Trade Scopes across Destinations: Evidence from Chinese Firm

Transcription

1 MPRA Munih Personal RePE Arhive Trade Sopes aross Destinations: Evidene from Chinese Firm Zhuang Miao and Yifan Li MGill University 15 Marh 2017 Online at MPRA Paper No , posted 10 September :23 UTC

2 Trade Cost and Export Diversifiation: Evidene from Chinese Firms Zhuang Miao MGill University Yifan Li MGill University September 9, 2017 Abstrat We examine how Chinese exporters adjust their number of exported varieties with respet to different harateristis of destination ountries and varying trade ost. Using the Chinese firm-level ustoms data from the years 2001 and 2006, we show that: (i) firms export fewer varieties (indexed by HS6 ode) to the destinations whih are with higher exhange rate volatility, farther from China, or impose higher import-tariff rate; (ii) in response to the tariff redution proess by the destination ountries after China entering to the WTO in 2001, the high produtivity firms expanded the export sope while the low produtivity firms redued it.with a theoretial framework whih onsiders firms optimization deision involving both prodution and export varieties, we explain all our empirial findings, highlighting the relation between the exhange rate volatility and the number of export varieties. China JEL Classifiation: F12 F14 F31 Key words: Multiprodut firm; Produt sope; Exhange rate volatility; Transportation distane; Tariff Redution; Corresponding author. Department of Eonomis, MGill University, Montreal, Canada, H3A 0G4. addresses: zhuang.miao@mail.mgill.a (Z. Miao) yifan.li4@mail.mgill.a (Y. Li) 1

3 1 Introdution 1.1 Bakground How harateristis of the destination ountries play roles in the deisions of export firms (e.g. entry deision, the prie or quantity strategies, and the number of produt varieties to produe and export) has been widely disussed in reent years. The importane of a firm s deision on how many produt varieties to produe and trade lies on two points: (i) as indiated by Melitz (2003), onsumers prefer diversity of produts and thus firms deisions on produt varieties are diretly orrelated with onsumers welfare; (ii) it helps to redue the market risks a firm faing by providing more varieties to the market. Indeed, if a firm offers only one variety to the market, it will fae relatively high level of demand risks (equal to individual risks of this variety); but if the firm sells plenty number of varieties and the individual risks of these varieties are independent, then the market risks faed by the firm will be redued. This point stands on firms performane as it points to the firms profits and their future growth, i.e. the low-risky firms are easier to borrow money from finanial institutions and more likely to make R&D investment. Several papers studies how firms adjust the number of prodution and export varieties in response to the trade liberalization, e.g. Qiu and Yu (2014), Lopresti (2016), or how the export varieties are adjusted with different levels of exhange rate volatility, e.g. Heriourt and Ponet (2015) and Berthou and Lionel (2013). However, the answers to some questions are still unknown or inomplete: (i) how a single firm adjusts the number of export varieties with varying transportation ost and tariff rate (empirial evidene is missing among the existing literature); (ii) what is the mehanism that exhange rate volatility affet firms deisions on export varieties; (iii) how the heterogeneous firms (differentiated in produtivity) adjust their export varieties in response to a tariff redution. Several papers did some empirial analysis on the third point. Qiu and Yu (2014) looked at how the heterogeneous firms adjust their total number of exported varieties (we define it as prodution sope) in response to the trade liberalization. Lopresti (2016) and Berthou and Lionel (2013) study how the firms adjust their export sopes in response to a trade liberalization. Berthou and Lionel (2013) uses the introdution of the Euro as the treatment to Frenh firms in deiding their export sope. Yet despite its impliations for firm prodution and export performane, there are fundamental differenes between produt sope and export sope, and how firms adjust the number of exported variables towards eah single market (defined as export sope) and how does this effet differ in firm heterogeneity in produtivity remain poorly understood. In this paper, we examine how Chinese exporters adjust their export sope (the number of produt varieties) with 2

4 respet to different harateristis of destination ountries and varying trade ost. Theoretially, we present a flexible heterogeneous firm trade model involving both prodution and export sopes deisions in regard to the exhange rate volatility, tariff rate and distane of the destination ountries. A firm s export sope is defined as the number of produt varieties exported to a speifi market. Speifially, we onsider two produts belong to different varieties if either they are produed by different firms, or if their HS6 lassifiation odes are different. Our researh fouses on the hange of the number of varieties (differentiated by the HS6 ode) provided by the single firm in response to the features of the destination ountries. Using the Chinese firm-level ustoms data from the years 2001 and 2006, we show that firms export fewer varieties (indexed by HS6 ode) to the destinations whih are with higher exhange rate volatility, more distane from the home ountry, or larger tariff rate; Moreover, we exploit China s aession to the WTO in 2001 and show that the high produtivity firms expanded the export sope while the low produtivity firms redued. Our researh is aimed to answer the following researh questions whih are analyzed rarely: (i) how do exporting firms adjust their export sopes to destination ountries that differ in exhange rate sheme and trade ost, i.e. import-tariff rate and transportation distane to the home ountry; (ii) how do heterogeneous firms with different produtivity levels adjust their export and prodution sopes in response to the trade liberalization. With the empirial analysis, we find three main results: (i) the firms export fewer varieties to the destinations with farther distane to China, higher tariff rate or higher exhange rate volatility; in response to a tariff-redution proess in a destination ountry, the high produtivity firms expand while the low produtivity firms redue their export sopes to this ountry; (iii) in response to the trade liberalization, the hange of prodution sope is insignifiant. Our theoretial framework illustrates how heterogeneous firms arrange export sope aross destinations with different harateristis. Following the method by Melitz and Ottaviano (2008), Dhingra (2013), and Qiu and Yu (2014), we onstrut a theoretial model to materialize our preliminary empirial findings to work. In the model, a firm deides prodution sope before realization of the market state and the export sope is thus adjusted based on the realization of the market state. Due to the different length of the proessing time, export sope dereases in the level of exhange rate volatility, transportation distane, and tariff rate of a destination ountry. Further, firm-level high produtive firms pursue ost-saving advantage in produing high quality produts than the low produtive firm. Guided by this oneptual framework, we analyze the operations of firms using transation-level data for China on the universe of export and import transations in 2001 and We take advantage of the observations of prie and sales for all of a firm s exports by destination and produt, whih allow us to examine the relationship between export 3

5 sope aross different destinations where the exhange rate volatility, distane and tariff rate differ. We perform several empirial exerises, and onlude that firms export fewer varieties (indexed by HS6 ode) to the destinations whih are with higher exhange rate volatility, more distane from the home ountry, or larger tariff rate. Moreover, we exploit China s aession to the WTO in 2001 and show that the high produtivity firms expanded the export sope while the low produtivity firms redued. 1.2 Related literature on exhange rate unertainty We ontribute to the international trade literature on the interation of destination markets harateristis and operations of multi-produt firms and on firm heterogeneity in effiieny. There is a rapidly inreasing number of papers studying how exporters behave reat to varying harateristis of the destination markets. Among the literature, the most studied topi is about the hanges of the firms hoies in prie and quality levels. However, to the best of our knowledge, very few papers studied how firms adjust their export sopes towards different harateristis of the destination markets, i.e. the exhange rate sheme, transportation distane and import tariff rate. Heriourt and Ponet (2015) studied how the Chinese firms adjust their export sopes reat to different exhange rate shemes among destination ountries. However, they didn t provide the theoretial model to explain their results. In addition, they didn t study the other variables of the trade ost that involved in our analysis, i.e. the transportation distane and the tariff rate. One of our researh targets is to fill up these gaps among the relevant studies by: providing theoretial foundations to the linkage between exhange rate volatility and export sope; and providing firm-level empirial evidene of the effet of transportation distane on export sope. We study the effets of two features of the destination markets, i.e. the exhange rate volatility and the distane to the home ountry, on the hoies of export sopes by firms. Among the previous papers, Heriourt and Ponet (2015), Sauer and Bohara (2001), and Aghion et al. (2009) studied how the exhange rate volatility in the destination markets would affet the performane of the export firms.these authors fous on the effet of exhange rate volatility on the firms prie, quality, and investment strategies. Among them, the study by Heriourt and Ponet (2015) is mostly related to ours. Heriourt and Ponet (2015) used the real effetive exhange rate to estimate the exhange rate risks in the target ountry, while many other papers using the nominal exhange rate to do the estimation. Our view is that the real exhange rate risk is more relevant than nominal exhange rate risk, beause the real exhange rate is the one 4

6 whih measures the deviation of the ountry s prie index from the world prie and thus reflets the demand unertainty in the market, while the movement of the nominal exhange rate annot fully reflet the demand hanges in the market if the hange of a urreny s value is fully translated in the prie and wage levels. The basi purpose of involving the exhange rate risks in our study is to use it as a measure of the demand unertainty in the destination market. In this ase, following Heriourt and Ponet (2015) we use the real effetive exhange rate to measure the demand unertainty of the market. Basially, there are two main findings in Heriourt and Ponet (2015): the exhange rate risks have negative effets on the firms entry deision, the trade volume, and export sopes to the foreign markets; the negative effets are more signifiant among firms that have severe finanial onstraints. We repeat their works of investigating the relation between exhange rate volatility and export sope with using both the real effetive and nominal exhange rate to onstrut our key independent variable. We find a similar result as theirs, i.e. the exhange rate volatility negatively affet the number of export varieties. As addressed by Heriourt and Ponet (2015), Greenaway and Kneller (2007) and Ethier (1973), the effet of exhange rate volatility is equivalent to that of an inrease in the variable and sunk osts in trading. However, none of the papers onstruted the theoretial frameworks to show how this mehanism works. To fill up this gap among Heriourt and Poet (2015) and other existing literature, we onstrut a theoretial framework to explain how the exhange rate volatility affets the firms hoies on export sopes. Some other reent papers whih studied the firms behaviors and market unertainty inlude Chen and Juvenal (2016), Berman et al. (2012), Nguyen (2012), Lopez and Nguyen (2015), and Bekes et al. (2015). Chen and Juvenal (2016) found that in response to the movements of the exhange rate, the prie of the high-quality produts hanges dramatially but the volume hanges insignifiantly. Berman et al. (2012) found results that are similar to Chen and Juvenal (2016) with the Frenh firm-level data between the years 1995 and Nguyen (2012) attempted to provide a theoretial explanation for the stylized fat that the firms entry some foreign markets shortly but then leave the market later on. He (2012) found the unertainties existing in the new markets fore the firms to make its entry deision before making the output supply deision for that market. Using Chile firm level data from 1995 to 2007, Lopez and Nguyen (2015) studied how the flutuation of the real exhange rate affets the input-importation by the Chilean plants. The paper found that the exhange rate movements redued the volume of the imports but didn t affet the firms deision on the importation. 5

7 1.3 Related literature on transportation ost On the effets of the geographi features on trading, the works by Bastos and Silva (2010), Manova and Zhang (2012), and Lugovskyy and Skiba (2016) are mostly related to our study. The first two papers looked at the firms prie strategies in the foreign markets with different distane to the home market. Using the Portugal firm level data, Bastos and Silva (2010) found that the plants tend to harge higher f.o.b. pries to the longer distane ountries, and their explanation for this phenomenon is that a greater proportion of high produtivity firms export to the farther markets with higher quality produts (usually higher prie). Contrarily, using the Chinese data, Manova and Zhang (2012) found the f.o.b. exporting prie dereases in the transportation distane with the sample of the poor destinations, but the relation turns to be positive with the rih destinations. Manova and Zhang (2012) did not provide an explanation for these results. In another paper, Lugovskyy and Skiba (2016) found the ontrary results to the findings by Manova and Zhang (2012) with the firm level data from nine Latin Amerian ountries, i.e. the distane elastiities of export prie is positive for the poor destinations but negative for the rih destinations. Lugovskyy and Skiba (2016) explained their results by onsidering two types of firms: the first type faes the fixed ost for eah speifi variety and the seond type faes the market-speifi fixed ost for eah variety. The first type of firms provide the same quality produts with the same prie aross the markets while the seond type of exporters provide more high quality produts to the riher ountries. However, the relation between the quality of the produts and the rih level of the destination will beome weak if the destination is far from the home ountry. That means the exporting proportion of the seond type of the produts is dereasing in the distane. For the rih ountries, the average prie of the seond type produts is higher than the first type produts due to the high quality of the seond type. In this ase, we will observe that within the group of the rih ountries, the average prie of the exporting produts is dereasing in the distane due to the redued proportion of the seond type produts. For the poor ountries, the situation will be opposite. As the prie of the seond type produts is lower than the first type, the average prie will be inreasing in the distane. Following the model by Qiu and Yu (2014), our theoretial framework fouses on the analysis of the hange of the export sope and doesn t inlude the variable indiating the quality of the produts. However, our results show a new soure for the ourrene of the negative impat of the distane, i.e. the export sope bridges the marginal ost of the produts and the distane of the market (only the low marginal ost produts are sold in the far markets due to the high transportation ost). The other related papers onerning the impats of the harateristis of the destinations on the exporting strategies 6

8 inlude Brambilla and Porto (2016), Di Comite et al. (2014), and Gorg (2016). With the multi-national data, Brambilla and Porto (2016) found that the high-inome ountries prefer to import produts from the plants with high average wage, indiating that the rih ountries prefer high-quality produts. Gorg (2016) reahed the same onlusion from the empirial evidene with the Hungarian firm level data. Di Comite et al. (2014) proved that the onsumers in different ountries have different preferenes on the same variety, and thus we will observe the prie of the same produts varies aross ountries. 1.4 Relation literature on trade liberalization and heterogeneous firms We also shed light on how the firms adjust the prodution and export sopes in response to the tariff redution by the importing destination ountries during the periods 2001 to The study of the impats of the trade liberalization on produt sope is not new. Most of them fous on the study of prodution sope, e.g. Qiu and Yu (2014) and Lopresti (2016), and few of them studied the export sope, e.g. Berthou and Fontagne (2013). Among all the relevant literature, the paper whih is mostly related to ours is Berthou and Fontagne (2013), whih studied how the Frenh exporters adjusted their export sopes reating to the introdution of the Euro in Berthou and Fontagne (2013) indiated that the introdution of the Euro will eliminate the nominal exhange rate volatility among the Euro-member ountries, and the proess is equivalent to the elimination of trade ost among the trade partners within the Euro distrit. Using the Frenh firm-level data, Berthou and Fontagne (2013) found that, in response to the introdution of the Euro, the high produtivity firms expanded their export sopes but the hange of the export sopes by the low produtivity firms was insignifiant. Our study is different from the Berthou and Fontagne (2013) in four points: firstly, our study fouses on the tariff redution ( the diret trade-ost redution) by the destination ountries rather than the finanial integration proess; seondly, our study makes the distintion between a firm s prodution sope and its export sopes to different destinations; thirdly, our empirial results show that the high produtivity firms expand their export sopes while the low produtivity firms shrink the export sope, however the hange of the prodution sopes are insignifiant; lastly, we onstrut a theoretial model to nest the hange of both the prodution and export sopes. Among the theoretial literature, with the exeption of Qiu and Zhou (2013), almost all the papers reahed the onlusion that the multiprodut firms redue the number of the export varieties in response to the trade liberalization, e.g. Bernard et al.(2011), Dhingra (2013), Ekel and Neary (2010), and Mayer et al. (2011). Consistent with the empirial findings of Berthou 7

9 and Fontagne (2013) and ours, Qiu and Zhou (2013) proved that the high produtivity firms may inrease the produt sopes in response to the trade liberalization. Similarly, Dhingra (2013) made a theoretial analysis and showed that in response to the market expansion (higher intensity of the market ompetition), the firms would shrink the produt sopes but invest more in proess innovation in order to redue the marginal ost. The differenes between our study and Qiu and Zhou (2013) lies on two points: firstly, Qiu and Zhou (2013) didn t distinguish the prodution sope by the firm and the export sope in eah firm-ountry-year transation, and their produt sope is supposed to be the same as the prodution sope in our analysis; seondly, Qiu and Zhou (2013) only reahed the theoretial results and didn t provide the empirial evidene,while our analysis is fully supported by the empirial results. Our theoretial model ontributes to the existing literature in two points: firstly, our model distinguishes the prodution sope and the export sope to eah single market; and seondly, we onsider the produt quality by the heterogeneous firms and the firm s ability to adapt to the inreased ompetition after the trade liberalization. The rest of the relevant literature whih studied the firms prodution sopes inlude the Goldberg et al. (2009) and Lo Turo and Maggioni (2015). The former paper analyzed how the firms adjust their prodution sopes in response to the trade liberalization (redution of import tariff). Using the Indian firm level data, they found that the redution of the import tariff inentives the firms to invest in the produt innovation and expand their prodution sopes. The latter paper investigated the relation between the firm s exporting status and the prodution sopes through the evidene from Turkey, and found that the exporting has a prominent role for firm produt innovation (expansion of the prodution sopes). The rest of the ontent is arranged as follows. Setion 2 introdues our data set and illustrates our main empirial results; Setion 3 develops the model and provide theoretial fundamentals to our empirial findings; Setion 4 heks the robustness of our empirial results. and Setion 5 onludes all our empirial and theoretial findings. 2 Empiris Reall that we have two researh targets: (i) provide firm level evidenes on the effets of trade ost and exhange rate volatility on firms deision on their export sopes; (ii) figure out how heterogeneous firms adjust their export and prodution sopes in response to trade liberalization. In this setion, we attempt to explore our researh questions using the Chinese firm-produt-level data. Firstly, we introdue our data set and disuss some stylized fats we found from 8

10 the data; seondly, we onstrut estimation models to explore our researh question; lastly, we summarize and briefly explain our empirial findings. In setion 3, we further explain our empirial findings with a theoretial framework whih distinguishes the export and prodution sopes. 2.1 Data Our first data set is retrieved from the repliable data set of Fan et al. (2015). The data set is part of the Chinese ustoms data whih is by the firm, six-digit Harmonized System (HS) ategory, trade mode and destination ountry, overing the years 2001 and 2006 (two years). For eah firm-produt-ountry observation, we observe the total nominal value and quantity exported as well as the exporting destination assoiated with eah transation. Counting the number of the different HS6 odes assoiated with eah firm-ountry transation, we onstrut the number of the varieties for eah firm to eah market, i.e. the export sope, and sum up the number of firm-produt transations as the prodution sope. Following Heriourt and Ponet (2015), we measure the exhange rate shemes using the monthly standard deviation of both the real effetive and nominal exhange rates for eah ountry at eah year. 1 The geographi database regarding the distanes between China and other ountries is retrieved from the website of the CEPII, and the tariff rate data omes from the database of the WTO. We use the straight distane between the largest ities of the two ountries as the distane of the two ountries. The tariff rate is ategorized by the ountry and the HS2 odes (industry level). The following graphs illustrate the orrelation between the firm-ountry-speifi trade sope and the exhange rate volatility and distane to the home ountry. 1 Real effetive exhange rate is omputed as the weighted average of exhange rate of a ountry s urreny in terms of a basket of urrenies with adjusting the inflation of the ountry. 9

11 Graph 1. Export Sope and Exhange Rate Volatility Notes: we use the yearly average level of exhange rate volatility aross all ountries in our sample to distinguish the high and low volatility levels; the export sope in the distribution is normalized as ln(export sope jιt ) ln(export sope jιt ), where ln(export sope jιt ) is the logged form of firm-ountry-year level of export sope (firm ι in market j at time t) and ln(export sope jιt ) is the average export sope aross all ountries for firm ι at time t; the dashed line denotes the export sopes towards the ountries with relatively high exhange rate volatility and the solid blue line denotes the export sopes towards the ountries with relatively low exhange rate volatility; the left graph illustrates the distribution for the year 2001 and the right graph illustrates the distribution for the year The diagram shows that the distribution for the export sope towards the high risk ountries is to the left of that assoiated with the risk-free ountries. That means the export sopes towards the risk ountries are relatively low. 10

12 Graph 2. Export Sope and Transportation Distane Notes: we use the average level of logged form of distane to China aross all ountries to distinguish the near and far ountries; the export sope in the distribution is normalized as ln(export sope jιt ) ln(export sope jιt ), where ln(export sope jιt ) is the logged form of firm-ountry-year level of export sope and ln(export sope jιt ) is the average export sope aross all ountries for firm ι at time t; the dashed line denotes the export sopes towards the ountries those are far from China and the solid blue line denotes the export sopes towards the ountries those are near to China; the left graph illustrates the distribution for the year 2001 and the right graph illustrates the distribution for the year The diagram shows that the distribution for the export sope towards the far ountries is to the left of that assoiated with the near ountries. That means the export sopes towards the far ountries are relatively low. Both diagrams above show negatively effets of the exhange rate volatility and transportation distane on firms export sopes. The distributions of export sope towards ountries whih onduting flexible exhange rate shemes or far away from China (denoted as solid blue lines) are to the left of those assoiated ountries whih onduting relatively stable exhange rate shemes or loated near to China (denoted as dashed red lines). Note that we didn t illustrate the distributions assoiating with varying tariff rates here, beause the tariff rate is in the ountry-varietyyear level, whih makes it diffiult to illustrate graphially. In next setion, we will show our speifi estimation methodologies. 11

13 2.2 Methodologies Following Heriourt and Ponet (2015), we speify our estimation model for the effet of the trade ost on export sope as follows. 2 Export sope ι jt = β 1 exhange rate volatilaty jt +β 2 distane j +β 3 tari f f rate jvt +X jt γ 1 +λ ιt γ 2 +η t +κ ν +ε ι jt where ι, j, v and t denote eah individual firm, destination ountry, industry (HS2 ode) and the time respetively. All variables are in logged form exept the exhange rate volatility and the tariff rate. Export sope ι jt measures the single firm s export sope assoiated with eah destination at eah year, whih is omputed as the (log) number of the varieties by the firm-ountry-year level, i.e. Export sope ι jt ln(number o f varieties) ι jt. The key explanatory variables inlude the exhange rate volatility (exhange rate volatilaty jt ), and the distane between China and destination j (distane j ). Exhange rate volatility is omputed as the yearly standard deviation of the exhange rate for ountry j at year t using the monthly data, and the distane to the home ountry is omputed as the logged value of distane between the largest ity in ountry j and the largest ity in China. X jt ontrols for the sale of sales for firm ι in market j and some other maro harateristis of market j, i.e. GDP, GDP per apita and prie index. 3 λ ιt ontrols the firm level harateristis, i.e. input-import duty, domesti output-import tariff, number of varieties of the imported inputs, TFP, wage rate, labor number, apital labor ratio, and industrial ompetition intensity. 4 The time (i.e. time-dependent unobservable or unmeasurable fators) and industry fixed effets (sale of produt sope varies aross industries) are ontrolled by η t and κ ν respetively. As our dependent variable is ount number, we estimate our model using both OLS and Poisson methods. To test the responses of the heterogeneous firms to the tariff redution, we onstrut the estimation models as follows. Export sope ι j = β tari f f rate ι j + θt FP ι tari f f rate ι j + λ ι γ + δ v + κ j + ε ι j 2 This onstrution is also theoretially speified by lemma 1 and proposition 1 in setion 3. Our model differs from Heriourt and Ponet (2015) by involving transportation distane and import tariff rate. 3 GDP and GDP per apita ontrol the firm-ountry level trade sale and the taste heterogeneity among ountries, e.g. the parameters in the preferene funtion may different aross ountries; prie index ontrols the effet of the inflation on the market demand in the destination market. 4 The firm level variables ontrol the ost and market power to initiate a new produt variety. 12

14 5 Prodution sope ι = β tari f f rate ι + θt FP ι tari f f rate ι + λ ι γ + δ v + ε ι where ι, j, and v denote eah individual firm, destination ountry, and the industry (HS2 ode) respetively. The hange of firm-ountry level export sope is onstruted as Export sope ι j ln(number o f varieties) ι j ; The hange of tariff rate is omputed in two ways. Firstly, we ompute the simple differene of the ountry-industry level of the tariff rate; and seondly we omputed the firm-ountry level of the [ average tariff rate using the ountry-variety ] level of tariff data weighted by the trade volume, i.e. tari f f rate ι j (tari f f rate i j ) volume i jιt / volume i jιt i I ι j i I h j (i distinguishes the variety with HS6 ode). We use both the trade volumes in 2001 and 2006 to onstrut our indiator. In summary, we have three different indiators for our export sope, i.e. (i) hange of the tariff rate in ountry-industry level, (ii) hange of firm-ountry level of average tariff rate weighted by the 2001 trade volume, and (iii) hange of firm-ountry level of average tariff rate weighted by the 2006 trade volume. The onstrution for prodution sope is similar as the export sope. The prodution sope is onstruted as the log value of the total number of export varieties by firm ι in the year t, i.e. Prodution sope ι ln(number o f varieties) ι. We onstrut two types of the tariff rates: hange of [ firm level of average tariff rate weighted ] by the trade volume in 2001 value and 2006 value, i.e. tari f f rate ι (tari f f rate i j ) volume i jιt / volume i jιt. The firm level ontrol variables ( aptured by λ ι ) i I ι i I h are the same as the previous model exept that they are in the hanged form (the reason for ontrolling eah variable has been disussed in previous text). δ v and κ j ontrol the industry and ountry level fixed effets respetively. The reason for ontrolling the ountry fixed effet is that some ountries hanged signifiantly in GDP, GDP per apita, prie index or other unobservable fators during the observation period 2001 to 2006, whih may alter the onsumption tastes for Chinese produts among these ountries. The details regarding the estimation results will be illustrated and disussed in next setion. 2.3 Results Tables 1a below shows the results of the first estimation model whih explores the effets of exhange rate volatility, transportation ost, and tariff on firms export sopes. 6 The results show signifiant negative oeffiients on the 5 We use both the ountry-industry level and firm-ountry average tariff rates in our regression. 6 The estimation results whih using the logged form of exhange rate volatility are reported in table 1b whih an be found in Appendix. 13

15 variables exhange rate volatility, the distane to home ountry, and the tariff rate, whih indiates a negative impat of these fators on export sope. Our estimation results are similar as Heriourt and Ponet (2015), exept that we involve the transportation distane and import tariff rate in our estimation. It is easy to understand the impat of the transportation distane and tariff, but the mehanism for the effet of exhange rate volatility is not obvious. As disussed previously, the intuition for this result is that when the destination s urreny depreiates, the firms need to redue their export sopes, but when the urreny appreiates, the firms annot expand their export sopes due to the onstraint of prodution apaity. In this ase, the average export sope towards the risky will be less than stable ountries. Table 1a. Trade ost and the Export Sope Dependent Variable: Export sope Panel A: Real Exhange Rate Volatility Panel B: Nominal Exhange Rate Volatility Distane *** *** *** *** *** *** ( ) ( ) ( ) ( ) ( ) ( ) Exhange rate volatility *** *** *** *** *** *** (0.0548) (0.0717) (0.0654) (0.0669) (0.0882) (0.0804) Tari f f rate *** *** *** *** *** *** (0.0303) (0.0383) (0.0333) (0.0297) (0.0372) (0.0324) Observations 143,644 94,226 94, , , ,428 R-squared Trade Volume Control NO NO YES NO NO YES Firm Level Controls NO YES YES NO YES YES Country Level Controls NO YES YES NO YES YES Time FE YES YES YES YES YES YES Industry FE YES YES YES YES YES YES Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Notes: All variables are in logged form exept the exhange rate volatility and tariff rate. The real exhange rate volatility is omputed as the yearly standard deviation using monthly data of real effetive exhange rate. 7 The nominal exhange rate volatility is omputed as the yearly standard deviation using monthly data of nominal exhange rate in terms of the U.S. dollars. Panel A shows the effets of trade ost on real exhange rate volatility and Panel B shows the results of nominal exhange rate volatility. The transportation distane is omputed as the geographi distane between the largest ity of two ountries. The tariff rate is the industry level importing tariff by the destination market. 8 The firm level ontrols inlude the variables TFP, labor number, apital to labor ratio, wage rate, average industrial level of import tariff rate imposed by China, number of imported inputs by eah firm and the industry s ompetition intensity. Trade volume ontrol is the aggregate trade volume in speifi ountry by the firm. The ountry level ontrols inlude the GDP, GDP per apita, and CPI. The estimation results show that the export sope is dereasing in transportation distane, tariff rate and exhange rate volatility. Another important finding is that during the observation period, i.e to 2006, the tariff rates on Chinese produts imposed by destination ountries are widely ut down, and the redution of trade ost inentivates high 7 The real effetive exhange rate is the weighted average of a ountry s urreny relative to an index or basket of other major urrenies, adjusted for the effets of inflation. 8 Industry is lassified by HS2 ode. 14

16 produtivity firms to expand their export sopes, however, the low produtivity firms redue their export sopes due to inrease of market ompetition (see table 2a below). This result is onsistent with Berthou and Fontagne (2013), who found that in response to the introdution of Euro (reduing trade ost from onverting between different urrenies), Frenh firms with high produtivity will expand their export sopes but the adjustment by low produtivity firms is not signifiant. Our explanation to this result lies on two hannels: firstly, due to the ost-saving advantage in improving quality by high produtivity firms, the high produtivity firms produe and export high quality produts while low produtivity firms produe and export low quality produts; seondly the firms with providing high quality produts would obtain more benefits from tariff-redution proess, but the produts offered by the low produtivity firms may be rowed out of market due to the inrease of market ompetition arisen by trade liberalization. Table 2a. Trade Liberalization, Heterogeneous Firms, and Change of the Export Sope Dependent Variable: Change of logged form of export sope between 2001 and 2006 Tariff Speifiation Panel A: Country-industry Level Panel B: Firm-ountry Level Average, Panel C: Firm-ountry Level Average, 2001 Weights 2006 Weights Tari f f *** 3.756*** *** 2.35** *** 2.55** (0.473) (0.922) (1.087) (0.255) (0.715) (1.15) (0.240) (0.710) (1.11) T FP 2001 Tari f f *** *** *** ** ** ** (0.206) (0.236) (0.167) (0.244) (0.162) (0.235) T FP *** *** *** *** *** *** *** *** *** ( ) ( ) ( ) (0.0107) ( ) (0.0108) (0.0107) ( ) (0.0108) Observations 6,351 11,359 6,351 4,849 8,850 4,849 4,849 8,850 4,849 R-squared Firm Level Controls YES NO YES YES NO YES YES NO YES Country FE YES YES YES YES YES YES YES YES YES Industry FE YES YES YES YES YES YES YES YES YES Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Notes: We use three indiators for the hange of the tariff rate, i.e. ountry-industry level, firm-ountry level weighted by trade volume in 2001, and firm-ountry level weighted by trade volume in The first indiator is omputed as the differene of the industry level importing-tariff rates imposed by the destination ountry in the years 2001 and The seond indiator is omputed as the weighted average of the hange of the tariff rate faed by a single firm in a speifi ountry between the years 2001 and 2006, using the firm-produt-ountry level of trade volume as the weight fator. The last indiator is similar as the seond one exept that it uses the trade volume in 2006 as the weight fator. The firm level ontrols inlude the hange values of logged labor number, apital labor ratio, logged wage rate, industrial level of input and output tariff rates imposed by China, number of imported inputs by eah firm and the industrial ompetition intensity. The estimation results show that in response to a tariff redution, the high produtivity firm expands while the low produtivity firms redues their export sopes. Among the regressions regarding the prodution sope, the oeffiients on tariff rate and produt of tariff rate and TFP are insignifiant or positive (see table 2b below), whih means that in response to the trade liberalization proess, firms didn t expand their prodution sopes or even redue them. We suppose result is aused through two ways: firstly, when the other ountries redue tariff to Chinese firms, China also redues trade obstales to foreign firms, and thus 15

17 the domesti ompetition intense inreases; seondly, during the observation periods, a plenty number of new-born firms entry the market, and these observations are not inluded in our regressions. 9 The old firms in the market fae the inrease of the market ompetition due to the entering of new-born firms and foreign firms, and thus they redue their prodution sope. The detailed explanation regarding this result an be found in setion 3.4. Table 2b. Trade Liberalization, Heterogeneous Firms, and Change of the Prodution Sope Dependent Variable: Change of logged form of prodution sope between 2001 and 2006 Tariff Speifiation Panel A: Firm Level Average, 2001 Weights Panel B: Firm Level Average, 2006 Weights Tari f f 0.857* (0.507) (1.52) (1.66) (0.475) (1.62) (1.85) T FP 2001 Tari f f (0.387) (0.424) (0.414) (0.466) T FP * (0.0285) (0.0213) (0.0267) (0.0285) (0.0213) (0.0267) Observations 816 1, , R-squared Firm Level Controls YES NO YES YES NO YES Country FE YES YES YES YES YES YES Industry FE YES YES YES YES YES YES Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Notes: We use two indiators for the hange of the tariff rate, i.e. firm level weighted by trade volume in 2001 and firm level weighted by trade volume in The first indiator is omputed as the weighted average of the hange of the tariff rate faed by a speifi firm between the years 2001 and 2006, using the firm-produt level of trade volume as the weight fator. The last indiator is similar as the seond one exept that it uses the trade volume in 2006 as the weight fator. The firm level ontrols inlude the hange values of TFP, labor number, apital labor ratio, wage rate, and the industrial ompetition intensity. The estimation results show that in response to a tariff redution, the firms may redue the prodution sope, but the orrelation is weak. Our results are similar as those in Qiu and Yu (2014), who found that in response to the trade liberalization, both high an low produtivity firms redue their prodution sopes and only the firms with low management ost (one type of fixed ost for produing new variety) inreases their prodution sopes. In summary, our empirial analysis has the following findings and ontributions to the existing literature: (i) our estimation is the first one to provide firm level evidene about the effet of the transportation ost on export diversifiation; (ii) our study is the first one to distinguish the prodution and export sope among relevant literature and study the effet of trade liberalization on export sopes among heterogeneous firms with different TFP levels; (iii) our study is the first one to find the evidene that in response to the tariff redution the high produtivity firms expand their export sopes. In Setion 4, we also try some other estimation methods or use alternative indiators as the dependent variable to onfirm the robustness of our empirial results. In next setion, we will disuss and explain our empirial results under a theoretial framework. 9 Reall that the variables in the regressions are in differene form. In this ase, the new-born firms are not inluded in our sample. 16

18 3 Theoretial frameworks How do multi-produt firms arrange their prodution and export sopes aross different markets? In this setion, we model multi-produt firms hoie on prodution and export faing exhange rate risks and different market harateristis. Three deisions that firms make in order to maximize profits in eah destination are highlighted: the optimal range of produts to produe, the optimal range of markets to enter, and the optimal produt varieties to export to eah market. We identify the key eonomi mehanisms that govern these deisions, and derive empirially testable preditions that allow us to validate them in the data. We highlight that the existene of exhange rate risks and a two-step deision making proess play ritial roles on observable firm outomes. We examine multi-produt exporters in a stylized oneptual framework with standard assumptions about underlying demand, prodution and market struture. Our theoretial framework is aimed to explain our main empirial findings: (i) the export sope dereases in the level of exhange rate volatility, transportation distane, and tariff rate of a destination ountry; (ii) in response to a tariff redution in a destination market, the high produtivity firms expand while the low produtivity firms redue their export sopes; (iii) in response to the tariff redution proess, the hange of the prodution sope of eah firm is not signifiant. Before our theoretial analysis, some questions regarding our empirial findings arise. Firstly, as most views, the firm is risk neutral, but how ould the risk-neutral agent behaves like a risk-aversion agent when faing the exhange rate risks aross many destination ountries. Seondly, why the empirial results are quite different between the two types of sopes, i.e export and prodution sopes. In this setion, we suppose a theoretial framework whih distinguishes some fundamental differenes between prodution and export sopes to explain all our findings. The main differene between our theoretial model and the ones among the previous literature is that we distinguish the number of produt varieties as two types, i.e. the prodution sope and export sope. The prodution sope refers to the total number of varieties exported by the firm and the export sope refers to the number of varieties exported to a speifi market by the firm. There are several differenes between these two sopes: (i) the expansion of prodution sope inurs higher level of fixed ost, e.g. R&D investment, produt s standardization, management ost and advertisement, while the enlargement of export sope is not neessary to inur larger fixed ost; (ii) it takes a relatively long period (at least several years) to initiate a new variety (expanding prodution sope) but the adjustment of export sope an be made eah year; (iii) due to the different length of the proessing time, the deision regarding the prodution sope is made before the realization of the market state while the export sope an be adjusted based 17

19 on the realization of the market state. With all these onsiderations, our intuitive explanations to the empirial findings are as follows. Consider a world with ountries onduting different exhange rate shemes. Some ountries ondut a relatively stable exhange rate poliy, e.g. fixed exhange rate sheme or pegged exhange rate sheme, while some other ountries ondut a relatively flexible exhange rate sheme. Like the tariff and the transportation ost, urreny depreiation would inur extra ost for firms to export while the urreny appreiation ould benefits the exporting firms. The firms need to take the flutuation of the exhange rate in the destination ountries into aount when they deide their prodution and export sopes. As disussed previously, it takes a long period to set up the prodution of a new variety. In this ase, the prodution sope an only be determined before the realization of the market states (urreny depreiation or appreiation), while the deision on the export sope an be made after the realization of the market ondition. In this ase, the prodution sope is deided aording to the preditions of market onditions over several years and the deision on the export sope to eah market an be made annually aording to the speifi ondition in that year. If the ountry s urreny depreiates, the firm will redue the total number of exported varieties to this ountry and fous in exporting the produts with highest demand or lowest marginal ost. In this ase, if the ountries are with low transportation ost, low tariff rate or experiening urreny appreiation, then the firm will be likely to export more varieties to these ountries. However, at most times, this extra exporting is onstrained by the prodution sope. As initialing a new variety inurs fixed ost, the prodution sope is determined aording to the demand of a typial ountry with average trade ost. For instane, if a new variety is only demanded by a small number of ountries whih are near to China, with low tariff rate, or experiening urreny appreiation, then this variety will not be produed just beause the market revenue annot over the fixed ost. In this ase, the firms will usually invest and produe the number of varieties just to meet the demand of an average market. In this ase, when some ountries experiene urreny appreiation, the firms ouldn t offer more varieties to these ountries due to the limitation of their prodution apaity. In this ase, the firms exports less when the market status goes bad (depreiation of destination s urreny) but they are prevent to export more when the status goes well (appreiation of destination s urreny). On average, the export sopes towards the ountries with the exhange rate risks will be less than those towards the ountries with relative stable exhange rate shemes. For the seond and third empirial findings, our intuitive explanations are as follows. The high produtive firms pursue ost-saving advantage in improving the quality level of produts than the low produtive firms. In this ase, 18

20 the high produtive firms will produe and export high quality produts while the low produtive firms will export low quality produts. Before the tariff-redution proess, the firms export the varieties that they are good at produing (with relatively low marginal ost). When the destination market redues their import tariff rates, then two direted effets will apply: firstly, the marginal revenue for eah produt will raise up and then it inentives the firms to expand their export sopes and start to export the varieties that require higher marginal ost; seondly, as the inrease of the exporting volume and varieties, the market ompetition in the destination market will be more intensive, whih will row out some varieties. As the pries and markups for high quality produts are higher than those for the low quality produts, the firms who exporting the high quality produts (high produtive firms) will reeive more tax returns by eah unit of produts in the trade-liberalization proess. In this ase, the high produtive firms are expeted to expand more of their export sope than the low produtive firms. At the same time, as the inrease of the exported varieties and volume, the ompetition among destination markets beome more intensity. The inrease of the market-ompetition intensity ould rowd out some varieties. If the rowding out effet is large enough, then the net effet on the low produtive firms will be negative. That means the low produtive firms will redue their export sope. Our empirial result further shows that in response to the trade liberalization, the hange of the prodution sopes for both types of firms is insignifiant. The explanation for this result is as follows. Firstly, the firms do not export their whole varieties to every ountry. The variety with the highest marginal ost is exported to the ountries where the tariff rates are relatively low. We all these ountries as marginal ountries and the rest of ountries as non-marginal ountries. If the tariff rate is not redued deeply enough, then the hange of the export sope in the non-marginal ountries won t touh the prodution onstraint. In another words, the firms are no neessary to expand their prodution sope to meet the hange of the market demand in the non-marginal ountries. Contrarily, if the tariff redution ours among the marginal ountries, the firms need to expand their prodution sope in order to meet the inreasing demand among these ountries. However, during our observation periods (2001 to 2006), the tariff redution ours mostly among the ountries whih impose relatively high level of initial tariff rates and the tariff redution among the marginal ountries is not signifiant. In this ase, the firms adjust little on their prodution sopes. Starting from next setion, we will show how our theoretial framework is onstruted and how it desribes our empirial findings. Generally, we propose two agents in our model, i.e. household and firm. Firstly, we exhibit and disuss our assumptions regarding the households and firms; then we reah the market equilibrium with solving the best strategies of firms in response to different market onditions; lastly, we explain our theoretial results intuitively. 19

21 3.1 Households Following Melitz and Ottaviano (2008), Dhingra (2013), and Qiu and Yu (2014), we assume the onsumers utility funtion for ountry j is the form of the quasi-linear preferene: U = q j0 + α q ji di 1 ( ) 2 i Ω j 2 β q ji di 1 i Ω j 2 i Ω γ q 2 jidi where q j0 is her onsumption of the numeraire good; Ω j is the set of all varieties sold in ountry j ; and q ji is the onsumption of variety i in ountry j. The quasi-linear preferene assumes a onstant marginal utility of the numeraire good (aptured by the first term), a dereasing marginal utility for the differentiate good (aptured by the seond and fourth terms with quadrati formula), and a measure of the ompetition among the differentiate produts (aptured by the third term). The quasi-linear preferene aptures the onsumption feature that onsumers ompare and deide purhasing amount among different varieties and also deide whether or not to buy a variety. For example, if the prie of one variety is relatively high ompared with other varieties, then the sales of this variety will be relatively low. If the prie of the variety inreases further, then the onsumers may deide not to buy this produt and save the money on onsumption of the numeraire good. The onsumer maximizes the utility subjet to the budget onstraint, i.e. p j0 q j0 + p ji q ji di M i Ω j where M is the inome of a typial onsumer. From the above it follows that the demand funtion for variety i in ountry j as: ( α q L ji = L j q ji = L j γ 1 γ p ji β ) γ Q j where Q j i Ω q jidi is an index of the onsumption of all the differentiated produts in ountry j; and L j denotes the population size of ountry j. Here we have a onventional assumption about the produt s variety. The variety omes from two soures: the 20

22 produts are provided by different firms; or the same firm provides the produts in different horizontal sopes. 3.2 Firms The heterogeneity of firms ome from the produtivity ϕ i when produing variety i, and i (0, + ). Firm-speifi produtivity for variety i is assumed to be given by ϕ i = κi r, where κ and r are firm-speifi general produtivity measurements, representing overall effiieny fators inluding management level, transferable tehnologies, et. The ost funtion for the firm ι is omposed of two parts: 10 [ ( ) ] C i j = q L j J h i Ω h ϕ jidi + F i di d j i F i is the sunk ost for the firm ι to be able to produe variety i. As a onventional assumption following Qiu and Yu (2014), we assume an non-dereasing marginal ost funtion in variety i. In this ase, we have r 1. Then we an write the expeted profit funtion for firm ι as followings: [ ( ] } Eπ ι = E ε j (1 τ j ) p ji q L ji t j )q Lji F i di d j j J h i Ω h ϕ i where ε j is the exhange rate in ountry j with the mean ε; τ j is the import tariff rate in ountry j ; and t j is the transportation ost to the destination j. Among previous relevant studies, some of them rely on the real effetive exhange rate (REER), e.g. Aizenman and Marion (1999) and Heriourt and Ponet (2015), while others studied the effet of the nominal exhange rate, e.g. Shnabl (2008). In our analysis, we involve both of the real effetive exhange rate and the nominal exhange rate. The real effetive exhange rate is omputed as the weighted average of exhange rate of a ountry s urreny in terms of a basket of urrenies with adjusting the inflation of the ountry; and the nominal exhange rate is measured with U.S. dollar. The firms make deisions on both the prie of eah variety in the speifi ountry and the horizontal sope of the produts they produed. We assume the sope deision is made before the realization of the exhange rate and the prie strategy is deided after observing the exhange rate. Reall from Setion 1, it takes long period and inurs fixed ost 10 Here, to avoid umbersome notation, we have omitted the subsript h in the symbol for the firm-speifi produtivity level. 21

23 to invest in a new variety but the deision regarding the export sope an be made eah period. In this ase, the firms an adjust their exported produts with speifi exhange rate level but the sope sunk ost is made with an expetation of the situations in the market. Without loss of generality, to simplify our analysis, we make the following onventional assumptions: ASSUMPTION 1. [1] The sunk ost for eah variety is the same, i.e. I 0 F idi = µi; [2] the transportation distane is bounded within a range, i.e. t j [0, t max ]; [3] the exhange rate in any ountry is bounded within a range with an idential mean value, i.e. ε j [ε min, ε max ] and Eε j = ε for j; [4] the tariff rate, aggregate variety and population size are onstant and idential aross ountries, i.e. τ j = τ,α βq j B j = B and L j = L for j; [5] the negative realization of the exhange rate doesn t ause the firm to exit the market, i.e. Bε min > t max. The firm faes a two-stage deision problem. In the first stage, the firm determines its prodution sope I. In the seond stage, it deides how many varieties to export to eah ountry. We must solve the problem using bakward indution, i.e. solve the results in the seond stage at first. In the seond stage, before solving for the export sope for eah destination market, we need to state the firm s optimal prie strategy (and the resulting quantity) for eah variety i for eah ountry j, onditional on variety i being made available for ountry j, i.e. p i j = q i j = max 0, 2 α β 2 Q j + max 2(1 τ)ϕ i ε j + 0, α 2 β 2 Q j + 2(1 τ)ϕ i ε j + } t j 2(1 τ)ε j } t j 2(1 τ)ε j With all the assumptions above and the solution for the prie strategy, we an rewrite the expeted profit funtion for the typial firm as follows. εmax Eπ ι = i Ω h j J h i j/(1 τ)b L 4γ (1 τ)ε j where i j ϕ i +t j and φ(ε j ;t j ) is the joint density funtion of ε j and t j. Next, we an solve the export sope in the seond stage as follows. [ (1 τ)bε j ϕ i t j ] 2 φ (ε j ;t j )dε j d jdi µi LEMMA 1. Given the prodution sope hosen in the first stage, the firm hooses the export sope towards the 22

24 stable market as: 0 i f t j > (1 τ)bε; i j = [(1 τ)bε tj ] ( )} 1 κ r < I i f (1 τ)bε t j > ˆt (ε, I); I i f t j ˆt (ε, I). and towards the risk market as: i j = 0 i f ε j < t j (1 τ)b ; [(1 τ)bε j t j ] ( κ )} < I i f t j (1 τ)b < ε j < ˆε (t j, I); I i f ˆε (t j, I) < ε j ε max. where ˆt (ε, I) (1 τ)bε ( ) κ I r and ˆε (t j, I) (1 τ)b + [ κ(1 τ)b ]Ir. See the proof in the Appendix. t j In what follows, we assume that t j belongs to the interval [0, t max ], and that the random variable ε j belongs to the interval [ε min, ε max ]. As (1 τ)bε min > t max, so that i j is always positive. Given the solutions for i and i j, we an obtain the lemma 1 below. LEMMA 2. Given the same transportation distane, the expeted exporting sope towards the markets with the exhange rate volatility is on average lower than that towards the ountries without the volatility, i.e. Ei t i t. See the proof in the Appendix. In the first stage, we an solve the optimal prodution sope whih is given by the following equation. N N tmax εmax 0 ˆε(t,I) ˆt(ε,I) 0 L [ (1 τ)bε ] } 2 4γ (1 τ)ε κ Ir t ρ (t)g ( ) t; S f dt+ L [ (1 τ)bε ] } 2 4γ (1 τ)ε κ Ir t [1 ρ (t)]φ (ε;t)h(t;s v )dεdt = µ where N is the total number of the destination ountries; g ( t; S f ) is the density funtion of the distane for ountries belonging to set S f and g(t; S v ) is the density funtion of the distane for ountries belonging to set S v ; and ρ (t) (0,1) 23

25 is the fration of the destination ountries that are loated at distane t and that have the fixed exhange regime. 11 Next, we need to solve the optimal prodution sope I in the first stage and show that Ei t < i t for some transportation distane t j. Consider the set S f of destination ountries that have a fixed exhange rate. (For simpliity, assume their exhange rate are the same, and all this fixed rate ε. Given I, the firm s marginal variety i = I are sold only in a subset of this set, i.e. in those ountries in S f with t j ˆt (ε, I). The aggregate profit (in these markets) for the marginal variety i = I is R f (I) = N ˆt(ε,I) 0 L [ (1 τ)bε ] } 2 4γ (1 τ)ε κ Ir t ρ (t)g ( ) t; S f dt where g ( t; S f ) is the density funtion of the distane for ountries belonging to set S f, and ρ (t) (0,1) is the fration of the destination ountries that are loated at distane t and that have the fixed exhange regime. LEMMA 3. R f (I) is a dereasing funtion of I. See the proof in the Appendix. Consider next the set S v of destination ountries that have a variable exhange rate. Given I and any t < t max, the Chinese firm s marginal variety i = I are sold in a subset of this set, i.e. in those ountries in S v with t j = t and with some realized exhange rate ε in the range ˆε (t,i) ε ε max. The aggregate expeted profit earned for this set of ountries (for this marginal variety i = I ) is R v (t, I) = [1 ρ (t)]n εmax ˆε(t,I) L [ (1 τ)bε ] } 2 4γ (1 τ)ε κ Ir t φ (ε;t)dε where φ (ε; t) is the onditional density funtion of the random variable ε (onditional on loation t ). Integrating over all possible loations, we get tmax R v (I) R v (t,i)h(t;s v )dt 0 where h(t;s v ) is the density funtion of distane for ountries belonging to set S v. LEMMA 4. R v (I) is a dereasing funtion of I. 11 S f is the set of the ountries that ondut the fixed exhange rate sheme and S V is the set of the ountries that ondut the volatile exhange rate sheme. 24

26 See the proof in the Appendix. Finally, expeted aggregate profit earned for the marginal variety i = I is R(I) R f (I) + R v (I) The funtion R(I) is dereasing in I, and I is the value of I suh that R(I ) = µ where µ is the sunk ost for eah variety. We assume that µ is not too big, so that I is positive. Thus, the equation that determines I is N N tmax εmax 0 ˆε(t,I) ˆt(ε,I) 0 L [ (1 τ)bε ] } 2 4γ (1 τ)ε κ Ir t ρ (t)g ( ) t; S f dt+ L [ (1 τ)bε ] } 2 4γ (1 τ)ε κ Ir t [1 ρ (t)]φ (ε;t)h(t;s v )dεdt = µ With all the lemmas and results above, we reah the proposition below. PROPOSITION 1. Given the transportation distane t [0, t max ], the expeted exporting sope towards the markets with the exhange rate volatility is on average lower than that towards the ountries without the volatility, i.e. Ei t i t, and for some transportation distane, the strit inequality holds, i.e. Ei t < i t if t [0, t). See the proof in the Appendix. Based on the results obtained above, we an further explain the empirial results about the trade sopes and the transportation ost. Assume the exhange volatility is zero, i.e. ε j = 1. From the equations R(I ) = µ and i } j min = [(1 τ)b tj ] κ } 1 r, I, we an easily see that the export sope in ountry j, i.e. i j, is dereasing in the transportation ost t j, and only the low marginal ost produts are exported to the far destinations. 3.3 Disussion on the theoretial results Intuitive explanation Graph 1. Exhange Rate Volatility, Transportation Cost and the Export Sope 25

27 The intuition for the result of the transportation ost and the trade sopes is very straightforward: the rising of the marginal ost lowers the marginal profits for eah variety and thus fores the firms to withdraw some varieties that they are not good at produing. The mehanism behinds the orrelation between the exhange rate risks and the trade sopes is a little omplex. As shown in the theoretial model, eah firm faes idential onsumers aross the nations.sine firms must inur sunk osts for eah produt variety, they will deide the total number of the invested sopes based on the average demand aross the world. However, among the ountries, the speifi number of the varieties for eah ountry may be different due to the variety of the exhange rate risks and the transportation ost. In the markets with highly volatile exhange rates, the firms will derease their export sopes if the destination s urreny depreiates a lot, however, when the urreny appreiates the firms may not be able to inrease their export sopes due to the onstraints by the pre-invested sopes. This intuitive explanation is similar to the one by Heriourt and Ponet (2015), who argued that the export volume will derease if the destination s urreny depreiates and this proess is equivalent to wasting part of the pre-invested sunk ost in the trade. In this ase, the firms would be averse to enter into the markets with highly exhange rate risks. Graph 1 above illustrates how the export sope towards the ountries with flexible exhange rate shemes is on average less than that towards the ountries with relatively stable exhange rate shemes. The red dashed lines label the upper and lower bounds for the export sopes towards the risk markets, i.e. when the urreny appreiates the export sope inrease and when the urreny depreiates the movement will be 26

28 opposite, and the red solid line refers to the average export sopes towards these ountries. The blue line denotes the export sopes towards the risk-free markets. All the export sopes are onstrained by the upper bound of the prodution sope. The horizontal line desribes the heterogeneous transportation ost among ountries (to simplify our analysis, without loss of generality we assume all ountries ondut the same import tariff in this setion). From the graph, we observe that the export sopes towards the risk ountries are on average less than or equal to those towards the risk-free ountries, and for some group of ountries (near to China) strit inequality holds Link with the previous literature: The role of finanial onstraints In the literature on the effet of exhange rate volatility on the firms overall investment, some authors mentioned the role of finanial onstraints (see, for example, the papers by Aghion et al. (2009) and Heriourt and Ponet (2015 ) ). However, they do not provide a model that sorts out how the exhange rate volatility negatively affeted the firm s export performane and how finanial onstraints reinfore suh effet. In this sub-setion, we show that our model (whih does not neessarily inlude finanial onstraints) an be re-interpreted as a model where firms fae finanial onstraints, and thus our model an be seen as omplementary to the literature on finanial onstraints in that our model nests the intuition of previous models in a two-stage framework where in the first stage firms deide on their prodution sope. Reall that the firm determines its optimal prodution sope based on the formula N N tmax εmax 0 ˆε(t,I) ˆt(ε,I) 0 L [ (1 τ)bε ] } 2 4γ (1 τ)ε κ Ir t ρ (t)g ( ) t; S f dt+ L [ (1 τ)bε ] } 2 4γ(1 τ)ε κ Ir t [1 ρ (t)]φ (ε;t)h(t;s v )dεdt = µ Let us re-interpret µ as U divided by ω, i.e. µ U ω, where U is the lumpy set up ost for eah variety, and omega is the presribed number of years over whih U has to be paid bak in equal yearly installments. Finanial onstraint may impat the deision in three ways: first, the length of the pay-bak period; seond, the osts of loans; and third, the eiling on loans. Consider the first fator. If firms do not fae finanial onstraints, they an make long-term investments. They an extend their investment to less profitable produt varieties, whih take longer time to reover the sunk ost. This means that on average ω would be bigger, i.e., µ would be smaller. The derease in µ will lead to an 27

29 inrease in the prodution sope. As disussed above, with a greater prodution sope, the effet of volatility on export performane will be redued. The seond fator is the ost of loans. Firms that fae more severe finanial onstraints will fae higher ost of finaning. For example, when a Chinese firm annot borrow from state-owned banks with relatively low interest rates, it has to find help from the private banks or the shadow banks, in whih ase it has to bear higher finanial osts. This means that the marginal ost of expanding the sope is higher. The third mehanism is that redit rationing in its strongest form an effetively eliminate the firm s plan to invest in new varieties. In this ase, the realized prodution sope would be below the optimal sope. All these mehanisms redue the prodution sope of the firm and weakens its ability to ope with exhange rate risks. 3.4 Tariff redution, produt variety and the heterogeneous firms As another important ontribution, we explore the impats of the redution of trade ost on the adjustment of export sopes by heterogeneous firms. Aording to the findings of Berthou and Fontagne (2013), in response to the introdution of Euro, high produtivity firms expanded their export sopes but the hanges among low produtivity firms were insignifiant. Berthou and Fontagne (2013) did not provide any theoretial model to explain these results. 12 Here, we would like to re-explore the issue studied by Berthou and Fontagne (2013) with another type of poliy hange, i.e. the tariff-redution proess. We also attempt to explain the results of Berthou and Fontagne (2013) and ours with a theoretial framework. The period under study, , orresponds both to a drasti inrease in Chinese foreign trade (e.g., the yearly export growth inreased by 50% over the period) and to a signifiant episode of trade liberalization. Following China s aession to the WTO in Deember 2001, the authorities undertook a series of important ommitments to open and liberalize the eonomy and to offer a more preditable environment for trade and foreign investment. In turn, foreign trade partners also gradually provide redued tariffs, non-tariff measures, lienses and quotas. We make use of this poliy variation in tariff redutions to apture the impat of the trade liberalization on the export sopes onduted by heterogeneous firms. In order to address issues of endogeneity, we must verify that tariffs were set independently of industries expeted exports and lobbying ativities. First, Branstetter and Lardy (2006) onfirm that China s aession 12 All the previous theoretial literature fous on the analysis of the prodution sope. 28

30 into WTO is mainly motivated by the domesti reform agenda and willingness to beome a market eonomy. Thus it is hard to believe that exporters would have expeted or have influene on the hange of foreign ountries tariff. Moreover, Brandt et al. (2012): the onvergene in tariffs is more likely to reflet a requirement from WTO to reah low tariffs in all setors rather than a seletive alloation of tariff redution in response to setor performanes or lobbying ativities. Lastly, there is a growing literature take advantage of China s aession into WTO, Fan et al. (2015); Bas and Strauss-Kahn (2015) for example, analyze exporters performane using this poliy variation. The first empirial finding is that when the import tariff rates are redued by destination ountries, the high produtivity firms will expand more export varieties than the low produtivity firms in the destination ountries. (See the table 2a in the previous setion) This may be due to the quality differene among the produts provided by the heterogeneous firms. When the tariff is ut down in the destination ountry, more firms and produt varieties will enter the market. In this ase, the ompetition tends to be higher and some varieties ould be rowded out of the market. In this setion, we will show that, ompared with the high quality produts, the low quality produts are easier to be rowded out of the market. In this ase, if the high produtivity firms ommit to provide high quality produts, then the expansion of their produt varieties will be less affeted by the rowding-out effet than the low produtivity firms. Another empirial finding is that among the prodution-sope regressions, the oeffiients on the hange of tariff rate are insignifiant or positive. That means during the trade liberalization proess, firms adjust little or even redue their prodution sopes. The intuition for this result is as follows. The tariff redution proess mostly takes plae in the ountries where the initial tariff rates are relatively high. (See the graph 2 below) From the data set, we observe that ountries with low initial tariff rates are unlikely to redue the tariff rate further, and we suppose this is due to two reasons: firstly, as the tariff rates in these ountries are already so low, there is little spae to ontinue utting down the tariff rates; seondly, the initial tariff rates among these ountries are low enough that meet the requirement of the WTO agreement, thus it is no neessary for them to redue their tariff rates further. As disussed in the previous setions, the number of exported varieties to the destination ountry is dereasing in the tariff rate of this ountry. In this ase, the export sope towards the high tariff ountries will not touh the onstraints of the prodution sope, i.e. i j < I ; however, the export sope to the low tariff ountries will be bounded by the prodution sope, i.e. i j = I. If the tariff redution only takes plae among the high tariff ountries, then we will observe that the firms inrease the ountry-speifi export sope but do not expand their prodution sope in response to the tariff-redution proess. In other words, the firms will have an inentive to make new investment in their prodution sopes only if the low tariff 29

31 ountries ut the trade osts substantially. j J L 4γ (1 τ j )ε [ (1 τ j )B j ε κ (I ) r t] 2 d j = µ where the olletion set J ollets the ountries where τ j 1 κ (I ) r +t B j ε. 13 The equation above determines the optimal prodution sope by a typial firm. As the profits in eah market should be positive, we an further obtain that the variety I only shows up in the market where the tariff rate is low enough, i.e. τ j 1 κ (I ) r +t B j ε i.e. τ j 1 κ (I ) r +t B j ε. If the tariff redution only takes plae among the ountries where the tariff rate is relatively high,, then the prodution sope annot be expanded. Graph 2. Initial Tariff Rate and the Change of the Tariff Rate In the rest part of this setion, we will fous on the disussion of the hange of export sopes with the following assumptions: firstly, the tariff redution takes plae among the ountries where the export sope is below the prodution sope; seondly, the expansion of the export sope is not onstrained by the initial prodution sope. To model all the empirial findings in this setion and simplify our analysis without loss of generality, we make the following additional assumptions: ASSUMPTION 2. [1] There are two levels of quality, i.e. z H and z L, and the parameter α of the utility funtion inreases in the quality level of the produt, i.e. α (z H ) > α (z L ) ; 13 Different from the previous model, we assume the transportation ost and the exhange rate sheme are idential aross ountries but the ompetition index is endogenous now. 30