ERIA-DP--70 ERIA Dscusson Paper Seres Intensve and Extensve Margns of South South North Trade: Frm-Level Evdence Ll Yan ING * Economc Research Insttute for ASEAN and East Asa (ERIA) and Unversty of Indonesa Maoje YU ER, Pekng Unversty September Abstract: The man value added of our paper s twofold. Frst, we construct a theoretcal framework on how South South trade wll affect productvty cut-offs. Second, we present emprcal exercses usng hghly dsaggregated data. Our model s based on the South South North trade framework. Usng a vertcal ntegraton among Southern countres (Indonesa and hna) and testng t by employng merged hnese frms and customs trade data, we fnd that three types of tarff reductons foregn tarff reductons, home output tarff reductons, and home nput tarff reductons sgnfcantly ncrease home country frm productvty and exports va extensve and ntensve margns. Our fndngs are robust usng ex-ante and expost productvty. Keywords: hna, Indonesa, Tarff, Exports, Manufacturng JEL lassfcaton: F1, F13, F14 * Ll Yan Ing s an Economst wth the Economc Research Insttute for ASEAN and East Asa (ERIA) and Lecturer at the Faculty of Economcs, Unversty of Indonesa; Maoje Yu s a Professor at ER, Pekng Unversty. The authors thank Kun Zh for hs excellent research assstance. All errors are ours.
1. Introducton How much can a country expand ts exports? It could ether export more n terms of quantty of goods (ntensve margns), more n terms of the varety of goods (extensve margns), or could move to a hgher level qualty of goods (Hummels and Klenow, 2005). The conventonal trade theorem predcts that a country wll export the good that uses ts abundant factor ntensvely. In the North South trade framework, ths mples that developed countres wll export captal-ntensve goods whle developng countres wll export labour-ntensve goods. As tarff declnes, trade grows not only between countres wth dfferent levels of ntensty of factors of producton but also between countres wth smlar levels. The Lnder hypothess clams that countres of smlar ncome per capta should trade more ntensely wth one another (Lnder, 1961). Takng an example of trade between two bg developng countres, hna and Indonesa, exports of goods (excludng ol and gas) from hna to Indonesa ncreased thrteen fold, from US$2.8 bllon n 2000 to US$36.9 bllon n 2014, and exports of goods (excludng ol and gas) from Indonesa to hna ncreased eghtfold, rsng from US$1.7 bllon to US$14.5 bllon over the same perod, wth the average purchasng power party based-ncome per capta of hna of US$7,200 beng comparable wth the US$7,224 of Indonesa n 2000 2014. Our paper manly focuses on, frst, how the nteracton of South South trade affects countres ntensve and extensve margns. Second, how South South trade affects the producton and export decsons n ther North South trade, amng to llustrate how fallng trade costs strengthen a country s comparatve advantage n the global supply chan. Secton 2 llustrates a theoretcal framework. Secton 3 detals data and data sources. Secton 4 presents emprcal fndngs. Secton 5 concludes. 1
2. Theoretcal Framework Framework of the Model The man dea of the model s as follows. We consder a followng trade pattern: a labour-abundant country such as hna mports raw materals or ntermedate nputs from Indonesa, combnes these wth domestc captal and labour factors to produce and export labour-ntensve products such as textle and garments. Our man nterest s to see how fallng trade costs strengthen the comparatve advantage of the domestc country (.e. hna) n the global supply chan. To fully capture the mpact of trade lberalsaton and ft wth related emprcal lterature, we consder the followng three dmensons of trade lberalsaton: () home (.e. hna) tarffs cut n fnal products such as textle and garments n hna; () tarffs cut n textle and garments of the foregn destnaton country (.e. Unted States [US]); and () hna s tarffs cut on ts ntermedate nputs mported from Indonesa (e.g. cotton). The frst two types of tarffs are blateral trade lberalsaton on fnal goods. The last one s trade lberalsaton on ntermedate nputs, a la Goldberg et al. (2010) and Topalova and Khandelwal (2011). Our model has the followng features. Frst, t s able to govern both comparatve advantage along wth Heckscher-Ohln and frm heterogenety as n Meltz (2003). Second, smlar to Bernard-Reddng-Schott (2007), we are able to show that when frms possess heterogeneous productvty, countres dffer n relatve factor abundance, and ndustres vary n factor ntensty, then fallng trade costs nduce reallocatons of resources both wthn and across ndustres and countres. But we extend the Bernard- Reddng-Schott s model by allowng nternatonal fragmentaton and vertcal ntegraton, followng Y (2010). In terms of trade lberalsaton, most of the exstng lterature consder only blateral tarff reductons on fnal goods, but here we also consder trade lberalsaton n ntermedate nputs to better ft wth the realty. Our model can be sketched by the followng model as n Y (2010) where regon 1 represents Indonesa, regon 2 refers to hna, and regon 3 refers to the US. 2
Fgure 1: Three Regons and Three Factors of Producton Regon 1 Intermedate goods Regon 2 Domestc ntermedate goods aptal labour Fnal goods Domestc sales Regon 3 Export Predcton of the Model Smlar to Bernard-Reddng-Schott (2007), we predct the followng: (1) Wth trade lberalsaton on fnal goods, the domestc productvty cut-off ponts n the home country wll ncrease n both ndustres. Moreover, the ndustry wth a comparatve advantage ncreases more. Thus, ths confrms the effect of tougher nternatonal competton. (2) Wth trade lberalsaton on fnal goods, the exportng productvty cut-off ponts n the home country wll decrease n both ndustres. Moreover, the ndustry wth a comparatve advantage decreases more. Ths leads to the dea of the access to larger foregn markets due to foregn tarff reductons (Lleeva and Trefler, 2010). These two fndngs can be llustrated n Fgure 2: 3
Fgure 2: Productvty ut-off as a Functon of Varable ost omparatve advantage ndustry φ A φ T φ X T φ (0, ) omparatve dsadvantage ndustry φ A φ T φ X T φ (0, ) A = autarky T=costly trade Moreover, we also propose new hypotheses whch are completely dffer from the exstng lterature: (3) Trade n both ndustres ncreases. If we restrct our research scope to sngleproduct frms, we should expect both extensve and ntensve margns to ncrease, along wth total ndustral trade value. (4) Wth trade lberalsaton on ntermedate goods, f mported ntermedate nputs are complemented wth labour, labour-ntensve ndustres wll expand and export more. The assumpton that mported ntermedate nputs are complementary wth labour fts wth realty well: when Foxcom mports more ntermedate nputs, t wll hre more workers to expand ts producton. Thus, the ncrease n mported ntermedate nputs results n an ncrease n labour endowment. As usual, by holdng output prce unchanged, an ncrease n a factor endowment wll ncrease the ndustral producton usng such factor ntensvely, as suggested by the Rybczynsk theorem n theory and supported by the real world as n the phenomenon of the Dutch dsease. 4
Set up Our model draws heavly from Bernard-Reddng-Schott (2007, RES) wth an extenson ncorporatng ntermedate nputs. onsder a world of three countres (US, hna, and Indonesa), three factors (captal, labour, and materals), two ndustres, and a contnuum of heterogeneous frms. The trade pattern s as follows: hna mports materals from Indonesa, combnes t wth domestc labour and captal to produce, and then exports a fnal good to the US. ountres are dentcal n terms of preferences and technologes but dffer n terms of factor endowments. Factors of producton are moble between ndustres wthn countres but mmoble across countres. Each ndustry uses three factors n producton. 2.1. onsumpton The representatve consumer s utlty depends on consumpton of the output of two ndustres (=1, 2), each of whch contans a large number of dfferentated varetes (ω) produced by heterogeneous frms. We assume that the upper ter of utlty determnng consumpton of the two ndustres output takes the obb-douglas form and the lower ter of utlty determnng consumpton of varetes takes the ES form, U, 1 (1) 1 2 1 2 1 2 where s a consumpton ndex defned over consumpton of ndvdual varetes q ( ) wth dual prce ndex P, defned over prces of varetes p ( ), 1/ q( ) d, 1 P p( ) d 1/1 (2) where 1/ (1 ) 1 s a constant elastcty of substtuton across varetes. We assume that the elastcty of substtuton between varetes s the same n the two ndustres. 5
2.2. Producton To produce a varety of goods, a frm uses all three factors: captal (K), labour (L), and materal (M). Let r, w, v denote the prce of captal, labour, and materal, respectvely. Materal s assumed to complement labour whch s supported by recent emprcal evdence (hen, Yu and Yu, 2014). As labour and materal are supposed to be complementary, we could reduce the three factors nto two: K and N, where N mn L, M. Accordngly, the equlbrum prce of factor N s denoted u=w+v. Producton nvolves a fxed and varable cost n each perod. Both fxed and varable costs use multple factors of producton whose ntensty of use vares across ndustres. All frms share the same overhead cost, but varable cost vares wth frm productvty, (0, ). The cost functon takes the obb-douglas form, Assume 1 2 q 1 ( f ) u r (3) 1 0, so that ndustry 2 s assumed to be captal ntensve relatve to ndustry 1. Let denote hna and A denote the US, We also assume that N / K N / K A A, so that the US s relatvely captal abundant. It s easy to see that, to hna, ndustry 1 s the ndustry wth comparatve advantage. Let the prce for factor N n hna be numerare, u 1. Frms can choose to sell n a domestc market d, or export to a foregn market x. Internatonal trade ncurs fxed and varable costs. The fxed cost of export uses both factor N and captal K wth the same factor ntenstes as producton. In addton, the frm may also face varable trade costs, whch take the standard ceberg form, whereby a fracton of 1 unts of a good must be shpped n ndustry n order for 1 unt to arrve. Proft maxmsaton that mples the equlbrum prces n the two markets satsfy: p p d 1 ( u ) ( r ) ( ) (4) 1 ( u ) ( r ) ( ) p ( ) (5) x d 6
Wth ths prcng rule, we can derve frms equlbrum revenue n the domestc and export markets: P ( u ) ( r ) 1 d ( ) R ( ) (6) 1 r P A A 1 x ( ) R ( ) 1 (7) ( u ) ( r ) r where stands for the share of expendture allocaton to an ndustry. R and A R denote aggregate expendture (equals aggregate revenue) n hna and the US. P and A P denote the ndustry prce ndex n hna and the US. Accordng to (6) and (7), equlbrum revenue n the export market s proportonal to that n the domestc market: P R (8) A A 1 1 x ( ) ( ) ( ) rd ( ) P R r Then the total revenue receved by a frm n hna s: ( ) d ( ) A A 1 1 rd ( )(1 ( ) ( )) P R r P R r f t does not export f t exports (9) The fxed producton costs mply that frms that export also sell ther products n the domestc market. Therefore, we may separate each frm s proft nto components earned from domestc sales, ( ), and foregn sales, ( ), where we apporton d the entre fxed producton cost to domestc proft and the fxed exportng cost to foregn proft: r ( ) d 1 d ( ) f( u ) ( r ) (10) r ( ) x 1 x ( ) fx( u ) ( r ) (11) x 7
where the fxed cost of exportng requres both factors whch are a complementary of labour and materal N, and captal K, f ( u ) ( r ) x 1. Total frm proft s gven by: ( ) ( ) max 0, ( ) (12) d x 2.3. Decson to produce and export To produce n an ndustry, frms should nvest a fxed entry cost, whch s thereafter sunk. The entry cost also uses factors N and K, so that the ndustry-sunk entry cost takes the form f ( u ) ( r ) e 1. After frms nvest a sunk cost to enter an ndustry, they draw ther productvty,, from a dstrbuton g( ), whch s assumed to be common across ndustres and countres. As n Meltz (2003), frms then face an exogenous probablty of death n each perod,. There are two productvty cut-offs, the producng productvty cut-off, *, above whch frms produce for the domestc market, and the exportng productvty cut-off, * x, above whch frms produce for both the domestc and export markets: r f u r * 1 d ( ) ( ) ( ) (13) r f u r * 1 x ( x ) x( ) ( ) (14) There s an equlbrum relatonshp between the two productvty cut-offs (see the proof n the Appendx):, where * * x P R f x A A P R f 1/ 1 (15) Frms decsons concernng producton for the domestc and foregn markets are summarsed as follows. Of the mass of frms, M e, that enter the ndustry each perod, a fracton, G * ( ), attan a productvty level suffcently low that they are unable to cover fxed producton costs and ext the ndustry mmedately; a fracton, 8
G * * ( x ) G( ), realse an ntermedate productvty level such that they are able to cover fxed producton costs and serve the domestc market but are not proftable to export; and a fracton, * 1 ( ) G x, reach a productvty level suffcently hgh that t s proftable to serve both domestc and foregn markets n equlbrum. Note that G( ) s a cumulatve dstrbuton functon for g( ). * The ex-ante probablty of successful entry s 1 G( ), and the ex-ante probablty of exportng condtonal on successful entry s: H * 1 G( x ) * 1 G( ) (16) 2.4. Free entry There s an unbounded compettve frnge of potental entrants, and n an equlbrum wth postve producton of both goods, we requre the expected value of entry, V, to equal the sunk entry cost n each ndustry. 1 G( ) ( ) ( ) ( ) 1 V d x fe u r, (17) where d and x denote the average proftablty n the domestc and export markets. It can be demonstrated that ( ), ( ), where d d x x x s the weghted average productvty of frms that sell domestcally and weghted average productvty of frms that export: 1 1 1/( 1) g d 1 G( ) x s the ( ) ( ( ) ) (18) 1 1 1/( 1) x g d 1 G( x) x ( ) ( ( ) ) (19) x 9
Then we can wrte the free entry condton as a functon of the two productvty cutoffs and model parameters (see the proof n the Appendx): f f V g d g d f 1 x 1 (( ) 1) ( ) (( ) 1) ( ) e (20) x x 2.5. Goods markets The steady state equlbrum s charactersed by a constant mass of frms enterng an ndustry n each perod, M, and a constant mass of frms producng wthn the e ndustry, M. Thus, n steady-state equlbrum, the mass of frms that enter and attan a productvty level suffcently hgh to produce must equal the mass of frms that de: (1 G( )) M M (21) e Usng the equlbrum prcng rule, the ndustry prce ndces can be wrtten as: P ( M ( p ( )) M ( p ( )) ) (22) 1 A A A A 1 1/(1 ) d d x In equlbrum, we also requre that the sum of domestc and foregn expendtures on domestc varetes equals the value of domestc producton (total ndustry revenue, for each ndustry and country: d 1 A d x 1 ( ) ( ) A P P R ) p ( ) p ( ) R R M R M (23) Wth free trade nto each ndustry, total ndustry revenue equals total labour payments: R u N r K (24) Requrng that equaton (23) holds for all countres and ndustres mples that the goods markets clear at the world level. 2.6. Factor markets Factor market clearng requres the demand for labour used n producton, export and entry equal factor supply as determned by countres endowments: 10
N1N2 N, K1K2 K, N N N N p x e K K K K (25) p x e where superscrpt p refers to a factor used n producton, superscrpt x refers to a factor used n export, and superscrpt e refers to a factor used n entry. Here we omt the country ndex for smplcty. 2.7. ostly trade equlbrum The costly trade equlbrum s referenced by a vector of 13 varables n hna and the US: { k 1, k 2, k 1, k 2, k 1, k 2, k 1 ( ), k 2 ( ), k 1 ( ), k 2 ( ), k, k, k x x P P p p p x p x u r R } for k {, A}. All other endogenous varables can be wrtten as functons of these quanttes. The equlbrum vector s determned by the followng equlbrum condtons for each country: frms prcng rule (equatons [4] and [5] for each ndustry), free entry condtons (equaton [20] for each sector), the relatonshp between the two productvty cut-offs (equaton [15] for each sector), factor market clearng condtons (equaton [25] for factor N and captal K), the values for the equlbrum prce ndces mpled by consumer and producer optmzaton (equaton [22] for each sector), and the world s expendture on country s varetes equals the value of ther producton (equaton [23] for each sector). Proposton 1. There exsts a unque costly trade equlbrum referenced by the par of equlbrum vectors, { ˆ, ˆ, ˆ, ˆ, Pˆ, Pˆ, pˆ ( ), pˆ ( ), pˆ ( ), pˆ ( ), uˆ, rˆ, Rˆ } k k k k k k k k k k k k k 1 2 1x 2x 1 2 1 2 1x 2x for k {, A}. Proposton 2. Assumng that hna s mport from Indonesa s not affected by the trade lberalsaton between hna and the US, then the openng of costly trade between hna and the US ncreases the steady-state zero-proft productvty cut-off n both ndustres. 11
(a) Other thngs beng equal, the ncreases n the steady-state zero-proft productvty cut-off s greater n a country s ndustry that has a comparatve advantage: A A 1 2 and 2 1. (b) Other thngs beng equal, the exportng productvty cut-off s closer to the zeroproft productvty cut-off n a country s ndustry that has a comparatve advantage: A A A A 1x / 1 2x / 2 and 2x / 2 1x / 1. (c) When trade s costly, only a subset of frms wll export. As a result, trade has a dfferental effect on the profts of exportng and non-exportng frms. Along wth movng from autarky to costly trade, the ex post profts of more productve exportng frms rse. Ths ncreases the expected value of entry n each ndustry because there s a postve ex-ante probablty of achevng a productvty suffcently hgh to export. Ths nduces more entry, and so rases the mass of actve frms n the ndustry. The ndustry becomes more compettve, and the ex post profts of low-productvty frms that only serve the domestc market are reduced. As a result, some low-productvty domestc frms no longer receve enough revenue to cover fxed producton costs and ext the ndustry. The zero-proft productvty cut-off rses. Profts n the export market are relatvely larger to profts n the domestc market n ndustres that have a comparatve advantage. Therefore, along wth the openng of trade, the ex post profts of more productve exportng frms rse more n ndustres that have a comparatve advantage. As a result, the expected value of enterng the ndustry rses further n ndustres that have a comparatve advantage, whch nduces relatvely more entry and leads to a larger ncrease n the zero-proft productvty cutoff n ndustres that have a comparatve advantage. Last, snce exportng s relatvely more attractve n ndustres that have a comparatve advantage, the exportng productvty les closer to the zero-proft productvty cut-off. For hna, ndustry 1 s the ndustry that has a comparatve advantage. Our model predcts that the zero-proft productvty cut-off wll ncrease more and the exportng productvty cut-off wll be closer to the zero-proft productvty cut-off n ndustry 1. 12
3. Data and data sources Our study focuses only on manufactured goods. Our data set s constructed by means of a merger of the Indonesa Survey of Industry and Export and Import at the frm and product levels wth hna s customs data (mports of hna by products from Indonesa). Below we present a bref ntroducton to our hnese frm-level producton data and customs transacton-level trade data. 3.1. hnese frm-level producton data The sample s derved from a rch frm-level panel data set that covers between 162,885 frms (n 2000) and 301,961 frms (n 2006). The data are collected and mantaned by hna's Natonal Bureau of Statstcs (NBS) n an annual survey of manufacturng enterprses. omplete nformaton on the three major accountng statements (.e. balance sheet, proft and loss account, and cash flow statement) s avalable. In bref, the data set covers two types of manufacturng frms all stateowned enterprses (SOEs) and non-soes whose annual sales exceed RMB5 mllon ($830,000). The data set ncludes more than 100 fnancal varables lsted n the man accountng statements of these frms. Although the data contan rch nformaton, some samples are stll nosy and are therefore msleadng, largely because of msreportng by some frms. Followng Feenstra, L, and Yu (2014), we clean the sample and omt outlers by usng the followng crtera. Frst, observatons wth mssng key fnancal varables (such as total assets, net value of fxed assets, sales, and gross value of the frm's output productvty) are excluded. Second, we drop frms wth fewer than eght workers snce they fall under a dfferent legal regme, as mentoned n Brandt, van Besebroeck, and Zhang (2012). We delete observatons accordng to the basc rules of the Generally Accepted Accountng Prncples (GAAP) f any of the followng are true: (1) lqud assets are greater than total assets; (2) total fxed assets are greater than total assets; (3) the net value of fxed assets s greater than total assets; (4) the frm's dentfcaton number s 13
mssng; or (5) an nvald establshed tme exsts (e.g. the openng month s later than December or earler than January). After applyng such a strngent flter to guarantee the qualty of the producton data, the fltered frm data are reduced by about 50 percent n each year. To ensure the precseness of the estmatons, we exclude such tradng companes from the sample n all estmatons. In partcular, frms wth names ncludng any hnese characters for tradng company or mportng and exportng company are excluded from the sample. 3.2. hnese producton-level trade data The extremely dsaggregated product-level trade transacton data are obtaned from hna's General Admnstraton of ustoms. It records a varety of nformaton for each tradng frm's product lst, ncludng tradng prce, quantty, and value at the HS eght-dgt level. More mportantly, the data nclude not only both mport and export data but also break down the data nto several specfc types of processng trade, such as processng wth assembly and processng wth nputs. Overall, when focusng on the hghly dsaggregated HS eght-dgt level, approxmately 35 percent of the 18,599,507 transacton-level observatons are ordnary trade, and 65 percent refer to processng trade. Smlar proportons are obtaned when measurng by trade volume: around 43 percent of trade volume comprses ordnary trade. Processng wth nputs accounts for around 30 percent, whereas processng wth assembly only s around 10 percent. The remanng 17 percent represents other types of processng trade, asde from assembly and processng wth nputs. 3.3. Indonesan producton-level trade data Our paper ams to see the mpact of mported ntermedates on a southern mportng country (.e., hna) from another southern country (.e. Indonesa) affectng the ntensve and extensve margns of the hostng southern country (.e. hna). To ths purpose, we also need frm-level producton data for Indonesa. We are able to access such data for Indonesa for the same sample perod of 2000 2006 covered by the hna data. 14
However, as n other papers, we face a serous challenge when we try to match hna s transacton-level customs data and Indonesa s product-level data sets. Admttedly, we know that specfc Indonesan frms export to hna; unfortunately, we do not know to whch hnese frms they export. Thus, we are not able to match the Indonesan manufacturng exportng frms and hnese manufacturng mportng frms one by one. To work around ths data challenge, we nstead rely on hnese transacton-level customs data n ths paper. As the hnese transacton-level customs data also report the mportng orgns, we thus focus on all mports from Indonesa, the largest developng country n the ASEAN countres. We frst select all sample members wth any mports from Indonesa. To make sure mport from Indonesa plays an mportant role for hnese mporters, we focus on frms wth large mports from Indonesa, especally those frms wth mport shares of more than 5 and 10 percent from Indonesa, respectvely. Last, to calculate and estmate frms total factor productvty (TFP), we need to merge manufacturng frm data and customs data. The detaled approach has been ntroduced n Yu and Tan (2012) and Yu (). In partcular, we use the hnese frm s name-year, zp code, and the last seven dgt of the telephone number to merge the two datasets. As dscussed n Yu (), our merged data skew toward larger tradng frms as the matched sample has more export, more sales, and even larger number of employees. 4. Emprcal fndngs Before formally examnng the nexus between trade lberalsaton and frm exports, we look at Table 1, whch reports the performance of overall exporters and exporters wth large mport shares from Indonesa. By comparng all hnese exportng frms, those exportng frms wth a sgnfcant mport share from Indonesa (.e. mports from Indonesa as a proporton of ther total mports) tend to have better performance n terms of export value, number of employees, and sales. In partcular, of a total of 70,369 hnese exportng frms durng 20002006, 1,387 exportng frms had more 15
than a 5 percent mport share from Indonesa and 995 frms had more than a 10 percent mport share from Indonesa. Although frms wth sgnfcant mports from Indonesa have better performance than those wthout, ths does not mply that the larger the mport share from Indonesa, the better the frm s performance wll be. For example, hnese frms wth more than a 10 percent mport share from Indonesa apparently export less to other countres than those wth more than a 5 percent mport share, suggestng that frm performance has no smple lnear relatonshp wth ts mport share from Indonesa. Table 1: Overall Exporters and Exporters wth Large Import Shares from Indonesa All Exportng Frms >5% Import Share >10% Import Share from Indonesa from Indonesa Varable Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Log Exports 9.664 1.694 10.515 1.683 10.466 1.720 Log Number of Employees 5.456 1.167 5.876 1.249 5.853 1.283 Log Sales 10.802 1.337 11.504 1.564 11.465 1.584 Number of Frms 70369 1387 995 Note: hnese exporters reported n ths table are large exporters by matchng hnese frm-level data and customs data from 2000 to2006. Table 2 presents the summary statstcs for some key varables used n the estmatons. We report smple-average hnese-ndustry-classfcaton (I) two-dgt ndustry-level output mport tarffs, and external tarffs mposed by hna s tradng partners. The external tarffs are smaller than output tarffs, as hna s major tradng partners are developed countres that tend to have lower mport tarffs due partly to the World Trade Organzaton s dscplne and partly to nternatonal trade agreements. We measure nput tarffs at the frm-level to capture the feature of zero mport tarff of processng mports. It s mportant to stress that frm-level nput tarffs are much lower than output tarffs (see Yu, for detaled dscussons). To ths end, we also construct the dummy of processng ndcator and fnd that around 27 percent of frms are processng mporters. Last, we report the frm s export scope and mport scope by countng the HS eght-dgt product lnes reported n hna s customs data. On average, 16
hnese frms export around 7 products to, but mport more than 21 products from, the rest of the world. Table 2: Statstcs Summary of Key Varables All Exporters >5% Import Shares >10% Import Shares from Indonesa from Indonesa Std. Std. Varable Mean Dev. Mean Dev. Mean Std. Dev. Exports 9.664 1.694 10.515 1.683 10.466 1.720 Home Output Tarffs (ndustry-level) 11.71 0.056 11.80 0.058 11.74 0.057 Foregn Industry Tarffs 9.60 0.048 10.13 0.050 10.02 0.049 Home Input Tarffs (frm-level) 2.554 4.255 1.536 3.135 1.561 3.256 Frm TFP (Olley-Pakes) 1.072 0.668 1.196 0.863 1.202 0.862 Foregn Indcator 0.569 0.495 0.774 0.419 0.763 0.426 SOE Indcator 0.021 0.142 0.013 0.113 0.013 0.114 Log Labour 5.456 1.167 5.876 1.249 5.853 1.283 Processng Indcator 0.271 0.445 0.513 0.500 0.490 0.500 Export Scope 7.421 10.990 8.640 11.127 8.254 10.855 Import Scope 20.595 37.301 26.358 41.646 23.819 39.358 By way of comparson, frm TFP ncreases from 1.07 for all hnese exporters to 1.19 for hnese exporters wth more than a 5 percent mport share from Indonesa and 1.20 for those wth more than a 10 percent mport share from Indonesa, suggestng that the hgher the mport share from Indonesa, the hgher the frm productvty wll be. It s also mportant to stress that the share of processng (ndcated by processng ndcator) s hgher for frms wth hgher mport shares from Indonesa than that of the average exportng frms. The frms wth hgher more than a 5 percent of mport share from Indonesa have 50 percent of processng actvtes compared to 27 percent for the average of all hnese exporters. 4.1. Trade lberalsaton and frm export Table 3 shows the estmatons of the mpact of trade lberalsaton on frm exports. olumns (1) (4) nclude hnese exporters wth more than a 10 percent mport share 17
from Indonesa whereas olumns (5) (7) nclude those frms wth more than a 5 percent mport share. Several mportant fndngs deserve to be hghlghted. Frst, the coeffcents of frm productvty are postve and sgnfcant n all estmates, ndcatng that frms wth hgh productvty tend to export more. More mportantly, the magntude of frm TFP ncreases wth the mport shares from Indonesa, suggestng that the effect of TFP on frm exports s more pronounced for frms wth more mports from man developng countres lke Indonesa. The economc ratonale s reasonably clear. As hnese frms mport more ntermedate nputs or raw materals from Indonesa, they are more lkely to engage n processng trade (as confrmed n Table 2) and hence export more. Wth more mported ntermedate goods, frms are able to employ the advantage of the combnaton of domestc nputs and mported nputs, as suggested by Halpern et al. (2011). Table 3: Estmates of Trade Lberalsaton on Frm Exports Regress and: Log Frm Exports (1) (2) (3) (4) (5) (6) (7) Import Share from Indonesa >10% >5% Home Output Tarffs -2.218* -2.699** -2.048* -2.005-2.509*** -1.845* -2.062* (Industry Level) (-1.65) (-2.50) (-1.81) (-1.56) (-2.74) (-1.93) (-1.95) Foregn Tarffs -2.914** -2.299** -2.042** -1.863* -2.12*** -1.88** -1.749** (Industry Level) (-2.19) (-2.36) (-2.09) (-1.81) (-2.62) (-2.32) (-2.04) Home Input Tarffs -0.051-0.055** -0.056** -0.059** -0.06*** -0.05*** -0.060*** (Frm Level) (-1.60) (-2.16) (-2.13) (-2.28) (-2.78) (-2.70) (-2.83) Frm TFP (Olley-Pakes) 0.304*** 0.158*** 0.140*** 0.144*** 0.108** 0.091* 0.099** (3.86) (3.03) (2.71) (2.75) (2.26) (1.94) (2.07) Foregn Indcator 0.033 0.086 0.100 0.185* 0.234** 0.238** (0.29) (0.75) (0.85) (1.80) (2.27) (2.29) SOE Indcator 0.749*** 0.920*** 0.939*** 0.833*** 1.013*** 1.031*** (4.54) (4.84) (4.38) (5.61) (6.10) (5.67) Log Frm Labour 0.891*** 0.895*** 0.903*** 0.890*** 0.890*** 0.897*** (23.59) (24.00) (23.97) (25.77) (26.02) (26.11) Processng Indcator 0.240** 0.253** 0.272** 0.213** 0.236*** 0.253*** (2.35) (2.42) (2.58) (2.44) (2.66) (2.83) Year Fxed Effects No No Yes Yes No Yes Yes Industry Fxed Effects No No No Yes No No Yes Observatons 743 743 743 743 1008 1008 1008 R-squared 0.04 0.47 0.48 0.49 0.46 0.47 0.47 Note: Numbers n parentheses are robust t-value. *(**, ***) denotes sgnfcance at 10% (5%, 1%). 18
Second and equally mportant, we fnd that trade lberalsaton wll boost exports. Ths s frm for all aspects of trade lberalsaton, ncludng output tarff reductons, nput tarff reductons, and foregn tarff reductons. Wth nput trade lberalsaton, frms are able to save costs n ntermedate nputs, and thus earn more proft. Smlarly, wth lower tradng partners tarffs, frms gan easer access to foregn markets and have more exports. By contrast, the role of output trade lberalsaton s dfferent. A large degree of output tarff reductons suggests tough mport competton effects from nternatonal markets. Thus, only effcent frms are able to survve n the markets. As effcent frms are larger and export more, we see negatve coeffcents of output tarffs. Last, SOEs tend to have more exports, and larger frms tend to export more. Also, processng frms have more exports, whch makes good sense as processng frms, by defnton, wll export all products to the foregn markets. 4.2. Trade lberalsaton and export and mport scope Table 3 examnes the ntensve margn of trade lberalsaton on frm exports. We now move to explore the mpact of trade lberalsaton on the extensve margn of exports. In partcular, we focus on the change n export and mport scopes. By defnton, followng Qu and Yu (2014), we defne a frm s export scope as the number of HS eght-dgt product exported by a hnese manufacturng frm. We consder the followng emprcal specfcaton: es t = β 0 + β 1 TFP t + β 2 OT t + β 3 IT t + β 4 ET t + θx t + ε t where es t s frm 's export product scope, TFP t s total factor productvty, OT t s (hnese) tarff level faced by the frm, IT t s nput mport tarff level faced by the frm, and ET t s the foregn tarff level faced by frm at year t. X s a vector of control varables, ncludng frm s sze, ownershp type (SOE, multnatonal frm, or others), and trade mode (processng or ordnary trade). Table 4 reports the count-data estmates of trade lberalsaton on frm export scope. As before, columns (1) (3) nclude a sample of hnese exporters wth more than 10 percent mport share from Indonesa and columns (4) (6) cover observatons of frms wth more than 5 percent mport share from Indonesa. 19
We start from the Posson estmates n whch the mean of export scope s presumed to equal ts varance. The Posson estmate n column (1) suggests that both home output tarffs and foregn trade lberalsaton decrease a frm s export scope. In addton, a frm s nput tarffs overall decrease export scope. Such fndngs are exactly consstent wth the fndngs of Qu and Yu (2014) whch covered the whole sample of hnese exporters. The economc ratonale of the postve coeffcent of output tarff s straghtforward. Lower output tarffs lead to tougher mport competton, whch n turn makes frm focus on ther compettve products. However, at frst glance, the postve coeffcent of foregn tarffs s counter-ntutve. However, ths s just because of the trade-off between postve shock and negatve shock rased by a tradng partner s tarff reductons. As clearly presented n Qu and Yu (2014), lower foregn tarffs has both postve and negatve shock effects on a frm s export scope. Once the negatve competton mpact domnates the postve one, export scope falls. However, the assumpton that the mean of the export scope equal ts varance seems too strong. Instead, we adopt the negatve bnomal estmates n column (2) for hnese exporters wth more than a 10 percent mport share from Indonesa and those n column (5) wth more than a 5 percent mport share from Indonesa. The negatve bnomal estmates are more attractve here as they allow the sample to exhbt a pattern of over-dsperson. However, one may have a concern that some other macroeconomc fluctuatons such as Renmnb apprecaton durng the sample perod, partcularly, after 2005 may affect a frm s export scope. In addton, other unspecfed factors such as a frm s manageral effcency, as ntroduced n Qu and Yu (2014), may also affect sad frm s extensve margn. We thus control for frm-specfc fxed effects and year-specfc fxed effects n columns (3) and (6). It turns out that the bnomal estmaton results n columns (2) (3) and (5) (6) are qualtatvely dentcal to ther counterparts n columns (1) and (4) wth Posson estmates. Thus, our estmatons are nsenstve to dfferent emprcal specfcatons. 20
Table 4: ount-data Estmates of Trade Lberalsaton on Frm Export Scope Regresson: Export Scope (1) (2) (3) (4) (5) (6) Econometrc Method Posson Negatve Bnomal Posson Negatve Bnomal Import Share from Indonesa > 10% >5% Home Output Tarffs 0.724*** 1.100** 0.942** 1.102*** 1.347*** 0.871*** (Industry Level) (4.75) (2.57) (2.36) (9.05) (3.79) (2.71) Foregn Tarffs 5.078*** 4.189*** 1.709*** 4.472*** 3.848*** 1.782*** (Industry Level) (21.68) (6.97) (3.05) (23.17) (7.60) (3.78) Home Input Tarffs -0.006-0.007 0.004-0.016*** -0.013* -0.001 (Frm Level) (-1.64) (-0.85) (0.45) (-4.87) (-1.90) (-0.13) Frm TFP (Olley-Pakes) 0.353*** 0.425*** 0.226*** 0.324*** 0.397*** 0.233*** (14.31) (5.53) (2.96) (15.37) (6.02) (3.84) Foregn Indcator -0.200*** -0.114-0.047-0.128*** -0.067-0.036 (-7.73) (-1.55) (-0.56) (-5.78) (-1.05) (-0.49) SOE Indcator 0.093-0.043 0.138-0.071-0.138-0.046 (1.20) (-0.17) (0.42) (-1.02) (-0.64) (-0.16) Log Frm Labour 0.187*** 0.187*** 0.202*** 0.222*** 0.222*** 0.201*** (20.87) (8.06) (7.11) (28.75) (10.92) (8.02) Processng Indcator -0.259*** -0.27*** -0.12*** -0.14*** -0.17*** -0.10*** (-10.82) (-4.50) (-2.63) (-7.40) (-3.41) (-2.65) Year-specfc Fxed Effects No No Yes No No Yes Frm-specfc Fxed Effects No No Yes No No Yes Observatons 948 948 948 1323 1323 1323 Note: Numbers n parentheses are robust t-value. *(**, ***) denotes sgnfcance at 10% (5%, 1%). In addton to the above fndngs, we also observe that large-szed frms have relatvely more export scope than average frms. Interestngly, compared to nonprocessng frms (.e. ordnary frms), processng frms seem to have less export scope. ombned wth the above fndngs that processng frms have relatvely hgher export value, as shown n Table 3, the mplcaton s clear: processng exporters reduce the varety of ther trade products but focus on ther core compettve products. Last, the negatve sgn of foregn ndcator suggests that multnatonal companes based n hna have less export scope. Such a fndng s consstent wth the fact that processng frms also have less export scope, as processng frms generally are subsdares of multnatonal companes, as documented n Da et al. (2012). Table 5 shows the mpact of trade lberalsaton on a frm s mport scope. Once agan, trade lberalsaton s measured over three dmensons: output tarffs reductons, 21
nput tarffs reductons, and foregn tarff reductons. olumns (1) and (3) of Table 5 are Posson estmates whereas the rest are negatve bnomal estmates. Meanwhle, columns (1) (3) are estmates for hnese exporters wth more than a 10 percent mport share from Indonesa whereas columns (4) (6) are frms wth more than a 5 percent mport share. Table 5: ount-data Estmates of Trade Lberalsaton on Frm Import Scope Regresson: Import Scope (1) (2) (3) (4) (5) (6) Econometrc Method Posson Neg. Bnomal Posson Neg. Bnomal Import Share from Indonesa > 10% >5% Home Output Tarffs -0.073-1.419*** -0.601*** -0.977*** -1.183*** -1.038** (Industry Level) (-0.49) (-13.96) (-5.98) (-8.10) (-14.87) (-2.52) Foregn Tarffs -2.214*** -1.164*** -0.439*** -2.415*** -1.469*** -0.135 (Industry Level) (-13.57) (-7.79) (-3.45) (-18.30) (-12.32) (-0.24) Home Input Tarffs 0.014*** 0.023*** 0.019*** 0.022*** 0.029*** 0.046*** (Frm Level) (7.41) (12.20) (10.28) (13.92) (18.86) (3.92) Frm TFP (Olley-Pakes) 0.260*** 0.271*** 0.192*** 0.340*** 0.346*** 0.540*** (16.36) (17.43) (11.70) (26.06) (27.68) (7.67) Foregn Indcator 1.221*** 1.249*** 1.143*** 1.168*** 1.224*** 1.116*** (54.47) (55.68) (46.68) (63.41) (65.98) (16.19) SOE Indcator -0.846*** -0.865*** -0.932*** -0.860*** -0.810*** -0.727*** (-8.66) (-10.33) (-7.93) (-10.33) (-11.50) (-2.92) Log Frm Labour 0.497*** 0.473*** 0.475*** 0.468*** 0.454*** 0.455*** (94.06) (93.53) (78.49) (107.16) (107.85) (20.67) Processng Indcator -0.108*** -0.128*** -0.096*** -0.074*** -0.097*** -0.067 (-7.31) (-8.93) (-9.13) (-6.18) (-8.42) (-1.14) Year-specfc Fxed Effects No No Yes No No Yes Frm-specfc Fxed Effects No No Yes No No Yes Observatons 948 948 948 1323 1323 1323 Note: Numbers n parentheses are robust t-value. *(**, ***) denotes sgnfcance at 10% (5%, 1%). Table 5 llustrates that foregn tarff reductons ncrease a frm s mport scope due to stmulated foregn demand and larger access to foregn markets. We also fnd that home output tarff reductons wll ncrease a frm s mport scope. The mplcaton s straghtforward. Wth a tougher mport competton, frms mport more foregn (Indonesan) varetes possbly due to better qualty. Strkngly enough, home nput tarff reductons are found to decrease frm s mport scope. As nput trade lberalsaton 22
may have cost-savng effects, t n turn ncreases frm proftablty and hence push frms to mport more one possble reason to nterpret such a counter-ntutve fndng. The frst one s due to the sample restrcton as our sample only covers large exportng frms. Wth large proftablty, large frms could nstead use mode domestc varetes or mport less number of varetes but of a hgher volume. 4.3. More robustness checks Table 6: Estmates of Trade Lberalsaton on Frm Productvty Import Share from Indonesa >10% >5% Regressand: Frm TFP (system GMM) (1) (2) (3) (4) Home Output Tarffs -1.177*** -0.666** -1.343*** -0.925*** (Industry Level) (-4.76) (-2.08) (-6.46) (-3.42) Foregn Tarffs -0.770*** -1.089*** -0.768*** -1.034*** (Industry Level) (-2.70) (-3.17) (-3.24) (-3.57) Home Input Tarffs 0.237 0.412 0.249 0.329 (Frm Level) (0.71) (0.95) (0.83) (0.84) Foregn Indcator 0.138 0.357** 0.064 0.209 (0.70) (2.22) (0.43) (1.63) SOE Indcator -0.002 0.028 0.016 0.038 (-0.05) (0.76) (0.60) (1.20) Log Frm Labour 0.067*** 0.067*** 0.069*** 0.063*** (6.92) (5.61) (8.27) (5.94) Processng Indcator -0.092*** -0.087*** -0.085*** -0.084*** (-3.61) (-2.62) (-3.89) (-2.98) Year-specfc Fxed Effects No Yes No Yes Frm-specfc Fxed Effects No Yes No Yes Observatons 828 828 1156 1156 R-squared 0.15 0.21 0.15 0.19 Note: Numbers n parentheses are robust t-value. *(**, ***) denotes sgnfcance at 10% (5%, 1%). So far we have used the augmented Olley-Pakes TFP to measure frm productvty. Although such a measured TFP has many advantages compared to other alternatve measures of productvty, as dscussed n Yu (), t also has two man dsadvantages. Frst, the Olley-Pakes TFP assumes that frms adjust captal nput when facng an exogenous shock. However, ths may not happen n hna, as hna s a labour- 23
abundant country and, hence, hnese frms fnd t much easer to adjust labour than captal. Second, the Olley-Pakes TFP does not allow output to have any seral correlatons, whch are very lkely to occur. For these reasons, the system-gmm TFP measure seems an deal complementary, as t has enough flexblty to allow for possble seral autocorrelatons. We hence use system-gmm TFP to check whether our results wll reman robust even when usng other measures of TFP. Table 6 pcks up ths comparson. Followng Yu (), we now move to dscuss whether trade lberalsaton boosts frm productvty for hnese exporters wth sgnfcant mport shares from Indonesa. Once agan, we consder frms wth 10 and 5 percent mport shares from Indonesa, respectvely. As n other studes, we fnd that both output trade lberalsaton and external trade lberalsaton boost frm productvty. However, we do not fnd that nput trade lberalsaton rases frm productvty. The mpact of home nput trade lberalsaton on frm productvty s nsgnfcant. Such fndngs are robust even we control for year-specfc fxed-effects and frm-specfc fxed-effects n Table 6 column (2) for frms wth 10 percent mport shares from Indonesa and n column (4) for those frms wth 5 percent correspondng mport shares. Ths rases a concern over the prevous estmates of the effects of trade lberalsaton on frm productvty. One may worry that our estmates above have some estmaton bas. To address ths concern, followng Feenstra, L, and Yu (2014), we dstngush between ex-ante TFP and ex-post TFP measures. 24
Table 7: Estmates of Trade Lberalsaton wth Ex-ante Frm Productvty Regress and: Log Exports Export Scope Import Scope Import Share from Indonesa >5% >5% >10% >5% (1) (2) (2) (4) Home Output Tarffs -0.708 0.682* 0.826* -1.218*** (Industry Level) (-0.78) (1.89) (1.95) (-2.86) Foregn Tarffs -1.936** 2.806*** 4.164*** 0.734 (Industry Level) (-2.36) (5.30) (6.97) (1.16) Home Input Tarffs -0.059*** -0.002-0.005 0.063*** (Frm Level) (-3.24) (-0.23) (-0.64) (5.36) Frm TFP (Olley-Pakes) -0.064 0.749*** 0.666*** 0.025 (-0.49) (9.16) (6.89) (0.27) Foregn Indcator 0.280*** -0.035-0.115 1.134*** (2.82) (-0.57) (-1.58) (16.22) SOE Indcator 0.304 0.052 0.061-0.512** (0.83) (0.26) (0.25) (-2.04) Log Frm Labour 0.893*** 0.247*** 0.236*** 0.471*** (28.39) (12.65) (10.05) (20.61) Processng Indcator 0.258*** -0.171*** -0.281*** -0.056 (3.26) (-3.38) (-4.60) (-0.95) Year-specfc Fxed Effects No Yes Yes Yes Frm-specfc Fxed Effects No Yes Yes Yes Observatons 1192 1324 949 1324 Note: Numbers n parentheses are robust t-value. *(**, ***) denotes sgnfcance at 10% (5%, 1%). The conventonal measures of TFP, ncludng our above TFP measure (nclusve of both Olley-Pakes and Sytem-GMM), s a Solow resdual that ncludes both unspecfed factors and producton productvty. In ths way, the measured TFP certanly correlates wth the error term. To avod such a shortcomng and to be closer wth the sprt of Meltz (2003) that emphasses more on the ex-ante random draw of frm productvty, we exactly follow Feenstra, L, and Yu (2014) and Qu and Yu () to construct an ex-ante TFP. Table 7 reports the estmaton results usng the ex-ante TFP measure. The regressand n column (1) s frm exports, whereas those n columns (2) and (3) are export scope, and that n column (4) s mport scope. Estmates n column (1) show that all types of trade lberalsaton boost frm exports, whch make good economc senses. Meanwhle, all estmates on export scope and mport score are consstent wth estmates wth ex post frm productvty presented n Tables 4 and 5. Thus, our man fndngs are robust when usng dfferent measures of TFP. 25
5. onclusons The man value added of our paper s twofold. Frst, we construct a theoretcal model to ncorporate North South South trade pattern. Our theoretcal model predcts that trade lberalsaton n North and South producton countres can boost frm exports. Second, we provde emprcal exercses usng very detaled and hghly dsaggregated hnese data to test such predctons. In partcular, we use both hnese frm-level producton and transacton-level trade data to examne the effects of three types of tarff reducton on frm export, frm productvty, and frm export and mport scope by consderng vertcal ntegraton among South, producton South and onsumpton north. Our fndngs assert that trade lberalsaton sgnfcantly boosts frm productvty, and hence rases frm exports va both extensve margn (.e. export scope and mport scope) and ntensve margn. Such fndngs are consstent wth our theoretcal fndngs. Moreover, our fndngs provde nsghtful polcy mplcatons. Frst, f deeper ntegraton between North and South can ncrease trade flows, governments n the South and North should provde more trade facltaton to make trade ntegraton possble. Second and equally mportant, we fnd that trade lberalsaton n the destnaton countres (most lkely n the North) and n the producton countres (most lkely n the South) boosts frm productvty and rases trade flows. Thus, t would be a wse strategy for tradng countres to cut ther tarffs, phase out other non-tarff barrers, and mprove transparency of non-tarff measures. To understand the exact channel or mechansm of the correlaton between home nput trade lberalsaton and mport scope wll be an nterestng ssue to explore n a future study. 26
References Bernard, A., S. Reddng, and P. Schott (2007), omparatve Advantage and Heterogeneous Frms, Revew of Economc Studes, 74, pp.31 66. Da, M., M. Matra, and M. Yu (2012), Unexceptonal Exporter Performance n hna? Role of Processng Trade, mmeo, Pekng Unversty. Feenstra, R., Z. L, and M. Yu (2014), Exports and redt onstrants under Incomplete Informaton: Theory and Applcaton to hna, Revew of Economcs and Statstcs, 96(4), pp.729 744. Felbermayr, G.J., and W. Kohler (2006), Explorng the ntensve and extensve margns of World Trade, Revew of World Economcs, 142(4), pp.642 674 Fernandes, A.M. (2007), Trade polcy, trade volumes and plant-level productvty n olumban manufacturng ndustres, Journal of Internatonal Economcs, 71(1), pp.52 71. Goldberg, P., A. Khandelwal, N. Pavcnk, and P. Topalova (2010), Imported Intermedate Inputs and Domestc Product Growth: Evdence from Inda, Quarterly Journal of Economcs, 125 (4), pp.1727 1767. Hummels, D., J. Ish, and K.-M. Y (2001), The nature of growth of vertcal specalzaton n world trade, Journal of Internatonal Economcs, 54, pp.75 96. Hummels, D. and P.L. Klenow (2005), The Varety and Qualty of a Naton s Exports, Amercan Economc Revew, 95(3), pp.704 723. Koopman, R., Z. Wang, and S.-J. We (2008), How much of hnese exports s really made n hna? Assessng domestc value added when processng trade s pervasve, NBER Workng Paper No. 14109. Lleeva, A. and D. Trelfer (2010), Improved Access to Foregn Markets Rases Plant- Level Productvty for Some Plants, Quarterly Journal of Economcs, 125(3), pp.1051 1099. Lnder, Staffan Burenstam (1961), An Essay on Trade and Transformaton, Stockholm: Almqvst and Wksell. Helpman, E., M. Meltz, and Y. Rubnsten (2008), Estmatng Trade Flows: Tradng Partners and Tradng Volumes, Quarterly Journal of Economcs, 123(2), pp.441-487. Meltz, M. (2003), The Impact of Trade on Intra-ndustry Reallocatons and Aggregate Industry Productvty, Econometrca 71(6), pp.1695 1725. Pavnck, N. (2002), Trade lberalsaton, Ext and Productvty mprovement: evdence from hlean Plants, Revew of Economc Studes, 69(1), pp.245 276. Qu, L.D. and M. Yu (2014), Manageral Effcency and Product Decson: Evdence from hnese Frms, mmeo, Pekng Unversty. 27
Tan, W. and M. Yu, Processng Trade, Export Intensty, and Input Trade Lberalzaton: Evdence from hnese Frms, Journal of Asa-Pacfc Economy, forthcomng. Topalova, P. and A. Khandelwal (2011), Trade Lberalzaton and Frm Productvty: The ase of Inda, Revew of Economcs and Statstcs 93(3), pp.995 1009. Xng, Y. and N. Detert (2010), How the Phone wdens the Unted States trade defct wth the People s Republc of hna, ADBI Workng Paper No. 257. Xng, Y. (2012), Processng trade, exchange rates, and hna s blateral trade balances, Journal of Asan Economcs, 23, pp.540 547. Y, K-M. (2010), an Mult-Stage Producton Explan the Home Bas n Trade?, Amercan Economc Revew, 100 (1), pp.364 393. Yu, M. (), Processng Trade, Tarff Reductons and Frm Productvty: Evdence from hnese Frms, Economc Journal 125 (June), pp.943 988. 28
Appendx Proof of equaton (15): ombnng equaton (6) and (13), we have: P (A1) ( u ) ( r ) * * 1 1 rd ( ) R ( ) f ( ) ( ) 1 u r ombnng equaton (7) and (14), we have: P (A2) A * * A x 1 1 rx ( x ) R ( ) f ( ) ( ) 1 x u r ( u ) ( r ) omparng equaton (A1) and (A2), we can fnd: P ( ) ( ) Rearrangng (A3), we get equaton (15): R * A A x 1 1 x * P R f f (A3) Proof of equaton (20):, where * * x P R f x A A P R f 1/( 1) (15) In equlbrum, the expected proft of entry should equal the entry cost, so we have: 1 1 V g d g d f u r 1 ( ( ) ( ) ) ( ( ) ( ) ) ( ) ( ) * d * x e x (A4) Where accordng to equaton (10): From equaton (6), t s easy to see: r ( ) d 1 d ( ) f( u ) ( r ) (10) r r ( ) ombnng equaton (A5) and (13), we have: Insert equaton (A6) nto equaton (10), then: d 1 ( ) * * (A5) d ( ) r f u r 1 1 d ( ) ( ) ( ) ( ) * (A6) 29