Essays on International EconomicsbyZhe ChenB.S., Beijing Normal University, 2007M.A., Peking University, 2010A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Economics)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)September 2015c© Zhe Chen 2015AbstractThis dissertation consists of three chapters. The first chapter examines howthe export intensity of Chinese firms is related to their productivity, andfinds a negative correlation within exporting firms. This pattern remainsrobust even when controlling for ownership, trade mode, factor intensity,and export subsidies. Firms have to pay marketing costs in order to reachcustomers. If the elasticity of these costs, with respect to the number ofconsumers in the domestic market, is higher than in foreign markets, thenmore productive firms sell relatively more in the domestic market comparedwith low-productivity firms. Using the marketing cost framework, this chap-ter further estimates the elasticity of marketing costs in each market andthen uses local market competition to explain the heterogeneity in elastic-ity across markets. When competition is tougher in a market, elasticity ishigher there.The second chapter investigates the exchange rate pass-through differ-ences in import prices between two trade modes and finds some robust empir-ical patterns. First, Chinese-owned assembly firms bear higher exchange ratepass-through than multinational firms. However, joint-owned and foreign-owned assembly firms bear less. Second, the exchange rate pass-through isgreater when firms import materials from developed countries. Third, as-sembly firms can bear higher exchange rate pass-through if they have highermarket shares or import less inputs. Finally, high financial development ishelpful for assembly firms to bear higher exchange rate pass-through.The last chapter discusses the impact of China’s rare earth policy ondownstream industries. When China implemented the tough restriction onrare earth exporting in 2010, it caused a significant price gap between thedomestic market and foreign markets. Therefore, downstream industriesin China enjoy cost advantages relative to foreign competitors. First, thispolicy has led a rapid increase in exports of Chinese rare earth downstreamfirms relative to other Chinese firms. The increase of exports is mainly dueto the price rather than quantity. Second, this chapter focuses on a typicaldownstream product – magnets, and finds that the exports of Chinese metalpermanent magnet also increase faster than exports of other countries.iiPrefaceChapter 1 Productivity, Marketing Costs and Export Intensity of this disser-tation is a joint work with Xiaonan Sun. Chapter 2 Financial Developmentand Exchange Rate Pass-through in Processing Trade is a joint work withJunjie Hong and Xiaonan Sun. Dr. Hong is a Professor from the Universityof International Business and Economics. Xiaonan Sun is a graduate stu-dent from Sauder School of Business at the University of British Columbia.I was highly involved throughout every stage of these chapters: collectingand preparing data, designing empirical models, carrying out estimation,organizing and presenting results, writing and editing the manuscript.iiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . xii1 Productivity, Marketing Costs and Export Intensity . . . 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.1 Related Literature . . . . . . . . . . . . . . . . . . . . 51.2 Some Stylized Facts . . . . . . . . . . . . . . . . . . . . . . . 61.2.1 Customs Data and Firm Data . . . . . . . . . . . . . 61.2.2 Data Summary . . . . . . . . . . . . . . . . . . . . . 61.2.3 Export Intensity and Productivity . . . . . . . . . . . 81.2.4 Export Intensity and Sales, Capital Intensity, and Num-ber of Employees . . . . . . . . . . . . . . . . . . . . 131.2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . 191.3 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . 191.3.1 Firm Level . . . . . . . . . . . . . . . . . . . . . . . . 191.3.2 Processing Trade and Ownership Effect . . . . . . . . 291.3.3 Differentiated Goods . . . . . . . . . . . . . . . . . . 301.3.4 Firm-Destination Level . . . . . . . . . . . . . . . . . 321.4 Marketing Cost Model . . . . . . . . . . . . . . . . . . . . . 341.4.1 Consumer Demand . . . . . . . . . . . . . . . . . . . 371.4.2 Marketing Costs . . . . . . . . . . . . . . . . . . . . . 381.4.3 The Firms’ Problem . . . . . . . . . . . . . . . . . . . 381.4.4 Productivity Distribution . . . . . . . . . . . . . . . . 39ivTable of Contents1.4.5 Firms’ Sales and Profit . . . . . . . . . . . . . . . . . 401.4.6 Firms’ Export/Domestic Sales Ratio . . . . . . . . . 411.4.7 Estimation . . . . . . . . . . . . . . . . . . . . . . . . 421.5 Additional Robustness Checks . . . . . . . . . . . . . . . . . 471.5.1 Advertisement Expenditure . . . . . . . . . . . . . . . 471.5.2 Market Share . . . . . . . . . . . . . . . . . . . . . . 481.5.3 Other Explanations . . . . . . . . . . . . . . . . . . . 521.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 602 Financial Development and Exchange Rate Pass-Through inProcessing Trade . . . . . . . . . . . . . . . . . . . . . . . . . . 612.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 612.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . 642.3 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 662.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682.4.1 Customs Data . . . . . . . . . . . . . . . . . . . . . . 682.4.2 Exchange Rate Data . . . . . . . . . . . . . . . . . . 682.4.3 Financial Data . . . . . . . . . . . . . . . . . . . . . . 692.4.4 Data Summary . . . . . . . . . . . . . . . . . . . . . 692.5 Exchange Rate Pass-Through and Trade Mode . . . . . . . . 732.5.1 The Product-Country Level . . . . . . . . . . . . . . 732.5.2 The Firm-Product-Country Level . . . . . . . . . . . 862.5.3 Other Factors . . . . . . . . . . . . . . . . . . . . . . 892.6 Exchange Rate Pass-Through and Financial Development . . 942.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 963 The Impact of China’s Rare Earth Policy on DownstreamIndustries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 993.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 993.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 1013.2.1 World Production of Rare Earth . . . . . . . . . . . . 1023.2.2 The Value Chain of Rare Earth . . . . . . . . . . . . 1023.2.3 China’s Rare Earth Policy . . . . . . . . . . . . . . . 1073.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1113.4 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . 1193.5 Case Study: Magnets . . . . . . . . . . . . . . . . . . . . . . 1263.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1313.6.1 Welfare Analysis . . . . . . . . . . . . . . . . . . . . . 1313.6.2 Long-Term Effect . . . . . . . . . . . . . . . . . . . . 1313.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132vTable of ContentsBibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .A Appendix to Chapter 1 . . . . . . . . . . . . . . . . . . . . . . 137A.1 Industry List . . . . . . . . . . . . . . . . . . . . . . . . . . . 137A.2 Productivity and Export/Domestic Sales Ratio: QuadraticRegression . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138B Appendix to Chapter 2 . . . . . . . . . . . . . . . . . . . . . . 141B.1 Two Stages Approach . . . . . . . . . . . . . . . . . . . . . . 141C Appendix to Chapter 3 . . . . . . . . . . . . . . . . . . . . . . 143C.1 The Treatment and Control Groups . . . . . . . . . . . . . . 143viList of Tables1.1 The Summary of Firm Data . . . . . . . . . . . . . . . . . . . 71.2 Percentage of Firms and Export Intensity . . . . . . . . . . . 81.3 The Summary of Merged Data . . . . . . . . . . . . . . . . . 91.4 Export/Domestic Sales Ratio and Productivity: Firm Level . 281.5 Export/Domestic Sales Ratio and Productivity Excluding Pro-cessing Trade: Firm Level . . . . . . . . . . . . . . . . . . . . 301.6 Export/Domestic Sales Ratio, Productivity and Ownership:Firm Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311.7 Export/Domestic Sales Ratio, Productivity and Differentiat-ed Goods: Firm Level . . . . . . . . . . . . . . . . . . . . . . 331.8 Export/Domestic Sales Ratio and Productivity: Firm-DestinationLevel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351.9 Export/Domestic Sales Ratio and Productivity For The TopTen Destinations . . . . . . . . . . . . . . . . . . . . . . . . . 361.10 The Parameters for The Top Ten Destinations . . . . . . . . 471.11 Export/Domestic Sales Ratio, Productivity and AdvertisingSpending: Firm Level . . . . . . . . . . . . . . . . . . . . . . 481.12 Export/Domestic Sales Ratio, Productivity and AdvertisingSpending: Firm-Destination Level . . . . . . . . . . . . . . . 491.13 Export/Domestic Sales Ratio, Productivity and Market Share:Firm-Destination Level . . . . . . . . . . . . . . . . . . . . . . 511.14 Export/Domestic Sales Ratio, Productivity and Market Size:Firm-Destination Level . . . . . . . . . . . . . . . . . . . . . . 541.15 Sales Ratio and Productivity (Benchmark: HKG) . . . . . . . 551.16 Price Ratio and Productivity (Benchmark: HKG) . . . . . . . 561.17 Quantity Ratio and Productivity (Benchmark: HKG) . . . . 571.18 Product Number Ratio and Productivity (Benchmark: HKG) 581.19 Value Ratio and Productivity (Benchmark: HKG) . . . . . . 592.1 Firm Number and Import Value by Trade Modes . . . . . . . 702.2 Ownership and Trade Mode . . . . . . . . . . . . . . . . . . . 71viiList of Tables2.3 Source of Origin, Firm Location and Product by Trade Mode 722.4 Exchange Rate Pass-Through and Trade Mode: Product-Country-Year Level . . . . . . . . . . . . . . . . . . . . . . . . 742.5 Quantity and Trade Mode: Product-Country-Year Level . . . 762.6 Exchange Rate Pass-Through and Trade Mode: Product-Country-Month Level . . . . . . . . . . . . . . . . . . . . . . 772.7 The Summary of Intermediary Companies . . . . . . . . . . . 802.8 Exchange Rate Pass-Through, Trade Mode and Intermedi-aries: Product-Country-Month Level . . . . . . . . . . . . . . 812.9 The Intermediaries and Ownership . . . . . . . . . . . . . . . 822.10 Exchange Rate Pass-Through and Trade Mode By Owner-ship: Product-Country-Month Level . . . . . . . . . . . . . . 832.11 Exchange Rate Pass-Through and Trade Mode By Source ofOrigin: Product-Country-Month Level . . . . . . . . . . . . . 852.12 Exchange Rate Pass-Through in Export Prices: Product-Country-Month Level . . . . . . . . . . . . . . . . . . . . . . 862.13 Exchange Rate Pass-Through and Trade Mode: Firm-Product-Country-Month Level . . . . . . . . . . . . . . . . . . . . . . 902.14 Exchange Rate Pass-Through and Imported Input . . . . . . 932.15 Exchange Rate Pass-Through and Market Share . . . . . . . 942.16 Exchange Rate Pass-Through and Local Financial Development 983.1 World Production of Rare Earth (2008-2013) . . . . . . . . . 1043.2 Usage Share of Rare Earth Elements by Applications . . . . . 1063.3 China’s Rare Earth Policy (1985-2015) . . . . . . . . . . . . . 1083.4 China’s Rare Earth Export Tariff Rates (2007-2011) . . . . . 1093.5 China’s Rare Earth Export and Production Quotas (2000-2014)1103.6 The Number of Firms Eligible to Export (2006-2014) . . . . . 1103.7 The Impact of China’s Rare Earth Policy on Exporting ofDownstream Products: Product Level . . . . . . . . . . . . . 1243.8 The Impact of China’s Rare Earth Policy on Exporting ofDownstream Products: Firm Level . . . . . . . . . . . . . . . 1253.9 The Impact of China’s Rare Earth Policy on Domestic RareEarth Magnet Producers Relative to Other Magnet Producers 1293.10 The Impact of China’s Rare Earth Policy on Chinese MetalMagnet Producers Relative to Other Countries’ Metal Mag-net Producers . . . . . . . . . . . . . . . . . . . . . . . . . . . 130A.1 Export/Domestic Sales Ratio and Productivity: Firm Level . 139viiiList of TablesA.2 Export/Domestic Sales Ratio and Productivity: Firm-DestinationLevel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140B.1 Exchange Rate Pass-Through and Local Financial Develop-ment: Two Stages Method . . . . . . . . . . . . . . . . . . . . 142C.1 The Treatment and Control Groups . . . . . . . . . . . . . . 143ixList of Figures1.1 The Correlation between Value-added Per Worker and TFP . 91.2 The Average Export Intensity and Productivity Percentile(All Firms) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111.3 The Average Export Intensity and Productivity Percentile(Only Exporting Firms) . . . . . . . . . . . . . . . . . . . . . 121.4 The Average Export Intensity and Productivity (Value-addedPer Worker) Percentile by Industries . . . . . . . . . . . . . . 141.5 The Average Export Intensity and Productivity (TFP) Per-centile by Industries . . . . . . . . . . . . . . . . . . . . . . . 151.6 The Average Export Intensity and Productivity (Value-addedPer Worker) Percentile by Ownership . . . . . . . . . . . . . . 161.7 The Average Export Intensity and Productivity (TFP) Per-centile by Ownership . . . . . . . . . . . . . . . . . . . . . . . 171.8 The Average Export Intensity and Productivity PercentileExcluding Processing Trade . . . . . . . . . . . . . . . . . . . 181.9 Firms’ Other Characteristics and Export Intensity . . . . . . 201.10 The Average Export Intensity and Productivity (Value-addedper worker) Percentile by Different Sales . . . . . . . . . . . . 211.11 The Average Export Intensity and Productivity (TFP) Per-centile by Different Sales . . . . . . . . . . . . . . . . . . . . . 221.12 The Average Export Intensity and Productivity (Value-addedper worker) Percentile by Different Capital/Labor Intensities 231.13 The Average Export Intensity and Productivity (TFP) Per-centile by Different Capital/Labor Intensities . . . . . . . . . 241.14 The Average Export Intensity and Productivity (Value-addedper worker) Percentile by Different Numbers of Employees . . 251.15 The Average Export Intensity and Productivity (TFP) Per-centile by Different Numbers of Employees . . . . . . . . . . . 261.16 The Productivity and The Extensive Margin Effect . . . . . . 431.17 Export Intensity and Productivity . . . . . . . . . . . . . . . 451.18 Sales and Productivity . . . . . . . . . . . . . . . . . . . . . . 46xList of Figures2.1 The Production Chains in “Pure Assembly” and “Import andAssembly” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 662.2 The Distribution Across Imported Product Varieties . . . . . 732.3 The Exchange Rate Pass-Through Over Times . . . . . . . . 782.4 The Distribution of Quality . . . . . . . . . . . . . . . . . . . 882.5 The Exchange Rate Pass-Through Differences Across Industries 912.6 The Exchange Rate Pass-Through Differences Over Times . . 922.7 The Distribution of Loan/GDP . . . . . . . . . . . . . . . . . 963.1 World Reserve of Rare Earth in 2010 . . . . . . . . . . . . . . 1033.2 Value Chain of Rare Earth . . . . . . . . . . . . . . . . . . . 1033.3 Usage of Rare Earth by Applications in 2014 . . . . . . . . . 1053.4 Export Quantity of Rare Earth (2000-2011) . . . . . . . . . . 1123.5 Production Quantity of Rare Earth (2000-2014) . . . . . . . . 1133.6 Average Export Price of Rare Earth (2000-2011) . . . . . . . 1143.7 Smuggling of Rare Earth (2000-2011) . . . . . . . . . . . . . . 1143.8 Share of Rare Earth Metals (2000-2011) . . . . . . . . . . . . 1163.9 Export Destinations of Rare Earth (2000-2011) . . . . . . . . 1163.10 Average Export Price of Rare Earth Across Destinations (2000-2011) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1173.11 Distribution of Rare Earth Export Quantities(2007-2011) . . 1183.12 Distribution of Rare Earth Export Values(2007-2011) . . . . . 1183.13 Domestic Price (2008-2014) . . . . . . . . . . . . . . . . . . . 1203.14 FOB/Domestic Price Ratio (2008-2011) . . . . . . . . . . . . 1213.15 Export Values of Downstream Products (2007-2011) . . . . . 1223.16 Export Values of Three Permanent Magnets (2000-2011) . . . 1273.17 Export Quantity of Three Permanent Magnets (2000-2011) . 127xiAcknowledgementsIt is always challenging to complete a Ph.D. dissertation. Thanks to thosewho supported me in past five years, I made it.I am extremely grateful to my supervisor Michael Devereux for his en-lightening guidance and helpful advice. I would like to thank Matilde Bom-bardini who has always been there in those difficult times when I was workingon my job market paper. I would also like to thank my dissertation commit-tee members Viktoria Hnatkovska and Tomasz Swiecki for their insightfulsuggestions and much encouragement. I thank my university (external)examiners: Brian Copeland, Sanghoon Lee and Loretta Fung for makinghelpful comments. I also enjoy the discussions with my coauthor: XiaonanSun.I thank my friend Zhengfei Yu for living through the Ph.D life togetherwith me. I also thank my friends Tzu-Ting Yang, Yawen Liang, Jun Ma,Haimin Zhang and Jinwen Xu for their support.Special thanks are owned to my mother. Without her, I would notpursue a Ph.D. degree in Economics. And I would like to express my deepestgratitude to my father, who have supported me a lot throughout my yearsof pursuing this degree.xiiChapter 1Productivity, MarketingCosts and Export Intensity1.1 IntroductionStarting with Melitz (2003), recent literature uses the fixed cost of export-ing to explain the interaction between a firm’s productivity and its exportstatus (whether it exports or not), but little research has been done on theinteraction between productivity and export intensity (the ratio betweenexport and total sales). This is surprising since exporting behaviours varyconsiderably, even within exporting firms. This chapter aims to fill this gapby examining how the productivity of Chinese firms is correlated with theirexport intensity.This chapter finds an inverted U-shaped correlation between firms’ pro-ductivity and their export intensity. Within exporting firms, however, pro-ductivity is negatively correlated with export intensity.1 In addition, wefind that this negative correlation is more significant if firms are domesticor produce highly differentiated goods. These empirical findings are surpris-ing, because they imply that some firms with low-productivity can breakinto foreign markets and that they sell relatively more of what they producein foreign markets, compared to more productive firms. Some existing stud-ies (see Dai et al., 2011; Defever and Rian˜o, 2012; Lu, 2010; Lu et al., 2010)also find that Chinese exporting firms are less productive than non-exportingfirms, and variously use trade mode, export subsidies, ownership or factorintensity to explain this puzzle. Unlike these studies, which only investigatethe correlation between productivity and export status, this chapter finds asignificantly negative correlation between productivity and export intensitywithin exporting firms. In addition, we find that the mechanisms in these1Figure 1.3 demonstrates that, when a firm’s productivity is lower than the turningpoint, there is a positive correlation between productivity and export intensity. Afterpassing the turning point, productivity is negatively correlated with export intensity, butthis inverted U-shaped correlation is not significant. In Appendix A.2, we investigate thiscorrelation more carefully.11.1. Introductionstudies cannot entirely explain our empirical findings.In this chapter, we do not investigate how productivity affects exportpropensity. Instead, given that the firms have already entered foreign mar-kets, we study how productivity affects their sales in domestic and foreignmarkets and then rationalize the negative correlation between productiv-ity and export intensity (or the export/domestic sales ratio).2 As far aswe know, Arkolakis (2010) and Eaton et al. (2011) are the only studiesthat investigate this correlation, and they use marketing costs to explainthe positive correlation between the productivity of French firms and theirexport/domestic sales ratios.We will follow Arlolakis’ model and further allow for marketing costheterogeneity across markets to rationalize our empirical findings. Firmshave to pay marketing costs to reach consumers. The marginal marketingcost of a firm is an increasing function of the number of consumers reachedand this number is a firm’s endogenous choice. A firm’s productivity thus hastwo effects on its sales: intensive and extensive margin effects. On the onehand, when productivity is higher, the marginal cost of production is lower,thus enabling a firm to offer a more competitive price to existing consumers,whose demand for the firm’s product then is higher, leading to higher sales.We call this the intensive margin effect. On the other hand, when a firm’sproductivity is higher, it can afford to spend more on marketing so as toreach more consumers. By attracting new customers, the firm again hashigher sales. We call this the extensive margin effect.In Melitz (2003), only the intensive margin effect is active and, for agiven firm, this effect is proportional across markets. Thus, there is no cor-relation between a firm’s productivity and its export/domestic sales ratio.In Arkolakis’ model, the intensive margin effect is the same, but since thenumber of consumers varies across markets for a given firm, the extensivemargin effects are nonproportional across markets. A firm’s productivityis thus correlated with its export/domestic sales ratio. Yet, Arkolakis as-sumes that the elasticity of marketing costs, with respect to the numberof consumers, is the same across markets and finds a positive correlationbetween the productivity of French firms and their export/domestic salesratio. This result is contrary to our findings about Chinese firms. Thus, we2In this chapter, export intensity is defined as the export/total sales ratio, and it ispositively related to a firm’s export/domestic sales ratio. The correlation between a firm’sproductivity and its export intensity has the same sign as the correlation between a firm’sproductivity and its export/domestic sales ratio. We only use export intensity to describeexporting behaviour in the figures and instead use export/domestic sales ratio in ourempirical regressions and structural model.21.1. Introductionextend Arkolakis’ model by relaxing his assumption about the marketingcost function and allowing the elasticity of marketing costs to vary acrossmarkets. This assumption may reflect factors such as different competitiveconditions in different markets.If the elasticity of marketing costs, with respect to the number of con-sumers, in a market is low, then all firms in the market will choose a sim-ilar number of consumers to target, regardless of their productivity, whichmeans that the extensive margin effect will be comparable across firms. Ifthe elasticity of marketing costs is high in a market, then firms with low-productivity will choose to reach out to a small number of consumers, whilehighly-productive firms will be able to afford to pay the higher marketingcosts needed to reach a large number of consumers. For that reason, the ex-tensive margin effect is much greater for highly-productive firms. When theelasticity in the domestic market is higher than in foreign markets, then thedifference in the extensive margin effect between high- and low-productivityfirms will be larger in the domestic market than in foreign markets. Thus,there will be a negative correlation between a firm’s productivity and itsexport/domestic sales ratio.Using the marketing costs framework, we further estimate the elasticityof marketing costs in each market. We indeed find that the elasticity ishigher in domestic than foreign markets. We then investigate which factorsmight explain the elasticity heterogeneity across markets, and we find thatthe negative correlation in a market is more significant if the market share ofChinese firms there is larger. Thus, we believe local market competition maybe one of the factors behind the heterogeneity of marketing elasticity. Morespecifically, we hypothesize that the firms’ marketing strategy is to reachmore consumers by investing so as to make them aware of the differencesbetween their products and those of their competitors. In this context, it ismore costly for firms to differentiate their products when they face strongmarket competition. If consumers perceive the products of all Chinese firmsto be similar, then in markets where the share of Chinese firms is high,expanding sales by way of marketing will be marginally more costly, i.e. theelasticity of marketing costs will be higher.To validate our proposed mechanism we also investigate the strength ofthe correlation between the firms’ export intensity and their productivityacross different industries. First, if firms export homogeneous goods, thenegative correlation between productivity and export intensity should beless significant. This is because the elasticity of marketing costs for homo-geneous goods should vary less across markets. Thus, the extensive margineffect of productivity should also be comparable across markets. Second,31.1. Introductionwe investigate how advertising expenditure affects this correlation by calcu-lating the degree of advertising expenditure (advertising spending/sales) foreach industry. If firms belong to an industry with a high degree of advertis-ing expenditure, then the marketing strategy will be more important, andthus the negative correlation should be more significant in that industry.Our regression analysis indeed confirms these two hypotheses.Finally, since we do not observe a firm’s marketing costs directly, wediscuss three alternative explanations for our empirical results and find thateach of them can match some but not all of the patterns in the data. Thefirst explanation is based on variable markups. Melitz and Ottaviano (2008)adopt a linear demand framework and endogenize differences in the tough-ness of competition across markets. When a firm’s productivity increases, itssales expand at different rates across markets depending on those markets’sizes. In our data we find that these auxiliary predictions based on Melitzand Ottaviano (2008) cannot entirely explain our empirical findings. First,we find that market size can only explain a small part of the negative corre-lation between the productivity of Chinese firms and their export/domesticsales ratios. Second, Melitz and Ottaviano (2008) cannot explain Chinesefirms’ pricing strategy. They assume a quadratic utility function, and thusthat the markups are heterogeneous across firms, due to varying productivi-ty. Instead the utility function in Arkolakis (2010) is in the form of constantelasticity of substitution (CES), and thus the markups are the same re-gardless of the firm’s productivity. Melitz and Ottaviano (2008) predictthat both the price and quantity ratios between two export destinations forthe same firm-product pair should be correlated with a firm’s productivity.Arkolakis (2010) predicts that only the quantity ratio between two exportdestinations will be correlated with a firm’s productivity and that the priceratio between two destinations should be uncorrelated with productivity.Using the Chinese data, we only find correlation between a firm’s quantityratio and its productivity and none between its price ratio and productivity.The second explanation for our results concerns the quality of products.Manova and Zhang (2012) find that more successful exporters use higherquality inputs to produce higher quality goods, and that firms vary thequality of their products across destinations by using inputs of differentquality levels. Thus, sales vary across markets due to the heterogeneousquality of products. They also argue that markups are heterogeneous acrossfirms. If this were true, then the price ratio between two export destinationsshould be correlated with a firm’s productivity, but we cannot find suchevidence in the Chinese data.The third explanation relates to the CES demand function with mar-41.1. Introductionket power. If a firm is in a market with just a few competitors, then itspass-through of costs to price will be low because it will be concerned withits market share. Therefore, if the number of firms is heterogeneous acrossmarkets, then a firm’s productivity will be correlated with its export inten-sity. However, this mechanism also implies a correlation between the priceratio and productivity.A fourth possible explanation is heterogeneous ranges of products acrossmarkets. 3 Both the range of products and sales per product would affect afirm’s total sales. We find that the product range accounts for 17% − 26%and sales per product accounts for 74% − 83% of the negative correlationbetween a firm’s productivity and export intensity.1.1.1 Related LiteratureThis chapter is closely related to two strands of literature. The first consistsof studies that investigate the exporting performance of Chinese firms. Luet al. (2010) find that, among foreign affiliates, exporters are less productivethan non-exporters in China. They argue that the fixed cost for foreign affil-iates in foreign markets is lower than that in the Chinese market. Dai et al.(2011) argue that the low productivity of exporting firms is entirely drivenby firms that engage only in export processing - the activity of assemblingtariff-exempted imported inputs into final goods for resale in foreign market-s. These firms are less productive than non-exporters and cause a decreasein the average productivity across all exporting firms. Lu (2010) uses fac-tor intensity to explain the productivity difference between exporting firmsand non-exporting firms. When countries differ in their factor endowment,sectors that are intensive in the locally abundant factor face higher compe-tition in the domestic market than in foreign markets. Since China has ahuge labour supply, domestic rather than foreign markets select the mostefficient firms in labour-intensive industries. Yue and Ju (2013) also findthat production in China from 1999 to 2007 became more capital-intensive,while exporting became more labour-intensive. Defever and Rian˜o (2012)argue that pure exporting firms are less productive than regular exportingfirms, due to export subsidies.The second strand of literature investigates the exporting behaviour offirms using data from developed countries. Bernard et al. (2000) use USdata to study why exporting plants only export a small fraction of theiroutput. Crino` and Epifani (2012), using Italian data, find that Total Factor3Melitz et al. (2014) argue that firms would have different product ranges across mar-kets due to the market competition.51.2. Some Stylized FactsProductivity (TFP) is strongly negatively correlated with the export shareto low-income destinations and uncorrelated with the export share to high-income destinations.The rest of this chapter is organized as follows. Section 2 describes theChinese data and presents some stylized facts obtained from it. Section 3addresses the empirical results. Section 4 presents a theoretical model to ex-plain the empirical patterns. In section 5, we conduct additional robustnesschecks. Finally, section 6 concludes this chapter.1.2 Some Stylized Facts1.2.1 Customs Data and Firm DataIn this chapter, we use two databases, the “Chinese Industrial EnterprisesDatabase” and the “Chinese Customs Export and Import Database,” from2000 to 2006. The first database is collected by China’s National Bureau ofStatistics and covers all state-owned industrial firms and non-state-ownedindustrial firms above a designated size.4 It includes information about thebalance sheet of the firms. The second database is collected by the ChineseCustoms Office and includes information on export and import value andvolume for each eight-digit harmonized system (HS), the exporting country,and the importing country. Both databases contain firm identification num-bers but they are completely different in the two databases. Therefore, weuse the firm name, telephone number, manager’s name, and postal code tomerge them.51.2.2 Data SummaryIn this chapter we only use the data for manufacturing firms. Due to theconcerns over measurement error, we drop firms whose value-added, capi-tal, sales or export values are negative or zero. We also drop small firmswith less than six employees and without valid postal codes. Following thecleaning of the data, there are over 1.3 million firms left in the “ChineseIndustrial Enterprises Database.” Of these, 29% are exporters. A summaryis presented in Table 1.1. 78% of the firms are Chinese-owned, either by thestate or private parties, but only a small fraction of these Chinese-owned4The industries include mining, manufacturing, the production and supply of electricpower, gas and water. The designated size requires firms to have sales of at least 5 millionChinese Yuan (about 0.8 million US dollars).5See Wang and Yu (2012) for a detailed description of the matching procedure.61.2. Some Stylized Factsfirms are exporters while over half of joint-owned and foreign-owned firmsengaging in exporting. This result is similar to that in Lu et al. (2010), andshows that it is easier for foreign-owned firms to export.Table 1.1: The Summary of Firm DataNumber of Firms Percent of ExporterPanel A: by Year:2000 134,775 26.52%2001 143,931 27.10%2002 155,005 28.05%2003 173,114 28.67%2004 231,249 31.18%2005 231,623 30.87%2006 258,586 29.19%Panel B: by Ownership:State-owned Firm 523,540 18.21%Private-owned Firm 517,403 20.99%Joint-owned Firm 146,129 56.40%Foreign-owned Firm 141,211 71.23%All Firms 1,328,283 29.13%Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This table summarizes the firm data. We only use the data from manufacturingfirms. We drop the firms when the value-added, capital, sales and export values arenegative or zero. We also drop small firms with less than six employees and without validpostal codes. Foreign-owned or jointly-owned firms include firms that are owned by HongKong, Macao, Taiwan and other foreign countries.We divide exporting firms into ten bins according to their export inten-sity. When a firm’s export intensity is between (i− 1)× 10% and i× 10%,the bin equals i (i=1,2,..10). As Table 1.2 shows, the distribution of firms’export intensity is polarized. Over 16% of these firms export less than 10%of their output and about 41% have a very high export intensity. Over halfof the foreign-owned or joint-owned firms export 90% of their output. Thisresult is different from that in Bernard et al. (2000), who find that aroundtwo-thirds of US exporters sell less than 10% of their output abroad.After merging the “Chinese Industrial Enterprises Database” with the“Chinese Customs Export and Import Database,” we have about 177, 396firms, which is about 45.85% of all exporting firms. The total export value of71.2. Some Stylized FactsTable 1.2: Percentage of Firms and Export IntensityExport Intensity Full Sample Chinese-owned Foreign-owned/Joint-owned(0, 0.1] 16.34% 21.92% 10.13%(0.1, 0.2] 7.48% 9.29% 5.46%(0.2, 0.3] 5.26% 6.02% 4.42%(0.3, 0.4] 4.63% 5.23% 3.96%(0.4, 0.5] 4.38% 4.76% 3.95%(0.5, 0.6] 4.29% 4.49% 4.08%(0.6, 0.7] 4.48% 4.50% 4.46%(0.7, 0.8] 5.14% 4.95% 5.36%(0.8, 0.9] 6.55% 6.10% 7.04%(0.9, 1] 41.44% 32.75% 51.13%100% 100% 100%Data Sources: The “ Chinese Industrial Enterprises Database” (2000-2006).Notes: This table shows the distribution of firms’ export intensity within exporting firms.We only use the data from manufacturing firms. We drop the firms when the value-added,capital, sales and export values are negative or zero. We also drop small firms with less than sixemployees and without valid postal codes. Chinese-owned firms include state-owned firms andprivate-owned firms. Foreign-owned or joint-owned firms include firms that are owned by HongKong, Macao, Taiwan and other foreign countries.the firms in the merged dataset comprises about 54.54% of that of all export-ing firms. Due to the limits of the matching method, we cannot perfectlymerge the two databases. Another reason for the mismatch is that somefirms export their products through intermediary firms. Thus, we can ob-serve these firms’ records in the “Chinese Industrial Enterprises Database,”but not in the “Chinese Customs Export and Import Database.”6 The dataare summarized in Table 1.3. Over half of these firms are engaged in theprocessing trade. This is one reason why the export intensity of Chinesefirms is unexpectedly high compared with that of US firms.1.2.3 Export Intensity and ProductivityWe use two ways of measuring Chinese firms’ productivity. The first is value-added per worker and the second is TFP, which is constructed following themethod in Levinsohn and Petrin (2003). Figure 1.1 shows that these twomeasurements are highly correlated at 0.76.First, we rank firms by their productivity and divide them into 100 per-centiles. Then, we calculate the average export intensity for each produc-tivity percentile. Finally, we determine the correlation between the average6Bai et al. (2013) study the indirect export behaviour of Chinese firms.81.2. Some Stylized FactsTable 1.3: The Summary of Merged DataFirm Number PercentBy Trade Mode:Non-Processing Trade 85,503 48.20%Processing Trade 91,893 51.80%All 177,396 100%Data Sources: The “Chinese Industrial Enterprises Database”(2000-2006) and the “Chinese Customs Export and ImportDatabase” (2000-2006).Notes: This table shows the summary of merged data. We onlyuse the data from manufacturing firms. We drop those firms,whose value-added, capital, sales and export values are negativeor zero. We also drop small firms with less than six employeesand without valid postal codes.Figure 1.1: The Correlation between Value-added Per Worker and TFPData Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the“Chinese Customs Export and Import Database” (2000-2006).Notes: This figure shows the correlation between two productivity measurements. Thex-axis is ln(TFP) and the y-axis is ln(value-added per worker). The correlation betweentwo measurements is 0.76.91.2. Some Stylized Factsexport intensity and the productivity percentile. As shown in Figure 1.2, thecorrelation between the firms’ productivity and the average export intensityis an inverted U-shaped curve.7 The average export intensity of firms withlow productivity is indeed lower, because most of them are non-exportingfirms. The average export intensity then increases with the firms’ productiv-ity. However, when the firms’ productivity passes a particular turning point,the average export intensity starts to decrease with the firms’ productivi-ty. The trend is more significant when we use value-added per worker tomeasure the firms’ productivity. In this chapter, we do not investigate howthe firms’ productivity affects their export propensity. Instead, given thatthe firms have already entered foreign markets, we study the effect of theexporting firms’ productivity on their sales in the domestic and foreign mar-kets respectively. Thus, we drop all non-exporting firms and pure exportingfirms (firms that export all of their output)8 and recalculate the correlations.Figure 1.3 shows the results. We find that the inverted U-shaped correla-tion still holds. However, the positive correlation before the turning pointis no longer significant. Less than 6% of the firms have lower productivitythan the turning point. Basically, Figure 1.3 shows a significantly negativecorrelation between the firms’ productivity and their export intensity. Inthe rest of this chapter, we focus on this negative correlation. In AppendixA.2 we investigate the inverted U-shaped correlation more carefully.Excluding the Factor Intensity EffectIn Figure 1.3, we pooled all exporting firms together. However, the negativepattern might differ across industries. As Lu (2010) shows, capital/labourintensity differences across industries will affect the correlation between afirm’s productivity and its export status. To exclude the industry effec-t, we rank the industries in terms of their capital/labour ratio. Then, wecalculate the correlation between the firms’ average export intensity andthe productivity percentile at every two-digit industry level. There are 307Our emperical result is consistent with previous studies using Chinese data. When weuse value-added per worker to measure productivity, exporting firms are less productivethan non-exporting firms. When we use TFP to measure productivity, exporting firmsare more productive.8There are 111, 052 firms that are pure exporters. 14% of them are state-owned, 26%are private-owned, 21% are joint-owned and 39% are foreign-owned. The exporting be-haviour of these pure exporters is different from that of regular exporting firms. Thus,we have to drop the former group. Defever and Rian˜o (2012) investigate their exportingbehaviour.101.2. Some Stylized FactsFigure 1.2: The Average Export Intensity and Productivity Percentile (AllFirms).1.15.2.25.3Export Intensity0 20 40 60 80 100Value-added Per Worker Percentile95% Confidence Interval Average ExportIntensity.05.1.15.2Export Intensity0 20 40 60 80 100TFP Percentile95% Confidence Interval Average ExportIntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity percentile (all firms). We rank all firms according to their productivity anddivide them into 100 percentiles. Then, we calculate the average export intensity offirms for each productivity percentile. The x-axis is the productivity percentile and they-axis is the average export intensity within the corresponding percentile.111.2. Some Stylized FactsFigure 1.3: The Average Export Intensity and Productivity Percentile (OnlyExporting Firms).2.3.4.5.6Export Intensity0 20 40 60 80 100Value-added Per Worker Percentile95% Confidence Interval Average Export Intensity.2.3.4.5.6Export Intensity0 20 40 60 80 100TFP Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity percentile (only exporting firms). We drop both non-exporting firms and pureexporters, which export all of their outputs. We rank all firms according to their pro-ductivity and divide them into 100 percentiles. Then, we calculate the average exportintensity of firms for each productivity percentile. The x-axis is the productivity per-centile and the y-axis is the average export intensity within the corresponding percentile.121.2. Some Stylized Factsindustries at the two-digit industry level.9 From the first industry to thelast one, the capital/labour ratio of the industry increases. The results arepresented in Figures 1.4 and 1.5. We can see that the negative correlationis robust across different industries. Thus, factor intensity does not explainthe empirical pattern we have found. The only two industries that are ab-normal are “Recycling of Waste and Scrap ”(Industry 9) and “Manufactureof Tobacco Products” (Industry 29). The “Recycling of Waste and Scrap ”industry is very small, accounting for just 0.02% of all firms.Excluding the Ownership EffectLu et al. (2010) find that the ownership of a firm also affects its exportingbehaviour. They find that, among foreign affiliates in China, exporters areless productive than non-exporters. Table 1.2 shows that foreign-owned andjoint-owned firms export most of their output. Defever and Rian˜o (2012)argue that foreign-owned firms receive export subsidies. Thus, these firm-s have a greater incentive to export than Chinese-owned firms. For thatreason, we divide our sets of firms into four groups according to their own-ership, namely state-owned, private-owned, joint-owned and foreign-owned.The results are presented in Figures 1.6 and 1.7. Again, the negative patternremains robust. Thus, ownership does not explain the negative pattern.Excluding the Trade Mode EffectDai et al. (2011) show firms that participate in the processing trade to beless productive than non-exporters. At the same time, these processingtrade firms have a high export intensity. Thus, in order to exclude thiseffect, we drop all firms that are engaged in the processing trade. Theresults, presented in Figure 1.8, show that the negative correlation betweenthe firms’ productivity and their export intensity is still robust. Thus, thetrade mode does not explain the negative pattern.1.2.4 Export Intensity and Sales, Capital Intensity, andNumber of EmployeesWe next relate the other characteristics of the firms to their export inten-sity. First, Figure 1.9 shows the relationship between the firms’ sales, cap-ital/labour ratio, and number of employees, and their export intensity. Wecan see that the export intensity of larger (either by sales or by number of9The names of these industries are given in Appendix A.1.131.2. Some Stylized FactsFigure 1.4: The Average Export Intensity and Productivity (Value-addedPer Worker) Percentile by Industries.2.4.6.80 20 40 60 80 100Leather.5.6.7.80 20 40 60 80 100Wearing Apparel.4.5.6.7.80 20 40 60 80 100Crafts.4.5.6.7.8.90 20 40 60 80 100Cutural Products.2.4.6.80 20 40 60 80 100Furniture.2.4.6.80 20 40 60 80 100Office Machinery.2.3.4.5.6.70 20 40 60 80 100Metal.35.4.45.5.55.60 20 40 60 80 100Textile0.2.4.6.810 20 40 60 80 100Recycling.2.3.4.50 20 40 60 80 100Universal Equipmentsl.15.2.25.3.35.40 20 40 60 80 100Special Equipments.3.4.5.6.7.80 20 40 60 80 100Wood.2.3.4.5.60 20 40 60 80 100Electric.2.3.4.5.6.70 20 40 60 80 100Plastic0.2.4.6.80 20 40 60 80 100Rubber.2.3.4.5.6.70 20 40 60 80 100Agricultural0.2.4.6.80 20 40 60 80 100Food.2.3.4.5.60 20 40 60 80 100Non-metal0.2.4.6.80 20 40 60 80 100Printing.3.4.5.6.70 20 40 60 80 100Telecommunication.2.3.4.50 20 40 60 80 100Transportation.1.2.3.4.5.60 20 40 60 80 100Pharmaceuticals0.2.4.60 20 40 60 80 100Drinking0.2.4.6.80 20 40 60 80 100Paper.2.25.3.35.4.450 20 40 60 80 100Raw Materials.1.2.3.4.5.60 20 40 60 80 100Non-ferrous Metal0.2.4.60 20 40 60 80 100Ferrous Metal0.2.4.6.80 20 40 60 80 100Fiber-.50.511.50 20 40 60 80 100Tobacco0.2.4.6.80 20 40 60 80 100Crude OilExport IntensityValue-added Per Worker Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity (value-added per worker) percentile by industries. We only include firms thatsell both in the domestic market and foreign markets. The industries are on 2-digits lev-el. From the first industry to the last one, the capital/labor ratio of industry increases.The names of these industries are in the Appendix A.1. The only two industries, whichare abnormal, are “Recycling of Waste and Scrap ”(industry 9) and “Manufacture ofTobacco Products” (industry 29).141.2. Some Stylized FactsFigure 1.5: The Average Export Intensity and Productivity (TFP) Per-centile by Industries.4.5.6.7.80 20 40 60 80 100Leather.4.5.6.7.80 20 40 60 80 100Wearing Apparel.3.4.5.6.7.80 20 40 60 80 100Crafts.4.5.6.7.80 20 40 60 80 100Cutural Products.2.4.6.80 20 40 60 80 100Furniture.2.4.6.80 20 40 60 80 100Office Machinery.3.4.5.6.70 20 40 60 80 100Metal.3.4.5.6.70 20 40 60 80 100Textile0.2.4.6.810 20 40 60 80 100Recycling.1.2.3.4.5.60 20 40 60 80 100Universal Equipments.1.2.3.4.50 20 40 60 80 100Special Equipments.2.4.6.80 20 40 60 80 100Wood.2.3.4.5.60 20 40 60 80 100Electric.2.3.4.5.6.70 20 40 60 80 100Plastic.2.4.6.80 20 40 60 80 100Rubber0.2.4.6.80 20 40 60 80 100Agricultural0.2.4.6.80 20 40 60 80 100Food.2.3.4.5.60 20 40 60 80 100Non-metal0.2.4.6.80 20 40 60 80 100Printing.4.45.5.55.6.650 20 40 60 80 100Telecommunication0.2.4.60 20 40 60 80 100Transportation0.2.4.60 20 40 60 80 100Pharmaceuticals0.2.4.6.80 20 40 60 80 100Drinking.1.2.3.4.5.60 20 40 60 80 100Paper.1.2.3.4.50 20 40 60 80 100Raw Materials.1.2.3.4.5.60 20 40 60 80 100Non-ferrous Metal0.2.4.6.80 20 40 60 80 100Ferrous Metal0.2.4.6.80 20 40 60 80 100Fibres-.50.510 20 40 60 80 100Tabacco0.2.4.6.80 20 40 60 80 100Crude OilExport IntensityTFP Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity (TFP) percentile by industries. We only include firms that sell both in thedomestic market and foreign markets. The industries are on 2-digits level. From thefirst industry to the last one, the capital/labor ratio of industry increases. The namesof these industries are in the Appendix A.1. The only two industries, which are ab-normal, are “Recycling of Waste and Scrap ”(industry 9) and “Manufacture of TobaccoProducts” (industry 29).151.2. Some Stylized FactsFigure 1.6: The Average Export Intensity and Productivity (Value-addedPer Worker) Percentile by Ownership.1.2.3.4.50 20 40 60 80 100State-owned.2.3.4.5.60 20 40 60 80 100Private-owned.2.3.4.5.6.70 20 40 60 80 100Joint-owned.3.4.5.6.70 20 40 60 80 100Foreign-ownedExport IntensityValue-added Per Worker Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity (value-added per worker) percentile by ownership. We only include firms thatsell both in the domestic market and foreign markets. Foreign-owned or joint-ownedfirms include firms that are owned by Hong Kong, Taiwan and Macao.161.2. Some Stylized FactsFigure 1.7: The Average Export Intensity and Productivity (TFP) Per-centile by Ownership.1.2.3.4.50 20 40 60 80 100State-owned.2.3.4.5.60 20 40 60 80 100Private-owned.2.3.4.5.60 20 40 60 80 100Joint-owned.4.5.6.70 20 40 60 80 100Foreign-ownedExport IntensityTFP Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity (TFP) percentile by ownership. We only include firms that sell both in thedomestic market and foreign markets. Foreign-owned or joint-owned firms include firmsthat are owned by Hong Kong, Taiwan and Macao.171.2. Some Stylized FactsFigure 1.8: The Average Export Intensity and Productivity Percentile Ex-cluding Processing Trade.2.3.4.5.6Export Intensity0 20 40 60 80 100Value-added Per Worker Percentile95% Confidence Interval Average Export Intensity.1.2.3.4.5.6Export Intensity0 20 40 60 80 100TFP Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the“Chinese Customs Export and Import Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity percentile excluding processing trade. We only include firms that sell both inthe domestic market and foreign markets.181.3. Empirical Analysisemployees) and capital-intensive firms is lower. This maybe because firms’productivity is usually positively related to their size and capital intensi-ty, and higher productivity leads to lower export intensity. One concernis whether the negative correlation between a firm’s export intensity andits productivity is robust across different sizes of firms. To investigate, wedivide the firms into three groups according to their sales, capital/labourratio and number of employees, respectively. Then we examine whether thefirms’ characteristics affect the negative pattern. Figures 1.10 and 1.11 showthe results for firms with different sales levels. Figures 1.12 and 1.13 showthe results for firms with different capital/labor ratios. Finally, Figures 1.14and 1.15 show the results for firms with different numbers of employees. Thenegative pattern remains robust but is more significant for larger firms.1.2.5 SummaryIn this section, we have demonstrated four stylized facts about Chinese ex-porting firms:1 There is an inverted U-shaped correlation between the firms’ produc-tivity and their export intensity.2 Within exporting firms, the firms’ productivity is negatively correlatedwith their export intensity. Even when we control for the effects of capi-tal/labour ratio, ownership, trade mode, and export subsidies, the negativecorrelation remains robust.3 The export intensity of larger and more capital-intensive firms is lower.4 The negative correlation between the firms’ productivity and their ex-port intensity is robust across different levels of firm size, capital/labour ra-tio, and number of employees, but this pattern is more significant for largerfirms.1.3 Empirical Analysis1.3.1 Firm LevelIn Figure 1.3, we showed that, for exporting firms, a firm’s productivity isnegatively correlated with its export intensity. However, we did not controlfor industry fixed effects or various firm characteristics. In this section, we191.3. Empirical AnalysisFigure 1.9: Firms’ Other Characteristics and Export Intensity.2.3.4.5.60 20 40 60 80 100Sale PercentileSale.3.4.5.6.70 20 40 60 80 100Capital/Labor Ratio PercentileCapital/Labor Ratio.2.3.4.5.60 20 40 60 80 100Employee Number PercentileEmployee NumberExport Intensity95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between firms’ other characteristics and exportintensity.Firms’ other characteristics include firms’ sales, capital/labor ratio and thenumber of employees. We only include firms that sell both in the domestic market andforeign markets.201.3. Empirical AnalysisFigure 1.10: The Average Export Intensity and Productivity (Value-addedper worker) Percentile by Different Sales.4.45.5.55.6.650 20 40 60 80 100Small Firm.3.4.5.60 20 40 60 80 100Median Firm.2.3.4.5.60 20 40 60 80 100Large FirmExport IntensityValue-added Per Worker Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity (value-added per worker) percentile by different sales. We only include firmsthat sell both in the domestic market and foreign markets.211.3. Empirical AnalysisFigure 1.11: The Average Export Intensity and Productivity (TFP) Per-centile by Different Sales.35.4.45.5.55.60 20 40 60 80 100Small Firm.35.4.45.5.55.60 20 40 60 80 100Median Firm.2.3.4.5.60 20 40 60 80 100Large FirmExport IntensityTFP Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity (TFP) percentile by different sales. We only include firms that sell both in thedomestic market and foreign markets.221.3. Empirical AnalysisFigure 1.12: The Average Export Intensity and Productivity (Value-addedper worker) Percentile by Different Capital/Labor Intensities.3.4.5.6.70 20 40 60 80 100Low Capital/Labor Intensity Firm.3.4.5.60 20 40 60 80 100Median Capital/Labor Intensity Firm.25.3.35.4.450 20 40 60 80 100High Capital/Labor Intensity FirmExport IntensityValue-added Per Worker Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity (value-added per worker) percentile by different capital/labor intensity. Weonly include firms that sell both in the domestic market and foreign markets.231.3. Empirical AnalysisFigure 1.13: The Average Export Intensity and Productivity (TFP) Per-centile by Different Capital/Labor Intensities.4.5.6.70 20 40 60 80 100Low Capital/Labor Intensity Firm.3.4.5.60 20 40 60 80 100Median Capital/Labor Intensity Firm.1.2.3.4.50 20 40 60 80 100High Capital/Labor Intensity FirmExport IntensityTFP Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity (TFP) percentile by different capital/labor intensity. We only include firmsthat sell both in the domestic market and foreign markets.241.3. Empirical AnalysisFigure 1.14: The Average Export Intensity and Productivity (Value-addedper worker) Percentile by Different Numbers of Employees.3.4.5.60 20 40 60 80 100Small Firm.3.4.5.6.70 20 40 60 80 100Median Firm.2.3.4.5.60 20 40 60 80 100Large FirmExport IntensityValue-added Per Worker Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity (value-added per worker) percentile by different numbers of employees. Weonly include firms that sell both in the domestic market and foreign markets.251.3. Empirical AnalysisFigure 1.15: The Average Export Intensity and Productivity (TFP) Per-centile by Different Numbers of Employees.2.3.4.5.60 20 40 60 80 100Small Firm.2.3.4.5.60 20 40 60 80 100Median Firm.2.3.4.5.60 20 40 60 80 100Large FirmExport IntensityTFP Percentile95% Confidence Interval Average Export IntensityData Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This figure shows the correlation between the average export intensity and pro-ductivity (TFP) percentile by different numbers of employees. We only include firmsthat sell both in the domestic market and foreign markets.261.3. Empirical Analysisinvestigate this correlation more carefully. Since, in Melitz’s model, a firm’sproductivity is considered as exogenous, we consider it our independentvariable, while our dependent variable is export intensity. The benchmarkregression10 isln(Export/Domestic Sales Ratioijkt) =α0 + α1ln(Pijkt) + other controls+ µjkt + ijkt(1.1)Here, i is the firm, j is the province in which firm i is located, k is theindustry to which firm i belongs, and t is the year. Pijkt is firm i’s productiv-ity (value-added per worker or TFP). Export/Domestic Sales Ratioijkt isa new indicator of firm i’s export behaviour. µjkt is the province-industry-year dummy. Other controls include the firm’s total sales and its capi-tal/labour ratio. Some existing studies find that Chinese firms have a com-parative advantage in exporting labour-intensive products. Thus, a firm’scapital/labour ratio might affect its export intensity. By including a firm’ssales, we can control the effect of a firm’s size. α1 measures the correlationbetween the firm’s productivity and its export/domestic sales ratio.10In Appendix A.2, we rewrite Regression 1.1 in quadratic form and investigate theinverted U-shaped correlation between the firms’ productivity and their export intensity.We don’t find strong evidences to support the quadratic form.271.3.EmpiricalAnalysisTable 1.4: Export/Domestic Sales Ratio and Productivity: Firm LevelDependent Variable: ln(Export/Domestic Sales)(1) (2) (3) (4) (5) (6) (7) (8)ln(Productivity) -0.202*** -0.143*** -0.168*** -0.155*** -0.170*** -0.143*** -0.146*** -0.266***(0.0167) (0.0140) (0.0135) (0.0178) (0.0206) (0.0192) (0.0180) (0.0230)ln(Capital/Labour Ratio) -0.133*** -0.173*** -0.169*** -0.153*** -0.201*** -0.222***(0.0165) (0.0169) (0.0154) (0.0161) (0.0168) (0.0157)ln(Sale) -0.0242 0.130***(0.0268) (0.0365)Constant 0.612*** 0.857*** 0.598*** 0.799*** 1.002*** 1.359*** 1.086*** 0.609**(0.0649) (0.0836) (0.0859) (0.274) (0.142) (0.161) (0.163) (0.279)Ownership FE X X X XProvince-Industry-Year FE X X X X X X X XCluster By Industry X X X X X X X XObservations 275,872 275,872 275,872 275,872 275,872 275,872 275,872 275,872R-squared 0.348 0.351 0.368 0.369 0.348 0.351 0.369 0.369Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This table shows the correlation between export/domestic sales ratio and productivity on the firm level.1. The productivity in the first four columns is value-added per worker. The productivity in last four columns is TFP.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.281.3. Empirical AnalysisWe use a firm’s export/domestic sales ratio instead of its export inten-sity to measure its exporting behaviour for two reasons. First, if there is ameasurement error regarding a firm’s sales, for example if the firm’s sales areovervalued, then the firm’s export intensity (export/total sales) will be lowerand its productivity11 higher. Thus, there is always a negative correlationbetween a firm’s productivity and its export intensity. In order to removethis concern, we define a new measurement to describe a firm’s export be-haviour, namely the export/domestic sales ratio. Second, in the structuremodel, it is easier to calculate a firm’s export/domestic sales ratio and ourstructure estimation is mainly based on this variable. The correlation be-tween a firm’s productivity and its export intensity will have the same signas the correlation between the firm’s productivity and its export/domesticsales ratio. Thus, using this new measurement will not change our conclu-sions.In Regression 1.1, we do not intend to suggest that there is a causal rela-tionship between firms’ productivity and their export/domestic sales ratio;instead, we are just showing that there is a correlation between them. Theresults are presented in Table 1.4. We find that α1 is negative and signifi-cant. If a firm’s labour productivity (value-added per worker) increases by10%, then its export/domestic sales ratio decreases by 1.55%. If a firm’s TF-P increases by 10%, then its export/domestic sales ratio decreases by 2.66%.This negative correlation means that, when a firm’s productivity is higher,it sells relatively more in the domestic market than in foreign markets. Inaddition, more capital-intensive firms have lower export intensity, which isconsistent with existing studies. China has comparative advantages in ex-porting labour-intensive products. The effect of a firm’s size is not clear. Ifwe use a firm’s labour productivity as our measure, then the firm’s size hasno effect on its export/domestic sales ratio. However, if we use a firm’s TFP,then the firm’s size has a significantly positive effect on its export/domesticsales ratio. Thus, the output of large firms relies more on foreign marketsthan does that of small firms.1.3.2 Processing Trade and Ownership EffectIn Table 1.5, we additionally drop all firms which are engaged in the process-ing trade. We find that the negative pattern remains robust. The negativecorrelation might also be different across different types of ownership. Thus,in Table 1.6, we show the results of adding the interaction term between the11To construct our measurement of a firm’s productivity, we use value-added data.Thus, measurement error in a firm’s sales might not be a serious problem.291.3. Empirical Analysisfirms’ productivity and a dummy for ownership (Domestic F irm) into Re-gression 1.1. Here, Domestic F irm is a dummy equals to 1 when the firmis state-owned or private-owned and 0 otherwise. The interaction term isnegative and insignificant when we use the value-added per worker to mea-sure a firm’s productivity but it is negative and significant when we useTFP to measure a firm’s productivity. Thus, if the firm is state-owned orprivate-owned, the negative correlation is larger. This means that, when theproductivity of Chinese-owned firms increases, their sales mainly rely on thedomestic market.Table 1.5: Export/Domestic Sales Ratio and Productivity Excluding Pro-cessing Trade: Firm LevelDependent Variable: ln(Export/Domestic Sales)(1) (2)ln(Productivity) -0.0653** -0.235***(0.0299) (0.0321)ln(Capital/Labour Ratio) -0.228*** -0.260***(0.0209) (0.0229)ln(Sale) -0.211*** -0.0493(0.0278) (0.0375)Constant 2.659*** 2.440***(0.298) (0.300)Ownership FE X XProvince-Industry-Year FE X XCluster By Industry X XObservations 69,691 69,691R-squared 0.472 0.474Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “ChineseCustoms Export and Import Database” (2000-2006).Notes: This table shows the correlation between export/domestic sales ratio and productivityexcluding processing trade on the firm level.1. The productivity in the column 1 is value-added per worker. The productivity in column 2is TFP.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.1.3.3 Differentiated GoodsNow, we investigate whether the degree of differentiation in the goods thatfirms export affects the negative pattern. We use the elasticity of substi-301.3. Empirical AnalysisTable 1.6: Export/Domestic Sales Ratio, Productivity and Ownership: FirmLevelDependent Variable: ln(Export/Domestic Sales)(1) (2)ln(Productivity) -0.146*** -0.191***(0.0255) (0.0248)ln(Productivity)× Domestic Firm Dummy -0.0190 -0.160***(0.0384) (0.0321)ln(Capital/Labour Ratio) -0.170*** -0.225***(0.0151) (0.0154)ln(Sale) -0.0240 0.140***(0.0269) (0.0367)Constant 0.833*** 1.100***(0.269) (0.239)Ownership FE X XProvince-Industry-Year FE X XCluster By Industry X XObservations 275,872 275,872R-squared 0.369 0.370Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006).Notes: This table shows the impact of ownership on the correlation between export/domestic salesratio and productivity.1. The productivity in the column 1 is value-added per worker. The productivity in column 2 isTFP.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.311.3. Empirical Analysistution between varieties from Soderbery (2013) to measure the degree ofdifferentiation of goods.Using US data, Soderbery (2013) estimates the elasticity of substitutionfor import goods at the HS8 level. Thus, we firstly have to aggregate theseelasticities to the firm level. We use the export value of each good as aweight to calculate the weighted elasticity for each firm. The higher theweighted elasticity, the lower is the degree of differentiation.Weighted Elasticityi=∑nExport V alue of Good n for firm iTotal Export of firm i× Elasticity of Good nAfter obtaining each firm’s degree of differentiation of goods, we divide thefirms into two groups: those exporting homogeneous goods and those ex-porting differentiated goods. We use a dummy to represent the firm’s typeand call it the homogeneous dummy. When a firm’s degree of differenti-ation is larger than the median, then we assign 1 to this dummy. Other-wise, we assign 0 to it. Then, we include the interaction term between thehomogeneous dummy and the firm’s productivity in Regression 1.1. Theresults are shown in Table 1.7. We find that the interaction term is positiveand significant when we use value-added per worker to measure a firm’s pro-ductivity but positive and insignificant when we use TFP as the measure.In addition, the coefficient of the homogeneous dummy is negative. Thus,when a firm exports homogeneous goods, its export/domestic sales ratio islower and the negative pattern is less significant.1.3.4 Firm-Destination LevelFirms might export their products to many countries. The pattern mightdiffer depending on the country of destination. Thus it is necessary to controlthe export destination effect. The export valuem/domestic sales ratio foreach market m is defined asExport V aluem/Domestic Sales Ratio =Export V alue to Country mDomestic SalesThen, the firm-destination level regression is given byln(Export V aluem/Domestic Sales Ratioijkt) =β0 + β1ln(Pijkt)+ other controls+ µmjkt+ mijkt(1.2)321.3. Empirical AnalysisTable 1.7: Export/Domestic Sales Ratio, Productivity and DifferentiatedGoods: Firm LevelDependent Variable: ln(Export/Domestic Sales)(1) (2)ln(Productivity) -0.272*** -0.354***(0.0330) (0.0404)ln(Productivity)× Homogeneous Dummy 0.0796** 0.0181(0.0330) (0.0315)Homogeneous Dummy -0.221 -0.0308(0.152) (0.234)ln(Capital/Labour Ratio) -0.150*** -0.229***(0.0277) (0.0302)ln(Sale) 0.0632* 0.254***(0.0364) (0.0494)Constant 0.575 0.242(0.396) (0.401)Ownership FE X XProvince-Industry-Year FE X XCluster By Industry X XObservations 80,147 80,147R-squared 0.419 0.420Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “ChineseCustoms Export and Import Database” (2000-2006).Notes: This table shows the impact of the good differentiation on the correlation betweenexport/domestic sales ratio and productivity.1. The productivity in the column 1 is value-added per worker. The productivity in column 2 isTFP.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.331.4. Marketing Cost ModelHere, µmjkt is the country-province-industry-year dummy. We also controlthe number of destinations at the firm level. The results are given in Table1.8 and are similar to those for the firm level Regression 1.1. If a firm’slabour productivity (value-added per worker) increases by 10%, then itsexport/domestic sales ratio decreases by 0.86%. If a firm’s TFP increasesby 10%, then its export intensity decreases by 2.57%. When we drop firmsthat are engaged in the processing trade, the result remains robust for TFPbut not for labour productivity. However, in Regression 1.2 we only controlthe export destination fixed effect. Next, we run regressions for each exportdestination. We rank the destinations according to the number of Chineseexporting firms that sell products in that country. We choose the top tenexport destinations: the US, Hong Kong, Japan, South Korea, Germany,the UK, Canada, Australia, Taiwan, and Italy. The results are given inTable 1.9. We find that the negative correlation remains robust for the topten most significant destinations.1.4 Marketing Cost ModelIn the last section, we showed some empirical patterns regarding the ex-porting behaviour of Chinese firms. In this section, we use the extendedmarketing cost model presented in Arkolakis (2010) to rationalize our find-ings. The fixed cost of exporting discussed in Melitz (2003) can explain afirm’s export propensity, but it cannot explain a firm’s low export value.Arkolakis introduces variable marketing costs into Melitz’s model to explainthe unexpected exporting behaviour of French firms. In Arkolakis’s model,firms have to pay marketing costs to reach consumers. The marginal mar-keting cost of a firm is an increasing function of the number of consumers itintends to reach. The number of consumers that a firm attempts to reachis an endogenous decision. A firm’s productivity thus has two effects on itssales: an intensive margin effect and an extensive margin effect. On the onehand, when a firm’s productivity increases, its marginal cost of productiondecreases. Thus, a firm can offer a competitive price to existing consumers.If the demand from existing consumers increases, then the firm’s sales in-crease. We call this the intensive margin effect. On the other hand, whena firm’s productivity increases, it can afford to spend more on marketingand reach more consumers. By attracting new consumers, the firm’s salesincrease as well. We call this the extensive margin effect.In Melitz’s model, a firm’s productivity only has an intensive margineffect and this effect is proportional across markets. Thus, there is no cor-341.4. Marketing Cost ModelTable 1.8: Export/Domestic Sales Ratio and Productivity: Firm-Destination LevelDependent Variable: ln(Export/Domestic Sales)(1) (2) (3) (4)ln(Productivity) -0.0823*** 0.0273 -0.229*** -0.115**(0.0311) (0.0449) (0.0355) (0.0452)ln(Capital/Labour Ratio) -0.148*** -0.202*** -0.184*** -0.205***(0.0231) (0.0307) (0.0226) (0.0319)ln(Sale) -0.339*** -0.527*** -0.189*** -0.430***(0.0385) (0.0365) (0.0517) (0.0478)ln(Destination Number) -0.0764 -0.0286 -0.0779 -0.0330(0.0496) (0.0545) (0.0495) (0.0541)Constant 1.043*** 2.697*** 0.850** 2.587***(0.394) (0.386) (0.400) (0.385)Ownership FE X X X XCountry-Province-Industry-Year FE X X X XCluster By Industry X X X XExclude Processing Trade X XObservations 1,098,287 560,850 1,098,287 560,850R-squared 0.585 0.649 0.585 0.650Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “ChineseCustoms Export and Import Database” (2000-2006).Notes: This table shows the correlation between export/domestic sales ratio and productivityon the firm-destination level.1. The productivity in the first two columns is value-added per worker. The productivity inlast two columns is TFP.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.351.4. Marketing Cost ModelTable 1.9: Export/Domestic Sales Ratio and Productivity For The Top TenDestinationsDependent Variable: Ln(Export/Domestic Sales)(1) (2) (3) (4)USA -0.262*** -0.0705 -0.372*** -0.242***(0.0369) (0.0595) (0.0406) (0.0739)HKG -0.222*** -0.058 -0.247*** -0.0789(0.0367) (0.0539) (0.0396) (0.0598)JPN -0.224*** 0.0111 -0.305*** -0.130**(0.0357) (0.0433) (0.0474) (0.0615)KOR -0.0736 0.0449 -0.196*** -0.0516(0.0474) (0.0574) (0.0559) (0.062)GER -0.0934** -0.00168 -0.202*** -0.127(0.038) (0.0559) (0.0548) (0.0788)GBR -0.0648 0.0861 -0.177*** -0.0368(0.0419) (0.0673) (0.0511) (0.0854)AUS -0.0875** -0.048 -0.245*** -0.179**(0.0439) (0.0723) (0.0549) (0.0853)CAN -0.146*** -0.0369 -0.232*** -0.147(0.0463) (0.0893) (0.0574) (0.109)TWN -0.204*** -0.0917 -0.279*** -0.188**(0.0493) (0.0756) (0.0575) (0.0815)ITA -0.0919** 0.00554 -0.218*** -0.0958(0.0445) (0.0587) (0.054) (0.0746)Ownership FE X X X XProvince-Industry-Year FE X X X XCluster By Industry X X X XExclude Processing Trade X XData Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “ChineseCustoms Export and Import Database” (2000-2006).Notes: This table shows the correlation between export/domestic sales ratio and productivityfor the top ten destinations.1. The productivity in the first two columns is value-added per worker. The productivity in lasttwo columns is TFP.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.3. USA United States; HKG Hong Kong; JPN Japan; KOR South Korea; GER Germany; GBR:United Kingdom; AUS Australia; CAN Canada; TWN Taiwan; ITA Italy.361.4. Marketing Cost Modelrelation between a firm’s productivity and its export/domestic sales ratio.In Arkolakis’s model, the intensive margin effect is also proportional acrossmarkets. However, the extensive margin effect is nonproportional acrossmarkets. Arkolakis assumes that the marketing cost function is the sameacross markets. He also assumes that exporting firms have to afford ice-berg costs, and thus, that the threshold of productivity for selling in foreignmarkets is higher than that for selling in the domestic market. Consideringthat the extensive margin effect is a concave function of productivity, themarginal extensive margin effect is always larger in foreign markets than inthe domestic market for a given level of productivity. Thus, Arkolakis usesthis mechanism to explain the positive correlation between French firms’productivity and their export/domestic sales ratio, but the result Arkolakisobtains is contrary to our findings about Chinese firms.Thus, we extend Arkolakis’ model by relaxing his assumption about themarketing cost function. We allow the elasticities of marketing costs acrossmarkets to be different. We think this is reasonable since the competitiveenvironments must be different between foreign markets and the domesticmarket. If the elasticity of marketing costs is small in a market, then allfirms in that market will tend to choose a similar number of consumers, re-gardless of their productivity, which means that the extensive margin effectwill be similar across firms. Melitz’s model is a special case of Arkolakis’smodel when the elasticity is 0. In this case, the extensive margin effect is ex-actly the same across all exporting firms. If the elasticity of marketing costsis high in a market, then low-productivity firms will choose to reach a smallnumber of consumers, but high-productivity firms will be able to afford themarketing costs necessary to reach a large number of consumers. The exten-sive margin effect will thus be much larger for high-productivity firms. Whenthe elasticity in the domestic market is higher than that in foreign markets,then the extensive margin effect difference between high-productivity andlow-productivity firms will be larger in the domestic market. Thus, there is anegative correlation between firms’ productivity and their export/domesticsales ratio.1.4.1 Consumer DemandSuppose a firm from country i only exports one product to country j. Thepopulation of country j is Lj and the utility function of a representativeconsumer is the CES function. Each firm is small and cannot affect the371.4. Marketing Cost Modelprice index. Then, the demand function of a firm isqij = Ljp−σijP 1−σjyj , σ > 1 (1.3)Here, pij is the price of this firm’s product. Pj is the price index in countryj. yj is the income in country j. The income yj has two sources: wage wjand the firm’s profit pij . Thus, yj = wj + pij .1.4.2 Marketing CostsWe assume that a firm must first advertise to consumers. This incurs a mar-keting cost. With probability n ∈ [0, 1], a consumer sees the advertisements.Only when consumers have seen the firm’s advertisements do they buy theproduct. Thus, nLj measures the real number of consumers in country j.The population number Lj is exogenous for a firm, but the firm can choosethe probability n of reaching consumers, in other words the number of con-sumers it wants to reach. We call n the advertisement intensity. Supposethe marketing costs of country i’s firm in country j are fij(n,Lj).fij(n,Lj) ={Lλj · 1−(1−n)1−κij1−κij , if κij ∈ [0, 1) ∪ (1,+∞)−Lλj · ln(1− n), if κij = 1(1.4)First, λ ∈ [0, 1] measures the market size effect. When λ = 1, eachadvertisement only reaches one consumer. Thus, the marketing cost perconsumer is constant. The total marketing cost fij increases with the marketsize Lj . When λ = 0, each advertisement reaches all consumers. Thus, thetotal marketing cost fij is uncorrelated with the market size (Lj). Whenthe market size Lj is larger, the marketing cost per consumer is lower.Second, the total marketing cost fij(n,Lj) is a convex function of n.κij measures the elasticity of the marketing cost with respect to n. If κij islarge, the total marketing cost fij(n,Lj) increases more quickly with respectto n. When κij = 0, a firm always chooses n = 0 or n = 1, which meansthat the firm either enters market j or it does not. This is Melitz’s model.1.4.3 The Firms’ ProblemIn order to produce q units of products, a firm has to hire qφ units of domesticlabour. φ is the firm’s productivity. Suppose the wage in country i is wi.381.4. Marketing Cost ModelThen, the production cost for a firm with productivity φ in country i toproduce q units of a product isCi(φ, q) =wiqφ(1.5)In addition, we assume that the iceberg cost between country i andcountry j is τij > 1 and τii = 1.Given the consumers’ demand (1.3), the firm’s marketing cost (1.4) andthe firm’s production cost (1.5), a firm’ profit, which is located in country iand sells products in country j, ispiij(pij , nij , φ) = nijLjyjp1−σijP 1−σj− nijLjyjp−σij τijwiP 1−σj φ− Lλj1− (1− nij)1−κij1− κij(1.6)Given productivity φ, firms choose the optimal price pij and advertisingintensity nij that maximize their profit (1.6). Then, we we havepij(φ) = σ˜τijwiφ, σ˜ =σσ − 1 (1.7)nij(φ) = max{1− (φ∗ijφ)σ−1κij , 0}, (φ∗ij)σ−1 =Lλ−1jyjσ(σ˜τijwi)1−σP 1−σj(1.8)φ∗ij is the threshold for exporting from country i to country j, which is notaffected by the elasticity of marketing costs κij . Suppose a firm from countryi has already entered market j. If κij is very large, then the value of nijis sensitive to this firm’s productivity φ. If κij is very small, then nij willalmost equal to 1, regardless of this firm’s productivity φ.1.4.4 Productivity DistributionSuppose the firms’ productivity distribution is a Pareto distribution, withthe probability distribution function (pdf), g(φ), and the cumulative distri-bution function (cdf), G(φ), as follows:gi(φ) = θbθiφθ+1, θ > σ − 1Gi(φ) = 1− bθiφθ, φ ∈ [bi,+∞)391.4. Marketing Cost Modelwhere bi is the lower bound of the firms’ productivity and θ is the Paretoindex. Thus, the conditional distribution of the productivity of firms fromcountry i exporting to country j isµij(φ) ={θ(φ∗ij)θφθ+1, if φ ≥ φ∗ij0, otherwise(1.9)1.4.5 Firms’ Sales and ProfitFrom (1.3), (1.7) and (1.8), we find that the sales of a firm with productivityφ, exporting from country i to country j, will berij(φ) = pij(φ)qij(φ) =σLλj ( φφ∗ij )σ−1[1− (φ∗ijφ )σ−1κij ], if φ ≥ φ∗ij0, if φ < φ∗ij(1.10)Integrating expression (1.10) across the pdf (1.9), we can obtain theaverage sales of firms exporting from country i to country jr¯ij = σLλj [11− 1/θ˜ −11− 1/(θ˜κ˜ij)] (1.11)Hereθ˜ =θσ − 1 , κ˜ij =κijκij − 1We can also prove that the Pareto distribution assumption implies thatmarketing costs are a constant share of a firm’s sales:m =θ − (σ − 1)θσProfits and wages can also be expressed as constant shares of income:pii = ηyi, wi = (1− η)yiwhere η = (σ − 1)/(θσ).401.4. Marketing Cost Model1.4.6 Firms’ Export/Domestic Sales RatioThe export/domestic sales ratio of a Chinese firm is defined as the ratio ofthe export value to domestic sales:Export/Domestic Sales Ratio(φ)=rcj(φ)rcc(φ)=σLλj (φφ∗cj)σ−1[1−(φ∗cjφ)σ−1κcj ]σLλc (φφ∗cc )σ−1[1−(φ∗ccφ)σ−1κcc ]if φ ≥ φ∗cj0, if φ < φ∗cj(1.12)Here, the subscript c refers to China. Thus, a firm’s export/domestic salesratio depends on the threshold of productivity for exporting to country j(φ∗cj) and that for selling in the domestic market (φ∗cc). It also depends onthe elasticity of marketing costs in country j (κcj) and the domestic market(κcc). The first terms of the numerator and denominator represent the in-tensive margin effect of productivity on the firm’s sales. If the productivityφ is higher, then a firm’s sales are larger. When a firm’s productivity in-creases, then the sales increase proportionally in the domestic market c andthe foreign market j. Thus, the intensive margin effect does not explain thecorrelation between a firm’s productivity and its export/domestic sales ra-tio. The second terms of the numerator and denominator are the extensivemargin effect of productivity on the firm’s sales. If the productivity φ ishigher, then a firm can reach more consumers. When the elasticity of mar-keting costs in the domestic market (κcc) is different from that in country j(κcj), then the extensive margin effect is nonproportional across markets.To simplify, we assume that, for Chinese firms, the threshold of pro-ductivity for selling in the Chinese market is lower than that for selling inforeign markets, which means that φ∗cc < φ∗cj . This assumption is reasonablesince most Chinese firms first sell in the domestic market and then enterforeign markets. Thus we have the following Proportion 1.Proposition 1:1 When κcj ≥ κcc, ∂ln(Export/Domestic Sales Ratio)∂lnφ > 0;2 When κcj < κcc, there is a φ∗(> φ∗cj) satisfies1κcj(φ∗cjφ∗ )σ−1κcj = 1κcc (φ∗ccφ∗ )σ−1κcc ,– when φ ∈ (φ∗cj , φ∗], ∂ln(Export/Domestic Sales Ratio)∂lnφ ≥ 0;411.4. Marketing Cost Model– when φ ∈ (φ∗,+∞], ∂ln(Export/Domestic Sales Ratio)∂lnφ < 0.Thus, when κcj ≥ κcc, there is always a positive correlation between afirm’s productivity and its export/domestic sales ratio. Arkolakis (2010)confirms this case. When κcj < κcc, the correlation between a firm’s pro-ductivity and its export/domestic sales ratio is an inverted U-shaped curve,as shown in Figure 1.3. φ∗ is the turning point. Among Chinese firms, φ∗is very close to φ∗cj , thus the positive correlation part is not significant.The intuition is as in Figure 1.16. Suppose κcj is small and κcc is large.The extensive margin effect in foreign markets initially increases very quicklybut later it stays almost constant, while the extensive margin effect in thedomestic market increases much more slowly but keeps increasing. Thus, ifφ ∈ (φ∗cj , φ∗], the extensive margin effect in foreign markets is larger thanthat in the domestic market. If φ ∈ (φ∗,+∞], then the extensive margineffect in foreign markets is smaller than that in the domestic market.1.4.7 EstimationMethodologySince the firms’ productivity follows a Pareto distribution, this implies thatφφ∗ij= (1−Prij)−1/θ. Here Prij denotes the productivity percentile of a firmexporting from country i to country j relative to other firms exporting fromcountry i to country j. Substituting this expression into (1.10) and (1.12),we haverij(Prij) = σLλj (1− Prij)−1θ˜ [1− (1− Prij)1θ˜κij ] (1.13)Export/Domestic Sales Ratio(φ) =rcj(Prcj)rcc(Prcc)=Lλj (1− Prcj)−1θ˜ [1− (1− Prcj)1θ˜κcj ]Lλc (1− Prcc)−1θ˜ [1− (1− Prcc)1θ˜κcc ](1.14)In order to avoid population Lj , we normalize (1.14) using the averagesales in the domestic market r¯cc (1.11) and the average sales in country j,421.4. Marketing Cost ModelFigure 1.16: The Productivity and The Extensive Margin Effect4 6 8 10 12 14 16 18 2000.20.40.60.811.21.41.61.8ProductivityExtensive Marginal Effect DomesticForeignForeign/Domestic RatioNotes: This figure shows the simulation results under some given parameters. Theextensive margin effect is 1 − (φ∗ijφ)σ−1κij as shown in (1.12). In this simulation, weassume φ∗cc = 3, φ∗cj = 4,σ−1κcc= 1 and σ−1κcj= 4.431.4. Marketing Cost Modelr¯cj (1.11). Then we haveNormalized Export/Domestic Sales Ratio(φ)=rcj/r¯cjrcc/r¯cc=11−1/θ˜ −11−1/(θ˜κ˜cc)11−1/θ˜ −11−1/(θ˜κ˜cj)(1− Prcj)−1θ˜ [1− (1− Prcj)1θ˜κcj ](1− Prcc)−1θ˜ [1− (1− Prcc)1θ˜κcc ](1.15)Based on (1.15), we use the non-linear least squares method to estimateθ˜, κcc for the domestic market and κcj for each country j. First, we rankthe firms according to their productivity in each market and each industry.Then, we assign each firm to their productivity percentile Prcj in marketj. Finally, using the firms’ normalized export/domestic sales ratio data andproductivity percentile data, we choose θ˜, κcc and κcj to minimize the squareof the difference between the left side and the right side of equation (1.15).ResultsWe only estimate the elasticities for China and the 10 export destinations.We use the TFP to measure the firms’ productivity and drop all firms thatare engaged in the processing trade. The results are shown in Table 1.10.We find that the κ in the domestic market is larger than that in foreignmarkets. Thus the extensive margin effect is larger in the domestic market.After obtaining these parameters, we use firm productivity percentile dataand (1.15) to calculate the firms’ simulated export/domestic sales ratio.Then, we calculate the average export intensity for each percentile of firms’productivity. The results are presented in Figure 1.17. We find that the realdata and the model simulation results are very similar.Given the pattern in the data, the estimated elasticity in the domesticmarket must be higher. As an over-identifying test, we calculate the averagesales for each percentile of firms’ productivity based on the estimated pa-rameters. If the model is correct, then the simulated sales should be similarto the actual sales. The results are presented in Figure 1.18. It seems thatthe model underestimates the level of the firms’ sales, but the results arestill similar.441.4. Marketing Cost ModelFigure 1.17: Export Intensity and Productivity0.2.4.6.8Average export intensity0 20 40 60 80 100percentile of ln(productivity)data modelUSA0.2.4.6.8Average export intensity0 20 40 60 80 100percentile of ln(productivity)data modelHKG.2.4.6.8Average export intensity0 20 40 60 80 100percentile of ln(productivity)data modelJPN0.2.4.6.8Average export intensity0 20 40 60 80 100percentile of ln(productivity)data modelKOR.2.4.6.81Average export intensity0 20 40 60 80 100percentile of ln(productivity)data modelGER.2.4.6.81Average export intensity0 20 40 60 80 100percentile of ln(productivity)data modelGBR.2.4.6.81Average export intensity0 20 40 60 80 100percentile of ln(productivity)data modelAUS.2.4.6.81Average export intensity0 20 40 60 80 100percentile of ln(productivity)data modelCAN.2.4.6.81Average export intensity0 20 40 60 80 100percentile of ln(productivity)data modelTWN.2.4.6.81Average export intensity0 20 40 60 80 100percentile of ln(productivity)data modelITANotes: This figure shows the real export intensity from the data and the simulatedexport intensity from the model.451.4. Marketing Cost ModelFigure 1.18: Sales and Productivity0123Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelUSA0.511.522.5Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelHKG0.511.522.5Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelJPN0123Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelKOR0123Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelGER0.511.522.5Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelGBR0.511.522.5Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelAUS0.511.522.5Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelCAN0.511.522.5Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelTWN0.511.522.5Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelITA0246810Average Sale0 20 40 60 80 100percentile of ln(productivity)data modelCHNNotes: This figure shows the real sales from the data and the simulated sales from themodel.461.5. Additional Robustness ChecksTable 1.10: The Parameters for The Top Ten DestinationsCoefficient 95% Confidence Intervalθ˜ 2.000 (2.000,2.000)κCHN 2.713 (2.631,2.864)κUSA 0.780 (0.775,0.788)κHKG 2.423 (2.344,2.594)κJPN 0.522 (0.519,0.527)κKOR 1.281 (1.264,1.312)κGER 0.492 (0.490,0.493)κGBR 0.406 (0.405,0.408)κAUS 0.443 (0.441,0.446)κCAN 0.364 (0.363,0.365)κTWN 1.067 (1.055,1.099)κITA 0.326 (0.325,0.327)Notes: The table shows the estimated parameters. We usebootstrap to calculate the confidence intervals. We redraw thesample 199 times.1.5 Additional Robustness Checks1.5.1 Advertisement ExpenditureThe marketing strategy is critical in some industries but not in others. Thus,if the marketing cost model can indeed explain the negative correlation be-tween the firms’ productivity and their export intensity (or export/domesticsales ratio), then the negative pattern might be different across industrieswith different levels of advertising intensity. There might be a steeper neg-ative correlation in those industries for which marketing strategy is impor-tant. We use the advertisement/sales ratio to measure the importance ofmarketing in an industry. The measurement is as follows:AdvertismentSales=Average Advertisment Spending in this InudstryAverage Sales in this Inudstry(1.16)Then, we include the interaction term between the advertisement/sales ratioand the firms’ productivity in Regressions 1.1 and 1.2. The results can beseen in Tables 1.11 and 1.12. We find that the interaction term is significantand negative. Thus, in those industries with high spending on advertising,the negative correlation is much steeper and a firm’s sales expand much471.5. Additional Robustness Checksmore quickly in the domestic market.Table 1.11: Export/Domestic Sales Ratio, Productivity and AdvertisingSpending: Firm LevelDependent Variable: ln(Export/Domestic Sales)(1) (2) (3) (4)ln(Productivity) -0.141*** -0.0438 -0.240*** -0.207***(0.0181) (0.0312) (0.0224) (0.0325)ln(Productivity)× Advertisement/Sales Ratio -7.037*** -10.72*** -11.28*** -12.27***(1.759) (3.353) (2.378) (4.092)ln(Capital/Labor Ratio) -0.170*** -0.229*** -0.222*** -0.260***(0.0154) (0.0209) (0.0157) (0.0229)ln(Sale) -0.0231 -0.208*** 0.130*** -0.0483(0.0268) (0.0277) (0.0366) (0.0377)Constant 0.789*** 2.645*** 0.588** 2.416***(0.272) (0.297) (0.276) (0.299)Ownership FE X X X XProvince-Industry-Year FE X X X XCluster By Industry X X X XExclude Processing Trade X XObservations 275,872 69,691 275,872 69,691R-squared 0.369 0.473 0.370 0.474Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “ChineseCustoms Export and Import Database” (2000-2006).Notes: This table shows the impact of advertising spending on the correlation between export/domesticsales ratio and productivity on the firm level.1. The productivity in the first two columns is value-added per worker. The productivity in last twocolumns is TFP.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.1.5.2 Market ShareIn previous chapters, we have showed that the elasticity of marketing costs ishigher in the domestic market, and then used the heterogeneity of elasticityacross markets to explain our empirical findings. In this section, we providesome empirical evidences regarding marketing costs and investigate whetherit can indeed explain our findings. It is difficult to measure the elasticity ofmarketing costs directly. In this chapter, we use the share of Chinese firmsin a market as a proxy.The purpose of a firm’s marketing strategy is to distinguish its productsfrom other products and then convince consumers to buy them. Thus, itis more difficult for a firm to achieve this objective when it is surrounded481.5. Additional Robustness ChecksTable 1.12: Export/Domestic Sales Ratio, Productivity and AdvertisingSpending: Firm-Destination LevelDependent Variable: ln(Export/Domestic Sales)(1) (2) (3) (4)ln(Productivity) -0.0688** 0.0420 -0.212*** -0.101**(0.0313) (0.0462) (0.0352) (0.0458)ln(Productivity)× Advertisement/Sales Ratio -7.241*** -9.202*** -8.402*** -8.036***(1.986) (3.141) (2.005) (2.828)ln(Capital/Labor Ratio) -0.149*** -0.202*** -0.184*** -0.205***(0.0230) (0.0307) (0.0225) (0.0319)ln(Sale) -0.338*** -0.526*** -0.188*** -0.429***(0.0384) (0.0366) (0.0517) (0.0479)ln(Destination Number) -0.0773 -0.0297 -0.0792 -0.0341(0.0496) (0.0544) (0.0495) (0.0541)Constant 1.039*** 2.701*** 0.849** 2.597***(0.390) (0.385) (0.394) (0.385)Ownership FE X X X XCountry-Province-Industry-Year FE X X X XCluster By Industry X X X XExclude Processing Trade X XObservations 1,098,287 560,850 1,098,287 560,850R-squared 0.585 0.649 0.586 0.650Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “ChineseCustoms Export and Import Database” (2000-2006).Notes: This table shows the impact of advertising spending on the correlation between export/domesticsales ratio and productivity on the firm-destination level.1. The productivity in the first two columns is value-added per worker. The productivity in last twocolumns is TFP.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.491.5. Additional Robustness Checksby similar firms. This means that the elasticity of marketing costs shouldbe higher in such markets. Consumers might consider the products of allChinese firms to be similar. If the market share of Chinese firms in a marketis high, this implies that Chinese products are welcome in this market asa whole. However, for an individual Chinese firm, the competition in thismarket is tough. Since the market share of Chinese firms is already highin this market, it is difficult for an individual Chinese firm to attract anew consumer, no matter from other Chinese firms or foreign competitors.Thus, the higher is the market share of Chinese firms, the tougher is thecompetition and the higher is the elasticity of marketing costs in that market.In this chapter, we don’t explain why the market shares of Chinese firms varyacross markets. Instead, we argue that the market share actually proxieslocal market competition for an individual Chinese firm. In future work, wecould make the market share endogenous.The market share of Chinese firms is defined asMarket Share =Import From ChinaApparent ConsumptionThe import data are taken from the “UN Comtrade Database,” which con-tains the annual trade data between each pair of countries at the SITCRevision 3 level. The consumption data come from the “Industrial Demand-Supply Balance Database (IDSB),” which is collected by UNIDO. Thisdatabase contains datasets based on the four-digit level of ISIC Revision 3for each country and each year. The apparent consumption in this databaseis calculated as follows:Apparent Consumption= Domestic Output+ Total Imports− Total ExportsSince there are some missing values for domestic output, total imports ortotal exports, only half of the observations can be used. We include theinteraction term between the firms’ productivity and their relative marketshare in Regression 1.2.The regression can be written asln(Export V aluem/Domestic Sales Ratioijkt)= γ0 + γ1ln(Pijkt) + γ2ln(Pijkt)×Relative Market Sharekm+ other controls+ µmjkt + mijkt(1.17)501.5.AdditionalRobustnessChecksTable 1.13: Export/Domestic Sales Ratio, Productivity and Market Share: Firm-Destination LevelDependent Variable: ln(Export/Domestic Sales)(1) (2) (3) (4) (5) (6) (7) (8)ln(Productivity) -0.114*** 0.0370 -0.256*** -0.0960*(0.0374) (0.0537) (0.0422) (0.0570)ln(Productivity)× Relative Market Share 0.0603 -0.285 0.0846 -0.0579 0.269* -0.0366 0.239** -0.126(0.180) (0.380) (0.127) (0.212) (0.146) (0.274) (0.116) (0.180)ln(Capital/Labor Ratio) -0.160*** -0.199*** -0.152*** -0.183*** -0.201*** -0.204*** -0.195*** -0.197***(0.0223) (0.0279) (0.0227) (0.0281) (0.0222) (0.0293) (0.0223) (0.0291)ln(Sale) -0.337*** -0.519*** -0.329*** -0.505*** -0.194*** -0.438*** -0.183*** -0.407***(0.0337) (0.0390) (0.0335) (0.0393) (0.0470) (0.0501) (0.0465) (0.0505)ln(Destination Number) -0.0345 -0.00139 -0.0429 -0.0135 -0.0340 -0.00517 -0.0457 -0.0221(0.0537) (0.0590) (0.0537) (0.0587) (0.0538) (0.0590) (0.0537) (0.0589)Constant 1.227*** 2.678*** 1.156*** 2.566*** 1.043*** 2.587*** 1.008*** 2.800***(0.332) (0.381) (0.301) (0.321) (0.336) (0.378) (0.219) (0.241)Country-Province-Industry-Year FE X X X X X X X XOwnership FE X X X X X X X XCluster By Industry X X X X X X X XExcluding Processing Trade X X X XIndustry Dummy × ln(Productivity) X X X XObservations 391,072 212,477 391,072 212,477 391,072 212,477 391,072 212,477R-squared 0.514 0.586 0.518 0.591 0.515 0.586 0.520 0.592Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “Chinese Customs Export and Import Database”(2000-2006).Notes: This table shows the impact of market share on the correlation between export/domestic sales ratio and productivity on the firm-destination level.1. The productivity in the first four columns is value-added per worker. The productivity in last four columns is TFP.2. We control the interaction term between the firms’ productivity and the industry dummy in column 3, 4, 7 and 8. By this way, we canexclude the fixed effect of industry.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.511.5. Additional Robustness ChecksRelative market share is the relative market share in market m withrespect to the Chinese market. When the relative market share equals 1,the competition in the destination is almost the same as that of Chinesemarket. If Arkolakis (2010) can explain our empirical findings, then γ2should be positive. The results are shown in Table 1.13. We find thatthe relative market share cannot explain the negative pattern when we usevalue-added per worker to measure the firms’ productivity, but when we useTFP, the effect of the relative market share is significant. When the relativemarket share equals 1, it almost explains all of the negative correlation.After dropping all firms that are engaged in the processing trade, the effectof the market share is no longer significant.1.5.3 Other ExplanationsWhen we use market share of the Chinese firms as a proxy for the marketingcost, it might become mixed up with the effect of the demand side. Actually,the mechanism in Melitz and Ottaviano (2008) provides a potential expla-nation for the demand side. Melitz and Ottaviano (2008) endogenize differ-ences in the “toughness” of competition across markets. Thus, when firms’productivity increases, their sales expand at different rates across markets,depending on market size. The difference between Melitz and Ottaviano(2008) and Arkolakis (2010) is that the former focus on the demand sidewhile the latter emphasizes the supply side. In this section, we investigatewhich one provides a better way to rationalize our empirical findings.First, we examine whether market size can explain our findings. We useconsumption to measure market size. Then, we include the interaction termbetween firms’ productivity and relative market size in Regression 1.2.The resulting regression can be written as follows:ln(Export V aluem/Domestic Sales Ratioijkt)= γ0 + γ1ln(Pijkt) + γ2ln(Pijkt)×Relative Market Sizekm+ other controls+ µmjkt + mijkt(1.18)Relative market size is the relative market size in market m with respect tothe Chinese market. When the relative market size equals 1, the market sizeof the destination is the same as that of Chinese market. If the mechanism inMelitz and Ottaviano (2008) explains our empirical findings, then γ2 shouldbe positive. The results are given in Table 1.14. We find that relative marketsize does not explain our findings. When we use value-added per worker tomeasure the firms’ productivity, the coefficient is not significant. When we521.5. Additional Robustness Checksuse TFP to measure the firms’ productivity, the coefficient is significant butthe effect of market size is too small. When the relative market size equals1, it only explains 6% of the negative correlation.Another difference between Melitz and Ottaviano (2008) and Arkolakis(2010) lies in their predictions about firms’ pricing strategies. Melitz andOttaviano (2008) assume a quadratic utility function and, thus, that thefirms’ markups differ across firms with different levels of productivity. InArkolakis (2010), since the utility function is in the CES form, the firms’markups are the same across all levels of productivity. Thus, Melitz andOttaviano (2008) predict that both the price ratio and the quantity ratiobetween two export destinations should be correlated with the firms’ pro-ductivity, while Arkolakis (2010) predicts that only the quantity ratio shouldbe correlated with the firms’ productivity.Since we have no product price and quantity data for the domestic mar-ket, we use Hong Kong as a benchmark market. From Table 1.10, we cansee that the elasticity of marketing costs in Hong Kong is very similar tothat in China. From Table 1.15, we can see that the firms’ ratio of exportsto country j divided by exports to Hong Kong is negatively correlated withtheir productivity. Thus, Hong Kong is indeed a good benchmark.We define the price (quantity) ratio at the HS8-destination level. Thedefinition is as follows:Price (Quantity) Ratio =Price (Quantity) to Country jPrice (Quantity) to Hong KongIn Table 1.16, we examine the correlation between the firms’ export priceratio and their productivity and do not find a significant correlation. InTable 1.17, we examine the correlation between the firms’ export quantityratio and their productivity and do find a significantly negative correlation.Thus, Arkolakis’s model is indeed a better way to explain our findings.A second potential explanation for our empirical findings relates to thequality of the products. Manova and Zhang (2012) find that more successfulexporters use higher quality inputs to produce higher quality goods, andfirms vary the quality of their products across destinations by using inputsof different quality levels. Thus, a firm’s sales will vary across markets dueto the differing quality of its products. Manova and Zhang (2012) also thatargue the markups are heterogeneous across firms. Therefore, the priceratio between two export destinations should be correlated with a firm’sproductivity. However, we can find no such evidence in the Chinese data.531.5.AdditionalRobustnessChecksTable 1.14: Export/Domestic Sales Ratio, Productivity and Market Size: Firm-Destination LevelDependent Variable: ln(Export/Domestic Sales)(1) (2) (3) (4) (5) (6) (7) (8)ln(Productivity) -0.100*** 0.0215 -0.244*** -0.100**(0.0303) (0.0451) (0.0388) (0.0504)ln(Productivity)× Relative Market Size -0.00702* -0.00939** -0.00180 -0.00348 0.0157** 0.00176 0.0229*** 0.00639(0.00388) (0.00430) (0.00486) (0.00438) (0.00759) (0.00444) (0.00638) (0.00540)ln(Capital/Labor Ratio) -0.160*** -0.199*** -0.152*** -0.183*** -0.201*** -0.204*** -0.195*** -0.197***(0.0222) (0.0278) (0.0226) (0.0280) (0.0222) (0.0290) (0.0222) (0.0289)ln(Sale) -0.337*** -0.519*** -0.329*** -0.505*** -0.192*** -0.439*** -0.183*** -0.407***(0.0337) (0.0390) (0.0334) (0.0392) (0.0470) (0.0501) (0.0464) (0.0505)ln(Destination Number) -0.0346 -0.00168 -0.0431 -0.0132 -0.0341 -0.00499 -0.0463 -0.0219(0.0538) (0.0592) (0.0538) (0.0587) (0.0539) (0.0591) (0.0537) (0.0588)Constant 1.220*** 2.677*** 1.155*** 2.562*** 1.032*** 2.589*** 1.032*** 2.810***(0.331) (0.384) (0.302) (0.322) (0.342) (0.379) (0.217) (0.243)Country-Province-Industry-Year FE X X X X X X X XOwnership FE X X X X X X X XCluster By Industry X X X X X X X XExcluding Processing Trade X X X XIndustry Dummy × ln(Productivity) X X X XObservations 391,677 212,825 391,677 212,825 391,677 212,825 391,677 212,825R-squared 0.515 0.586 0.518 0.591 0.515 0.586 0.520 0.592Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “Chinese Customs Export and Import Database”(2000-2006).Notes: This table shows the impact of market size on the correlation between export/domestic sales ratio and productivity on the firm-destinationlevel.1. The productivity in the first four columns is value-added per worker. The productivity in last four columns is TFP.2. We control the interaction term between the firms’ productivity and the industry dummy in column 3, 4, 7 and 8. By this way, we canexclude the fixed effect of industry.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.541.5.AdditionalRobustnessChecksTable 1.15: Sales Ratio and Productivity (Benchmark: HKG)Dependent Variable: ln(Sales Ratio)(1) (2) (3) (4) (5) (6) (7) (8)ln(Productivity) -0.0750* -0.0820** -0.0463 -0.0934** -0.0641** -0.0656** -0.0519** -0.0837**(0.0383) (0.0328) (0.0307) (0.0475) (0.0280) (0.0258) (0.0238) (0.0338)ln(Capital/Labor Ratio) -0.0825*** -0.0668* -0.0886*** -0.0853**(0.0263) (0.0348) (0.0261) (0.0335)Constant -0.183 -1.932 -1.756 -4.027 0.00527 -1.741 -1.501 -3.764(0.183) (1.377) (1.390) (2.612) (0.210) (1.350) (1.379) (2.601)Country FE X XProduct(HS02) FE X X X X X XCountry-Province-Industry-Year FE X X X X X XOwnership fixed effect X X X X X XExclude Processing Trade X XCluster By Industry X X X X X X X XObservations 441,946 441,946 441,946 110,086 441,946 441,946 441,946 110,086R-squared 0.036 0.360 0.360 0.516 0.036 0.360 0.360 0.516Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “Chinese Customs Export and Import Database”(2000-2006).Notes: This table shows the correlation between sales ratio and productivity on the firm-destination level. We calculate the export value on HS8level for the same firm.1. The productivity in the first four columns is value-added per worker. The productivity in last four columns is TFP.2. The sales ratio is Export V alue to Country jExport V alue to Hong Kongat firm-product level. Here other countries (regions) include US, Japan, South Korea,Germany, UK, Canada, Italy, Australia and Taiwan. These countries (regions) are the top 10 destinations of Chinese exporting firms.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.551.5. Additional Robustness ChecksTable 1.16: Price Ratio and Productivity (Benchmark: HKG)Dependent Variable:ln(Price Ratio)(1) (2) (3) (4)ln(Productivity) -0.0170*** -0.00974 -0.00316 0.00600(0.00514) (0.00955) (0.00464) (0.00842)Constant 0.109*** 0.714** 0.0651* 0.635**(0.0264) (0.288) (0.0360) (0.290)Country FE X XCountry-Province-Industry-Year FE X XOwnership FE X XProduct(HS02) FE X XCluster By Industry X X X XObservations 109,743 109,743 109,743 109,743R-squared 0.001 0.431 0.001 0.431Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “ChineseCustoms Export and Import Database” (2000-2006).Notes: This table shows the correlation between price ratio and productivity on the firm-destination level. We calculate the export prices on HS08 level for the same firm.1. The productivity in the first four columns is value-added per worker. The productivity inlast four columns is TFP.2. The price ratio is Export Price to Country jExport Price to Hong Kongat firm-product level. Here other countries(regions) include US, Japan, South Korea, Germany, UK, Canada, Italy, Australia and Taiwan.These countries (regions) are the top 10 destinations of Chinese exporting firms.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.A third possible explanation relies on the CES demand function withmarket power. If a firm is in a market with few competitors, its pass-through of costs to price will be less, because it will be concerned with itsmarket share. Thus, if the number of firms is heterogeneous across markets,then a firm’s productivity will be correlated with its export intensity. How-ever, this mechanism also implies a correlation between the price ratio andproductivity.A fourth possible explanation is heterogeneous ranges of products acrossmarkets. Both the range of products and sales per product would affect afirm’s total sales. In this chapter, we mainly consider the impact of produc-tivity and market competition on sales per product. If a high productivefirm has more kinds of products and a higher fraction of these products isonly sold in the domestic market, then we also might observe a negativecorrelation between a firm’s productivity and its export intensity. Due tothe data limitation, we do not have firms’ each product sales in the domes-tic market. As we have done before, we use Hong Kong as a benchmarkmarket. We define the product number ratio between other countries and561.5. Additional Robustness ChecksTable 1.17: Quantity Ratio and Productivity (Benchmark: HKG)Dependent Variable: ln(Quantity Ratio)(1) (2) (3) (4)ln(Productivity) -0.154*** -0.118** -0.152*** -0.103***(0.0374) (0.0502) (0.0319) (0.0386)Constant 0.429** -4.822* 1.210 -4.586*(0.177) (2.680) (0.909) (2.666)Country FE X XCountry-Province-Industry-Year FE X XOwnership FE X XProduct(HS02) FE X XCluster By Industry X X X XObservations 109,743 109,743 109,743 109,743R-squared 0.026 0.515 0.056 0.515Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “ChineseCustoms Export and Import Database” (2000-2006).Notes: This table shows the correlation between quantity ratio and productivity on the firm-destination level. We calculate the export quantity on HS08 level for the same firm.1. The productivity in the first four columns is value-added per worker. The productivity in lastfour columns is TFP.2. The quantity ratio is Export Quantity to Country jExport Quantity to Hong Kongat firm-product level. Here othercountries include US, Japan, South Korea, Germany, UK, Canada, Italy, Australia and Taiwan.These are the top 10 destinations of Chinese exporting firms.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.571.5. Additional Robustness ChecksTable 1.18: Product Number Ratio and Productivity (Benchmark: HKG)Dependent Variable: ln(Number Ratio)(1) (2) (3) (4)productivity1 -0.0174** -0.0265* -0.0332*** -0.0340**(0.00868) (0.0152) (0.00963) (0.0139)Constant -0.0189 0.0639 0.150** 0.209**(0.0396) (0.0637) (0.0624) (0.0967)Country FE X XCountry-Province-Industry-Year FE X XOwnership FE X XProduct(HS02) FE X XCluster By Industry X X X XObservations 93,212 93,212 93,212 93,212R-squared 0.016 0.538 0.018 0.539Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the“Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the correlation between product number ratio and productivity onthe firm-destination level. We identify the product on HS08 level.1. The productivity in the first four columns is value-added per worker. The productivity inlast four columns is TFP.2. The product number ratio is Product Number to Country jProduct Number to Hong Kongat firm level. Here othercountries include US, Japan, South Korea, Germany, UK, Canada, Italy, Australia and Taiwan.These are the top 10 destinations of Chinese exporting firms.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.581.5. Additional Robustness ChecksTable 1.19: Value Ratio and Productivity (Benchmark: HKG)Dependent Variable: ln(Quantity Ratio)(1) (2) (3) (4)ln(Productivity) -0.170*** -0.127*** -0.151*** -0.0954***(0.0360) (0.0473) (0.0365) (0.0356)Constant 0.530*** -4.121 0.956*** -3.971(0.172) (2.624) (0.276) (2.609)Country FE X XCountry-Province-Industry-Year FE X XOwnership FE X XProduct(HS02) FE X XCluster By Industry X X X XObservations 110,086 110,086 110,086 110,086R-squared 0.028 0.516 0.028 0.516Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “ChineseCustoms Export and Import Database” (2000-2006).Notes: This table shows the correlation between value ratio and productivity on the firm-destination level. We calculate the export value on HS08 level for the same firm.1. The productivity in the first four columns is value-added per worker. The productivity inlast four columns is TFP.2. The value ratio is Export Quantity to Country jExport Quantity to Hong Kongat firm-product level. Here other coun-tries include US, Japan, South Korea, Germany, UK, Canada, Italy, Australia and Taiwan.These are the top 10 destinations of Chinese exporting firms.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.591.6. ConclusionHong Kong at the firm level. The definition is as follows:Product Number Ratio =Product Number to Country jProduct Number to Hong KongIn Table 1.18, we examine the correlation between the firms’ product numberratio and their productivity and find a significant negative correlation. Thus,the product ranges across market indeed can partially explain our empiricalfinding. In Table 1.19, we examine the correlation between the firms’ exportvalue ratio and their productivity for the same product and the coefficientis much larger than that of Table 1.18. When a firm’s productivity is 10%higher, the product number ratio is 0.27% − 0.34% lower and the exportvalue ratio per product is 0.95% − 1.27% lower. Thus, the product rangeaccounts for 17% − 26% and sales per product accounts for 74% − 83% ofthe negative correlation between a firm’s productivity and export intensity.1.6 ConclusionIn this chapter, we find a new empirical pattern between Chinese firms’productivity and their export intensity. Although high-productivity firmsexport more to foreign markets, most of their outputs are still sold in thedomestic market. Among exporting firms, there is a significantly negativecorrelation between the firms’ productivity and their export intensity. Thus,at least for Chinese firms, the firms’ total sales are enlarged by selling morein the domestic market but not by exporting more to foreign markets. Thisresult is different from the results of studies using data from French firms(see Arkolakis, 2010; Eaton et al., 2011). Thus, the exporting behaviour offirms in developing countries is worth investigating more carefully.In Melitz’s model, the fixed cost of exporting can only explain a firm’sexport propensity. Arkolakis’s marketing cost model offers us a mechanismfor rationalizing the correlation between a firm’s productivity and its exportintensity. Yet, more work is needed to investigate which factors determinelocal marketing costs. In this chapter, we use market competition to explainthe marketing cost heterogeneity across markets. We assume that marketcompetition is exogenous, which determined the elasticity of marketing cost-s. However, with more firms entering the market, market competition mightalso change. In this chapter, we do not consider this dynamic evolution. Infuture work, we could make marketing competition endogenous.60Chapter 2Financial Development andExchange Rate Pass-Throughin Processing Trade2.1 IntroductionWith the development of economic globalization, an increasing number ofproduction processes are transferred from developed countries to developingcountries for lower costs. By participating in the processing trade, develop-ing countries become one part of the global production chain. Especially,due to its success in the processing trade, China in particular has becomethe world’s largest trading country. The processing trade in China has ac-counted for about one third of its total trade and most of its trade surplus.12Since firms in the processing trade import a large amount of inputs, they aremore vulnerable to the exchange rate fluctuations than ordinary firms. Inaddition, the financial condition for processing firms is usually worse thanordinary firms, thus the exchange rate fluctuation brings a higher liquiditypressure on processing firms. Thus, it is worth to investigate how processingfirms to be exposed to exchange rate risk.Suppose firms are risk neutral, then firms that bear higher exchange ratepass-through take more exchange rate risks. In this chapter, we investigatethe exchange rate pass-through in import prices in Chinese processing trade.In addition, we examine the impacts of other factors on the exchange ratepass-through, such as the imported inputs ratio, the firm’s market share andlocal financial development.In the processing trade, local assembly firms first obtain production or-ders from multinational firms, then import raw materials from abroad, pro-12According to the China General Administration of Customs report, in 2014 the valueof the processing trade is 1,409 billion U.S. dollars, which accounts for 33% of China’stotal trade values. The trade surplus of the processing trade is 360 billion U.S. dollars,which accounts for 94% of China’s total trade surplus.612.1. Introductioncess them, and finally export the finished goods back to multinational firms.The raw materials can be either imported by multinational firms or by localassembly firms. When materials are imported by multinational firms, it iscalled “pure assembly” (PA). When materials are imported by local assem-bly firms, it is called “import and assembly” (IA). In “pure assembly”, localassembly firms are not responsible for importing materials and only earnthe processing service fees. Thus, the profit in this trade mode is low. In“import and assembly”, local assembly firms own the imported materialsand thus can claim more profit from the production process. Although theprofits in “import and assembly” are higher, a large number of assemblyfirms in China are still engaged in “pure assembly”. Existing studies eitheruse the property right theory or financial constraint of the firm to explainthis phenomenon. In this chapter, we present that the inability to bearexchange rate risks might be another potential explanation.When assembly firms are risk neutral, they would like the input costs(Chinese Yuan) to be fixed. This implies that assembly firms prefer the ex-change rate pass-through in import prices (Chinese Yuan) to be 0 when theexchange rate fluctuates. Most assembly firms in China are financial con-strained, the fluctuation of input costs would cause liquidity risks for them.Thus, some assembly firms, which cannot bear exchange rate risks, haveto choose “pure assembly”. In this case, multinational firms bear exchangerate risks. When assembly firms choose “import and assembly”, they canearn more profit. However, the disadvantage is to taking exchange rate risksby themselves.In order to get the exchange rate pass-through for both assembly firmsand multinational firms, we need their import prices. It is usually difficultto directly observe the import prices of multinational firms and assemblyfirms at the same time. However, the special structure of the processingtrade in China provides us an opportunity to investigate the price level andexchange rate pass-through differences between them. As mentioned, in“pure assembly”, multinational firms import raw materials and in “importand assembly”, local assembly firms import raw materials. From the ChineseCustoms data we can distinguish these two trade modes and thus get theimport prices of both multinational firms and local assembly firms.First, we find that exchange rate pass-through differences depend on thefirm’s ownership. Chinese-owned assembly firms bear higher exchange ratepass-through than multinational firms. However joint-owned and foreign-owned assembly firms bear less exchange rate pass-through than multina-tional firms. Even if we exclude the effect of the intermediary companiesand control the quality of imported materials, the pattern remains robust.622.1. IntroductionWe believe that joint-owned and foreign-owned assembly firms are usuallyowned by large multinational firms and thus they have good knowledge ofthe international market. Chinese-owned assembly firms cannot efficient-ly use the international market and thus they bear greater exchange ratepass-through. In addition, we find that the exchange rate pass-through isgreater when firms import materials from developed countries. When firmsimport materials from developing countries, they are free of exchange ratepass-through. We posit that firms are in a weak bargaining position whenthey import materials from developed countries, and thus they have to payhigher prices and bear more exchange rate pass-through. When firms importmaterials from developing countries, the disadvantage disappears and thusthe exchange rate pass-through is not significant. Over time, the exchangerate pass-through for assembly firms in China becomes lower. This impliesthat assembly firms become more competitive than before when importingmaterials from the international market.Second, we find that assembly firms can bear higher exchange rate pass-through if they have higher market shares. Usually a higher market shareimplies a better bargaining position and thus these firms will take less ex-change rate pass-through. However, the result in our finding is contrary tothis prediction. Goldberg and Tille (2013) show a bargain model betweenimporters and exporters and argue that the higher bargaining power implieslower import price but more exchange rate risk. Our result verifies theirprediction. When the market share is higher, the import price is lower butthe exchange rate pass-through is higher. In addition, we find that the ratioof imported inputs also affects the exchange rate pass-through. Assemblyfirms use both domestic and imported materials as inputs. When the ratioof imported inputs is lower, assembly firms take higher exchange rate pass-through. This is because assembly firms are less sensitive to the exchangerate fluctuations when the ratio of imported inputs is low. Thus, these firmscan bear greater exchange rate pass-through.Finally, we find that high financial development is helpful for assem-bly firms to bear higher exchange rate pass-through. When firms are lo-cated in financially developed prefectures, they have two benefits: First,they have access to more financial tools to hedge the exchange rate fluc-tuations. Some papers (Do¨hring, 2008; Takatoshi et al., 2013) argue thatfinancial hedging is a substitute strategy with direct pass-through. Whenfirms can hedge the exchange rate risks, they can bear greater exchangerate pass-through. Second, the developed financial sector is also helpful fordecreasing the borrowing costs of firms. Thus, these firms have less finan-cial constraints. Strasser (2013) finds that financially constrained exporting632.2. Literature Reviewfirms pass-through exchange rate changes to prices at almost twice the rateof unconstrained exporting firms. Thus, unconstrained importing firms canbear greater exchange rate pass-through. These two benefits can explain thepositive correlation between local financial development and exchange ratepass-through.As far as we know, this is the first study to directly compare the ex-change rate pass-through differences between assembly firms in developingcountries and multinational firms. Another contribution of this chapter is itsinvestigation of how assembly firms react to the exchange rate fluctuationsfrom the perspective of local financial development. Unlike the previousliterature that uses aggregate data or surveys, this chapter uses detailedfirm-product level trade data to show that local financial development ishelpful for assembly firms to bear greater exchange rate pass-through. Forthose firms in developing countries, hedging the exchange rate fluctuationsis critical for their profits. Thus, the findings in this chapter have some pol-icy implications. Local governments should support the development of thefinancial sector so that local assembly firms can better mitigate the impactof exchange rate fluctuations.2.2 Literature ReviewThere are four strands of literatures that are related to this chapter. Thefirst strand addresses the choice of processing trade modes. Some literaturestudies these choices from the perspective of multinational firms. They usethe property right theory of the firm (Feenstra and Hanson, 2005; Fernan-des and Tang, 2012) to explain multinational firms’ decisions on controllingmaterial purchases. They argue that firms prefer internalization becauseownership of materials is a source of power when contracts are incomplete.Other papers use the financial constraint (Manova and Yu, 2012) to ex-plain the choice of processing trade modes from the perspective of assemblyfirms. They find that limited access to capital prevents assembly firms up-grading from “pure assembly” to “import and assembly”. In this chapter,we present another possible mechanism to explain outsourcing decisions ofmultinational firms from the perspective of imported material costs. Whenassembly firms import materials, they have to pay higher import prices andbear greater exchange rate pass-though than multinational firms. This in-effective use of the international market might prevent local assembly firmsfrom choosing the “import and assembly” trade mode.The second strand concerns bargaining between exporters and importers.642.2. Literature ReviewGoldberg and Tille (2013) show that a party has a higher effective bargainingweight when it is large or more risk tolerant. A higher effective bargainingweight of importers relative to exporters in turn translates into lower im-port prices and greater exchange rate pass-through into import prices. Inour chapter, we indeed find that larger market share and local financialdevelopment help firms to bear greater exchange rate pass-through.The third strand of literature studies how financial constraints affectfirm’s responses to exchange rate fluctuations. Strasser (2013) finds that fi-nancially constrained firms’ pass-through exchange rate changes to prices atalmost twice the rate of unconstrained firms. He´ricourt and Poncet (2013)find that a firm’s exported value decreases for destinations with a higherexchange rate volatility and that this effect is magnified for financially vul-nerable firms. As Manova and Yu (2012) have shown, the assembly firms in“pure assembly” are more likely to be financially constrained firms. Thus,they cannot bear much exchange rate pass-through and have to choose “pureassembly”.The last strand of literature studies how firms mitigate the impact ofexchange rate fluctuations. Firms can employ three kinds of tools: opera-tion hedging strategies, financial hedging strategies, and direct pass-throughto customers. Allayannis et al. (2001) point out that operation-hedging s-trategies only benefit shareholders when used in combination with financialhedging strategies. Bartram et al. (2010) find that firms pass through part ofcurrency changes to customers and also utilize both operational and financialhedges. Pass-through and operational hedging both reduce exchange rateexposure by 10− 15% while financial hedging decreases exposure by about40%. Do¨hring (2008) and Takatoshi et al. (2013) investigate exchange raterisk managements of European firms and Japanese firms respectively. Theyfind that firms with higher sales and greater dependency on foreign marketsmore actively engage in currency hedging. In addition, domestic-currencyinvoicing and hedging are, under certain circumstances, complementary s-trategies. In this chapter, we also find that financial hedging is a substitutestrategy with directly pass-through. When assembly firms can obtain morefinancial support, they can bear greater exchange rate pass-through.The rest of the chapter is organized as follows. Section 3 introducesthe background of Chinese processing trade. Section 4 describes the data.Section 5 discusses the relationship between the exchange rate pass-throughand trade modes. Section 6 examines the impact of financial developmenton exchange rate pass-through. Finally, section 7 concludes this chapter.652.3. Background2.3 BackgroundIn processing trade, the assembly firms obtain raw materials from abroad,process them locally, and then export the value-added goods. Most firmsthat are engaged in Chinese processing trade do not have their own brandsor responsibilities for marketing in foreign countries. Thus, these assemblyfirms are only in charge of the production process.There are two processing trade modes in China: “pure assembly” and“import and assembly”. The distinction is that in “pure assembly” trademode, the assembly firm does not take ownership of either the importedmaterials or the final goods, and hence, plays a fairly passive role. Thevalue addition it creates is only the processing service fee. In the “import andassembly” trade mode, the assembly firm plays a more active role because itcontrols the imported materials process and holds ownership of the importedmaterials and the final goods.Figure 2.1: The Production Chains in “Pure Assembly” and “Import andAssembly”Pure Assembly (PA) Import Price Export Price Sell Free Transfer Free Transfer Import and Assembly (IA) Import Price Export Price Sell Sell Material Exporters Multinational Firms Assembly Firms in China Multinational Firms Material Exporters Assembly Firms in China Multinational Firms Notes: This figure shows the production chains in the “pure assembly” and “importand assembly” trade modes. First, an assembly firm in China signs an export contractwith a multinational firm. Then, these two firms decide who is in the charge of theimported materials. In the “pure assembly” trade mode, the multinational firm will buymaterials and then transfer them to the assembly firm for free. Although the materialsare free, the multinational firm still needs to report the value of these materials toChinese Customs. This is the “Import Price” observed in the customs data. In the“import and assembly” trade mode, the assembly firm buys materials by itself. We canalso observe the “Import Price” observed in the customs data. After processing thesematerials, the assembly firm either returns or sells the final good to the multinationalfirm. This is the “Export Price” observed in the customs data.Figure 2.1 shows the production chains for the “pure assembly” and“import and assembly” trade modes. First, the assembly firm signs anexport contract with a multinational firm. Then, these firms decide who isin charge of the imported materials. In the “pure assembly” trade mode,the multinational firm buys materials then transfers them to the assembly662.3. Backgroundfirm for free. Although the materials are free, the multinational firm stillneeds to report the values of these materials to Chinese Customs. This is the“Import Price” observed in the customs data. In the “import and assembly”trade mode, the assembly firm buys materials by itself. We can also observethe “Import Price” in the customs data. After processing these materials,the assembly firm either returns or sells the final goods to the multinationalfirm. This is the “Export Price” observed in the customs data. Thus, inthe “pure assembly” trade mode, the multinational firm only outsourcesthe assembly process but controls the purchase of materials. While in the“import and assembly” trade mode, the multinational firm outsources boththe assembly process and purchase of materials.In the Chinese Customs data, we can observe the import price betweenmultinational firms and assembly firms in “pure assembly” and the importprice between materials exporters and assembly firms in “import and assem-bly”. Unfortunately, we cannot directly observe the prices between materialexporters and multinational firms. However, it is reasonable to assume thatmultinational firms do not have incentive to misreport the costs of materialsto Chinese Customs. First, in “pure assembly”, the ownership of the ma-terials belongs to multinational firms. Hence, it is not necessary for themto hide the price information to assembly firms. In some legal disputes, themultinational firms might have incentive to inflate the prices to get morecompensation. However, if the inflation is not related to the exchange ratefluctuations, then there is no concern about the exchange rate pass-through.Second, the import and export are tariff free for the processing trade inChina. Multinational firms cannot get tariff benefits from misreporting toChinese Customs. Third, assembly firms only charge the processing servicefees in “pure assembly”. Multinational firms cannot evade corporation taxby misreporting import or export prices.If the transaction is intra-firm trade, then firms have incentive to misre-port the import or export prices for tax benefits. However, this argumentworks in both trade modes. Due to the data limitation, we cannot discussthis concern in this chapter. In the rest of this chapter, we assume thatthe price between materials exporters and multinational firms is the sameas the transfer price between multinational firms and assembly firms. Thus,whether in “pure assembly” or “import and assembly”, the import pricesboth measure the costs of importing materials. We can then investigate howassembly firms and multinational firms react to the exchange rate fluctua-tion.In “pure assembly”, multinational firms own final goods. Thus, they donot care about the exchange rate pass-through in export prices. However,672.4. Datasince multinational firms need to import raw materials from other suppliers,they indeed care about the exchange rate pass-through in import prices. In“import and assembly”, local assembly firms import raw materials and sellfinal goods. Thus, assembly firms care about the exchange rate pass-throughboth in import and export prices. Usually the export price is pre-decidedwhen the contract is signed, and then raw materials are imported. Thus, nomatter multinational firms or assembly firms import materials, they regardthe export price as given. Then both kinds of firms wish the exchange ratepass-through in import prices to be low.2.4 Data2.4.1 Customs DataThis chapter uses the “Chinese Customs Export and Import Database” from2000 to 2006, which is reported on a monthly basis and collected by the Chi-nese Customs Office. This database includes all transaction information onexport and import value and quantity for each eight-digit harmonized system(HS), the exporting country, the importing country, firm ownership (state-owned, private-owned, joint-owned and foreign-owned), and trade modes(ordinary, pure assembly, import and assembly). This database does notdirectly provide any price information. However, we can divide the value ofthe import good by the quantity to get the unit value price. In this chap-ter, we use the unit value price at the HS8 level. Some imported goods aresold in the domestic market and others are used as the intermediate inputs.Thus, it is necessary to distinguish the usage of imported goods. In Chineseprocessing trade, the imported goods must be used as inputs to producefinal goods. Thus, we do not mix the exchange rate pass-through of thesetwo kinds of imported goods.2.4.2 Exchange Rate DataThe nominal exchange rate data and consumer price indices are collectedfrom the International Financial Statistics (IFS), which are on monthly ba-sis.13 The real exchange rate (RERjt) between country j and China at timet is defined as the foreign currency price per Chinese Yuan (NERjt) timesthe Chinese consumer price index (CPI) divided by the foreign CPI, which13The CPI data of Australia and New Zealand are on quarterly basis. The CPI andnominal exchange rate data of Taiwan are collected from National Statistics, Taiwan.682.4. Datais as follows:RERjt = NERjt × CPIChina,t/CPIjtTherefore, an increase in the real exchange rate (RERjt) implies an appre-ciation of the Chinese Yuan.2.4.3 Financial DataIn practice, it is difficult to measure financial development given its com-plexity and multi-dimensionality. This chapter uses the ratio between loanand gross domestic product (GDP) to measure financial development on theprefecture level. The loan includes both enterprise and resident loan. Thefinancial data is collected from “China City Statistical Yearbook”, whichincludes 287 prefectures and is from 2003 to 2006.2.4.4 Data SummaryThe customs data in this chapter includes all import transactions of Chineseprocessing trade from 2000 to 2006. The number of transaction is over 27million, which covers 208 countries and regions, and 6, 973 kinds of goods.In 2000, there are 33, 285 firms that are engaged in the processing tradeand these firms import 766 billion Chinese Yuan of goods. In 2006, the firmnumber increases to 48, 493 and the value of import goods increases to 2, 550billion Chinese Yuan.Table 2.1 shows the firm numbers and import values by trade modes.About 15%− 18% of firms are engaged in the “pure assembly” trade modeand over 70% of firms are engaged in the “import and assembly” trade mode.About 12% of firms participate in both processing trade modes. By firmnumber, shares of firms in different trade modes remain stable. By importvalue, the share of firms engaged in both trade modes decreases over timeand the share of firms only engaged in “import and assembly” increasesover time. This implies that assembly firms in China are updating alongwith the production chain. More and more firms are only engaged in moreprofitable trade mode – “import and assembly”. Table 2.2 shows that mostChinese-owned assembly firms are engaged in the “pure assembly” trademode. Most joint-owned and foreign-owned firms are engaged in the “importand assembly” trade mode. In some sense, this implies that Chinese-ownedassembly firms have some disadvantages in getting raw materials from theinternational market.Panel A of Table 2.3 shows the top six sources of origin and Panel Bshows the assembly firms’ locations. Most processing trades happen be-692.4. DataTable 2.1: Firm Number and Import Value by Trade ModesYear Trade Mode1 Firm Number Share2 Import Value3 Share42000 PA 4,850 14.57% 97 12.65%IA 24,265 72.9% 458 59.71%Both PA and IA 4,170 12.53% 212 27.64%Either PA or IA 33,285 100% 767 100%2001 PA 5,552 15.95% 105 12.38%IA 24,637 70.89% 507 59.79%Both PA and IA 4,563 13.13% 236 27.83%Either PA or IA 34,752 100% 848 100%2002 PA 6,267 17.13% 95 9.88%IA 25,801 70.55% 610 63.4%Both PA and IA 4,507 12.32% 257 26.72%Either PA or IA 36,575 100% 962 100%2003 PA 6,950 17.38% 180 13.35%IA 28,288 70.77% 927 68.77%Both PA and IA 4,732 11.84% 241 17.88%Either PA or IA 39,970 100% 1,348 100%2004 PA 7,829 17.11% 245 13.35%IA 31,030 70.18% 1260 68.66%Both PA and IA 5,353 12.11% 330 17.98%Either PA or IA 44,212 100% 1,835 100%2005 PA 8,689 18.27% 289 12.90%IA 33,382 70.18% 1,540 68.72%Both PA and IA 5,492 11.55% 412 18.38%Either PA or IA 47,563 100% 2,241 100%2006 PA 8,916 18.38% 449 17.66%IA 34,220 70.56% 1,760 69.24%Both PA and IA 5,357 11.06% 333 13.10%Either PA or IA 48,493 100% 2,542 100%Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the firm numbers and import values by the trade modes.1. “PA” is “pure assembly” and “IA” is “import and assembly”.2. The share is measured by the firm number ratio between the sub-sample (PA, IA, both PAand IA, either PA or IA) and the full sample.3. The import value is the total import value of firms in this sub-sample and the unit is in billionChinese Yuan.4. The share is measured by the value ratio between the sub-sample (PA, IA, both PA and IA,either PA or IA) and the full sample.702.4. DataTable 2.2: Ownership and Trade ModeOwnership Pure Assembly Import and AssemblyValue* Share Value* ShareChinese-Owned 1,370 71.50% 547 28.50%Joint-Owned 319 13.05% 2,120 86.95%Foreign-Owned 959 15.63% 5,180 84.37%Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the ownership shares by the trade modes.* The unit is in billion Chinese Yuan.tween China and East Asian countries.14 China is like a processing transferstation, which imports intermediate goods from other East Asian countries,assembles them locally, then exports the final goods to developed countries.There are 31 provinces in China and we also find that over 94% of assemblyfirms are located in eight coastal provinces. Especially, assembly firms inGuangdong operate almost half of processing trade. In Panel C of Table 2.3,we additionally investigate the differences of product varieties between twoprocessing trade modes. The product varieties are at the HS2 level, whichincludes 98 kinds of product categories. We find that both trade modesimport similar products. Chapter 8515 is the primary imported materialwhich accounts for 31.89% of all imports in “pure assembly” and accountsfor 45.46% of all imports in “import and assembly”. Figure 2.2 shows thedistribution of imported product varieties within this category. In this cat-egory, there are 295 kinds of products on the HS6 level. The x-axis is theproduct variety and the y-axis is the import value share of this product.We find that the bar graphs are similar between the two processing trademodes and the correlation of these shares is 0.97. Thus, we conclude thatby location, source of origin and product categories, there are no significantdifferences between the two processing trade modes. The only significantdifference between them is ownership.14When an assembly firm imports materials from a bonded area in China, the source oforigin is recorded as China.15Chapter 85 is “electrical machinery and equipment and parts thereof; sound recordersand reproducers, television image and sound recorders and reproducers, and parts andaccessories of such articles”.712.4. DataTable 2.3: Source of Origin, Firm Location and Product by Trade ModePure Assembly Import and AssemblyPanel A: Source of Origin ShareTaiwan 21.51% Japan 18.59%Japan 16.96% Taiwan 18.43%South Korea 16.16% South Korea 13.53%China 11.10% China 12.96%Hong Kong 5.68% Untied States 5.17%Untied States 4.92% Hong Kong 4.72%Others 23.67% Others 26.6%Panel B: Firm Location Share1Guangdong 52.79% Guangdong 44.95%Jiangsu 21.93% Jiangsu 16.60%Shanghai 7.74% Shanghai 13.35%Shandong 6.48% Tianjin 4.63%Liaoning 2.76% Shandong 4.55%Zhejiang 2.19% Fujian 3.83%Fujian 2.13% Liaoning 3.52%Tianjin 1.42% Zhejiang 3.00%Others 2.56% Others 5.57%Panel C: Imported Product Share2Electrical machinery andequipment (chapter 85)31.89% Electrical machinery andequipment (chapter 85)45.46%Optical, photographic,cinematographic (chapter90)10.43% Nuclear reactors, boilers,machinery and mechanicalappliances (chapter 84)9.72%Plastics (chapter 39) 9.11% Optical, photographic,cinematographic (chapter90)8.33%Nuclear reactors, boilers,machinery and mechanicalappliances (chapter 84)3.88% Plastics (chapter 39) 6.95%Others 44.69% Others 29.54%Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the source of origin, firm locations and product categories by thetrade modes. In this table, the share is measured by the value ratio between the sub-sampleand the full sample.1. The location is on the province level, which include 31 provinces.2. The product is on the HS2 level, which include 98 kinds of product categories.722.5. Exchange Rate Pass-Through and Trade ModeFigure 2.2: The Distribution Across Imported Product Varieties0.1.2.3.4Pure Assembly Import and AssemblyShareData Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This figure shows the distribution of imported product varieties within the “elec-trical machinery and equipment and parts thereof; sound recorders and reproducers,television image and sound recorders and reproducers, and parts and accessories of sucharticles” category (Chapter 85). Within this category, there are 295 kinds of productson the HS6 level. The x-axis is the product varieties and the y-axis is the import valueshare of this product. The correlation of these shares between two trade modes is 0.97.2.5 Exchange Rate Pass-Through and TradeModeIn this section, we examine the exchange rate pass-through differences inimport prices between the two processing trade modes: “pure assembly”and “import and assembly”.2.5.1 The Product-Country LevelWe examine the exchange rate pass-through on the product-country level.First, we aggregate all import transactions to the HS8 level by country-modepair for each year. Then we calculate the average price of each product-country-mode pair for each year. The benchmark regression is as follows:ln(Pijt) = α0 + α1 ln(RERjt) + α2 ln(RERjt)×Modeijt + α3 Modeijt+ µij + λt + ijt(2.1)Here Pijt is the import price (Chinese Yuan) of product i from countryj at year t. RERjt is the real exchange rate between country j and Chinaat year t. Modeijt is a dummy for the trade mode. For the same product, it732.5. Exchange Rate Pass-Through and Trade Modecan be traded under both trade modes. If the product is traded under the“pure assembly” mode, then Modeijt is 0; otherwise, it is 1. µij measuresproduct-country fixed effect and λt measures year fixed effect. The coeffi-cient α1 measures the exchange rate pass-through for the “pure assembly”trade mode. The coefficient α2 measures the exchange rate pass-throughdifferences between the two trade modes. When α2 is positive, it meansthat the exchange rate pass-through is larger in the “pure assembly” trademode. The coefficient α3 measures the price differences between the twotrade modes.Table 2.4: Exchange Rate Pass-Through and Trade Mode: Product-Country-Year LevelDependent Variable: ln(Price)(1) (2) (3) (4)ln(Exchange Rate) -0.213 -0.185 -0.156 -0.13(0.026)*** (0.026)*** (0.028)*** (0.027)***ln(Exchange Rate) × Trade Mode -0.027 -0.03(0.002)*** (0.003)***Trade Mode 0.552 0.56(0.012)*** (0.014)***Constant 3.47 3.153 3.513 3.188(0.008)*** (0.011)*** (0.013)*** (0.015)***Product-Country FE X X X XYear FE X X X XCluster By Product X X X XExclude HKG & USA X XObservations 519,001 519,001 443,083 443,083R-squared 0.832 0.843 0.838 0.847Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the import prices and exchange rate pass-through differences betweentwo trade modes. HKG Hong Kong; USA the United States.1. The product is on the HS8 level. The trade mode is a dummy. If the product is traded underthe “pure assembly” mode, it is 0; otherwise, it is 1.2. The price is in Chinese Yuan and the exchange rate is the real exchange rate between thesource of origin and China. An increase in the real exchange rate implies an appreciation of theChinese Yuan.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.The result is presented in Table 2.4. Column 1 shows that the exchangerate pass-through is 0.213. On average, when Chinese Yuan appreciates by10%, the import price (Chinese Yuan) will decrease by 2.13%. In column2, we additionally control the trade mode effect and the interaction term742.5. Exchange Rate Pass-Through and Trade Modebetween the real exchange rate and the trade mode. It shows that theexchange rate pass-through in the “pure assembly” trade mode is 0.185.When Chinese Yuan appreciates by 10%, the import prices (Chinese Yuan)will decrease by 1.85%. In “import and assembly”, the exchange rate passthrough is even higher, at 0.212. When Chinese Yuan appreciates by 10%,the import price (Chinese Yuan) will decrease by 2.12%. At the same time,the import price is also 73.6% higher in the “import and assembly” trademode. In sum, the result in Table 2.4 shows that when the assembly firms inChina import materials themselves, they have to pay higher import pricesand bear larger exchange rate pass-through. In this sense, the assemblyfirms in China are disadvantaged in getting intermediate goods from theinternational market relative to multinational firms.There are several mechanisms that may explain this result. First, theassembly firms in China might have weak bargaining power, and thus theyhave to pay higher prices and bear more exchange rate risks. Second, thequantity per transaction can also affect the import price. The more prod-ucts being imported, the lower the price a firm might pay. To investigatethis channel, we first calculate the average quantity of transactions for eachproduct-country-mode pair. Then we examine whether the average quantityis significantly different across trade modes. Table 2.5 shows that the aver-age quantity is higher in the “import and assembly” trade mode. Therefore,if the quantity channel works, the import price should be lower in the “im-port and assembly” trade mode. Thus, the quantity per transaction cannotexplain our result. Third, the significant price differences between these twotrade modes could be the result of differences in the quality of the product.In Panel C of Table 2.3 and Figure 2.2, we have shown that there are no sig-nificant differences between imported goods under the two processing trademodes. However, this does not mean that the quality of the product is thesame under the two processing trade modes. Even within the HS8 level, theproducts are still very different. On the product-country level, we cannotdiscuss this issue. In section 2.5.2, we go through the product quality indepth. We find that the pattern remains robust even excluding the qualityeffect. Fourth, the multinational firms might not buy intermediate goodsfrom other firms, and they actually transfer intermediate goods within thefirm boundary. In “pure assembly”, the imported materials are owned bymultinational firms and thus, the import prices are not necessary to be thereal international market price. This mechanism can explain how the price islower in “pure assembly”. As we have discussed in section 2.3, multinationalfirms cannot get tax benefits from misreporting, so this concern may not bea serious problem. In addition, even if the import prices are misreported,752.5. Exchange Rate Pass-Through and Trade Modethey are not necessarily related with exchange rate fluctuations. Finally, theinvoice of currency can also affect the exchange rate pass-through. Supposethe import price is fixed and the transaction is invoiced in Chinese Yuan,then the exchange rate pass-through should be close to 0. Contrarily, sup-pose the transaction is invoiced in foreign currency, then the exchange ratepass-through should be close to 1. If the invoices of currency are significant-ly different between these two trade modes, the exchange rate pass-throughwill be different. Due to the data limitation, we cannot discuss this issue inthis chapter.Table 2.5: Quantity and Trade Mode: Product-Country-Year LevelDependent Variable: ln(Average Quantity)Trade Mode 0.226(0.013)***Constant 6.920(0.012)***Product-Country FE XYear FE XCluster By Product XObservations 559,964R-squared 0.792Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the import quantity differences between two trade modes.1. The product is on the HS8 level. The trade mode is a dummy. If the productis traded under the “pure assembly” mode, it is 0; otherwise, it is 1.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.Exclude United States and Hong KongUntil 2005, the value of Chinese Yuan was pegged to the U.S. dollar. Thus,the exchange rate between China and the United States is fixed. The realexchange rate fluctuation between them only reflects the CPI fluctuation.In column 3 and 4 of Table 2.4, we exclude all import transactions from theUnited States from the sample. In the top ten sources of origin, Hong Kongalso pegged its currency to the U.S. dollar. Thus, we also drop all importtransactions from Hong Kong from the sample. The result is very similarwith that in the full sample, but the exchange rate pass-through is smaller.762.5. Exchange Rate Pass-Through and Trade ModeMonthly FrequencyThe volatility of exchange rate on the annual level is not large. To betterinvestigate the exchange rate pass-through in import prices, we re-run theregression (2.1) on the monthly level. The result is shown in Table 2.6. Wecan find that the results are robust and the assembly firms in China still haveto pay higher import prices and bear larger exchange rate pass-through.Table 2.6: Exchange Rate Pass-Through and Trade Mode: Product-Country-Month LevelDependent Variable: ln(Price)(1) (2) (3) (4)ln(Exchange Rate) -0.307 -0.250 -0.257 -0.196(0.024)*** (0.024)*** (0.024)*** (0.025)***ln(Exchange Rate) × Trade Mode -0.036 -0.038(0.002)*** (0.003)***Trade Mode 0.596 0.605(0.013)*** (0.014)***Constant 3.418 3.063 4.198 3.133(0.012)*** (0.015)*** (0.097)*** (0.022)***Product-Country FE X X X XYear FE X X X XCluster By Product X X X XExclude HKG & USA X XObservations 3,074,426 3,074,426 2,553,332 2,553,332R-squared 0.788 0.801 0.801 0.807Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the import prices and exchange rate pass-through differences betweentwo trade modes. HKG Hong Kong; USA the United States.1. The product is on the HS8 level. The trade mode is a dummy. If the product is traded underthe “pure assembly” mode, it is 0; otherwise, it is 1.2. The price is in Chinese Yuan and the exchange rate is the real exchange rate between thesource of origin and China. An increase in the real exchange rate implies an appreciation of theChinese Yuan.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.In Chinese Customs data, we can only observe the arrival month ofimport products at the ports. One concern is that there is a time lag betweenthe signing of import contracts and the arrival of those products. If thetime lag is greater than one month, then we should examine the effect ofexchange rate in the last month instead of in the current month. To checkthis problem, we include twelve exchange rate lags to examine the totalexchange rate pass-through. The result is shown in Figure 2.3. The exchange772.5. Exchange Rate Pass-Through and Trade ModeFigure 2.3: The Exchange Rate Pass-Through Over Times0.3070.346Exchange Rate Pass-Through0 4 8 12Time LagsData Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This figure shows the exchange rate pass-through over times. The x-axis is thetime and the y-axis is the exchange rate pass-through. When the time lags increasesfrom 0 to 12, the exchange rate pass-through rises from 0.307 to 0.346.rate pass-through in current month is 0.307 and it increases to 0.346 whenwe include twelve-month lags. Over twelve months, the exchange rate pass-through only increases 12.7%, so the time lag is not a large issue. In therest of this chapter, we only examine the exchange rate pass-through of thecurrent month.Intermediary CompanyIn the processing trade, some firms are pure import-export companies thatdo not produce any products. These firms only provide intermediary ser-vices between domestic producers and foreign buyers. Thus, it is necessaryto distinguish these firms from other ordinary assembly firms. FollowingManova and Yu (2012), we use the keywords in firms’ names to identifythem.16 The summary is presented in Table 2.7. In Panel A, it shows thatthe number of intermediary companies remains stable from 2000 to 2006. In2000 the share of intermediary companies is 7.8%, and in 2006 this numberdecreases to 7.25%. However, the total values that are imported by inter-16The keywords that we use are “jingmao”, “jinchukou”, “maoyi”, “kemao”, “waimao”,“jiagongzhuangpeifuwusongsi”, “waijingfazhan” and “duiwaijingjifazhan”.782.5. Exchange Rate Pass-Through and Trade Modemediary companies decrease significantly. In 2000 intermediary companiesimported around 20% of total goods and in 2006 these firms only imported8.41% of total goods. This implies that the role of intermediary companiesis in the decline. Some assembly firms do not need the intermediary servicesany more and they can directly establish connections with multinationalfirms. Panel B and Panel C show that the intermediary companies are dif-ferent from other assembly firms in processing trade modes. Around 75% ofnon-intermediary companies are engaged in “import and assembly” and theimport value is more than 80%. Only 58% of intermediary companies areengaged in “import and assembly” and the total value is less than 14%. Thisimplies that the firms that cooperate with intermediary companies prefer toparticipate in the ”pure assembly” trade mode. Some small firms cannotdirectly get orders from multinational firms, and intermediary companiescan supply such matching services. At the same time, these small firms donot have an international market network or they cannot bear the exchangerate risks and thus they are only engaged in “pure assembly”.In the first three columns of Table 2.8, we exclude all intermediary firmsfrom our sample. This shows that the assembly firms in China still have topay higher import prices, but the price is only 12.4% higher than that ofmultinational firms. Moreover, the exchange rate pass-through is now lowerin the “import and assembly” trade mode. In the last three columns of Table2.8, we investigate the exchange rate pass-through and price differences onlyfor intermediaries. The assembly firms in China have to pay a much higherprice, which is 1.6 times higher than that of multinational firms. At thesame time, the assembly firms in China also bear higher exchange rate pass-through.792.5. Exchange Rate Pass-Through and Trade ModeTable 2.7: The Summary of Intermediary CompaniesShare of Firm Number Value1Panel A: Full SampleYear Non-Intermediary Intermediary Non-Intermediary Intermediary2000 92.20% 7.80% 612 1542001 92.34% 7.66% 683 1642002 92.87% 7.13% 794 1692003 93.08% 6.92% 1,160 1852004 93.01% 6.99% 1,620 2122005 94.17% 5.83% 2,030 2112006 97.75% 7.25% 2,330 215Panel B: Non-IntermediaryYear PA IA PA IA2000 24.57% 85.78% 163 5562001 26.77% 84.30% 184 6192002 27.31% 83.03% 192 7222003 27.13% 83.06% 247 1,0602004 27.83% 82.70% 378 1,5102005 28.34% 81.93% 504 1,8702006 27.49% 82.24% 582 2,030Panel C: IntermediaryYear PA IA PA IA2000 57.11% 81.12% 146 1142001 57.42% 80.74% 157 1232002 57.41% 80.70% 161 1452003 57.39% 76.65% 175 1042004 56.28% 76.89% 197 82.42005 55.13% 76.17% 197 77.82006 54.36% 73.43% 201 66.3Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the summary of intermediary companies.1. The unit is in billion Chinese Yuan.PA Pure Assembly. IA Import and Assembly.802.5.ExchangeRatePass-ThroughandTradeModeTable 2.8: Exchange Rate Pass-Through, Trade Mode and Intermediaries: Product-Country-Month LevelDependent Variable: ln(Price)Non-intermediaries Intermediaries(1) (2) (3) (4) (5) (6)ln(Exchange Rate) -0.250 -0.262 -0.258 -0.281 -0.173 -0.063(0.026)*** (0.026)*** (0.029)*** (0.032)*** (0.030)*** (0.034)*ln(Exchange Rate) × Trade Mode 0.013 0.014 -0.085 -0.086(0.003)*** (0.003)*** (0.007)*** (0.009)***Trade Mode 0.117 0.107 0.986 1.001(0.012)*** (0.013)*** (0.034)*** (0.042)***Constant 3.722 3.642 3.769 2.828 2.594 2.600(0.016)*** (0.019)*** (0.028)*** (0.019)*** (0.019)*** (0.030)***Product-Country FE X X X X X XYear-Month FE X X X X X XCluster By Product X X X X X XExclude HKG & USA X XObservations 2,478,936 2,478,936 2,061,117 1,208,708 1,208,708 972,327R-squared 0.804 0.805 0.812 0.804 0.805 0.812Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the import prices and exchange rate pass-through differences between two trade modes. HKG Hong Kong; USAthe United States.1. The product is on the HS8 level. The trade mode is a dummy. If the product is traded under the “pure assembly” mode, it is 0; otherwise,it is 1.2. The price is in Chinese Yuan and the exchange rate is the real exchange rate between the source of origin and China. An increase in thereal exchange rate implies an appreciation of the Chinese Yuan.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.812.5.ExchangeRatePass-ThroughandTradeModeTable 2.9: The Intermediaries and OwnershipNon-intermediaries IntermediariesYear Chinese-owned Joint-owned Foreign-owned Chinese-owned Joint-owned Foreign-owned2000 7.10% 39.45% 53.45% 99.80% 0.14% 0.06%2001 7.28% 37.28% 55.44% 99.77% 0.15% 0.08%2002 6.84% 32.59% 60.57% 99.82% 0.09% 0.09%2003 6.33% 28.23% 65.44% 99.83% 0.04% 0.13%2004 6.39% 25.80% 67.81% 99.85% 0.03% 0.12%2005 6.88% 22.82% 70.30% 99.78% 0.02% 0.20%2006 6.71% 21.14% 72.15% 99.74% 0.01% 0.25%Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the import value share of ownership for both non-intermediary and intermediary companies.822.5.ExchangeRatePass-ThroughandTradeModeTable 2.10: Exchange Rate Pass-Through and Trade Mode By Ownership: Product-Country-Month LevelDependent Variable: ln(Price)Exclude IntermediariesChinese-owned Joint-owned Foreign-owned Chinese-owned Joint-owned Foreign-ownedln(Exchange Rate) -0.199 -0.266 -0.292 -0.142 -0.267 -0.292(0.028)*** (0.035)*** (0.033)*** (0.044)*** (0.035)*** (0.033)***ln(Exchange Rate) × Trade Mode -0.083 0.027 0.032 -0.067 0.027 0.032(0.006)*** (0.005)*** (0.004)*** (0.006)*** (0.005)*** (0.004)***Trade Mode 1.028 -0.018 -0.031 0.745 -0.019 -0.031(0.028)*** (0.018) (0.014)** (0.027)*** (0.018) (0.014)**Constant 2.816 3.753 3.719 3.483 3.753 3.720(0.018)*** (0.032)*** (0.027)*** (0.046)*** (0.032)*** (0.027)***Product-Country FE X X X X X XYear-Month FE X X X X X XCluster By Product X X X X X XObservations 1,473,291 1,235,637 1,821,255 601,895 1,234,785 1,820,227R-squared 0.833 0.809 0.789 0.858 0.809 0.789Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the import prices and exchange rate pass-through differences between two trade modes by ownership.1. The product is on the HS8 level. The trade mode is a dummy. If the product is traded under the “pure assembly” mode, it is 0; otherwise,it is 1.2. The price is Chinese Yuan and the exchange rate is the real exchange rate between the source of origin and China. An increase in the realexchange rate implies an appreciation of the Chinese Yuan.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.832.5. Exchange Rate Pass-Through and Trade ModeOwnershipTable 2.2 shows that most Chinese-owned firms are engaged in “pure assem-bly” and Table 2.7 shows that the intermediary companies are also mainlyengaged in “pure assembly”. There must be some connections between thesetwo facts. In Table 2.9, we show that most intermediaries are Chinese-ownedfirms. Thus, it is necessary to distinguish all assembly firms by their owner-ship. Table 2.10 shows that firms’ ownership indeed matters. For Chinese-owned assembly firms, the import price and exchange rate pass-through areboth higher in the “pure assembly” trade mode. For joint-owned or foreign-owned assembly firms, the import price and exchange rate pass-through areboth lower in the “pure assembly” trade mode. Even though we excludethe intermediary companies, the result is still robust. Thus, the greaterexchange rate pass-through in “import and assembly” is mainly driven byChinese-owned firms.Source of OriginExisting studies argue that the quality of product is different across countriesand the bargaining position of assembly firms might vary across countries.Thus, we divide sources of origin into two groups: developed and developingcountries (regions).17 Then we investigate the exchange rate pass-throughfor these two groups respectively. Table 2.11 shows that the result is robustfor developed countries (regions). However, the exchange rate pass-throughin the two trade modes is not significant for developing countries (regions).This means that the exchange rate risk is almost zero for assembly firms inChina and there is no significant difference between the two trade modes.One possibility is that assembly firms in China have strong bargaining powerwhen they import intermediate goods from developing countries (regions).Exchange Rate Pass-Through in Export PriceIn Section 2.3, we mentioned that the export price is pre-decided when thecontract is signed. Since there is a time lag between shipping final goods andsigning a contract, the export price should be uncorrelated with the exchangerate at the time of shipping. To verify it, we regress the export price onexchange rate and the result is in Table 2.12. It shows that the exchangerate pass-through are not significant for Chinese-owned and foreign-owned17We exclude Hong Kong and the United States from our sample due to China’s dollarpeg policy. The developed countries (regions) include OECD countries plus Singapore andTaiwan.842.5. Exchange Rate Pass-Through and Trade ModeTable 2.11: Exchange Rate Pass-Through and Trade Mode By Source ofOrigin: Product-Country-Month LevelDependent Variable: ln(Price)Exclude IntermediariesDeveloped Developing Developed DevelopingPanel A: Chinese-owned Firmsln(Exchange Rate) -0.114 0.039 -0.081 -0.133(0.036)*** (0.050) (0.055) (0.103)ln(Exchange Rate) × Trade Mode -0.097 0.016 -0.060 -0.004(0.007)*** (0.009)* (0.007)*** (0.012)Trade Mode 1.061 0.532 0.693 0.447(0.033)*** (0.048)*** (0.032)*** (0.055)***Panel B: Joint-owned Firmsln(Exchange Rate) -0.331 0.089 -0.332 0.090(0.043)*** (0.065) (0.043)*** (0.065)ln(Exchange Rate) × Trade Mode 0.035 -0.006 0.035 -0.006(0.006)*** (0.009) (0.006)*** (0.009)Trade Mode -0.042 0.065 -0.043 0.065(0.021)** (0.045) (0.021)** (0.045)Panel C: Foreign-owned Firmsln(Exchange Rate) -0.379 0.060 -0.379 0.061(0.040)*** (0.074) (0.040)*** (0.074)ln(Exchange Rate) × Trade Mode 0.041 0.011 0.040 0.011(0.004)*** (0.007)* (0.004)*** (0.007)*Trade Mode -0.063 -0.014 -0.062 -0.014(0.016)*** (0.031) (0.016)*** (0.032)Product-Country FE X X X XYear-Month FE X X X XCluster By Product X X X XData Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the import prices and exchange rate pass-through differences betweentwo trade modes by source of origin. The developed countries (regions) include OECD countriesplus Singapore and Taiwan. We exclude Hong Kong and the United States from our sample.1. The product is on the HS8 level. The trade mode is a dummy. If the product is traded underthe “pure assembly” mode, it is 0; otherwise, it is 1.2. The price is Chinese Yuan and the exchange rate is the real exchange rate between the sourceof origin and China. An increase in the real exchange rate implies an appreciation of the ChineseYuan.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.852.5. Exchange Rate Pass-Through and Trade Modeassembly firms. For joint-owned assembly firms, it is significant but thesize is smaller than the exchange rate pass-through in import prices. Thus,we think the exchange rate risks for assembly firms mostly come from theimported materials rather then exported final goods. A surprising finding isthat the average export price of Chinese-owned assembly firms is higher thanthat of multinational firms, while the average export price of joint-ownedand foreign-owned assembly firms is lower than that of multinational firms.Table 2.12: Exchange Rate Pass-Through in Export Prices: Product-Country-Month LevelDependent Variable: ln(Export Price)Chinese-owned Joint-owned Foreign-ownedln(Exchange Rate) 0.060 -0.113 0.010(0.042) (0.027)*** (0.022)Trade Mode 0.275 -0.197 -0.171(0.040)*** (0.032)*** (0.034)***Constant 3.424 3.983 3.726(0.040)*** (0.030)*** (0.030)***Product-Country FE X X XYear-Month FE X X XCluster By Product X X XObservations 538235 1025285 1729160R-squared 0.922 0.915 0.865Data Sources: The “Chinese Customs Export and Import Database”(2000-2006).Notes: This table shows how export prices react to exchange rate fluctua-tions.1. The product is on the HS8 level. The trade mode is a dummy. If theproduct is traded under the “pure assembly” mode, it is 0; otherwise, it is1.2. The price is Chinese Yuan and the exchange rate is the real exchange ratebetween the export destination and China. An increase in the real exchangerate implies an appreciation of the Chinese Yuan.3. We exclude intermediary firms, and exports to the United States andHong Kong.4. Standard errors in parentheses. **Significant at 5%; ***significant at1%.2.5.2 The Firm-Product-Country LevelIn order to control the quality of imported materials, now we examine theexchange rate pass-through on the firm-product-country level. Using theexport transaction information, we first construct the quality index. In our862.5. Exchange Rate Pass-Through and Trade Modedata, we cannot directly observe the product quality, but we can use theproduct price to proxy it. This construction is based on two assumptions:Assumption 1 : When the export price of product is higher, then itsquality is higher.Assumption 2 : High quality product needs high quality inputs.In Chinese Customs data, we can observe all import and export transac-tions, but we cannot match the export products with the imported materialsthat are used. Thus, we can only construct the quality index on the firmlevel as follows:1 For each product (HS8), we calculate the average exporting price Pitat year t. Here, i is the product.2 For each firm-product pair, we calculate the average export price Piftat year t. Here f is the firm.3 The quality index of firm f at year t is defined asQualityft =∑iViftVftPiftPitHere, Vift is the export value for product i of firm f at year t. Vft is theexport values of firm f at year t.Thus, the quality index measures the exporting product quality of thefirm relative to the average quality. Under two assumptions, this quali-ty index is higher, and assembly firms import higher quality inputs. Thedistribution of quality is shown in Figure 2.4. In order to exclude the ef-fect of outliers, we either use the logarithm of quality index or exclude thetop/bottom 5% observations. The results are similar. The average qualityin “pure assembly” is significantly greater than that in “import and assem-bly”. One possible explanation is that, for some high quality materials,local assembly firms cannot get them from the international market, andthus multinational firms prefer to purchase these materials by themselves.This result is consistent with Tables 2.10 and 2.12.Then we control the quality of products and re-run regression (2.1) onthe firm level. The result is shown in Table 2.13. We find that the exchangerate pass-through is much smaller on the firm level. For Chinese-owned firm-s, the exchange rate pass-through is only 3.7%-8.4%. For joint-owned andforeign-owned firms, the exchange rate is also around 10%. In addition, highquality of inputs indeed implies the high import price. The result about theexchange rate pass-through difference between assembly firms and multina-tional firms remains robust. Chinese-owned assembly firms still need to bearlarger exchange rate pass-through and joint-owned/foreign owned assembly872.5. Exchange Rate Pass-Through and Trade ModeFigure 2.4: The Distribution of Quality0.2.4.6.8Kernel Density-10 -5 0 5 10 15ln(Quality)PA IAT test for the mean of two samples: Ha: PA>IA Pr(T>t)=0.00ln(Quality)0.2.4.6.8Kernel Density0 5 10QualityPA IAT test for the mean of two samples: Ha: PA>IA Pr(T>t)=0.00Exclude Top/Bottom 5%Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This figure shows the distribution of product quality on the firm level. Thex-axis is the relative quality and the y-axis is the kernel density. To exclude the effect ofoutliers, we use the logarithm of quality in the left graph and exclude top/bottom 5%observations from the sample in the right graph.firms take less exchange rate pass-through. After controlling quality, the im-port prices of joint-owned and foreign-owned assemly firms are higher thanthat of multinational firms, which is contrary with the result in Table 2.10.IndustryAcross industries, the exchange rate pass-through might be different. Wedivide the imported products into different industries at the HS2 level. Wedrop the industries if they include less than 50,000 observations. In the end,we have 34 kinds of industries. Then we re-run regression (2.1) for each in-dustry and the results are shown in Figure 2.5. For joint-owned and foreign-owned assembly firms, their exchange rate pass-through is always lower inmost industries. For Chinese-owned assembly firms, the exchange rate pass-through differences vary across industries. In chemistry and electron relatedindustries, the exchange rate pass-through is larger. In textile related indus-tries, the exchange rate pass-through is lower. This implies that Chinese-owned assembly firms have advantages in getting textile-related materials,but have disadvantages in getting chemistry and electron-related materials.882.5. Exchange Rate Pass-Through and Trade ModeChanges Over TimeWhen assembly firms become familiar with the international market, theexchange rate pass-through will change over time. We use rolling regressionand twenty months as the window to investigate this issue. The result isshown in Figure 2.6. After 2002, the exchange rate pass-through for allkinds of assembly firms becomes lower relative to multinational firms. Thismeans that assembly firms bear less exchange rate risks than before.2.5.3 Other FactorsNow we investigate other factors that can affect the exchange rate pass-through, such as the ratio of imported inputs and market share. In “pureassembly”, the imported materials within the same assembly firm can besupplied by different multinational firms. Since we could not observe whichmultinational firm operates this transaction, we cannot get the importedinputs ratio and the market share of multinational firms. Thus, we can onlyinvestigate the impact within “import and assembly”.Imported InputsProcessing trade does not require all raw materials to be imported, and someinputs can be bought from the domestic market.18 If the imported materialsonly account for a small number of the total inputs, then it is possiblethat firms are less sensitive to exchange rate fluctuations. In addition, ifthe value-added is large, then the firm might have enough profit to bearhigher exchange rate pass-through. Thus, we define an index to proxy theimportance of the imported materials. The definition is as follows:Inputft =Imported InputsftTotal InputsftTotal InputsftTotal OutputsftThe first part is the ratio between imported inputs and total inputs. Itmeasures the importance of imported inputs in total inputs. The secondpart is the ratio between inputs and outputs. It measures the value-addedof the firm. When the input index is higher, the firm should be less sensitiveto the exchange rate fluctuations and thus can bear higher exchange ratepass-through. The result in Table 2.14 indeed verifies our prediction.18In processing trade, some inputs must be imported from abroad. However, there isno definite rule about the minimum ratio of imported inputs. The defining of processingtrade depend on local customs.892.5. Exchange Rate Pass-Through and Trade ModeTable 2.13: Exchange Rate Pass-Through and Trade Mode: Firm-Product-Country-Month LevelDependent Variable: ln(Price)Chinese-owned Joint-owned Foreign-ownedln(Exchange Rate) -0.037 -0.108 -0.038(0.037) (0.027)*** (0.024)ln(Exchange Rate) × Trade Mode -0.047 0.026 0.038(0.006)*** (0.005)*** (0.003)***Trade Mode 0.453 0.119 0.192(0.032)*** (0.015)*** (0.017)***ln(Quality) 0.146 0.247 0.263(0.009)*** (0.010)*** (0.008)***Constant 3.737 3.083 3.258(0.085)*** (0.110)*** (0.055)***Product-Country FE X X XYear-Month FE X X XProvince FE X X XCluster By Product X X XObservations 1,352,420 4,856,936 10,732,303R-squared 0.798 0.723 0.713Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the import prices and exchange rate pass-through differences betweentwo trade modes. We exclude the intermediary companies from our sample.1. The product is on the HS8 level. The trade mode is a dummy. If the product is traded underthe “pure assembly” mode, it is 0; otherwise, it is 1.2. The price is Chinese Yuan and the exchange rate is the real exchange rate between thesource of origin and China. An increase in the real exchange rate implies an appreciation of theChinese Yuan.3. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.902.5.ExchangeRatePass-ThroughandTradeModeFigure 2.5: The Exchange Rate Pass-Through Differences Across IndustriesINORGANIC CHEM, ORG 28ORGANIC CHEMICALS 29TANNING OR DYEING EXTRACTS, DY 32SOAPS, WAXES, SCOURING PRODUCT 34ALBUMINODIAL SUB, STARCHES, GL 35MISCELLANEOUS CHEMICAL PRODUCT 38PLASTICS & ARTICALES THEREOF 39RUBBERS & ARTICLES THEREOF 40RAW HIDES & SKINS & LEATHER 41WOOD & ARTICLES OF WOOD, WOOD 44PAPER & PAPERBOARD, ARTICLES 48PRINTED BOOKS, NEWSPAPERS, PIC 49WOOL & FINE OR COARSE ANIMAL H 51COTTON, INC. YARNS & WOVEN FAB 52MAN?MADE FILAMENTS, INC. YARNS 54MAN?MADE STAPLE FIBERS, INC. Y 55WADDING, FELT & NONWOVENS, SPE 56SPECIAL WOVEN FABRICS, TUFTED 58IMPREGNATED, COATED, COVERED 59KNITTED OR CROCHETED FABRICS 60ARTICLES OF APPAREL & CLOTHING 61ARTICLES OF APPAREL & CLOTHING 62GLASS & GLASSWARE 70PEARLS, STONES, PREC. METALS 71IRON & STEEL 72RTICLES OF IRON OR STEEL 73COPPER & ARTICLES THEREOF 74ALUMINUM & ARTICLES THEREOF 76TIN & ARTICLES THEREOF 80MISCELLANEOUS ARTICLES OF BASE 83NUCLEAR REACTORS, BOILERS, MAC 84ELECTRICAL MACHINERY & EQUIP. 85OPTICAL, PHOTOGRAPHIC, CINEMAT 90MISCELLANEOUS MANUFACTURED ART 96-.5 0 .5 1 -.5 0 .5 1 -.5 0 .5 1Chinese-Owned Joint-Owned Foreign-OwnedConfidence Interval (95%) CoefficientHS2Data Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This figure shows the exchange rate pass-through differences across industries. We exclude those industries with less than50,000 observations. The x-axis is the exchange rate pass-through and the y-axis is the industries on the HS2 level.912.5. Exchange Rate Pass-Through and Trade ModeFigure 2.6: The Exchange Rate Pass-Through Differences Over Times-.050.05Exchange Rate Pass-Through Differences2001-1 2002-1 2003-1 2004-1 2005-1 2006-1TimeChinese-owned Joint-ownedForeign-ownedData Sources: The “Chinese Customs Export and Import Database” (2000-2006).Notes: This figure shows the exchange rate pass-through differences over times. Thex-axis is the time and the y-axis is the exchange rate pass-through. We use the rollingregression to get the exchange rate pass-through and the window is 24 months.Market ShareWe define the market share of firm f on product i in country j at time t asMarket Shareijft =Import Quantityijft∑f∈F Import QuantityijftHere, Import Quantityijft is the import quantity. F is the set of all thefirms which import product i from country j at year t.The result is shown in Table 2.15. For Chinese-owned assembly firms,the effect of market share is not significant. For joint-owned and foreign-owned assembly firms, higher market share implies higher exchange ratepass-through. This result is contrary to previous studies. Usually the highermarket share implies better bargaining position and thus these firms willtake less exchange rate pass-through. However, Goldberg and Tille (2013)show a bargain model between importers and exporters. They argue thatthe higher bargaining power implies lower import price but more exchangerate risks and our result verifies their prediction.Another possible explanation is related with Baldwin and Krugman(1989) and Froot and Klemperer (1989). In this chapter, we only considerthe static decision of firms. Actually, firms’ pricing strategy is dynamic.Froot and Klemperer (1989) argue that firms’ current market share mat-ters for their future profit. Thus, a firm would like to take more exchange922.5. Exchange Rate Pass-Through and Trade ModeTable 2.14: Exchange Rate Pass-Through and Imported InputDependent Variable: ln(Price)Chinese-owned Joint-owned Foreign-ownedln(Exchange Rate) -0.454 -0.258 -0.242(0.068)*** (0.038)*** (0.033)***ln(Exchange Rate) × Imported Input 0.036 0.056 0.028(0.011)*** (0.009)*** (0.006)***Imported Input 0.182 0.211 0.442(0.041)*** (0.040)*** (0.036)***ln(Quality) 0.058 0.230 0.261(0.014)*** (0.011)*** (0.009)***Constant 4.900 2.953 3.113(0.116)*** (0.176)*** (0.085)***Product-Country FE X X XYear-Month FE X X XProvince FE X X XCluster By Product X X XObservations 452,182 3,054,362 6,871,574R-squared 0.832 0.740 0.729Data Sources: The “Chinese Customs Export and Import Database” (2000-2006) and the“China City Statistical Yearbook” (2003-2006).Notes: This table shows the impact of value-added on the exchange rate pass-through within“import and assembly”. We exclude the intermediary companies from our sample. We alsoexclude those firms that with negative value-added.1. The product is on the HS8 level. The trade mode is a dummy. If the product is tradedunder the “pure assembly” mode, it is 0; otherwise, it is 1.2. The price is Chinese Yuan and the exchange rate is the real exchange rate between thesource of origin and China. An increase in the real exchange rate implies an appreciation ofthe Chinese Yuan.3 The import input is defined on the firm-year level.4. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.932.6. Exchange Rate Pass-Through and Financial Developmentrate pass-through if the exchange shock is permanent. This mechanism alsocan explain why large firms bear higher exchange rate pass-through. It isbecause large firms wish to keep market share by changing prices.Table 2.15: Exchange Rate Pass-Through and Market ShareDependent Variable: ln(Price)Chinese-owned Joint-owned Foreign-ownedln(Exchange Rate) -0.349 -0.135 -0.152(0.065)*** (0.035)*** (0.029)***ln(Exchange Rate) × Market Share 0.016 -0.027 -0.023(0.014) (0.008)*** (0.006)***Market Share -0.956 -1.041 -1.038(0.052)*** (0.032)*** (0.028)***ln(Quality) 0.065 0.226 0.249(0.013)*** (0.010)*** (0.008)***Constant 5.010 3.148 3.380(0.109)*** (0.146)*** (0.075)***Product-Country FE X X XYear-Month FE X X XProvince FE X X XCluster By Product X X XObservations 516,126 3,575,781 8,563,852R-squared 0.841 0.740 0.731Data Sources: The “Chinese Customs Export and Import Database” (2000-2006) andthe “China City Statistical Yearbook” (2003-2006).Notes: This table shows the impact of market share on the exchange rate pass-throughwithin “import and assembly”. We exclude the intermediary companies from our sample.1. The product is on the HS8 level. The trade mode is a dummy. If the product is tradedunder the “pure assembly” mode, it is 0; otherwise, it is 1.2. The price is Chinese Yuan and the exchange rate is the real exchange rate between thesource of origin and China. An increase in the real exchange rate implies an appreciationof the Chinese Yuan.3 The market share is defined as the quantity ratio between firm i and all firms on thefirm-product-year level.4. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.2.6 Exchange Rate Pass-Through and FinancialDevelopmentIn this section, we examine the impact of local financial development on theexchange rate pass-through. When firms are located in financially devel-oped prefectures, they can get two benefits. First, they can get access to942.6. Exchange Rate Pass-Through and Financial Developmentmore financial tools to hedge the exchange rate fluctuations. Some papers(Do¨hring, 2008; Takatoshi et al., 2013) argue that the financial hedging is asubstitute strategy with direct pass-through. When firms can hedge the ex-change rate risks, they can bear greater exchange rate pass-through. Second,the developed financial sector is also helpful to decrease the borrowing costsof firms. Thus, these firms have less financial constraints. Strasser (2013)finds that financially constrained exporting firms pass-through exchange ratechanges to prices at almost twice the rate of unconstrained exporting firms.Thus, for unconstrained importing firms, they can bear greater exchangerate pass-through. Based on these two channels, we have that:Proposition 1: When local financial development is higher, the ex-change rate pass-through is higher.We measure local financial development at the prefecture-year level.19We use Loan/GDP to measure financial development and Figure 2.7 showsthe distribution of Loan/GDP for each year. It shows that the distributionof Loan/GDP remains stable over time.According to the median value of Loan/GDP for each year, we dividethe prefectures into two groups: financially developed and undeveloped pre-fectures. FinDf is a dummy for firm f . If firm f is located in a financiallydeveloped prefecture, then FinDm is 1. Otherwise, it is 0. Unfortunately,we could only identify the locations of assembly firms in China but not theorigins of multinational firms. Thus, we have to assume all multinationalfirms are identical. In order to test Proposition 1, we include FinDf andthe interaction terms in regression (2.1).20ln(Pijft) = β0 + β1 ln(RERjt) + β2 Modeijft + β3 FinDf+β4 ln(RERjt)×Modeijft + β5 ln(RERjt)× FinDf+β6 Modeijft × FinDf + β7 ln(RERjt)×Modeijft × FinDf+Zft + µij + λt + ijt(2.2)Here, i is the product, j is the source of origin, f is the firm and t is thetime. In Zft, we control other characteristics on the firm and prefecturelevel: the average quality of exporting product on the firm level, GDP per19Since we only use the customs data from 2003 to 2006, there is a sample selectionproblem. We check regression (2.1) both for the full sample and this sub-sample, and donot find any significant differences. Thus, we think the sample selection is not a largeconcern.20In Appendix B, we use the two-stages approach to test the Proposition 1.952.7. ConclusionFigure 2.7: The Distribution of Loan/GDP012301230 1 2 3 0 1 2 32003 20042005 2006DensityKernel Density of Loan/GDPDensityLoan/GDPData Sources: The “China City Statistical Yearbook” (2003-2006).Notes: This figure shows the distribution of Loan/GDP. The x-axis is the Loan/GDPand the y-axis is the density.capital and population on the prefecture level. The coefficient β4 measuresthe exchange rate pass-through differences between the two trade modes infinancially undeveloped prefectures. The coefficient β4 + β7 measures theexchange rate pass-through differences in financially developed prefectures.If Proposition 1 is correct, then β7 should be negative.The result is in Table 2.16. It shows that β7 is negative and signif-icant if firms are state-owned, joint-owned or foreign-owned. Especially,state-owned assembly firms benefit mostly from local financial development.However, it is not significant for private-owned firms.2.7 ConclusionIn this chapter, we discuss the exchange rate pass-through differences be-tween two processing trade modes and find some interesting patterns. Chinese-owned assembly firms have to bear higher exchange rate pass-through thanmultinational firms. However, joint-owned and foreign-owned assembly firm-s bear less exchange rate pass-through. In addition, we find that the importinput ratio, market share and local financial development also affect theexchange rate pass-through for assembly firms.962.7. ConclusionDue to the data limitation, we cannot distinguish the intra-firm transac-tions from inter-firm transactions. Thus, the price differences between thetwo trade modes in this chapter might be accurate. However, if the price ofintra-firm transactions is not related to the exchange rate fluctuations, thenit will not affect the conclusion about exchange rate pass-through.972.7. ConclusionTable 2.16: Exchange Rate Pass-Through and Local Financial DevelopmentDependent Variable: ln(Price)State-owned Private-owned Joint-owned Foreign-ownedln(Exchange Rate) -0.405 0.127 -0.250 -0.079(0.077)*** (0.110) (0.041)*** (0.037)**Trade Mode 0.268 -0.014 0.120 0.070(0.075)*** (0.080) (0.031)*** (0.046)Financial Development -0.034 -0.207 -0.040 -0.114(0.062) (0.084)** (0.030) (0.037)***ln(Exchange Rate) × TradeMode0.024 0.019 0.054 0.076(0.016) (0.018) (0.011)*** (0.009)***ln(Exchange Rate) ×Financial Development0.040 0.035 0.030 0.015(0.013)*** (0.015)** (0.009)*** (0.009)*Trade Mode × FinancialDevelopment0.133 0.240 -0.010 0.128(0.079)* (0.084)*** (0.033) (0.042)***ln(Exchange Rate) × TradeMode× Financial Development-0.073 -0.016 -0.032 -0.038(0.017)*** (0.016) (0.010)*** (0.010)***ln(Quality) 0.020 0.193 0.189 0.209(0.014) (0.019)*** (0.011)*** (0.008)***ln(GDP per Capital) -0.012 -0.394 -0.070 -0.123(0.060) (0.110)*** (0.037)* (0.024)***ln(Population) 0.286 -0.080 0.116 0.055(0.072)*** (0.090) (0.077) (0.026)**Constant 3.053 7.758 4.098 4.496(0.844)*** (1.510)*** (0.624)*** (0.321)***Product-Country FE X X X XYear-Month FE X X X XPrefecture FE X X X XCluster By Product X X X XObservations 349717 158961 1625771 4217880R-squared 0.864 0.783 0.745 0.724Data Sources: The “Chinese Customs Export and Import Database” (2000-2006) and the“China City Statistical Yearbook” (2003-2006).Notes: This table shows the impact of local financial development on the exchange rate-passthrough differences between two trade modes.1. The product is on the HS8 level. The trade mode is a dummy. If the product is tradedunder the “pure assembly” mode, it is 0; otherwise, it is 1.2. The price is Chinese Yuan and the exchange rate is the real exchange rate between thesource of origin and China. An increase in the real exchange rate implies an appreciation ofthe Chinese Yuan.3. The financial development is a dummy. If the firm is located in high financial developmentcounty, it is 1; otherwise, it is 0.4. We exclude intermediary firms, imports from the United States and Hong Kong.5. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.98Chapter 3The Impact of China’s RareEarth Policy on DownstreamIndustries3.1 IntroductionWhen a government intends to protect domestic firms against internationalcompetition, the import tariff is considered to be an efficient tool. However,this tool is not as powerful as before since multilateral trade contracts haveresulted in smaller import tariffs in recent decades. Thus, the governmenthas to choose alternative protectionist trade policies (Anderson and Schmitt,2003; Bown and Crowley, 2014; Garred, 2015; Lima˜o and Tovar, 2011).21The export restriction on raw materials is one of these trade policies, whichis applied principally by developing countries to meet economic and non-economic goals (Bonarriva et al., 2009). In this chapter, we particularlyinvestigate the impact of China’s rare earth policy on Chinese downstreamindustries.Rare earth is a group of metal elements that is essential for many oftoday’s most important high-tech products and technologies. China is themain producer of rare earth, controlling about 97% of the market in 2010(USGS, 2011). Thus, when China greatly decreased the export quota of rareearth in 2010, the price of rare earth in the international market increasedsharply. This had led to serious industry concerns and fears in foreign coun-tries. Although China also decreased the production quota of rare earthin the domestic market, the decline of the production quota was not asmuch as that of the export quota. This discriminating trade policy causeda significant price gap of rare earth between the Chinese market and for-21Rodrik (2004) discusses the range of constraints for less developed countries in currentmultilateral system. He argues that these constraints do not bind and the governmentstill can implement industrial policies by using various tools. These tools include importquotas, anti-dumping legislation, administrative barriers, export subsidies and export re-strictions.993.1. Introductioneign markets. Hence, the downstream firms in China enjoy cost advantagesrelative to foreign competitors. In the short run, this policy could enlargethe market share of Chinese downstream firms. In addition, foreign down-stream firms might have to offshore the production to China in favour of astable supply of rare earth. In the long run, the benefits from the exportrestriction might disappear. First, other countries will produce rare earthto make up for the decreasing import from China. Second, new technologywill be developed and rare earth might be substituted by other materials. Inthis chapter, we narrow our investigation on the short-term effect of China’srare earth policy. Due to the data limitation, the long-term effect is left forfuture work.First, we briefly describe the evolution of China’s rare earth policy andargue that the unexpected decline of export quotas in 2010 is an exogenousshock to downstream firms. The Chinese government believed that theprice of rare earth was too low due to disorderly competition among Chinesemining firms. Thus, the Chinese government attempted to raise the price byregulating supply of rare earth. In 2010, China forced many small domesticmining firms to exit the market and decreased the production quotas. Inparticular, China decreased the export quotas by 40%. As a result, theprices of rare earth experienced a rapid growth in both markets. However,the price in the international market increased faster than the domesticmarket.Second, we use the difference-in-difference (DID) method to examinewhether the exports of downstream products from China benefit from thispolicy. We find that this policy indeed stimulates the export values of rareearth downstream products by 1.25 times relative to that of other similarproducts. This is mainly due to the rise in the price rather than to the quan-tity. In addition, we find that the effects vary across downstream products.Phosphorus, glass polishing, and magnets are the main products that benefitfrom this policy. This policy has no significant effect on other downstreamproducts.Third, we choose magnets as a special case and investigate the impactof export restriction policy in greater detail. Using the DID method, weexamine two aspects: whether Chinese rare earth magnet exports benefitfrom this policy relative to other kinds of magnet exports from China andwhether Chinese rare earth magnet exports benefit from this policy relativeto rare earth magnet exports from other countries. We find positive resultsfor both examinations: the export values of Chinese rare earth magnetsincreased by 44% relative to other kinds of magnets while the export valuesof Chinese metal magnets increased by 43% relative to export values from1003.2. Backgroundother countries. The increase in exports is also due to the rising price, notto the quantity.Finally, we discuss the welfare effect of this policy. We show how welfareis redistributed among domestic rare earth producers and downstream firms,domestic and foreign downstream firms. We also distinguish the short-termand long-term effects.Since the export restriction is rare, very few empirical studies investi-gate this topic.22 Our chapter is mainly related to Garred (2015), who findsthat China increases the export tariff to partly restore its industrial protec-tion policy, which is restricted by the World Trade Organization (WTO).Garred (2015) either uses the firm-level or aggregated product-level data toestimate the impact of export tariff. In this chapter, we focus on rare earthdownstream industries with refined data that allows us to discuss in depththe effect of export restriction policies.In addition, Sanyal et al. (2013) develop a stochastic frontier modellingapproach to the gravity equation for rare earth trade between China andits trading partners. They find that the total export losses of China dueto export restriction almost tripled from 2001 to 2009. Mu¨ller et al. (2012)investigated the share price reactions for Chinese rare earth suppliers, Unit-ed States rare earth users, and the rest of the world rare earth refiners byfocusing on export quota announcements (so-called MOFCOM announce-ments) by China. They do not find a wealth transfer in connection withthe MOFCOM announcements. Apart from these studies, we examine theimpact on downstream firms instead of mining firms.The rest of this chapter is organized as follows. Section 2 introducesthe background of China’s rare earth policy. Section 3 describes the data.Section 4 investigates the impact of China’s rare earth policy on downstreamindustries. Section 5 focuses on a special downstream industry – magnets.Section 6 discusses the welfare and long-term effects. Finally, section 7concludes this chapter.3.2 BackgroundRare earth is a group of metal elements, among which fifteen belong to thechemical group called lanthanides. The other two are yttrium (Y) and s-22Most of these studies focus on the impact of export tax on raw materials producersand government revenue. See Turner et al. (2008) for Russian roundwood, Hasan et al.(2001) for Indonesia palm oil and Warr (2001) for Thailand’s rice and Sanyal et al. (2013)for Chinese rare earth.1013.2. Backgroundcandium (Sc). Lanthanides consist of the following elements: lanthanum(La), cerium (Ce), praseodymium (Pr), neodymium (Nd), promethium (P-m), samarium (Sm), europium (Eu), gadolinium (Gd), terbium (Tb), dys-prosium (Dy), holmium (Ho), erbium (Er), thulium (Tm), ytterbium (Tb),and lutetium (Lu).3.2.1 World Production of Rare EarthFigure 3.1 depicts the estimated world reserve of rare earth in 2010 and Ta-ble 3.1 lists the estimated world production of rare earth from 2008 to 2013.China, Russia and the United States have the largest reserves, which accountfor three quarters of the world reserves. Due to the supply of cheaper rareearth from China, however, Russia and the United States stopped miningrare earth from 2008 to 2011. China, India, Malaysia and Brazil were themain producers of rare earth during that period. In particular, China sup-plied over 95% of the rare earth in the market. In 2010, China implementeda strict control policy on rare earth production and the production of rareearth decreased by 20% in 2011. Other countries such as Australia, Russiaand the United States had to resume the rare earth production to make upfor the decreasing import from China. It usually takes three to five years toimplement new projects of rare earth. Thus, the imbalance between demandand supply continued from 2011 to 2013. The rare earth market is still dom-inated by China: in 2013, the production of China accounted for 86% of theworld production. Thus, China has a market power in rare earth marketand its policy change can significantly affect the supply of rare earth in ashort time. However, in the long run, the supply from other countries canmake up the decline of production in China. China might lose the marketpower at that time.3.2.2 The Value Chain of Rare EarthFigure 3.2 illustrates the value chain of rare earth. We divide the productsinto three categories along the value chain: raw materials, intermediategoods and final goods. We regard the intermediate and final goods as thedownstream goods of rare earth.Raw materials include rare earth ores, rare earth compounds (oxides,chlorides, fluorides and carbonates) and rare earth metals. Rare earth pro-ducers first separate rare earth ores from other ores, then process the oresinto compounds. Depending on the isolation techniques, the compounds arein the form of oxides, chlorides, fluorides or carbonates. Finally, rare earth1023.2. BackgroundFigure 3.1: World Reserve of Rare Earth in 20101.4%48.4%2.7%19.4%16.7%11.4%Australia ChinaIndia Other countriesRussia United StatesData Source: U.S. Geological Survey (USGS), Mineral Commodity Summaries,January 2011.Notes: This figure shows the estimated world reserve of rare earth in 2010. The totalworld reserve is around 110 million metric tons.Figure 3.2: Value Chain of Rare Earth Raw Materials • Rare earth ore • Rare earth compounds (oxide, chlorides, fluorides and carbonates) • Rare earth metal Intermediate Goods • Magnets • Batteries • Metallurgy • Auto catalytic converters • Catalyst refineries • Glass polishing • Glass additives • Phoshpors • Ceramics • Others Final Goods • Advanced optics & other glass products • Audio equipment • Communications & electronics • Consumer electronics • Defense technologies • Oil refining • Electric power • Energy efficient lighting • ... Data Source: “The Economic Benefits of North American Rare Earths Industry” byAmerican Chemistry Council, April 2014.Notes: This figure shows the value chain of rare earth. We divide the products into threecategories along the value chain: raw materials, intermediate goods and final goods.China has the export restriction on the raw materials. We regard the intermediate andfinal goods as the downstream goods of rare earth.1033.2. BackgroundTable 3.1: World Production of Rare Earth (2008-2013)Country Mine Productione2008 2009 2010 2011 2012 2013Australia – – – 2,200 3,200 2,000Brazil 650 550 550 250 140 330China 120,000 129,000 130,000 105,000 100,000 95,000India 2,700 2,700 2,800 2,800 2,900 2,900Malaysia 380 350 300 280 100 180Russia – – – – 2,400 2,500Thailand – – – – – 800United States – – – – 800 5,500Vietnam – – – – 220 220Other NA NA NA NA NA NAWorld Total 124,000 133,000 133,000 111,000 111,000 111,000Data Source: U.S. Geological Survey (USGS), Mineral Commodity Summaries, 2009-2014.Notes: This table shows the estimated world production of rare earth from 2008 to 2013.The mine production is in metric tons.e Estimated. NA Not available. – Zero.metals are purified from these compounds. The Chinese government onlysets up the export quota on the total quantity of raw materials, regardlessof the varieties. Thus, rare earth ores are rarely exported; most rare earthexports are in the form of compounds and metals.Raw materials are mainly used in producing nine kinds of intermediategoods: magnets, batteries, metallurgy, auto catalytic converters, catalystrefineries, glass polishing, glass additives, phosphorus and ceramics (Amer-ican Chemistry Council, 2014). Figure 3.3 shows the amount of rare earthused by these applications in 2014. The usage is measured in metric tonsand the total usage of rare earth was 182, 000 metric tons in 2014. Magnets,glass polishing, batteries and catalyst refineries are the main applications,and use 72% of rare earth. Every application only uses some of the rareearth elements. Table 3.2 shows the detailed usage of rare earth elementsby application. Panel A shows the quantity usage shares, collected by theLynas Corporation Ltd. It shows that lanthanum, cerium, praseodymium,neodymium and yttrium are the most commonly used rare earth elements.Based on the prices of rare earth metals in 2008, we calculated the value us-age shares of rare earth elements. Due to the data limitation, we only have1043.2. BackgroundFigure 3.3: Usage of Rare Earth by Applications in 20147%16%13%2%5%15%28%7%1% 6%Auto catalytic converters BatteriesCatalyst refineries CeramicsGlass additives Glass polishingMagnets MetallurgyOther PhosphorsData Source: Report from Lynas Corporation Ltd.Notes: This figure shows the quantity usage shares of rare earth by applications in 2014.The usage is measured in metric tons and the total usage of rare earth is 182, 000 metrictons.the spot prices of lanthanum, cerium, praseodymium, neodymium, terbium,dysprosium and yttrium in the Shanghai market. Thus, the value usageshares only measure the costs of these seven rare earth elements. We findthat the value share of rare earth is similar to the quantity share.Intermediate goods are used to produce final goods: advanced opticsand other glass products, audio equipment, communications and electron-ics, consumer electronics, defence technologies, oil refining, electric power,energy efficient lighting and so on. Although rare earth are essential formany high-tech final goods, the cost of rare earth only accounts for a smallproportion of total costs, thus, the fluctuation of rare earth prices might nothave a significant impact on the cost of final goods. Yet, the cost of rareearth accounts for a large proportion of the cost of intermediate goods. Forexample, over 30% of the total costs of magnets come from rare earth. Thus,in this chapter we only examine the impact of China’s rare earth policy onintermediate goods.1053.2.BackgroundTable 3.2: Usage Share of Rare Earth Elements by ApplicationsApplication La Ce Pr Nd Sm Eu Gd Tb Dy Y OtherPanel A: Quantity ShareMagnets – – 23.4 69.4 – – 2 0.2 5 – –Batteries 50 33.4 3.3 10 3.3 – – – – – –Metallurgy 26 52 5.5 16.5 – – – – – – –Auto catalytic converters 5 90 2 3 – – – – – – –Catalyst refineries 90 10 – – – – – – – – –Glass polishing 31.5 65 3.5 – – – – – – – –Glass additives 24 66 1 3 – – – – – 2 4Phosphors 8.5 11 – – – 4.9 1.8 4.6 – 69.2 –Ceramics 17 12 6 12 – – – – – 53 –Other 19 39 4 15 2 – 1 – – 19 –Panel B: Value ShareMagnets – – 19.5 59.7 NA NA NA 3.9 16.9 – NABatteries 39.4 22.2 9.3 29.1 NA NA NA – – – NAMetallurgy 17.3 29.2 13.1 40.4 NA NA NA – – – NAAuto catalytic converters 5 76.6 7.2 11.2 NA NA NA – – – NACatalyst refineries 91.4 8.6 – – NA NA NA – – – NAGlass polishing 31.9 55.5 12.6 – NA NA NA – – – NAGlass additives 23.3 54.1 3.5 10.7 NA NA NA – – 8.4 NAPhosphors 1.2 1.3 – – NA NA NA 54.6 – 42.9 NACeramics 5.3 3.1 6.7 13.7 NA NA NA – – 71.2 NAOther 9.3 16.2 7 27.1 NA NA NA – – 40.4 NAData Source: Report from Lynas Corporation Ltd.Notes: This table shows the usage shares of rare earth elements by applications (percentage). We calculate the valueusage share of rare earth based on the prices of rare earth metals in 2008. Due to the data limitation, we only have thespots prices of La, Ce, Pr, Nd, Tb, Dy and Y in Shanghai Market. The price of La is 57,260, Ce is 48,335, Pr is 204,566,Nd is 211,212, Tb is 4,749,570, Dy is 834,423 and Y is 248,466. The price is in Chinese Yuan/metric ton.NA Not available. – Zero. La lanthanum, Ce cerium, Pr praseodymium, Nd neodymium, Sm samarium, Eu europium,Gd gadolinium, Tb terbium, Dy dysprosium, Y yttrium.1063.2. Background3.2.3 China’s Rare Earth PolicyTable 3.3 shows the evolution of China’s rare earth policy. China uses threetools to manage rare earth export: export tariffs, export quotas and exportlicenses.From 1985 to 2003, China encouraged the exporting of rare earth throughvalue-added tax (VAT) rebates. Every firm pays VAT at the rate of 17%. Inorder to boost the exports of some special products, China usually refunds allor part of the VAT to exporting firms of these products. The VAT rebate forrare earth was either 13% (rare earth metal) or 17% (rare earth compounds).Due to the export promotion policy of rare earth, China gradually becamethe largest producer in the rare earth market. From 2004 to 2006, Chinastopped the policy that supported rare earth exports, and the VAT rebateskept decreasing year by year. In January 2004, the VAT rebate of rare earthmetal was abolished and the VAT rebate of rare earth compounds decreasedto 5%. In May 2005, the VAT rebates of all kinds of rare earth elementswere abolished. Since November 2006, China began to restrict the exportingof rare earth by charging export tariffs, which increased year by year. Table3.4 shows the export tariff rates of China’s rare earth elements from 2007 to2011. From 2007 to 2011, the export tariff rates of most rare earth increasedfrom 10% to 25%.In 1999, China began to set the export quota of rare earth. Especially, in2010 the export quota of rare earth decreased almost 40% from the precedingyear. The first column of Table 3.5 shows China’s rare earth export quotasfrom 2000 to 2014. In 2005, the export quota was 65,000 metric tons, butafter 2010 it was only around 30,000. In the domestic market, China alsoset the production quotas of rare earth. Two departments are in charge ofthe rare earth production: Chinese Ministry of Industry and InformationTechnology (MI) and Chinese Ministry of Land and Resources (ML). In2006, the ML began to set production quotas and in 2008, the MI did thesame. The columns 2 and 3 of Table 3.5 show the production quotas set byMI and ML respectively. At first, the production quotas set by these twodepartments were not the same. Thus, the Chinese government could noteffectively regulate rare earth production. In 2010, the Chinese governmentimplemented a strict control policy on rare earth production and these twoministries began to coordinate production quotas each other. In 2010 theproduction quota decreased 20% from the preceding year. After that, theproduction quota remained stable.1073.2.BackgroundTable 3.3: China’s Rare Earth Policy (1985-2015)Period VAT* Rebate Export Tariff Export QuotaPeriod 1: Export Promotion1985 - 1998 13% or 17% - NA1999 - 2003 13% or 17% - 50,000 - 55,000Period 2: Export Control2004 - 2005 5% - 50,000 - 60,0002006 - 2009 - 10%, 15% or 25% 50,000 - 60,000Period 3: Strict Export Control2010 - 2014 - 15% or 25% 30,000Period 4: Neither Export Control Nor Promotion2015 - - NAData Source: China Customs Import and Export Tariff Department and China Ministry of Commerce.Notes: This table shows the evolution of China’s rare earth policy. From 1985 to 2003, the VAT rebate of rare earth is either 13% (rareearth metals) or 17% (rare earth compounds). In January 2004, the VAT rebate of rare earth metal was abolished and the VAT rebate ofrare earth compounds decreased to 5%. In May 2005, the VAT rebates of all kinds of rare earth were abolished. Table 3.4 shows China’srare earth export tariff from 2007 to 2011. Table 3.5 shows China’s rare earth export quotas from 2000 to 2014.* VAT is the abbreviation of “value-added tax”.NA Not available. - Zero.The unit of export quota is in metric tons.1083.2. BackgroundTable 3.4: China’s Rare Earth Export Tariff Rates (2007-2011)Commodity Export Tariff Rate (%)2007 2008 2009 2010 2011Panel A: Rare Earth Ores:Ores of rare earth metals 15 15 15 15 15Panel B: Rare Earth Compounds:Cerium oxide, hydroxide, carbonate andother compounds10 15 15 15 15Yttrium oxide, Europium oxide, Dyxpro-sium oxide, Terbium oxide10 25 25 25 25Lanthanum oxide, Neodymium oxide 10 15 15 15 15Other rare earth oxide 10 15 15 15 15Terbium chloride, Edrophonium chlorde NA 25 25 25 25Lanthanum chloride NA NA NA NA 25Other rare earth chloride 10 15 15 15 15Terbium fluoride, Dysprosium fluoride,Lanthanum fluorideNA NA NA NA 15Other rare earth fluoride 10 15 15 15 15Lanthanum carbonate NA 15 15 15 15Terbium Carbonate, Dysprosium carbon-ateNA 25 25 25 25Other rare earth carbonate 10 15 15 15 15Ferro phosphorus NA NA NA 20 20Containing by weight more than 10% ofrate each elementsNA NA NA NA 25Panel A: Rare Earth Metals:Neodymium 10 15 15 15 25Dysprosium, Terbium 10 25 25 25 25Lanthanum, Cerium NA NA NA NA 25Other rare earth metals, mixed, unmixed 10 25 25 25 25Data Sources: China Customs Import and Export Tariff Department.Notes: This table shows China’s rare earth export tariff rates from 2007 to 2011.1 In order to better control the export of rare earth, China kept creating new HS codes for somevarieties of rare earth from 2007 to 2011. NA Not available. It means that this variety has noHS code at that time.1093.2. BackgroundTable 3.5: China’s Rare Earth Export and Production Quotas (2000-2014)Year Export QuotaMC Production QuotaMI Production QuotaML2000 55,294 NA NA2001 52,941 NA NA2002 NA NA NA2003 47,059 NA NA2004 52,941 NA NA2005 65,680 NA NA2006 61,070 NA 86,6202007 59,643 NA 87,0202008 49,990 119,500 90,1802009 50,146 110,700 87,6202010 30,258 89,200 89,2002011 30,246 93,800 93,8002012 30,996 93,800 93,8002013 30,999 93,800 93,8002014 30,610 105,000 105,000Data Sources: China Ministry of Commerce, China Ministry of Industry andInformation Technology and China Ministry of Land and Resources.Notes: This table shows China’s rare earth export and production quotas from 2000 to2014. The unit is metric tons.MC China Ministry of Commerce. MI China Ministry of Industry and InformationTechnology. ML China Ministry of Land and Resources. NA Not available.Table 3.6: The Number of Firms Eligible to Export (2006-2014)2006 2007 2008 2009 2010 2011 2012 2013 2014Number of Firms 47 41 24 23 22 32 33 29 28Data Sources: China Ministry of Commerce.Notes: This table shows the number of firms that are allowed to export rare earth from 2006 to 2014.In 2011, one of these firms is subject to audit; in 2012, twenty-one of these firms are subject to audit;in 2013, four of these firms are subject to audit.1103.3. DataIn addition, China issues export licenses of rare earth to exporting firms.Usually these firms are large state-owned firms. Table 3.6 shows the numberof firms that are allowed to export rare earth. From 2006 to 2014, thenumber decreased from 47 to 28. By this way, Chinese government couldmore efficiently regular the exports of rare earth.According to the General Agreement on Tariffs and Trade (GATT) 1994Article XI (1),23 export quotas are generally prohibited by the WTO. How-ever, in principle, export tariffs are not prohibited and about one third ofWTO members impose export tariffs (Piermartini, 2004). Despite the risk ofbreaking the rules of the WTO, China cited GATT 1994 Article XX GeneralException (b) and (g)24 to defend its export quantity limitation policy. InMarch 2012, the United States, the European Union (EU) and Japan fileda joint complaint to the WTO that argued that export quotas, tariffs andlimitations imposed by China for rare earth were in breach of its obligationsto the WTO. In March 2014, the WTO ruled in favour of the joint com-plaint, so in 2015 China decided to end rare earth export tariffs and removeexport quotas.3.3 DataIn China, a unified market for rare earth does not exist.25 Rare earth istraded according to market demand and the buyers/sellers themselves ne-gotiate the price. Some companies release the contract prices of rare earthon a daily basis, and traders can use this information as the reference pricefor contracts. In this chapter, we use the spot prices from the Shanghaimarket to proxy the domestic prices of rare earth. The information in thisdatabase was collected by “Steel Home”26 from 2008 to 2015. It coversthe spot prices in the Shanghai market for 10 kinds of rare earth com-pounds and 13 kinds of rare earth metals for every business day. The rare23GATT 1994 Article XI (1): “no prohibitions or restrictions other than duties, taxes orother charges, whether made effective through quotas, import or export licences or othermeasures, shall be instituted or maintained by any contracting party on the importationof any product of the territory of any other contracting party or on the exportation orsale for export of any product destined for the territory of any other contracting party”.24GATT 1994 Article XX General Exception (b): “necessary to protect human, animalor plant life or health” and (g): “relating to the conservation of exhaustible naturalresources if such measures are made effective in conjunction with restrictions on domesticproduction or consumption”.25The rare earth is not an exchange traded commodity. The first exchange for rareearth in China was “Baotou Rare Earth Products Exchange,” which opened in 2014.26This database was obtained from Bloomberg.1113.3. DataFigure 3.4: Export Quantity of Rare Earth (2000-2011)203040506070Quantity2000 2005 2010YearReal Export Export QuotaData Sources: Chinese Customs Export and Import Database (2000-2011).Notes: This figure shows the real export quantities and export quotas of rare earth fromChina. The unit of quantity is in thousand metric tons. Since 2005, the export quotasbegan to decline and the real export declined as well.earth compounds include rare earth carbonate, lanthanum oxide, ceriumoxide, neodymium oxide, praseodymium oxide, terbium oxide, dysprosiumoxide, europium oxide, yttrium oxide, and praseodymium-neodymium ox-ide. The rare earth metals include lanthanum metal, praseodymium met-al, neodymium metal, cerium metal, terbium metal, dysprosium metal, yt-trium metal, praseodymium-neodymium alloy, praseodymium-neodymium-dysprosium, battery grade mischmetal, cerium-rich metal, lanthanum-richmetal, and dy-fe alloy.The second database used in this chapter is the “Chinese Customs Ex-port and Import Database” from 2000 to 2011. This was collected by theChinese Customs Office and includes information on export and import val-ues, as well as volumes categorized by the eight-digit harmonized system (H-S), the exporting country, and the importing country. From this database,we can get the free on board (FOB) prices of rare earth for each country.We also can get the export values and quantities of downstream productsto each country. The third database is the “UN Comtrade Database.” Thisdatabase includes the trade flow information (quantity, value) between eachcountry at the six-digit HS level.Figure 3.4 shows the “real” export quantities27 (reported by Chinese27Here, the “real” export quantity is the quantity reported by Chinese Customs. The1123.3. DataFigure 3.5: Production Quantity of Rare Earth (2000-2014)6080100120140Quantity2000 2005 2010 2015YearProduction Quota by ML Real ProductionProduction Quota by MIData Sources: China Ministry of Industry and Information Technology and ChinaMinistry of Land and Resources.Notes: This figure shows the real production and production quotas of rare earth. Theunit of quantity is in thousand metric tons. In China, two departments (MI and ML) arein charge of the rare earth production. Thus, the management of rare earth productionis inefficient in China. In 2010, China implemented the tougher controls on rare earthproduction and these two departments began to coordinate with each other.MI China Ministry of Industry and Information Technology. ML China Ministry ofLand and Resources.Customs) and export quotas of rare earth from 2000 to 2011. Before 2005,the “real” export quantities were always larger than the export quotas – wethink it is because the export quotas were not strictly implemented. China’srare earth policy at that time was to encourage exporting, thus, customshad no incentive to strictly implement the export quotas. In 2005 China’srare earth policy changed and the export quotas began to decline. At thesame time, the “real” export quantities declined as well, particularly in 2011,when they declined sharply. Thus, we can regard the exporting of rare earthafter 2005 as exogenous: it was determined by the government rather thanfirms. Figure 3.5 shows the real production and production quotas of rareearth. From 2000 to 2005, the real production was increased. Again, thereal production was larger than the production quotas, no matter whether itwas set by the ML or the MI. Since there are two departments (MI and ML)that are in charge of controlling rare earth production, the management ofthis sector is inefficient. Since 2010, China implemented strict productioncontrol and the real production of rare earth began to decline.We observe only one export transaction of rare earth ores from 2000statistics do not include export by smuggling, so the data does not accurately reflect actualquantities.1133.3. DataFigure 3.6: Average Export Price of Rare Earth (2000-2011)050100150200250Average2000 2005 2010YearCompounds MetalsData Sources: Chinese Customs Export and Import Database (2000-2011).Notes: This figure shows the average export prices of rare earth compounds and metals.The unit of price is U.S. dollar/kg.Figure 3.7: Smuggling of Rare Earth (2000-2011)20406080100Quantity2000 2005 2010YearExoprt Quantity Report By ChinaImport Quantity Report By Other CountriesData Sources: UN Comtrade Database (2000-2011).Notes: This figure shows the export quantity of rare earth reported by China and importquantity of rare earth reported by other countries. The gap of these two quantity ismainly due to the smuggling. The unit of quantity is in thousand metric tons.1143.3. Datato 2011. Perhaps, the profits from exporting rare earth ores were too lowand mining firms preferred to export processed rare earth products. Thus,we only calculate the average export prices for rare earth compounds andmetals, shown in Figure 3.6. The price trends for these two kinds of rawmaterials are similar. Since 2005, the average price of rare earth began torise year by year and it increased sharply in 2011. There are two reasonsthat might explain this phenomenon. First, the export quantity reportedby Chinese Customs was not the real export quantity since the smuggling ofrare earth was significant. Figure 3.7 contrasts the export quantity of rareearth reported by China with the import quantity of rare earth reported byother countries. The gap between these two quantities was mainly due tosmuggling: smuggling was serious from 2004 to 2007 and from 2010 to 2011.If we include smuggling, the supply of rare earth from China had increaseduntil 2007. Thus, the price of rare earth did not rise greatly before 2007. In2008 and 2009, the smuggling almost disappeared, which can be explainedby the weak demand. As evidence, Figure 3.6 shows that the price of rareearth declined in 2008 and 2009. In 2010 and 2011, even if we consider thesmuggling of rare earth, the total export quantity was significantly smallerthan before, which explains the rapidly rising prices. Second, the low priceelasticity of rare earth might also explain this price trend. Rare earth isessential for many high-tech products and demand is stable. Before 2010,the total supply of rare earth could meet the demand and thus the priceremained stable, but in 2010, China implemented tough controls restrictingexports of rare earth. This policy led to serious concerns of foreign countries.Since rare earth is difficult to be replaced by other materials, the uncertaintyof China’s rare earth policy makes foreign downstream firms bid much higherprices for rare earth. Thus, together, the low price elasticity and panic moodexplain the rapidly rising prices after 2010.Since the export quotas of rare earth decreased substantially in 2010,firms would have preferred to export rare earth metals than rare earth com-pounds. Figure 3.8 shows the export share of rare earth metals from 2000to 2011. We do not find strong evidence to support this hypothesis. Fig-ure 3.9 shows the export destinations of rare earth: Japan, the EuropeanUnion and the United States are the main destinations. Figure 3.10 showsthe average export price of rare earth across those main destinations. Theprice trends are very similar across destinations. Figures 3.11 and 3.12 showthe distribution of rare earth exports. We find that many small firms wereforced to exit the market due to the stricter regulations. Thus, the benefitsare concentrated to the large firms.Usually a country that has a market power only implements the export1153.3. DataFigure 3.8: Share of Rare Earth Metals (2000-2011)10%40%30%20%Share2000 2005 2010YearQuantity ValueData Sources: Chinese Customs Export and Import Database (2000-2011).Notes: This figure shows the export share of rare earth metals from 2000 to 2011.Figure 3.9: Export Destinations of Rare Earth (2000-2011)020406080Export Quantity2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011EU JapanSouth Korea OthersTaiwan United StatesData Sources: Chinese Customs Export and Import Database (2000-2011).Notes: This figure shows the export destinations of rare earth. The unit of quantity isthousand metric tons.1163.3. DataFigure 3.10: Average Export Price of Rare Earth Across Destinations (2000-2011)050100150200250Average Price2006 2007 2008 2009 2010 2011YearCompound Price0100200300Average Price2006 2007 2008 2009 2010 2011YearEU JapanOthers United StatesMetal PriceData Sources: Chinese Customs Export and Import Database (2000-2011).Notes: This figure shows the average export price of rare earth across destinations. Theunit of quantity is U.S. dollar/kg.1173.3. DataFigure 3.11: Distribution of Rare Earth Export Quantities(2007-2011)0.2.40.2.40 5 10 150 5 10 15 0 5 10 152007 2008 20092010 2011DensityQuantityData Sources: Chinese Customs Export and Import Database (2007-2011).Notes: This figure shows the distribution of rare earth export quantities on the firmlevel. The unit of quantity is in kilograms and we use the logarithm form.Figure 3.12: Distribution of Rare Earth Export Values(2007-2011)0.1.2.3.40.1.2.3.40 5 10 15 200 5 10 15 20 0 5 10 15 202007 2008 20092010 2011DensityValueGraphs by yearData Sources: Chinese Customs Export and Import Database (2007-2011).Notes: This figure shows the distribution of rare earth export values on the firm level.The unit of quantity is in U.S. dollar and we use the logarithm form.1183.4. Empirical Analysisrestriction and no control on the domestic market. Then the price in theinternational market will increase and the price in the domestic market willdecrease. However, since China also regulated the domestic production, theprice of rare earth in the domestic market also increased. We use the spotprices in the Shanghai market to proxy the domestic prices of rare earth.Figure 3.13 shows the prices of selected rare earth elements in the Shanghaimarket from 2008 to 2014. The price in domestic market increased sharplyin 2011 and then decreased gradually. This implies that China’s rare earthpolicy affected the price of rare earth in the short run but had no effectin the long run. Due to the discriminating trade policy, there is a largeprice gap of rare earth between the Chinese market and foreign markets.We use the FOB prices to proxy the international prices. Because of thedata limitation, we only have four kinds of metal prices (lanthanum, cerium,neodymium, yttrium) and one kind of oxide price (neodymium). Figure 3.14shows the FOB/Domestic price ratios. We can find that the FOB prices weremuch higher than those in the domestic market. Specially, the price ratioincreased substantially after 2010.3.4 Empirical AnalysisIn this section, we investigate the impact of China’s rare earth policy ondownstream industries. As shown in Figure 3.2, rare earth is mainly usedto produce nine kinds of intermediate goods. Using the name of the ap-plications as key-words,28 we identified the products that use rare earth asinputs. We ended up with 17 kinds of products at the HS8 level.First, we use the DID method to examine whether China’s rare earthpolicy stimulates the exporting of rare earth related products. We regardthe 17 rare earth related products as the treatment group. Then, within thesame HS4 categories, we pick similar products to form the control group.The control group includes 22 kinds of products.29 Figure 3.15 shows theexporting value trends of the treatment group and the control group. Sincethe variances of the export values are large across products, we normalizethem by using the ratio between current values and export values in 2007.For metallurgy (3606) and ceramic (6906, 6914), we cannot find similarproducts within the same HS4 categories. Thus, we drop these productsfrom our sample. For other applications, the effects of rare earth policy28The key words we used are: “rare earth” + “magnet, battery, metallurgy, auto cat-alytic converters, catalyst refinery, glass polishing, glass additive, phosphor or ceramics”.29In Appendix C.1, we list the HS8 codes for the treatment group and the control group.1193.4. Empirical AnalysisFigure 3.13: Domestic Price (2008-2014)0500100015002008 2010 2012 2014Dy50010001500200025002008 2010 2012 2014Eu050010001500200025002008 2010 2012 2014Tb102030402008 2010 2012 2014Y051015202008 2010 2012 2014La051015202008 2010 2012 2014Ce0501001502002008 2010 2012 2014NdPriceYearOxide MetalData Sources: Steel Home (2008-2014).Notes: This figure shows the domestic price in the Shanghai Market. The unit ofquantity is U.S. dollar/kg.Dy dysprosium, Eu Europium, Tb ytterbium, La Lanthanum, Ce cerium, Nd neodymi-um, Y yttrium.1203.4. Empirical AnalysisFigure 3.14: FOB/Domestic Price Ratio (2008-2011)11.522.532008 2009 2010 2011yearNd22.533.544.52008 2009 2010 2011yearY24682008 2009 2010 2011yearLa2345672008 2009 2010 2011yearCeFOB/Domestic Price RatioYearOxide MetalData Sources: Chinese Customs Export and Import Database (2008-2011) and SteelHome (2008-2011).Notes: This figure shows the FOB/Domestic price ratios.La lanthanum, Ce cerium, Nd neodymium, Y yttrium.are more significant for phosphorus (3206), glass polishing (3405), catalyst(3815), and magnet (8505).Since China’s rare earth policy had a significant effect on export pricesof rare earth after 2010, we divide the time into two periods: 2007-2009 and2010-2011. The benchmark regression is as follows:exportit/exporti,2007 =α0 + α1 treatment dummyi + α2 time dummyt+ α3 treatment dummyi × time dummyt+ it(3.1)The exportit is the export values of product i in year t. The treatment dummyis 1 if the product is related to rare earth; otherwise it is 0. The time dummytis 0 if the year is 2007, 2008 or 2009; otherwise it is 1. We also control theindustry fixed effect on the HS4 level. α3 measures the impact of China’srare earth policy on downstream products.China’s rare earth policy has two effects on domestic downstream firms.On the one hand, domestic downstream firms enjoy cost advantages from thediscriminating trade policy relative to foreign competitors. Thus, this policy1213.4. Empirical AnalysisFigure 3.15: Export Values of Downstream Products (2007-2011)0102030010203001020302007 2008 2009 2010 2011 2007 2008 2009 2010 20112007 2008 2009 2010 2011 2007 2008 2009 2010 2011Phosphor (3206) Glass Polishing (3405) Metallurgy (3606) Catalyst (3815)Ceramic (6902) Ceramic (6906) Ceramic (6914) Magnet (8505)Battery (8506) Battery (8507)Control TreatmentExport Values (2007==1)YearData Sources: Chinese Customs Export and Import Database (2007-2011).Notes: This figure shows the export values of downstream products. We use the exportvalues in 2007 as the benchmark. The numbers within brackets are HS4 code.1223.4. Empirical Analysiscould enlarge the market share of Chinese rare earth downstream products,and the exports will increase relative to other similar products. On the otherhand, the decline of production quotas pushed up domestic prices of rareearth. The input costs of domestic downstream firms also increased, andthey were forced to sell products at higher prices. Then the demand of rareearth downstream products will decrease. In addition, producers of finalgoods will turn to other similar products as substitutes. Thus, the exportsof rare earth downstream products will decrease relative to other similarproducts. The overall effect is ambiguous.The result (see Table 3.7) shows that the export restriction policy stim-ulates the export values by 1.25 times of the export values in 2007 for thetreatment groups. In column 2, we assume the policy was implemented in2011 instead of 2010 and find that the effect is much larger. It shows thatthe export restriction policy stimulates the export values by 2.32 times ofthe export values in 2007 for the treatment groups. This result is consistentwith Figure 3.6, which shows that the price increases was much more sig-nificant in 2011. In column 3, we additionally assume that the policy wasimplemented in 2008 and find no differences between treatment and controlgroups. This implied that there were no significantly differences betweentreatment and control groups in 2008. Their time trends diverged in 2010,when the policy was implemented. In columns 4 to 6, we find that the ex-port restriction policy had no significant effect on the export quantity. Thisimplies that the rise of exports is mainly due to the rising price rather thanquantity. In Table 3.8, we examine the effect at the firm level and the resultsremain consistent.Thus, the positive effect of China’s rare earth policy dominated the neg-ative effect. Rare earth is essential for many high-tech products. The specialproperty of rare earth elements makes it difficult to be replaced. Thus, theprice elastic of rare earth downstream products is low. Even though theprices of rare earth downstream products increased, the demand for themremained stable. Thus, the export price increased but the export quantitydid not decrease.1233.4.EmpiricalAnalysisTable 3.7: The Impact of China’s Rare Earth Policy on Exporting of Downstream Products: Product Levelexport value/export value2007 export quantity/export quantity2007(1) (2) (3) (4) (5) (6)treatment dummy 0.00834 0.0388 -0.0408* 0.117 0.0789 -0.0358(0.134) (0.146) (0.0191) (0.294) (0.226) (0.0211)time dummy 0.530*** 0.636*** 0.365 0.0871 0.0883 -0.0560(0.111) (0.115) (0.315) (0.192) (0.133) (0.187)treatment dummy × time dummy 1.254* 2.321* 0.0311 -0.0777 0.0422 0.926(0.545) (1.163) (0.390) (0.428) (0.312) (0.744)Constant 1.089*** 1.174*** 1.022*** 1.067*** 1.084*** 1.021***(0.115) (0.0961) (0.112) (0.0915) (0.0706) (0.176)Industry FE X X X X X XCluster by Industry X X X X X XPolicy Not Implemented 2007-2009 2007-2010 2007 2007-2009 2007-2010 2007Policy Implemented 2010-2011 2011 2008 2010-2011 2011 2008Observations 134 134 55 134 134 55R-squared 0.244 0.360 0.243 0.006 0.006 0.226Data Sources: Chinese Customs Export and Import Database (2007-2011)Notes: This table shows the impact of rare earth policy on the exports of downstream products in China. We use the exports in 2007 asthe benchmark.1 The treatment group is the rare earth related products. The control group is other products within the same HS4 code.2 Standard errors in parentheses. **Significant at 5%; ***significant at 1%.1243.4.EmpiricalAnalysisTable 3.8: The Impact of China’s Rare Earth Policy on Exporting of Downstream Products: Firm Levelln(export value) ln(export quantity)(1) (2) (3) (4) (5) (6) (7) (8)treatment dummy 1.277*** 0.518 -0.549* -1.430*** 0.0342 0.0364 -0.158*** -0.155***(0.356) (0.348) (0.266) (0.0904) (0.0393) (0.0434) (0.0437) (0.0465)time dummy 0.0618 0.0574 -0.0611 -0.0676 1.368*** 0.612* -0.464 -1.342***(0.0426) (0.0452) (0.0403) (0.0457) (0.385) (0.343) (0.275) (0.0609)treatment dummy × time dummy 0.551** 0.555* 0.356* 0.363* 0.647* 0.645 0.288 0.285(0.254) (0.277) (0.171) (0.188) (0.341) (0.373) (0.189) (0.206)Constant 11.28*** 10.82*** 9.595*** 10.44*** 11.29*** 10.84*** 9.602*** 10.44***(0.103) (0.865) (0.253) (1.074) (0.0940) (0.868) (0.253) (1.073)Industry FE X X X X X X X XFirm FE X X X XCluster by Product X X X X X X X XPolicy Not Implemented 2007-2009 2007-2009 2007-2010 2007-2010 2007-2009 2007-2009 2007-2010 2007-2010Policy Implemented 2010-2011 2010-2011 2011 2011 2010-2011 2010-2011 2011 2011Observations 4,184 4,184 4,184 4,184 4,184 4,184 4,184 4,184R-squared 0.088 0.751 0.009 0.757 0.087 0.750 0.009 0.757Data Sources: Chinese Customs Export and Import Database (2007-2011)Notes: This table shows the impact of rare earth policy on the exports of downstream products in China.1 The treatment group is the rare earth related products. The control group is other products within the same HS4 code.2 Standard errors in parentheses. **Significant at 5%; ***significant at 1%.1253.5. Case Study: Magnets3.5 Case Study: MagnetsIn last section, we used key-words to identify the downstream productsrelated to rare earth, but this method is not accurate. In this section,we choose one special downstream product – magnets and investigate theimpact of China’s rare earth policy in greater detail. Magnets are the mainapplication of rare earth. According to Figure 3.3, it consumes 28% of rareearth in 2014. In the Chinese Customs HS Code, there are three kinds ofpermanent magnets: “rare earth” (85051110), “other metal” (85051190) and“nonmetal” (85051900). Only the first one uses rare earth as input.Figure 3.16 shows the export values of three permanent magnets from2000 to 2011. Before 2010, the growth trends of these three permanent mag-nets were very similar, but since 2010, the growth of rare earth permanentmagnet has been much faster than that of the other two permanent mag-nets. Thus, the producers of rare earth permanent magnet in China indeedbenefited from the decline of rare earth export quotas. Figure 3.17 showsthe export quantities of three permanent magnets from 2000 to 2011. Wefind that there are no significant differences between the three permanentmagnets.First, we use the DID method to examine whether China’s rare earthpolicy stimulates the exports of the rare earth permanent magnet producersin China relative to other magnet producers. The rare earth permanentmagnet producers constitute the treatment group and other two permanentmagnets producers form the control group. We divide the time into twoperiods: 2008-2009 and 2010-2011. The benchmark regression is as follows:ln(exportijkt) =β0 + β1 treatment dummyi + β2 time dummyt+ β3 treatment dummyi × time dummyt+ φjt + θk + ijkt(3.2)The exportijkt is the export value of product i from firm k to country j atyear t. The treatment dummy is 1 if the product is a rare earth permanentmagnet; otherwise it is 0. The time dummyt is 0 if year is 2008 or 2009;otherwise, it is 1. φjt is the country-year fixed effect and θk is the firm fixedeffect. β3 measures the impact of China’s rare earth policy on domestic rareearth permanent magnet producers.The result is shown in Table 3.9. The first three columns show the effecton export values and the remaining columns show the effect on export quan-tities. Column 1 and 4 use both metal permanent magnets and nonmetalpermanent magnets as the control group. Column 2 and 5 only use other1263.5. Case Study: MagnetsFigure 3.16: Export Values of Three Permanent Magnets (2000-2011)0123Export Value2000 2005 2010YearRare Earth Permanent Magnet Metal Permanent MagnetNonmetal Permanent MagnetData Sources: Chinese Customs Export and Import Database (2000-2011).Notes: This figure shows the export values of three permanent magnets. The values arein billion U.S. dollars.Figure 3.17: Export Quantity of Three Permanent Magnets (2000-2011)0.05.1.15Export Quantity2000 2005 2010YearRare Earth Permanent Magnet Metal Permanent MagnetNonmetal Permanent MagnetData Sources: Chinese Customs Export and Import Database (2000-2011).Notes: This figure shows the export quantities of three permanent magnets. The quan-tity are in million metric tons.1273.5. Case Study: Magnetsmetal permanent magnets as the control group. Column 3 and 6 only usenonmetal permanent magnets as the control group. The rare earth poli-cy raises the export values by 43%∼57%, depending on the control group.However, the rare earth policy only stimulates the export quantity whenwe use metal permanent magnet as the control group: it raises the exportquantity by 14%. Since the prices of rare earth increased dramatically, theprices of rare earth permanent magnets also increased. This explains whythe growth of export values is faster than the growth in quantity.Second, we use the DID method to examine whether China’s rare earthpolicy stimulates the exports by rare earth permanent magnet producersin China relative to producers in other countries. We obtain trade flowdata from the “UN Comtrade Database.” It includes the import and exportvalues and quantities on the HS6 level. The code for permanent magnetsis 850511, which includes both “rare earth permanent magnet” (85051110)and “other metal permanent magnet” (85051190). Since the export quotawould benefit Chinese firms, China is assigned to the treatment group andother countries are assigned to the control group. Then we divide the timeinto two periods: 2008-2009 and 2010-2011. The benchmark regression is asfollows:ln(exportijt) =β0 + β1 treatment dummyi + β2 time dummyt+ β3 treatment dummyi × time dummyt+ φi + θj + ρt + ijt(3.3)The exportijt is the export values from country i to country j at year t.The treatment dummy is 1 if the export country is China; otherwise it is0. The time dummyt is 0 if the year is 2007, 2008 or 2009; otherwise, it is1. φi is the export country fixed effect and θj is the import country fixedeffect. ρt is the time fixed effect. β3 measures the impact of China’s rareearth policy on domestic rare earth permanent magnet producers relative toforeign competitors.The result is shown in Table 3.10. We find that the exporting restrictionindeed stimulates the exports by Chinese firms, but it is only significantfor the export values and not for the export quantity. The export valuesof Chinese metal magnet increase 43% relative to export values from othercountries. In addition, if we assume the policy was implemented in 2011,the impact is not significant.1283.5.CaseStudy:MagnetsTable 3.9: The Impact of China’s Rare Earth Policy on Domestic Rare Earth Magnet Producers Relative to OtherMagnet Producersln(export value) ln(export quantity)(1) (2) (3) (4) (5) (6)treatment dummy 0.238*** 0.158 0.387*** -1.083*** -0.832*** -1.631***(0.0859) (0.0989) (0.101) (0.0896) (0.0991) (0.109)time dummy 0.495*** 0.871*** 0.219 -0.0222 0.0305 -0.137(0.114) (0.190) (0.155) (0.118) (0.179) (0.145)treatment dummy × time dummy 0.366*** 0.362*** 0.455*** 0.0768 0.134** 0.0347(0.0557) (0.0674) (0.0687) (0.0566) (0.0642) (0.0729)Constant 8.161*** 7.819*** 8.570*** 6.512*** 5.985*** 7.190***(0.101) (0.158) (0.126) (0.110) (0.150) (0.124)Country-Year FE X X X X X XFirm FE X X X X X XCluster by Country X X X X X XControl Group Both Metal Nonmetal Both Metal NonmetalObservations 47,010 30,583 23,994 47,010 30,583 23,994R-squared 0.567 0.593 0.584 0.529 0.527 0.568Data Sources: Chinese Customs Export and Import Database (2008-2011)Notes: This table shows the impact of rare earth policy on the domestic rare earth permanent magnet exporting relative othermagnet exporting in China.1 The treatment group is Chinese rare earth permanent magnet export. The control group is other permanent magnet export.2 Standard errors in parentheses. **Significant at 5%; ***significant at 1%.1293.5.CaseStudy:MagnetsTable 3.10: The Impact of China’s Rare Earth Policy on Chinese Metal Magnet Producers Relative to OtherCountries’ Metal Magnet Producersln(export value) ln(export quantity)(1) (2) (3) (4) (5) (6)treatment dummy 3.579*** 3.676*** 3.708*** 5.066*** 5.149*** 5.378***(0.210) (0.201) (0.248) (0.223) (0.211) (0.271)treatment dummy × time dummy 0.362** 0.261 -0.215 0.0935 -0.204 -0.471**(0.158) (0.194) (0.207) (0.167) (0.186) (0.221)Constant 9.042*** 9.042*** 9.237*** 5.620*** 5.620*** 5.855***(0.0142) (0.0141) (0.0154) (0.0145) (0.0144) (0.0169)Import Country-Year FE X X X X X XCluster by Import Country X X X X X XPolicy Not Implemented 2007-2009 2007-2010 2007 2007-2009 2007-2010 2007Policy Implemented 2010-2011 2011 2008 2010-2011 2011 2008Observations 8,696 8,696 3,207 8,696 8,696 3,207R-squared 0.254 0.254 0.264 0.271 0.271 0.278Data Sources: UN Comtrade Database (2008-2011)Notes: This table shows the impact of rare earth policy on Chinese rare earth permanent magnet exporting relative to other countries’rare earth permanent magnet exporting.1 The treatment group is Chinese metal permanent magnet export. The control group is other countries’ metal permanent magnetexport.2 Standard errors in parentheses. **Significant at 5%; ***significant at 1%.1303.6. Discussion3.6 Discussion3.6.1 Welfare AnalysisIn this section, we discuss how welfare is redistributed among domestic rareearth producers and downstream firms, domestic and foreign downstreamfirms.Before 2006, the market of rare earth was a free competition market. Therare earth producers only gained little profits due to the tough competition.Since the Chinese government began to regulate the production of rare earth,rare earth producers could charge higher prices. Thus, the total welfare ofrare earth producers was improved. However, many small mining firmswere forced to exit the market and the production were concentrated tolarge firms. Thus, within the rare earth industry, China’s rare earth policyhad the positive effect on the welfare of large firms but the negative effecton the welfare of small firms.As we have showed, China’s rare earth policy promoted the export valuesof domestic downstream firms. Since the input costs also increased, theimpact on the profits was ambiguous. We will investigate this issue in futurework. In addition, the market shares among downstream firms were indeedredistributed. Chinese downstream firms obtained larger market shares thanbefore. Thus, in term of market share, Chinese downstream firms gainedwelfare improvement and foreign downstream firms bore welfare loss.3.6.2 Long-Term EffectIn the short run, Chinese downstream firms benefit from the export re-striction policy. This is because the supply of rare earth and demand ofrare earth downstream products are inelastic in a short time. As Table 3.1shown, other countries will produce rare earth to make up the decline ofimports from China. Thus, in the long run, the supply of rare earth willbecome more elastic. In addition, as the development of new technology,new substitutes will partially replace rare earth. Thus, in the long run, thedemand will also become more elastic. When the supply increases and thedemand decreases, the price of rare earth will fall. As Figure 3.13 shown,the price of rare earth in 2014 almost fell back to the price level in 2009.Thus, in the long run, when China lost its market power in the rare earthmarket, then the export restriction policy would have negative effect on bothdomestic rare earth producers and downstream firms.1313.7. Conclusion3.7 ConclusionIn this chapter, we discuss the impact of China’s rare earth policy on down-stream industries. When China implemented tough restrictions on rare earthexporting in 2010, it caused serious fears in foreign countries and a hugeprice rise in the international market. First, we find that the exports ofChinese rare earth downstream firms increased significantly relative to oth-er Chinese firms. In addition, we find that the increase of exports was dueto price rather than quantity. Second, we focus on a typical downstreamproduct – magnets. We find that exports from Chinese metal permanentmagnet producers also benefited from this policy relative to exports fromother countries.Due to the data limitation, we can only discuss the exports of firms. Infuture work, we have three potential extensions. First, we could investigatethe impact of the export restriction policy on firms’ profits and investments.Second, we could examine whether this policy attract FDI in rare earthrelated industries. Third, as we have only examined the short-term effect,we would investigate the long-term effect in future work.132BibliographyAllayannis, G., Ihrig, J., and Weston, J. P. (2001). Exchange-Rate Hedg-ing: Financial versus Operational Strategies. American Economic Review,91(2):391–395.American Chemistry Council (2014). The Economic Benefits of the NorthAmerican Rare Earths Industry. Technical Report April.Anderson, S. P. and Schmitt, N. (2003). Nontariff Barriers and Trade Lib-eralization. Economic Inquiry, 41(1):80–97.Arkolakis, C. (2010). Market Penetration Costs and the New ConsumersMargin in International Trade.Bai, X., Krishna, K., and Ma, H. (2013). How you export matters: exportmode, learning and productivity in China. manuscript.Baldwin, R. and Krugman, P. (1989). Persistent Trade Effects of LargeExchange Rate Shocks. Quarterly Journal of Economics, 104(4):635–654.Bartram, S. M., Brown, G. W., and Minton, B. a. (2010). Resolving theexposure puzzle: The many facets of exchange rate exposure. Journal ofFinancial Economics, 95(2):148–173.Bernard, A., Eaton, J., Jenson, J., and Kortum, S. (2000). Plants andproductivity in international trade. American Economic Review.Bonarriva, J., Koscielski, M., and Wilson, E. (2009). Export Controls: anoverview of their use, economic effects, and treatment in the global tradingsystem. Office of Industries Working Paper.Bown, C. P. and Crowley, M. a. (2014). Emerging Economies, Trade Policy,and Macroeconomic Shocks. Journal of Development Economics, 111:261–273.Crino`, R. and Epifani, P. (2012). Productivity, quality and export behaviour.The Economic Journal, 122:1206–1243.133BibliographyDai, M., Maitra, M., and Yu, M. (2011). Unexceptional exporter perfor-mance in China? The role of processing trade. Manuscript, ColumbiaUniversity.Defever, F. and Rian˜o, A. (2012). China’s pure exporter subsidies.manuscript.Do¨hring, B. (2008). Hedging and invoicing strategies to reduce exchangerate exposure-a euro-area perspective. Economics papers.Eaton, J., Kortum, S., and Kramarz, F. (2011). An anatomy of internationaltrade: Evidence from French firms. Econometrica, 79(5):1453–1498.Feenstra, R. and Hanson, G. (2005). Ownership and control in outsourcingto China: estimating the property-rights theory of the firm. The QuarterlyJournal of Economics, 120(2):729–761.Fernandes, A. P. and Tang, H. (2012). Determinants of vertical integra-tion in export processing: Theory and evidence from China. Journal ofDevelopment Economics, 99(2):396–414.Froot, K. and Klemperer, P. (1989). Exchange rate pass-through whenmarket share matters. The American Economic Review.Garred, J. (2015). Capturing the Value Chain : The Persistence of TradePolicy in China After WTO Accession.Goldberg, L. and Tille, C. C. (2013). A Bargaining Theory of Trade Invoicingand Pricing. NBER Working Paper Series.Hasan, M. F., Reed, M. R., and Marchant, M. a. (2001). Effects of an ExportTax on Competitiveness : The Case of the Indonesian Palm Oil Industry.Journal of Economic Development, 26(2):77–90.He´ricourt, J. and Poncet, S. (2013). Exchange rate volatility, financial con-straints and trade: empirical evidence from Chinese firms. CESifo Work-ing Paper.Levinsohn, J. and Petrin, A. (2003). Estimating production functions usinginputs to control for unobservables. The Review of Economic Studies,70(2):317–341.Lima˜o, N. and Tovar, P. (2011). Policy choice: Theory and evidence fromcommitment via international trade agreements. Journal of InternationalEconomics, 85(2):186–205.134BibliographyLu, D. (2010). Exceptional exporter performance? Evidence from Chinesemanufacturing firms. Manuscript, University of Chicago.Lu, J., Lu, Y., and Tao, Z. (2010). Exporting behavior of foreign affiliates:Theory and evidence. Journal of International Economics, 81(2):197–205.Lynas Corporation Ltd (2010). ASX Mid Caps Conference Presentation.Technical report.Manova, K. and Yu, Z. (2012). Firms and credit constraints along the value-added chain: processing trade in China. NBER Working Paper Series.Manova, K. and Zhang, Z. (2012). Export prices across firms and destina-tions. The Quarterly Journal of Economics, 127(1):379–436.Melitz, M., Mayer, T., and Ottaviano, G. I. (2014). Market size, compe-tition, and the product mix of exporters. American Economic Review,104:495–536.Melitz, M. J. (2003). The impact of trade on intra-industry reallocationsand aggregate industry productivity. Econometrica, 71(6):1695–1725.Melitz, M. J. and Ottaviano, G. I. P. (2008). Market size, trade, and pro-ductivity. Review of Economic Studies, 75(1):295–316.Mu¨ller, M., Schweizer, D., and Seiler, V. (2012). Wealth Effects of RareEarth Prices and Chinas Rare Earth Elements Policy.Piermartini, R. (2004). The Role of Export Taxes in the Field of Primary.Rodrik, D. (2004). Industrial policy for the twenty-first century. TechnicalReport September.Sanyal, P., Brady, P. V., and Vugrin, E. D. (2013). The Impact of TradeCosts on Rare Earth Exports : A Stochastic Frontier Estimation Ap-proach. 2013.Soderbery, A. (2013). Estimating import supply and demand elasticities:analysis and implications. Available at SSRN 2272749.Strasser, G. (2013). Exchange rate pass-through and credit constraints.Journal of Monetary Economics, 60(1):25–38.Takatoshi, I., Satoshi, K., Kiyotaka, S., and Junko, S. (2013). Exchange RateRisk Management of Export Firms: New findings from a questionnairesurvey. RIETI Discussion Paper Series.135Tse, B. P.-k. and Survey, U. S. G. (2011). China s Rare-Earth Industry.Turner, J. a., Buongiorno, J., Katz, A., and Zhu, S. (2008). Implicationsof the Russian roundwood export tax for the Russian and global woodproducts sectors. Scandinavian Journal of Forest Research, 23(2):154–166.USGS (2009). U.S. Geological Survey Mineral Commodity Summaries 2009.Technical report.USGS (2010). U.S. Geological Survey Mineral Commodity Summaries 2010.Technical report.USGS (2011). U.S. Geological Survey Mineral Commodity Summaries 2011.Technical report.USGS (2012). U.S. Geological Survey Mineral Commodity Summaries 2012.Technical report.USGS (2013). U.S. Geological Survey Mineral Commodity Summaries 2013.Technical report.USGS (2014). U.S. Geological Survey Mineral Commodity Summaries 2014.Technical report.Wang, Z. and Yu, Z. (2012). Trading partners, traded products and firmperformances of Chinas exporter-importers: Does processing trade makea difference? The World Economy, 35(12):1795–1824.Warr, P. G. (2001). Welfare Effects of an Export Tax: Thailand’s RicePremium. American Journal of Agricultural Economics, 83(4):903–920.Yue, V. and Ju, J. (2013). A unified model of structural adjustments and in-ternational trade: theory and evidence from China. 2013 Meeting Papers,2012:1–54.Zepf, V. (2010). Rare Earth Elements: A New Approach to the Nexus ofSupply, Demand and Use. PhD thesis.136Appendix AAppendix to Chapter 1A.1 Industry ListThere are 30 manufacturing industries at two-digit level. The name list ofthese industries is as follows:1. Manufacture of Leather, Fur Apparel, Feather and Products (3.40%30)2. Manufacture of Wearing Apparel, Footwear and Headwear (8.01%)3. Manufacture of Arts and Crafts and Other Manufacturing (2.91%)4. Manufacture of Cultural, Educational and Sporting Products (2.38%)5. Furniture Manufacturing (1.38%)6. Manufacture of Instruments and Appliances, Culture-related and Of-fice Machinery (2.02%)7. Manufacture of Metal Products (5.62%)8. Manufacture of Textile (12.22%)9. Recycling of Waste and Scrap (0.02%)10. Universal Equipments Manufacturing (7.67%)11. Manufacture of Special Equipments (4.03%)12. Manufacture of Wood and Articles of Wood, Bamboo, Bine, PalmFibre, Straw and Grass (1.74%)13. Manufacture of Electric Machines and Equipments (6.82%)14. Manufacture of Plastic Products (4.53%)15. Manufacture of Rubber Products (1.35%)30The number in the bracket is the proportion of firms in this industry.137A.2. Productivity and Export/Domestic Sales Ratio: Quadratic Regression16. Processing Industry of Agricultural and Subsidiary Food (1.30%)17. Manufacture of Food Products (2.09%)18. Manufacture of Non-metal Products (4.54%)19. Printing and Reproduction of Recorded Media (0.76%)20. Manufacture of Telecommunication Equipments, Computers and Oth-er Electric Equipments (5.94%)21. Manufacture of Transportation Equipments (4.31%)22. Manufacture of Pharmaceuticals (2.02%)23. Manufacture of Drinking Products (0.87%)24. Manufacture of Pulp, Paper, Paperboard and Articles of Paper andPaperboard (1.47%)25. Chemical Raw Materials and Manufacture of Other Basic ChemicalRaw Materials (6.50%)26. Manufacture and Casting of Non-ferrous Metals (1.19%)27. Manufacture and Casting of Ferrous Metals (1.02%)28. Manufacture of Chemical Fibres (0.41%)29. Manufacture of Tobacco Products (0.11%)30. Processing of Crude Oil, Coking and Nuclear Fuel (0.42%)A.2 Productivity and Export/Domestic SalesRatio: Quadratic RegressionFigure 1.3 showed that the correlation between firms’ productivity and theirexport intensity is an inverted U-shaped curve. However, the positive cor-relation before the turning point is not significant. Thus, we use a linearregression model to describe the correlation. In this section we instead usea quadratic regression to address the inverted U-shaped correlation. Theregression can be written as follows:ln(Export/Domestic Sales Ratioijkt) = η0 + η1 ln(Pijkt) + η2 ln(Pijkt)2+ other controls+ µjkt + ijkt(A.1)138A.2. Productivity and Export/Domestic Sales Ratio: Quadratic RegressionThe results are shown in Tables A.1 and A.2. We find that the coefficientof ln(Productivity)2 is always negative and significant. But the coefficientof ln(Productivity) is only significant when we use a firm’s TFP as ourmeasurement and the regression is at the firm level. Thus, a linear regressionis a better way to address the correlation between the firms’ export/domesticsales ratio and their productivity.Table A.1: Export/Domestic Sales Ratio and Productivity: Firm LevelDependent Variable: ln(Export/Domestic Sales)(1) (2) (3) (4)ln(Productivity) 0.0114 0.0225 0.189*** 0.459***(0.0301) (0.0650) (0.0700) (0.121)ln(Productivity)2 -0.0220*** -0.0117 -0.0359*** -0.0547***(0.00427) (0.00972) (0.00579) (0.00999)ln(Capital/Labour Ratio) -0.165*** -0.223*** -0.221*** -0.255***(0.0154) (0.0211) (0.0157) (0.0226)ln(Sale) -0.0218 -0.208*** 0.172*** 0.0141(0.0269) (0.0281) (0.0380) (0.0422)Constant 0.467* 2.456*** -1.218*** -0.367(0.271) (0.325) (0.376) (0.608)Ownership FE X X X XProvince-Industry-Year FE X X X XCluster By Industry X X X XExclude Processing Trade X XObservations 275,872 69,907 275,872 69,907R-squared 0.369 0.472 0.370 0.474Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the“Chinese Customs Export and Import Database” (2000-2006).Notes: This table shows the correlation between export/domestic sales ratio and pro-ductivity at the firm level.1. We use value-added per worker to measure a firm’s productivity in the first two anduse TFP in last two columns.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.139A.2. Productivity and Export/Domestic Sales Ratio: Quadratic RegressionTable A.2: Export/Domestic Sales Ratio and Productivity: Firm-Destination LevelDependent Variable: ln(Export/Domestic Sale)(1) (2) (3) (4)ln(Productivity) 0.0440 -0.0137 0.225 0.318(0.0672) (0.117) (0.138) (0.223)ln(Productivity)2 -0.0160* 0.00519 -0.0336*** -0.0325*(0.00962) (0.0172) (0.0103) (0.0179)ln(Capital/Labour Ratio) -0.145*** -0.202*** -0.183*** -0.203***(0.0233) (0.0311) (0.0226) (0.0320)ln(Sale) -0.338*** -0.528*** -0.153*** -0.400***(0.0386) (0.0368) (0.0528) (0.0523)ln(Destination Number) -0.0780 -0.0283 -0.0829* -0.0375(0.0496) (0.0548) (0.0496) (0.0546)Constant 0.782** 2.781*** -1.016* 0.864(0.393) (0.437) (0.578) (0.887)Ownership FE X X X XDestination-Province-Industry-Year FE X X X XCluster By Industry X X X XExclude Processing Trade X XObservations 1,098,287 560,850 1,098,287 560,850R-squared 0.585 0.649 0.586 0.650Data Sources: The “Chinese Industrial Enterprises Database” (2000-2006) and the “ChineseCustoms Export and Import Database” (2000-2006).Notes: This table shows the correlation between export/domestic sales ratio and productivity atthe firm-destination level.1. We use value-added per worker to measure a firm’s productivity in the first two and use TFPin last two columns.2. Standard errors in parentheses. **Significant at 5%; ***significant at 1%.140Appendix BAppendix to Chapter 2B.1 Two Stages ApproachIn this section, we use a two-stages approach to test the Proposition 1. Inthe first stage, we run Regression 2.1 for each prefecture n at year t to getα2,nt. Then in the second stage, we use local financial development FinDntto explain α2,nt.α2,nt =γ1 + γ2FinDnt + nt (B.1)In the second stage, we have to correct heteroscedasticity problem usinga weighted least squares (WLS) method. The standard errors of the coeffi-cients α2,nt in the first stage are used as weights to run Regression B.1. Ifthe Proposition 1 is correct, then γ2 should be negative.Some prefectures only have a small number of processing trade transac-tions. The first stage is not credible for these places. Thus, we only keepthose prefectures whose transaction numbers are larger than 5000 per year.After processing the data, we have 147 prefecture-year pairs left.The result is in Table B.1. In column 1, we could see that the localfinancial development indeed increases the exchange rate pass-through. Incolumn 2, we additional control the year fixed effect and the result is robust.Thus in column 3, we control GDP per capital. However, the impact onexchange rate pass-through is not significant. We think it might becauseGDP per capital has a strong correlation with local financial development.When we control both of them, the effect of local financial developmentbecomes insignificant.141B.1. Two Stages ApproachTable B.1: Exchange Rate Pass-Through and Local Financial Development:Two Stages MethodExchange Rate Pass-Through Coefficient(1) (2) (3)Loan/GDP -0.018 -0.016 -0.013(0.008)** (0.009)* (0.010)GDP per Capital -0.007(0.010)Constant 0.025 0.011 0.077(0.010)** (0.014) (0.100)Year FE X XObservations 147 147 147R-squared 0.030 0.047 0.050Data Sources: The “Chinese Customs Export and Import Database”(2000-2006) and the “China City Statistical Yearbook” (2003-2006).Notes: This table shows the impact of local financial development on theexchange rate pass-through differences between two trade modes using thetwo stages approach.1. Standard errors in parentheses. **Significant at 5%; ***significant at1%.142Appendix CAppendix to Chapter 3C.1 The Treatment and Control GroupsTable C.1 lists the HS8 codes of downstream products for the treatment andcontrol groups.Table C.1: The Treatment and Control GroupsIntermediate Goods Treatment ControlMagnet 85051110 85051190, 85051900Battery 85065000, 85068000, 85061011, 85061012,85073000, 85074000, 85061019, 85061090,85075000, 85076000 85063000, 85064000,85066000, 85071000,85072000, 85078030,85078090Ceramic 69141000, 69149000, 69021000, 6902200069029000, 69060000Phosphor 32065000 32042000Glass Polishing 34059000 34051000, 34052000,34053000, 34054000Catalyst 38159000, 38151900 38151100, 38151200Metallurgy 36069011, 36069019Notes: This table shows the HS8 codes for the treatment and control groups.143