 2007. The decline occur mainly during 2002-2007 eriod, which corresonds to the first 5 years of China s entry to the WTO. Our estimates indicate that the HIY method aears to incorrectly estimate both the level and the trend in domestic versus foreign content in the PRC s exorts. What accounts for the difference between ours and HIY aroaches? There are at least three factors drive the change of foreign content of the country s gross exorts: (1 the relative roortions of its total imorts used as intermediate inuts in roducing rocessing exorts and domestic sales & normal exorts; (2 the share of rocessing exorts in its total exorts; and (3 the sector comosition of its exorts. Because rocessing exorts tend to use substantially more imorted inuts, and rocessing exorts account for a maor share of China s total exorts, the HIY indicator is likely to substantially underestimate the true for foreign content in China s exorts. This exlains the level of domestic content by our measure is much lower than that of the HIY indicator. On the other hand, as exorting firms (both those roducing for normal exorts and those for rocessing exorts gradually increase their intermediate inuts sourcing from firms within China or multinationals move their ustream roduction to be near their downstream roduction, the extent of domestic content in exorts rose over time. This exactly is what has haened since China oining the WTO. However, because exorts from industries with relatively lower domestic content often grow faster due to dramatic inflow of foreign direct investment, the comosition of a countries total exorts may lay as an offsetting factor to reduce the share of domestic value-added in the country s gross exorts thus slow down the increase of domestic value-added share in a country s total exorts. As Chinese government start to reduce the olicy incentives for 16

both Foreign Invested Enterrises (FIE and rocessing exorts at the end of 2006, we are observing a trend of increasing domestic contents in Chinese exorts as China continue its industrial ugrading in the years to come. Our interretation is confirmed by DVA shares for rocessing and normal exorts estimated searately (Table 3. There is a more than 10 ercentage oint increase in the total foreign value-added share for domestic sales and normal exorts between 1997 and 2007, which is consistent with the trend indicated by HIY measure. However, in rocessing exorts we see that more domestic-roduced inuts were used, domestic value-added share increased from 20.7% in 1997 to 37.0% in 2007, u more than 16 ercentage oint. Because rocessing exorts still constitute more than 50% of China s total exorts in 2007, which resulting the weighted average total domestic value-added share went u during the decades. There are conflicting forces at work. On the one hand, as domestic inut suliers increase their quality over time, and multinationals move more and more of their ustream roduction into China, exorting firms may decide to increase local sourcing of their inuts. On the other hand, the reductions in the country s trade barriers also encourage exorting firms to use more imorted inuts. These two oosing forces artially offset each other. However, on net, the domestic content share in China s exorts aears to be on the rise. Looking ahead, the share of imorted content in exorts could fall or rise, deending on the relative seed with which domestic inut suliers and multi-nationals can ste u their quality and variety versus the extent of additional reductions in the cost of using imorted inuts. (insert table 3 here We erform a number of robustness checks on the sensitivity of our main results to alternative ways of setting the initial values of the variables and the share arameters of imort use. First, we initialize v0 and d v0 by aortioning the observed direct value added in a sector to rocessing exorts and other final demands based on their resective ortions in the sector s total outut. Second, we initialize v0 either at the residuals imlied by equation (24 if the residuals are ositive, or by following the revious alternative if the residuals are non-ositive. Third, when we artition imorts into 17

different users, we use the average of a three-year eriod (revious, current, and following years rather than ust one year s statistics. Fourth, we exeriment with 0% versus 10% annual dereciation rate for caital goods. These variations roduce relatively little change in the main results. For examle, the estimated share of domestic value added in manufacturing exorts lies in a relatively narrow range between?% and?% in 2002, and between?% and?% in 2007. 3.3 Domestic content in exorts by firm ownershi Since foreign-invested firms account for over half of China s exorts, comare the share of domestic content in exorts between them and other Chinese firms may hel us better to understand why there is a raising trend of domestic value-added share in China s total exorts. However, there is no information on searate inut-outut coefficients by firm ownershi, our estimation assumes that they are the same on this dimension. The variation in the share of domestic content comes from different degrees of reliance on rocessing exorts with a sector, and differences in the sector comosition of their total exorts. Estimates of the domestic content shares by firm ownershi are resented in Table 4. The results show that exorts by wholly foreign owned enterrises exhibit the lowest share of domestic valued-added (at 33.4% in 2002 and increase to 44.1% in 2007, followed by Sino-foreign oint venture comanies (about 44% in both 2002 and 2007. Exorts from Chinese rivate enterrises embodied the highest domestic content shares (83.9% and 80.8% in 2002 and 2007, resectively, while those from the state-owned firms were in the middle (about 70% in both years. Note that these estimates reresent the best guesses based on currently available information; better estimates can be derived once information on I/O coefficients by firm ownershi becomes available. The most noticeable feature of this table is the raising domestic contents in exort roduced by foreign invested firms, their DVA share increased more than 10 ercentage oints between 2002 to 2007(There is a 10.5% ercentage oint increase for wholly Foreign Owned firms, and a 13.6 ercentage oint increase for oint ventures, while DVA share of Chinese domestic firms basically stay steady during the same eriod. This indicates the raise of domestic contents in Chinese exorts during the first 5 years of 18

China in WTO is mainly due to FIE sourcing more of their intermediate inuts within China. This is consistent with the observation of more and more multinationals move their ustream roduction to China getting closer to their downstream roduction already in China and FIE using more and more intermediate inuts from local suliers. Further exam the change of domestic value-added share across firms with the same ownershi but engage in rocessing and non-rocessing exorts reveals that firms engage in normal exorts and domestic sales used more imorted intermediate inuts while firms engage in rocessing exorts sourced more of their intermediate inuts domestically, regardless their ownershi tyes. Therefore, we can conclude that the raise of domestic value-added content share in Chinese exorts are maorly caused by China s rocessing exorts using more locally sulied intermediate inuts in 2007 than that in 2002 6. If this trend continues, the difference in the intensity of using imorted inuts for rocessing and non-rocessing exorts will be further reduced in next decades. (Insert table 4 here 3.4 Domestic content by sector To see if there are interesting atterns at the sector level that hels to exlain the decline trend of imorted contents in China s total exorts, and further assess whether the increasing domestic value-added share reflect actual ugrade of Chinese industrial structure, Tables 5 and 6 reort, in ascending order on domestic content share, the decomosition in China manufacturing exorts by industry in 2002 and 2007 resectively, together with shares of rocessing and foreign invested enterrises exorts in each sector s exorts as well as the sector s share in China s total merchandise exorts. We choose to reort the results from 2002 and 2007 not only because we would like use the latest I/O table released but also due to these two benchmark tables are consistently classified on most recent Chinese industry classifications, which simlifies issues involved in overtime comarison. Similar results for 1997 are omitted to save sace. Among the 57 manufacturing industries in the table, 15 have a share of domestic value-added in their exorts less than 50 ercent in 2002, and collectively account for 6 Domestic value-added is a concet consistent with Gross Domestic roducts, which measures the value creation roduction activities occurring with national boarder regardless firm ownershis. 19

nearly 35 ercent of China s merchandise exorts that year. Many low-dva industries are likely to be labeled as relatively sohisticated, such as telecommunication equiment, electronic comuter, measuring instruments, and electronic devices. A common feature of these industries is that rocessing exorts account for over two-thirds of their exorts and foreign invested enterrises layed an overwhelming role. In 2007, the number of industry with less than 50 ercent domestic contents in their exorts declined into 10, but their exorts count for more than 32 ercent of China s total merchandise exorts and these low-dva industry are more concentrated in high-tech sectors. There are 11 industries in the to 15 low-dva industries in 2002 still remain in the15 to low-dva industries in 2007. (insert table 5 and table 6 here The next 18 industries in table 6 have their share of domestic value-added in the range of 51 to 65 ercent; they collectively accounted for 28 ercent of China s total merchandise exorts in 2002. Several labor-intensive sectors are in this grou, such as furniture, toys and sorts roducts, Leather, fur, down and related roducts. The remaining 24 industries have relatively high shares of domestic value-added. However, they as a grou roduced only less than 30 ercent of China s total merchandise exorts in 2002. Aarel, the country s largest labor intensive exorting industry, which by itself was resonsible for 7 ercent of the country s total merchandise exorts in 2002, is at the to of this grou with a share of domestic content at 66 ercent. The 12 industries at the bottom of Table 6 with DVA share more than 75 ercent collectively roduced only less 10 ercent of China s total merchandise exorts in 2002. The weights of high-dva industries in China s exorts increased significantly in 2007. There are, the number of industries with DVA share more than 75 ercent increase to 25 in 2007 (bottom of Table 7, and their exorts constitute more than 30 ercent of China s total merchandise exorts in 2007. Among these High-DVA industries, we not only see the traditional labor industries such as furniture, textiles and aarel still lay an significant role (account for more than half of these high DVA-sector exorts, but also the increasing role of heavy and caital intensive industries such as automobile, industrial machinery and rolling steel (account for nearly one thirds of these high-dva sector s 20

exorts. The data clearly indicate China s industrial ugrade is real and FIEs have layed very imortant role in this rocess. 3.5 DVA shares in Chinese exorts by trading artners By assuming domestic and foreign value added shares in er unit gross exorts are the same for all destination countries in each IO industry and exort regime, we can further estimate the domestic value-added share in China s exorts to each of its maor trading artners. The decomosition results for China s total merchandise exorts to each of its maor trading artners are reorted in Table 7 in increasing order of the estimated domestic value-added share. Note, however, the variation by destination in this method is caused solely by China s exort structure to each of its trading artners (exorts to each individual country/region vary by sector and trade regime structures, not the direct inut intensities of imorted intermediates in roducing such exorts. Hong Kong, the United States, Singaore, Taiwan and Malaysia are at the to of the table in both 2002 and 2007, with less than or about 50/60 ercent of China s domestic value-added embodied in its exorts to these five destinations in 2002 and 2007 resectively. Since a large ortion of Chinese exorts to Hong Kong are re-exorts to the United States, the U.S. remain as China s largest exorts market in both 2002 and 2007. The lower domestic value-added share in its exorts to the U.S. may artially exlain why Chinese exorts continued their raid exansion in the U.S. market desite RMB gradually areciation since July 2005. China s exorts to the U.S. have started to slow down since 2008 when China return to eg exchange its rate to U.S. dollars because the world financial crisis, likely because of other macro economic factors and olicy measures adoted by the Chinese government during last year. 7 (Insert table 7 here Another interesting feature in Table 8 is that China s exorts to develoing countries embody much higher domestic valued added than its exorts to OECD countries, but exorts with higher domestic value-added (more than two thirds of the gross value of its exorts constituted of less than 13 ercent of its total exorts of goods 7 China has taken a series of olicy measures to change its references to rocessing trade and foreign invested enterrises since the end of 2006. 21

in 2002, this share raised to about 20 ercent in 2007. It is also interesting to note that the domestic value-added share in China s exorts to high income country was increased between 2002 and 2007, while they are declined for exorts to develoing countries (slit at the row of Mexico in the middle of table 8 indicating relatively more local sulied inuts were used to making exorts to high income countries while relative more imorted inuts were used in roducing exorts to develoing countries. 4. The accuracy of Content Share Estimates As we discussed earlier, the estimation rocedure resented in this aer is a formalization and extension of calibration methods widely used in macroeconomics and CGE modeling. Different from econometric model, evaluation on the accuracy of the calibration largely rely on a benchmark data set from the real world. However, it is not very easy to find such a benchmark data set to assess the accuracy of results from our estimation method because the "true" IO account that searately trace rocessing exorts and other roduction transactions in national economy is rarely exist. Fortunately, Mexico s statistical agency, the Instituto Nacional de Estadística, Geografía e Informática (INEGI, has comlied 2003 benchmark IO table based on economic census, which have searate accounts for Mexico domestic and Maquiladora industries 8, which is the maor exorts romotion rogram in Mexico. The table includes national roduction of goods and services classified under Mexico s 2002 3- and 4-digit NAICS, inuts urchased in the domestic and Maquiladora industries, and imorts from the rest of the world by both economies. The domestic and foreign content share comuted directly from this secial Mexico IO table rovides a reference benchmark to test the erformance of the estimation method we roosed. Based on exorts and imort statistics for Maquiladora industries in the World Trade Atlas and Mexico's aggregate 2003 IO table, we imlemented the same quadratic rogramming model that generates domestic/foreign content estimates for Chinese exorts, to comute domestic and foreign value-added share in Mexico manufacturing exorts, and reort the estimation results in Table 8. The three anels in table 8 lists direct/total domestic/foreign value-added share for normal, rocessing and total exorts resectively. To quantitatively assess know how much each set of value-added share estimates differs from the "true" share data comuted directly from the Mexico IO table with 8 We are grateful to INEGI for roviding us with the inut-outut table. 22

a searate rocessing trade account, we reort three tye metrics in the three bottom row of table 8. The first bottom row list the absolute difference between the estimated share and the "true" shares comuted directly from Mexico IO table with a searate rocessing trade account for manufacture as a whole. The errors for various share estimates seems less than 3 ercentage oint. However, it is the roortionate errors and not the absolute errors that matter; therefore, we comuted the "Mean Absolute Percentage Error" with resect to the true shares as follows: 100 s i s0i i=1 MAPE = n s0 n i=1 i where s i is the estimated share and s0 i share is the reference share for industry i. The resulted index number is reorted in the second bottom row. The error range from 4% to 17% for normal exorts, 14% to 28% for rocessing exorts and 12% to 15% for total exorts. Numbers in the last row of table 8 are the correlation coefficients between the estimated share and the reference shares, they reveal that our estimates are highly correlated with the true shares comuted directly from Mexico IO table with a searate rocessing trade account for normal and total exorts, while the correlation are lower for rocessing exorts. 5. Concluding Remarks Segmentation of roduction across countries allows for reductions in roduction costs and more efficient allocation of resources. The oening-u of China has likely facilitated this rocess. A quantitative assessment of the extent of its articiation in global roduction chains allows us to get a better gras of many olicy questions, including the effect of an exchange rate change on bilateral trade balances. In this aer, we first resent a general framework in assessing the shares of domestic and foreign value added in a country s exorts when tariff-favored rocessing exorts are ervasive. This formula nests the existing best known aroach (HIY, 2001 as a secial case. Because some of the I/O coefficients called for by the new formula are not readily available from conventional I/O tables, we roose an easy-to-relicate mathematical rogramming rocedure to estimate these coefficients by combining information from detailed trade statistics (which records rocessing and normal 23