UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

How Chinese firms link with foreign markets Sun, Xiaonan 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2017_september_sun_xiaonan.pdf [ 1.59MB ]
JSON: 24-1.0354484.json
JSON-LD: 24-1.0354484-ld.json
RDF/XML (Pretty): 24-1.0354484-rdf.xml
RDF/JSON: 24-1.0354484-rdf.json
Turtle: 24-1.0354484-turtle.txt
N-Triples: 24-1.0354484-rdf-ntriples.txt
Original Record: 24-1.0354484-source.json
Full Text

Full Text

How Chinese Firms Link with ForeignMarketsbyXiaonan SunB.A. (Economics), University of International Business and Economics, 2009M.A. (Economics), University of International Business and Economics, 2011A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Business Administration)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)August 2017c© Xiaonan Sun 2017AbstractThis dissertation consists of three essays, each of which studies a unique perspective of how Chinesefirms participate in foreign markets. The first essay investigates a government regulation thatrestricts the use of intermediaries linking domestic producers and foreign buyers. A model isdeveloped to describe the matches between automakers and intermediaries. It shows that theregulation leads to market division, inefficiencies in matching, and double marginalization. Themodel predictions coincide with a number of stylized facts: a strong decline in the number ofauto intermediaries, assortative matching, export price increases for intermediaries, and substantialchurning in the sets of intermediaries registered by the automakers. Welfare analysis shows thatthis regulation benefits automakers while intermediaries are made worse off.The second essay studies the sales allocation of Chinese exporters between domestic and foreignmarkets. The exports to domestic sales ratio is decreasing in firm productivity. Heterogeneousmarketing cost elasticities are introduced in the model to rationalize this empirical phenomenon.A higher marketing cost elasticity domestically gives rise to a larger sales expansion in the homemarket as firm productivity increases. Empirical tests of the model predictions provide evidenceto support the role of marketing costs and are inconsistent with alternative explanations related tovariable markups and product quality.The third essay examines the persistence of trade relationships from a historical point of view. Itstudies the effect of treaty linkages established between Chinese cities and foreign countries duringthe 19th century on China’s trade today. Evidence shows that trade is higher among the group ofcountries and cities that were involved in treaty arrangements. These higher levels of trade maybe due to complementary in industry structure and business knowledge developed in the treatyport era. An alternative explanation is that today’s trade is due to the relatively high levels ofdevelopment of the trading partners and is unrelated to treaty ports.iiLay SummaryThis dissertation studies how Chinese firms participate in foreign markets from three perspectives:Regulation limiting exports through trade intermediaries, relative sales to domestic and foreignmarkets, and historical linkages through treaty agreements. First, a model describing the matchesbetween automakers and intermediaries demonstrates that a Chinese registration requirement couldhave created inefficiencies. Second, firms of higher productivity are shown to be less export orientedand sell more in the home market. This is because the ability to accumulate customers in thedomestic market is higher for more efficient firms. Finally, an examination of treaty linkagesestablished between Chinese cities and foreign countries during the 19th century shows persistentimpacts on current trade among participants in the treaty arrangements.iiiPrefaceThe essay in Chapter 3 is joint research with Zhe Chen. Both of us worked on all aspects of thepaper which includes the identification of the research question, development of the theoreticalanalysis, execution of the empirical work and writing of the manuscript.The research project in Chapter 4 is co-authored with Keith Head, John Ries and Junjie Hong.A version of Chapter 4 has been published (Head et al., 2015). I initiated the research idea ofstudying the impact of treaty ports, collected the data from various sources, performed econometricanalysis, and participated in writing the paper.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 China’s export registration in the automobile industry . . . . . . . . . . . . . . . 32.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2 Data and stylized facts of the industry . . . . . . . . . . . . . . . . . . . . . . . . . 62.2.1 Fact 1: Many intermediaries have zero or small auto export orders. . . . . . 82.2.2 Fact 2: Net exit of intermediaries post regulation. . . . . . . . . . . . . . . . 92.2.3 Fact 3: Automakers are more likely to list same province intermediaries. . . 102.2.4 Fact 4: Automakers rarely share intermediaries on the list. . . . . . . . . . . 132.2.5 Fact 5: High drop rate and relationship to previous order size. . . . . . . . . 142.2.6 Fact 6: Export prices of intermediaries rise post regulation. . . . . . . . . . 152.2.7 Fact 7: Positive assortative matching among automakers and intermediaries. 162.3 A model of automaker-intermediary matches . . . . . . . . . . . . . . . . . . . . . . 182.3.1 Model setup and assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . 182.3.2 Pre-regulation: pricing and order allocation . . . . . . . . . . . . . . . . . . 212.3.3 Post-regulation: pricing and order allocation . . . . . . . . . . . . . . . . . . 212.3.4 Post-regulation: welfare analysis . . . . . . . . . . . . . . . . . . . . . . . . . 232.3.5 Bayesian updating process . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242.4 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.4.1 Choice of parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26vTable of Contents2.4.2 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323 Productivity, market penetration and allocation of sales . . . . . . . . . . . . . . 343.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.2 Data and stylized facts of sales allocation . . . . . . . . . . . . . . . . . . . . . . . . 363.2.1 Data and summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 363.2.2 Firm productivity and export intensity . . . . . . . . . . . . . . . . . . . . . 383.2.3 Firm productivity and sales ratio (firm level) . . . . . . . . . . . . . . . . . . 413.2.4 Firm productivity and sales ratio (firm-destination level) . . . . . . . . . . . 453.3 A model with heterogeneous marketing cost elasticities . . . . . . . . . . . . . . . . 463.3.1 Marketing cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473.3.2 Consumer demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473.3.3 Firm problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483.3.4 Sales ratio between foreign and domestic market . . . . . . . . . . . . . . . . 503.3.5 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513.4 Evidence on the effect of marketing cost . . . . . . . . . . . . . . . . . . . . . . . . . 523.4.1 Advertisement expenditure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533.4.2 Alternative explanations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 574 The legacy of 19th century treaties on the current trade of Chinese cities . . . 594.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 594.2 Treaty Ports, Concessions, and Leased Territories . . . . . . . . . . . . . . . . . . . 614.3 Data description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654.4 Specification and results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 674.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 755 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78Appendix A Appendix for Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83A.1 Registration policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83A.2 Comparing staying and exiting intermediaries . . . . . . . . . . . . . . . . . . . . . 83A.3 Notes for Bayesian statistics and quality distribution . . . . . . . . . . . . . . . . . 86A.4 Intermediary/automaker ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86Appendix B Appendix for Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89B.1 Correlation between value-added per worker and TFP . . . . . . . . . . . . . . . . . 89B.2 Productivity and sales ratio: quadratic regression . . . . . . . . . . . . . . . . . . . 89viTable of ContentsB.3 Industry list and firm share . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90B.4 Firm productivity and sales ratio: top 10 destinations . . . . . . . . . . . . . . . . . 90B.5 Proof of Proposition 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102Appendix C Appendix for Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104viiList of Tables2.1 Automobile export by HS4 product category . . . . . . . . . . . . . . . . . . . . . . 72.2 Share of intermediaries with small auto exports . . . . . . . . . . . . . . . . . . . . 82.3 Exiting vs surviving intermediaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.4 Probability of automaker-intermediary matches 2007 (Y=1 if match) . . . . . . . . . 132.5 No. of automaker-intermediaries pairs . . . . . . . . . . . . . . . . . . . . . . . . . . 142.6 Probability of being listed again by any automaker . . . . . . . . . . . . . . . . . . . 152.7 Assortative matching Y = ln(Qi0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.8 Summary of stylized facts and model features . . . . . . . . . . . . . . . . . . . . . 172.9 Description of model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262.10 Exiting vs surviving Intermediaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.11 # of Automaker-Intermediaries pairs . . . . . . . . . . . . . . . . . . . . . . . . . . . 293.1 Share of exporters by year or ownership . . . . . . . . . . . . . . . . . . . . . . . . . 373.2 Distribution of export intensity by ownership . . . . . . . . . . . . . . . . . . . . . . 383.3 Trade mode: ordinary vs processing . . . . . . . . . . . . . . . . . . . . . . . . . . . 403.4 Firm productivity and sales ratio (export/domestic) . . . . . . . . . . . . . . . . . . 433.5 Firm productivity and sales ratio (robustness) . . . . . . . . . . . . . . . . . . . . . . 443.6 Firm productivity and sales ratio: firm-destination level . . . . . . . . . . . . . . . . 463.7 Estimates of marketing cost elasticities . . . . . . . . . . . . . . . . . . . . . . . . . . 523.8 Effect of advertisement expenditure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543.9 Effect of relative market size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.10 Firm productivity and sales ratio: Hong Kong . . . . . . . . . . . . . . . . . . . . . . 563.11 Price ratio vs quantity ratio: Hong Kong . . . . . . . . . . . . . . . . . . . . . . . . . 574.1 Top trading partners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654.2 Number of city linkages, by type and recipient country . . . . . . . . . . . . . . . . . 664.3 Bilateral treaty linkage effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 694.4 Restricted sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 704.5 Bilateral and multilateral effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724.6 Probit prediction of treaty recipients and hosts . . . . . . . . . . . . . . . . . . . . . 734.7 Placebos and propensities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74A.1 Exiting vs staying Intermediaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84viiiList of TablesA.2 Exiting vs staying Intermediaries (cont.) . . . . . . . . . . . . . . . . . . . . . . . . . 84A.3 Discrete hazard model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85B.1 Firm productivity and export sales . . . . . . . . . . . . . . . . . . . . . . . . . . . 90B.2 Firm productivity and sales ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91B.3 Firm productivity and sales ratio (value-added/worker): firm-destination level . . . . 92B.4 Effect of advertisement expenditure: value-add/worker . . . . . . . . . . . . . . . . . 93B.5 Effect of relative market size (value-added per worker) . . . . . . . . . . . . . . . . . 94B.6 Firm productivity and sales ratio: value-added/worker (HK) . . . . . . . . . . . . . 95B.7 Quadratic productivity: firm level . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96B.8 Quadratic productivity: firm-destination level . . . . . . . . . . . . . . . . . . . . . . 97B.9 List of industries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98B.10 Firm productivity and sales ratio: top 10 . . . . . . . . . . . . . . . . . . . . . . . . 101C.1 Description of treaty arrangements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105C.2 Treaty linkages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106C.3 Probit regression predictions for countries . . . . . . . . . . . . . . . . . . . . . . . . 107ixList of Figures2.1 Number of auto exporters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.2 Geographic locations of automakers (top) and intermediaries (bottom) in 2011 . . . 112.3 Distance in terms of number of provinces apart . . . . . . . . . . . . . . . . . . . . . 122.4 Long-term (left) vs Hit-and-run (right) relationships . . . . . . . . . . . . . . . . . . 122.5 # Automaker per Intermediary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.6 Average price changes of intermediaries . . . . . . . . . . . . . . . . . . . . . . . . . 162.7 Intermediary demand shock (simulation) vs order size (data) . . . . . . . . . . . . . 272.8 Number of intermediaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.9 Price change of intermediaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.10 Profit of automakers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.11 Total profit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313.1 Average export intensity and productivity percentile . . . . . . . . . . . . . . . . . . 393.2 Average export intensity and TFP percentile (Textile vs Electric machines) . . . . . 403.3 Export intensity and productivity (excluding processing trade) . . . . . . . . . . . . 413.4 Average export intensity and TFP percentile by ownership . . . . . . . . . . . . . . 423.5 Productivity and extensive margin effect n(φ) . . . . . . . . . . . . . . . . . . . . . . 494.1 Geographic distribution of host cities . . . . . . . . . . . . . . . . . . . . . . . . . . . 674.2 Bilateral and group linkages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71A.1 µi ∼ Gamma(2, 1) and Ai ∼ Poisson(µi) . . . . . . . . . . . . . . . . . . . . . . . . 87A.2 Direct export automaker/Automakers who list intermediaries) (mDir/mSlot) . . . . 88A.3 Number of auto-exporters: available vs registration . . . . . . . . . . . . . . . . . . 88B.1 The Correlation between Value-added Per Worker and TFP . . . . . . . . . . . . . . 89B.2 Average export intensity and productivity (value-added per worker) . . . . . . . . . 99B.3 Average export intensity and productivity (TFP) . . . . . . . . . . . . . . . . . . . . 100xAcknowledgementsI would like to express my special gratitude to my supervisors, Keith Head and John Ries fortheir invaluable guidance throughout my graduate studies. Your advices have been most usefulto help me grow as a researcher. I appreciate your great patience, insightful comments as well asencouraging words that keep me motivated.I am very grateful to my graduate advisors, Sanghoon Lee and Barbara Spencer, who are notonly helpful but also caring. They are always willing to have discussions and provide feedbacks tomy research projects. I would like to thank Matilde Bombardini for accepting to join my committee,providing numerous suggestions, and organizing the Trade Study Group that greatly improves theknowledge sharing and bonding between faculty members and graduate students. I am also verygrateful to Ralph Winter who not only provides constructive suggestions to my research but alsooffers the most cozy venue for the Research Methodology course. Many thanks to other facultymembers of the Strategy and Business Economics Division and participants in the Trade StudyGroup, your comments, suggestions, and acts of kindness are most appreciated.Special thanks to my parents who always provide me with full support and great comfort.xiChapter 1IntroductionThis dissertation is a collection of three essays at the intersection of International Trade and Indus-trial Organization. Although the topics are diverse, they share the common objective of studyinghow Chinese firms link with foreign markets. The first essay investigates the impact of a governmentregulation that restricts the use of intermediaries linking domestic producers and foreign buyers.The second essay studies the rationale behind the sales allocation of Chinese exporters betweendomestic and foreign markets. The third essay examines the persistence of trade relationships froma historical point of view.The first essay of this dissertation is “China’s export registration in the automobile industry:Effects on manufacturer-intermediary match efficiency.” In 2007, China implemented a policy re-quiring automobile producers to distribute through at most three trade intermediaries and listtheir intermediaries on a registry. Motivated by the registration requirements and granularity inthe order sizes handled by most intermediaries, this paper develops a model to describe the matchesbetween automakers and intermediaries. The model shows market division arises endogenously dueto the regulation. It creates inefficiencies in matching and double marginalization. The modelpredictions coincide with a number of stylized facts: a strong decline in the number of auto inter-mediaries, assortative matching, export price increases for intermediaries, and substantial churningin the sets of intermediaries registered by the automakers. Welfare analysis in terms of total profitshows that this regulation benefits automakers, especially those relatively less efficient ones whileintermediaries are made worse off.The second essay is titled “Productivity, Market Penetration and Allocation of Sales.” Thischapter examines the sales allocation of Chinese exporters. We demonstrate that highly produc-tive firms are less export oriented compared with less productive ones. This negative correlationbetween firm productivity and export intensity among exporters remains robust when we controlfirm ownership, factor intensity, and rule out impacts of processing trade. In order to rationalizeour empirical findings, we extend Arkolakis (2010) model to allow marketing cost elasticities to beheterogeneous across markets. A higher marketing cost elasticity domestically gives rise to a fastersales expansion in the home market as firm productivity grows. The fact that this negative corre-lation is more pronounced among firms who belong to advertising intensive industries supports themodel predictions. Further evidence helps to rule out alternative explanations such as the effectsof variable markup and product quality.The title of the third essay is “The legacy of 19th century treaties on the current trade ofChinese cities.” It examines the effect of treaty linkages established between Chinese cities and1Chapter 1. Introductionforeign countries during the 19th century on China’s trade today. Hypothesis is made that thesehistorical arrangements created relationship-specific capital that continues to facilitate trade. Inthe full sample of bilateral trade between 335 cities and 212 countries, there are significant linkageeffects. However, ensuing analysis indicates that greater trade among cities and countries who arelinked by treaties largely reflects the propensity of higher income partners to trade more with eachother. Even after controlling for this propensity, trade is higher among the group of countries andcities that were involved in treaty arrangements. We refer to this higher trade as group effectsand propose two explanations. First, participation in a treaty arrangement changed the industrialstructure of cities party to a treaty in a manner that continues to facilitate economic exchange today.Second, the experience gained through participation in a treaty arrangement created knowledge thathas passed down through generations. The results are consistent with the proposition that bilaterallinkages promote multilateral trade by generating group effects.Because each essay investigates a different topic and the chapters are self-contained. I thusleave a more comprehensive discussion of the research question and contribution to the introductionspecific to each chapter.2Chapter 2China’s export registration in theautomobile industry: Effects onmanufacturer-intermediary matchefficiency2.1 IntroductionChina’s auto industry has experienced rapid growth over the past fifteen years.1 While most ofthis new production was destined for the Chinese market—exports are only 5% of production—automobile exports have risen rapidly as well. Accompanying the growth in exports was a dra-matic rise in the number of intermediaries exporting automobiles. Over 700 intermediaries withauto exports were identified from China’s Customs records in 2006. In response to this surge inauto exporting intermediaries, the Chinese government introduced a registration policy where automanufacturers were required to list and use at most three intermediaries. This paper investigateshow the structure and performance of China’s automobile export sector reacted to the regulation.It provides a model of exporting through intermediaries and explores the efficiency consequencesof the regulation.Unlike developed countries where big brand passenger cars dominate the export market, China’sautomobile exports mainly consist of commercial vehicles. Its top destinations include Russia,Iran, Ukraine, Vietnam, and other developing economies. China’s automobile industry is veryfragmented. There were at least 250 motor vehicle manufacturers (auto-parts makers not included)in China when the registration policy took place. The registration policy, initiated in 2007, requiresautomakers who intend to export automobiles the next year to register with the government andto authorize at most three intermediaries as their export agents. Automakers are allowed to exportdirectly by themselves as well. The policy did not put restrictions on automaker or intermediaryqualifications. That is, any automaker with a valid production permit could register and list anyintermediary they prefer.The registration list reveals a unique set of information on matches between automakers andintermediaries. It also highlights a series of stylized facts that inform my modeling decisions.Specifically, automakers typically list three intermediaries while most intermediaries are not listed1Its share of global motor vehicle production surged from less than 5% in 2000 to almost 25% in 2011.32.1. Introductionby multiple automakers. The automaker-intermediary partnerships are more likely to form amongclosely located pairs although churning on the list occurs frequently. Intermediaries with largerprevious export orders have higher probability of being listed again. Among the matched pairs,positive assortative matching is observed where larger automakers are matched with intermediariescapable of bringing more export orders. Moreover, the regulation triggers changes in the intermedi-ary industry as the number of auto intermediaries declines and the export prices of surviving onesincrease.Motivated by the fragmented export orders handled by intermediaries and institutional infor-mation of the regulation, a model is developed to emphasize the impact of regulation on automaker-intermediary matches. Intermediaries are heterogeneous in quality and higher quality draws areassociated with larger demand shocks. They obtain export orders and contract with an automakerto fulfill them. Automakers (or sellers) produce homogeneous products and engage in Bertrandcompetition for orders of intermediaries. Each intermediary sources from his lowest cost automakerbefore the regulation. The registration policy restricts the number of intermediaries that each au-tomaker could list on the registry to three. The listing occurs before the realization of intermediarydemand shocks and before prices are offered by automakers. As a result, market division is endoge-nously implemented by the regulation. Each automaker lists a different set of intermediaries asan equilibrium outcome. Relatively less efficient automakers benefit from this regulation becauseintermediaries who are unable to contract with the lowest cost automaker are pushed towards thehigher cost ones. Surviving intermediaries face higher wholesale prices and the lowest quality onesexit the market. Inefficiencies are generated due to market partition, double marginalization, andpotential existence of non-assortative matching equilibria.This paper is related to a number of studies examining the role of intermediaries in internationaltrade. Both Ahn et al. (2011) and Crozet et al. (2013) show that intermediated share of exports arehigher for more difficult destination markets and extends the Melitz (2003) model to allow economiesof scale in export intermediation. Akerman et al. (2010), investigates wholesalers’ advantages inhandling multiple goods due to economies of scope. Blum et al. (2009) examines the matchesbetween importers and exporters to find that importing intermediaries help small exporters savebilateral match costs. All these papers model homogeneous intermediaries who serve as a meansto avoid fixed costs. In my model, intermediaries are heterogeneous in quality. High qualityintermediaries are connected to a large pool of potential buyers while low quality ones have positiveprobabilities of receiving zero export orders. Rather than having supply at hand and searchingfor customers abroad, intermediaries in my model obtain orders and then search for domesticautomakers to fill them.A common challenge faced by researchers studying the role of trade intermediaries is the iden-tification of intermediaries. As Blum et al. (2009) points out that using Customs data it is difficultto identify with precision importing intermediaries from final consumers. It is equally hard todistinguish manufacturers from exporting intermediaries. Therefore, Ahn et al. (2011) utilizes firmname orthography as indications of their business type while Bai et al. (2017a) infers intermedi-42.1. Introductionaries from direct exporting manufacturers by examining the discrepancy between customs recordsand self-report export values in national surveys. Even when intermediaries are identified, whichproducers they source from remain largely unknown. By focusing on the automobile industry, thispaper is able to identify automakers and intermediaries with great precision. Further, the exactlisting of automaker-intermediary pairs on the registry provides a unique opportunity to examinethe linkages between export intermediaries and their domestic suppliers.This research also contributes to the growing literature documenting matching between interna-tional sellers and buyers (Blum et al., 2010; Bernard et al., 2014; Eslava et al., 2015; Tybout et al.,2016). Most of these studies have intermediaries in terms of wholesalers or retailers on either orboth sides of the matches across boarders. Blum et al. (2010) shows that at least one party in therelationship is large and Bernard et al. (2014) finds negative assortative matching in terms of firmconnections. This paper presents a slightly different perspective: low cost (or large) automakersare predicted to be favored by intermediaries of various sizes in the pre-regulation scenario of themodel; and positive assortative matching in terms of firm sizes is observed among post-regulationautomaker-intermediary pairs. Another pattern found by previous matching literature is the highprobability of breaking relationships (e.g., Tybout et al., 2016). Consistent with Tybout et al.(2016), this paper also shows high pair level drop rate among automakers and intermediaries overtime. While previous research shows seller-buyer matching patterns under free market conditions,analyses in this paper add to the literature by considering effects of a registration policy limitingmatch opportunities and its impact on equilibrium and efficiency.In order to rationalize the prevalence of intermediaries with extremely small or even zero autoexport orders, this paper takes advantage of recent developments in modeling with granularity(e.g., Eaton et al., 2012; Head et al., 2016). In particular, Eaton et al. (2012) assume that thenumber of exporters is the realization of a Poisson random variable instead of a fixed measureof firms. Zero trade flows in their model imply that no firm happened to be efficient enough.Following their wisdom, intermediaries are assumed to receive demand shocks from a randomPoisson distribution based on their quality parameters. Lower quality intermediaries are morelikely to have bad demand shocks and generate zero export orders. This method helps to reconcilethe existence of intermediaries with zero or small amount of auto export orders. Dynamics emergebecause automakers update priors about intermediary quality based on realizations of order draws.The rest of the paper is organized as follows. Section 2.2 first describes information revealed bythe registration lists and introduces other sources of data used in this paper. Then, stylized factsfeaturing China’s automobile exports and impacts of regulation are shown. In section 2.3, a modelis proposed to rationalize the data patterns and capture the matching process among automakersand intermediaries both before and after the registration policy. Before showing simulation results,Section 2.4 explains the choice of key parameters of the model. The last section concludes thepaper.52.2. Data and stylized facts of the industry2.2 Data and stylized facts of the industryThe data set used in this paper features the registration list of automaker and intermediary pairsfrom 2007 to 2011. They are combined with transaction level export data from Chinese Customsrecords and firm level balance sheet information from manufacturer surveys of the same period.Before 2004, domestically owned firms below a registered capital threshold were prohibitedfrom direct exporting. They had to export through state owned intermediaries who had monopolytrading rights. As part of China’s commitment to joining the WTO these restrictions were removedby 2004 (Bai et al., 2017a; Chen and Li, 2014). The liberalization of trading rights also facilitatesentry of intermediaries. From 1999 to 2000, private firms with larger than 10 million USD exportsin the previous year are allowed to export directly and export products of other firms (i.e., beingan intermediary). After 2001, any domestic firm with over 3 million RMB registered capital canbe a trade intermediary (Chen and Li, 2014). The auto registration policy was going against theprevious trend of liberalizing who can export in a more subtle way.According to the regulation, automakers who intend to export automobiles the next year arerequired to register by the end of the current year. On the registration, automakers are allowedto list at most three intermediaries as their authorized agents to export for them. Intermediariesnot listed by an automaker are not allowed to export that automaker’s vehicles. Automakers couldalso export directly by themselves.2The registration lists since 2007 are available on the government website as public notices. Forautomakers, the lists include name (in Chinese), location (i.e., province), Customs identification(ID) number (if available3), and general product categories4. The lists also include name andCustoms ID number of intermediaries. If an automaker only plans to export directly, he couldregister himself without listing any intermediary. The majority (89%) of automakers put at leastone intermediary on their lists. The key characteristic of the registration list is that it containsinformation on the matched partnerships among automakers and intermediaries. That is, we knowthe suppliers for each intermediary and that intermediary’s export records. But a drawback is thatif an intermediary was listed by multiple automakers, we can not attribute his auto exports to aspecific automaker.The registration list also reveals unique information about the ownership linkages between au-tomakers and intermediaries. Examined by orthography, about 17% of intermediaries share part oftheir firm names with automakers who listed them. They are defined as linked intermediaries forthey tend to either be owned by the corresponding automakers or belong to the same CorporateGroup as the automaker. For instance, First Automobile Works Import and Export (FAWIE) is asubsidiary of FAW Group and facilitates trade activities of other FAW subsidiary auto manufac-turers such as FAW Tianjin Xiali.5 Linked intermediaries account for about one third of total auto2Please refer to Appendix A.1 for detailed descriptions of the regulation.3The Customs ID number is available for automakers who export some or all of their products directly by them-selves.4The product categories include passenger cars, trucks, buses, and chassis.5Hu et al. (2014) also observes the existence of big corporate groups in Chinese automobile industry. But instead62.2. Data and stylized facts of the industryexports and are more specialized in auto products. The stylized facts described in this paper willtake the special feature of ownership linkages into consideration.This registration policy is designated for automobiles. Firms who export auto parts are notrestricted by this regulation. To be specific, the Chinese government provided a product list whichcontains 83 CN 10-digit (covering 16 HS 6-digit) product varieties.6 Table 2.1 shows China’sautomobile export by a more aggregated HS 4-digit product category.7 Column (1) presents thatthe total export value of regulated automobiles increased from 0.13 million USD in 2000 to over8 million USD in 2011. Trucks account for about 50% of China’s automobile exports followed bycars (26%) and buses (18%).Table 2.1: Automobile export by HS4 product categoryTotVal Truck% Car% Bus% Tractor% Chassis%Year (1) (2) (3) (4) (5) (6)2000 0.13 0.48 0.23 0.23 0.01 0.052001 0.18 0.38 0.22 0.34 0.01 0.052002 0.17 0.48 0.18 0.23 0.01 0.102003 0.30 0.56 0.21 0.14 0.02 0.072004 0.57 0.49 0.30 0.14 0.03 0.042005 1.52 0.38 0.38 0.12 0.08 0.032006 2.83 0.42 0.34 0.15 0.07 0.022007 6.38 0.43 0.33 0.14 0.08 0.022008 9.19 0.46 0.33 0.12 0.07 0.022009 4.23 0.55 0.18 0.15 0.10 0.032010 5.51 0.48 0.19 0.19 0.13 0.022011 8.78 0.49 0.22 0.18 0.11 0.01TotVal = total value of automobile exports in billion US dollars.Firms on the registration lists are first combined with export data collected by the ChineseCustoms Office using their Customs ID numbers. The transaction level Customs data containinformation on firm name, export value, quantity, product at eight-digit level, destination country,indicator of processing trade, and etc. Then, characteristics of automakers are obtained from theChinese Industrial Enterprises Survey database. It is collected by National Bureau of Statistics andcovers all state-owned manufacturing firms and non-state-owned manufacturers above certain sizethresholds. Firm level balance sheet information such as sales, employment, and capital investmentare included.8 Sales of automakers are extracted from this dataset to demonstrate assortativematching.Now I turn to describing the empirical facts of automobile exporters in China and matchingof export intermediaries, they focus on passenger vehicles and the competition structure of the industry.6The Chinese Customs use CN10 standard which contains more detailed product specifications than HS 6-digit.Also, motorcycles are not included in this regulation.7There are 5 HS 4-digit level product categories: tractor (HS8701), bus (HS8702), car (HS8703), truck (HS8704)and chassis (HS8706).8The Survey sample has a different firm identification system from the Customs records, and therefore, firms arematched by orthography when automaker characteristics are required.72.2. Data and stylized facts of the industrypatterns among automakers and intermediaries.2.2.1 Fact 1: Many intermediaries have zero or small auto export orders.I start with the size distribution of registered intermediaries. There are a large number of inter-mediaries with small or even zero amount of auto export orders. Table 2.2 shows that on average44% of intermediaries on the list do not have auto exports the following year and for around 29%of them, auto exports account for less than half of their total export values.As Bernard et al. (2011) pointed out that intermediaries handle more product varieties thandirect exporting manufactures, auto intermediaries on the registry also export other products suchas auto parts or electronics. Patterns in Table 2.2 suggest that automobile is not the main productcategory for the majority of intermediaries in China. It is also consistent with anecdotal evidencethat the business presence of intermediaries in a foreign country will attract local buyers withoccasional demand.There are 979 distinct intermediaries listed on the registry over the five years. About 40%of these intermediaries are only listed once and their chances of bringing positive orders is lessthan 30%.9 This finding is consistent with what Ganapati (2016) observes from the other sideof the relationship: smaller buyers predominantly deal with wholesalers instead of manufactures.Similarly, if we compare the size distribution of auto export orders between intermediaries and directexporting automakers, only 3% of intermediaries export over 100 million USD in 2007 compared to6.3% for automakers. At the same time, 43% of intermediaries have export orders below 1 millionUSD compared to 36% among automakers.Table 2.2: Share of intermediaries with small auto exportsYear #Intm AutoExp=0 AutoShr<0.52007 294 0.31 0.362008 474 0.42 0.272009 510 0.55 0.262010 540 0.53 0.262011 389 0.41 0.28There are 979 distinct intermediaries on the list overthe five years.In order to capture the highly skewed distribution of auto export orders, especially to ensurethat intermediaries have a positive probability to receive zero orders, granularity is introducedinto the model. That is, intermediaries are assumed to be endowed with heterogeneous qualities.These qualities can be considered as networks or business connections with potential customers.Intermediary-specific demand shocks are drawn based on their qualities. The granularity of ex-port orders indicates that the order generating process follows a discrete distribution with highprobability of receiving zero or small orders.9There are 8 intermediaries listed throughout the period who never bring any orders.82.2. Data and stylized facts of the industry2.2.2 Fact 2: Net exit of intermediaries post regulation.The second stylized fact is the decreasing number of auto intermediaries after the registrationpolicy and describes the characteristics of those who exit. As presented in Figure 2.1, the numberof intermediaries with positive auto exports peaked in 2006 and then dropped by almost a half andstabilized after 2009.10 In order to demonstrate that the sharp decrease of auto intermediaries isnot an effect of any macro economic shock to the global automobile market, Figure 2.1 also depictsthe number of direct exporting automakers. The steady increase of direct exporting automakersbefore regulation and stable auto exporters post regulation form a stark contrast to the trend ofintermediaries.11Figure 2.1: Number of auto exportersWho are the exiting intermediaries? The first column of Table 2.3 compares the auto exportquantity of surviving and exiting intermediaries before and after the registration policy.12 Survivingintermediaries had four times the exports of exiting ones13 and the gap increased post regulation.On average, intermediaries export around 70% more after the registration policy. The remainingcolumns report results from a discrete hazard model which focuses on the probability of exit.10How to identify intermediaries before policy? The cleanest way is to use the registration list. However, interme-diaries shaken out by the policy can not be found on the list. Therefore, orthography provides additional informationto identify intermediaries. We worry that orthography is not precise and might generate errors. In oder to makethe numbers comparable pre and post regulation, orthography is also used to identify intermediaries in addition tothe list after the registration policy. To remain consistence of the comparison, the same method is used to identifyautomakers as well. In sum, the number of automakers and intermediaries available in Figure 2.1 are identified byboth registration list and orthography.11The increasing number of intermediaries before 2006 mainly results from the deregulation of export trading rights.That is, before 1999 SOE intermediaries have monopoly trading rights. From 1999 to 2000, manufacturing firms wereallowed to set up privately owned intermediaries if they exported more than 10 million USD in the previous year.Since 2001, domestic manufacturers were allowed to set up trade intermediaries as long as its registered capital wasover 3 million RMB. (5 million RMB for non-manufacturers.) Details of the policy can be found in Chen and Li(2014).12Other aspects of intermediary characteristics—not directly related with model predictions—are also comparedand the results are presented in Appendix A.2.13qstay/qexit = e1.417 = 4.1292.2. Data and stylized facts of the industryColumn (2) and (3) of Table 2.3 show that intermediaries with fewer auto exports are more likely toexit especially in the post policy period. However, the exit probability does not change significantlypost regulation. The last two columns include additional control variables such as price, numberof destination markets, number of products, and auto export share and their interactions with thepost dummy. Intermediary’s order size remains negatively related to exit while the scale of theeffect becomes smaller.Table 2.3: Exiting vs surviving intermediariesY=ln(q) Y=Prob(exit)(1) (2) (3) (4) (5)OLS Hzd LP Hzd Logit Hzd LP Hzd Logitexit -1.417a(0.111)post 0.567a -0.028 0.097 0.068 0.541(0.121) (0.025) (0.126) (0.104) (0.591)exit×post -0.663a(0.142)ln(q) -0.073a -0.324a -0.040a -0.178a(0.004) (0.023) (0.007) (0.035)post×ln(q) -0.020a -0.189a -0.023b -0.175a(0.006) (0.038) (0.010) (0.053)Control N N Y YN 4197 4197 4197 4197 4197R2 0.160 0.161 0.240Standard errors in parentheses. c p < 0.1, b p < 0.05, a p < 0.01.Firm-year level regression with firm clusters.Appendix A.2 presents the regression results with the coefficients on control vari-ables as price, the number of products, the number of markets, auto export shareas well as their interactions with the post dummy.2.2.3 Fact 3: Automakers are more likely to list same province intermediaries.Figure 2.2 displays the geographic locations of automakers (top) and intermediaries (bottom) onthe 2011 registry. There are two clusters of automakers around Hubei and Shandong provinces.Meanwhile, a slightly higher share of intermediaries are also found close to these clusters.14Among the matched pairs on the registry, automakers are more likely to list intermediaries thatare located in the same region.15 Figure 2.3(a) illustrates that 38.16% of matched pairs are from thesame province. And it is rare for automakers to list intermediaries located five or more provinces14The northwest province Xinjiang turns out to be exceptional in hosting intermediaries. The reason is that the topdestinations of auto exports from China are Russia, Iran, and Ukraine. Intermediaries choose to located in Xinjiangto get close to their customers. Similar fact is also described in Ganapati (2016) where wholesalers are found to shipproducts mainly to nearby destinations.15Location information of automakers and intermediaries are extracted from their Customs IDs (i.e., first 2 digitsfor provinces and first 4 digits for cities). 88% of automakers and all intermediaries on the list have Customs IDs.102.2. Data and stylized facts of the industryFigure 2.2: Geographic locations of automakers (top) and intermediaries (bottom) in 2011away. Although linked pairs are more likely to locate close to each other, as Figure 2.3(b) showsthat the large share of same province match is not driven by linked pairs.16In addition, Figure 2.4 shows that long-term relationships tend to be local. That is, almost60% of those pairs who maintain their relationships throughout the sample period are from thesame province. On the contrary, hit-and-run relationships—automaker only list an intermediaryonce—present fewer same province matches.Next, I investigate to what extent does locating in the same province or same city affect theprobability of automaker-intermediary matches.17 The first two columns of Table 2.4 show OLSregression results of distance in terms of same city, same province, and number of provinces apart16As a reference point, random matches lead to less than 5% same province pairs (assuming intermediaries can bere-drawn). Prob(same prov) =∑jNmjNm× NijNi.17For simplicity, I focus on the initial year of the registration list. According to the model, subsequent adjustmentsto the registration lists only result from updating the expectation of intermediary qualities. Robustness checks showsimilar patterns for other years.112.2. Data and stylized facts of the industryFigure 2.3: Distance in terms of number of provinces apart(a) All matched pairs (b) Non-linked pairsFigure 2.4: Long-term (left) vs Hit-and-run (right) relationshipsdummies on the probability of automaker-intermediary matches. Automaker and intermediaryfixed effects capture all the characteristics that affects the attractiveness of a firm. Being in thesame province increases the probability of match by 5.3 percentage points while being in the samecity further boosts the matching probability by 14.2 percentage points. The effects decay as thetwo partners become further apart. Coefficients from logit regressions suggest similar results. Theodds ratio of pair match is 6.8 times larger if they locate in the same city.18The model incorporates cost advantage of same region matches. The marginal cost of supplyingan intermediary located in a different province is higher by τ compared with selling to same provinceintermediaries. To what extent same province is valued depends on the relative size of τ anddifferences in production costs.18The scale of same city effect (exp(1.918) = 6.8) is much smaller than the same province effect due to the reasonthat same city effect is in addition to the same province effect. The OLS coefficients also suggest a larger odds ratiofor same province dummies. That is, the OLS predicted same province odds ratio = 0.062/(1−0.062)0.009/1−0.009 = 7.3 (where0.062 = 0.009 + 0.053) and same city odds ratio = 0.204/(1−0.204)0.062/(1−0.062) = 3.9 (where 0.204 = 0.062 + 0.142).122.2. Data and stylized facts of the industryTable 2.4: Probability of automaker-intermediary matches 2007 (Y=1 if match)(1) (2) (3) (4)OLS LI=0 Logit LI=0Same city 0.142a 0.129a 1.918a 1.788a(0.019) (0.020) (0.214) (0.228)Same province 0.053a 0.057a 14.006a 14.365a(0.006) (0.007) (0.464) (0.670)Adjacent province 0.010a 0.013a 11.369a 11.790a(0.004) (0.004) (0.507) (0.738)1-province apart 0.009b 0.011b 11.011a 11.368a(0.004) (0.004) (0.598) (0.676)2-provinces apart 0.008b 0.010b 10.908a 11.144a(0.004) (0.004) (0.535) (0.715)3-provinces apart 0.006c 0.008b 10.742a 11.095a(0.003) (0.004) (0.529) (0.722)4-provinces apart 0.005 0.007c 10.536a 10.940a(0.004) (0.004) (0.560) (0.743)N 45746 40158 45746 40158Y¯ 0.009 0.009 0.009 0.009R2 0.052 0.045Standard errors in parenthesis. c p <0.1, b p <0.05, a p <0.01Automaker and intermediary fixed effects included. The longest dis-tance between any two of Chinese provinces is 5 provinces apart andthat is the omitted group in the regressions. Column (2) and (4) restrictthe sample to non-linked (LI=0) pairs.2.2.4 Fact 4: Automakers rarely share intermediaries on the list.In this subsection, the matching pattern among automakers and intermediaries on the registrationlist is examined. Most automakers list three (the maximum allowed by the regulation) intermedi-aries with the registry and their lists rarely overlap. Figure 2.5 depicts the number of automakersper intermediary and shows that 70% of intermediaries are listed by one automaker (instead of mul-tiple automakers). In other words, each automaker generally lists a different set of intermediariesfrom other automakers.19 This pattern is similar to what Blum et al. (2009) found with Chileanexporter and Colombian importer pairs. Blum et al. (2009) shows that most importers source fromone exporter while only a few importers buy from multiple exporters.20This matching result—each automaker listing a different set of intermediaries—is consistentwith predictions of the proposed model. Intuitively, Bertrand competition among automakers givesadvantage to the lowest cost automaker. A pair-specific search cost ensures that each intermediaryhas his lowest cost supplier which may be different from other intermediaries. Having in mindeach intermediary’s preference, automakers avoid listing intermediaries listed by other lower-cost19Those listed by multiple automakers tend to be linked intermediaries—subsidiaries of the automaker or theCorporate Group.20In Blum et al. (2009), the mean of Exporters per Importer is 1.6 and the median is 1.132.2. Data and stylized facts of the industryautomakers. This leads to the result that automakers rarely share intermediaries.Figure 2.5: # Automaker per Intermediary2.2.5 Fact 5: High drop rate and relationship to previous order size.This subsection examines the persistence of automaker-intermediary partnerships over time andexplores the importance of previous orders brought by an intermediary on his chances of beinglisted again. Table 2.5 shows that the pair-level drop rate ranges from 48% to 59% indicatingfrequent churning of automaker-intermediary partnerships on the registry. Tybout et al. (2016)discovered similar patterns of non-trivial probability of eliminated connections when investigat-ing the transitional relationship between wholesale exporters worldwide and Colombia footwearimporters.Table 2.5: No. of automaker-intermediaries pairsYear New Continue Dropped Total DropRate2007 . . 245 490 0.502008 566 245 391 811 0.482009 481 420 437 901 0.492010 463 464 550 927 0.592011 250 377 . 627I hypothesize that intermediaries are more likely to be dropped when they have lower quality,draw zero demand and fail to bring orders in the previous year. In other words, intermediaries withhigher realized demand shocks previously are more likely to bring larger orders and stay on the list.Table 2.6 shows that an intermediary’s probability of being listed by any automaker is positivelyaffected by his previous order size. Columns (3) and (4) take the linkage between automakers andintermediaries into account. As expected, linked intermediaries are more likely to be listed andtheir previous performances in bringing orders play a smaller role in automakers’ choice decisions142.2. Data and stylized facts of the industrycompared with non-linked intermediaries. As a robustness check, the last two columns restrict thesample to non-linked intermediaries only. The results confirm that among non-linked intermediariesthe impact of previous order sizes on chances of being listed is larger than the whole sample.The model allows uncertainty in intermediary order draws each period. Specifically, sincequalities of intermediaries are not observed, automakers are allowed to update their expectations ofintermediary quality µi following a Bayesian updating process based on previous realized demandshocks of each intermediary. As a result of the uncertainty aspect of the model, substantial churningin the sets of intermediaries listed by automakers is observed. Higher order sizes in the previousyears reflect intermediary quality and are valued by automakers when they choose whom to put ontheir lists.Table 2.6: Probability of being listed again by any automaker(1) (2) (3) (4) (5) (6)OLS Logit Logit Logit OLS LI=0 Logit LI=0qit−1 0.030a 2.044b 1.870b 4.373a 0.034a 4.365a(0.005) (0.901) (0.862) (1.471) (0.009) (1.470)LI 0.781a 1.302a(0.265) (0.280)qit−1×LI -4.094a(1.480)N 1823 1823 1823 1823 1664 1664R2 0.036 0.032Standard errors in parentheses. c p < 0.1, b p < 0.05, a p < 0.01.Intermediary-year level regression with year fixed effects and intermediary clusters.2.2.6 Fact 6: Export prices of intermediaries rise post regulation.Export prices of intermediaries are investigated with firm-market and firm-market-product fixedeffects respectively. Figure 2.6(a) traces the average price of intermediaries at destination marketlevel over time. With firm-market fixed effect, it demonstrates that the same intermediary wouldcharge on average 50% (i.e., e0.4=1.49) more post policy in markets he previously exported to. Andthis price increase is not fully driven by changes in product composition as shown in Figure 2.6(b)when an additional product layer is taken into consideration. In other words, when firm-market-product dummies are controlled in the regressions, the price increase post policy could possiblyreflect higher unit cost, higher quality or higher markup rather than extensive margin effects. Themodel proposed in the next section predicts cost changes channeled through matching mechanismand structure of competition while holding product quality unchanged.152.2. Data and stylized facts of the industryFigure 2.6: Average price changes of intermediaries(a) Firm-market-year level (b) Firm-market-product-year level2.2.7 Fact 7: Positive assortative matching among automakers andintermediaries.The last stylized fact describes the characteristics of the matched partners. Positive assortativematching is observed from Table 2.7. That is, automakers with larger overall sales are matchedwith intermediaries who generate larger amount of auto export orders. Note that the overall salesof automakers include a variety of products, especially auto parts, besides completely assembledautomobiles. One issue is that the sales of automakers may contribute to the export volume oftheir matched intermediaries. In order to partly address this issue, the lagged export value ofintermediary i (ln(Qi0)) right before he matches with any automaker is used as the dependentvariable. Similarly, the pre-match sales ln(Sales0) is used to represent the size of automakers.Columns (2) & (3) take automaker-intermediary linkages into consideration. Linked pairs bynature are more likely to be matched with each other. Despite of shrinking slightly in scale, positiveassortative matching still exists when the sample is restricted to non-linked pairs. Distance controlsin terms of Same city and Same province dummies are included in the regressions. They no longerhave significant effects because only matched pairs are taken into consideration.Moreover, sales of intermediaries might be correlated with automakers sales over time whenmatches are permanent. To eliminate this built-in relationship, I restrict the sample to newlymatched pairs. Results are shown in the last three columns of Table 2.7 and positive assortativematching remains strong. The scale of the coefficients become slightly smaller suggesting theexistence of carried over sales between matched pairs over time. Linked pairs tend to have stablelong-term relationships and are largely excluded from the new match sample. Bernard et al. (2014),in contrast, find negative assortative matching in terms of connectivity among Norwegian exportersand importers from other countries. That is, a well connected exporter has on average buyers withvery few connections to other sellers. This paper focuses on matching in terms of firm size insteadof connections since the latter is restricted by the registration policy.162.2. Data and stylized facts of the industryTable 2.7: Assortative matching Y = ln(Qi0)(1) (2) (3) (4) (5) (6)All All LI=0 NewMatch NewMatch LI=0ln(Sales0) 0.481a 0.411a 0.436a 0.406a 0.396a 0.397a(0.045) (0.044) (0.044) (0.060) (0.060) (0.060)Same province 0.261 0.248 0.369 0.118 0.111 0.120(0.225) (0.222) (0.240) (0.285) (0.285) (0.298)Same city 0.291 0.103 0.133 0.502 0.484 0.545(0.290) (0.290) (0.314) (0.428) (0.427) (0.434)LIis 1.676a 0.466(0.277) (0.602)N 1348 1348 1169 494 494 469R2 0.126 0.179 0.118 0.102 0.104 0.098Standard errors in parentheses. c p < 0.1, b p < 0.05, a p < 0.01.The dependent variable Y = ln(Qi0) stands for the order size of intermediary i before he matcheswith any automaker in the next period. Similarly, ln(Sales0) represents the size of automaker spre-match.Automaker-intermediary-year level regression with year fixed effects and pair clusters.Table 2.8 summarizes the stylized facts and relates each of them to a brief description of mod-eling features. The model developed in the next section is motivated by institutional information(i.e., timing of the registration and realization of export orders) and Fact 1 (granular orders re-ceived by intermediaries). In particular, intermediaries are endowed with heterogeneous quality µi.Intermediary-specific shock, Ai, is distributed Poisson21 with parameter µi and can be zero. Anintermediary with a demand shock of zero will not bring any export orders. The other facts areconsistent with model predictions.Table 2.8: Summary of stylized facts and model featuresStylized facts Model featuresFact 1 Granular orders (2.2.1) Poisson process with intermediary quality µi.Fact 2 Small firms exit (2.2.2) Automakers pick stronger intermediaries.Fact 3 Same province match (2.2.3) Higher probability to list same province intermediaries.Fact 4 1:m match (2.2.4) Market partition arises endogenously from the model.Fact 5 High drop rate (2.2.5) Uncertain intermediary quality with Bayesian updates.Fact 6 Price increase (2.2.6) Policy reduces match efficiency and induces doublemarginalization.Fact 7 Assortative matching (2.2.7) Assortative matching always characterizes some equilibriumand with certain parameters it captures all equilibria.21The Poisson distribution can generate frequent zero and small discrete order sizes. That is, the order generationprocess can be modeled by a binomial distribution with two parameters capturing the probability of obtainingpositive foreign orders and number of intermediaries in the market. It converges towards the Poisson distributionas the number of intermediaries becomes large while the probability of receiving positive orders remains low. Theheterogeneous quality µi predicts more skewed right distribution than the Poisson with homogeneous quality. Thisis consistent with Figure A model of automaker-intermediary matches2.3 A model of automaker-intermediary matchesA model featuring the matching process among automakers and intermediaries is developed in thissection. I start with the basic setup and assumptions of the model. Pre-regulation order allocationand pricing strategies are discussed next. Then I introduce the matching and pricing game withregistration requirements. Results based on Nash Equilibrium are analyzed.2.3.1 Model setup and assumptionsAutomakers (or sellers) with production cost c0s are assumed to produce homogeneous products.The marginal cost of automaker s selling to an intermediary i depends on the geographic locationsof the pair, i.e.,cs = c0s + τDiswhere Dis equals 0 if they locate in the same region and 1 if in different regions. There is no capacityconstraint for automakers. The market structure is assumed to be Bertrand competition amongautomakers for orders brought by intermediaries. That is, automakers make simultaneous offers toeach intermediary i and intermediaries always favor the lowest price offered by automakers.22Intermediaries are endowed with quality µi which is higher if an intermediary has a larger busi-ness network or is better connected to potential automobile buyers abroad. Ai is an intermediary-specific demand shock that depends on µi and it is randomly drawn from the Poisson distribution,i.e.,Ai ∼ Poisson(µi).Intermediaries are monopolistic competitors with CES demand in the foreign market.23 Supposethe price intermediary i offers to foreign buyers is pi(s), then the demand for that intermediarybecomesqi(s) = Ai · pi(s)−σ ·XPσ−1where P = [∑iAi · pi(s)1−σ]1/(1−σ) and X is the total expenditure of automobiles in the foreignmarket. The realization of intermediary-specific demand shock Ai introduces randomness so thatdemand is not pre-determined by price (which is a function of production cost). Intermediariescould receive zero export orders purely because of a bad demand shock rather than sourcing froma high cost producer. This implication is in line with Eaton et al. (2012).22Automakers do not compete with each other by offering price-quantity menu contracts to intermediaries. Thelowest cost automaker could always offer the second lowest price to intermediaries with any quantity they bringpre-regulation, making other offers unattractive to intermediaries. Such menu competition has not been observedpractically, suggesting it is not used. It might be too complicated to make the offers and too difficult to enforce thecontracts.23Intermediaries and direct exporting automakers serve different consumers and they each fit a different niche inthe export market. Since the main emphasis of this paper is the matching among automakers and intermediaries, Iwill focus on indirect exports of intermediaries only and assume away its interactions with direct exporters.182.3. A model of automaker-intermediary matchesMaximizing its profit, intermediary i will price atpi(s) =σσ − 1 · psThat is, intermediary i charges a constant markup to its marginal cost—price offered by automakers, ps.24Export transactions in this model are initiated by foreign buyers and automakers are unawareof the occasional demand abroad. Intermediaries are the ones who connect with foreign buyers andreceive the demand shocks modeled by Ai. They secure the demand and then contract with theautomaker offering the lowest cost of filling the order. Therefore, intermediaries play an active rolein bringing orders to automakers rather than passively serving producers and economize on fixedcosts of exporting as modeled in previous studies.25 Given little brand recognition of Chinese auto-mobiles and a large number of automakers, it is reasonable for foreign buyers to hire intermediariesto help with finding the best supplier.26Consistent with the pre-regulation environment, the number of intermediaries is assumed to beat least three times larger than the number of automakers, i.e., N > 3M . If s is the only automakerwho competes for intermediary i, that automaker sets the monopoly price. If multiple automakerscompete for the same intermediary, the automakers engage in Bertrand competition and set theprice at the minimum between the second lowest and monopoly price. The following two scenariosanalyze demand, pricing strategies, and profits for both automakers and intermediaries in the abovetwo situations.Scenario 1: s is the only automaker who deals with i. (Monopoly)In this situation, automaker s maximizes its profit and sets monopoly price for i:ps =σσ − 1 · csGiven ps, the price intermediary i charges to foreign buyer is:pi(s) =σσ − 1 · ps =(σσ − 1)2· cs24No fixed cost of entry is assumed for intermediaries because cost associated with business presence in a foreignmarket is usually country-specific rather than product-specific. Moreover, even if the fixed cost is product-specific,for the majority of intermediaries the share of auto exports is very low (as shown in Section 2.2.1) which indicates arather small fixed cost easy to be covered.25For example, in Ahn et al. (2011) firms could choose to pay a fixed cost to match with an intermediary and getaccess to all foreign markets. Intermediaries are assumed to be homogeneous in their model.26Anecdotal evidence is provided by agents working in Chinese trading firms. The role of intermediary tends tobe agents hired by foreign buyers in search of qualified suppliers. This type of business model is not uncommon inChina and Midler (2010) records a rich variety of illustrations.192.3. A model of automaker-intermediary matchesDemand received by intermediary i is:qi(s) = Ai ·(σσ − 1 · cs)−σ· κM = Ai · c−σs · κ2Mwhere κ = ( σσ−1)−σ and M = XPσ−1. This means that foreign orders received by intermediary idepends on its realized quality Ai as well as the efficiency (or cost) of its matched automaker. Themultiplicative demand consists of both Ai and cs also indicates that high quality intermediariesbring more demand for automakers and especially those with low costs. The profit of automaker sdecreases with its cost cs and increases with intermediary quality Ai, i.e.,pis(i) = (ps − cs) · qi(s) = 1σ − 1 · c1−σs ·Ai · κ2M.Similarly, the profit of intermediary i becomespii(s) =1σ − 1 · ps · qi(s) =σ(σ − 1)2 · c1−σs ·Ai · κ2M.Scenario 2: intermediary i deals with multiple automakers. (Bertrand)If multiple automakers favor the same intermediary i, they engage in Bertrand competition. Ifautomaker s has the lowest cost, it will set price to the cost of that intermediary i’s next lowestcost automaker s′, i.e.,ps = cs′ .Consider the situation where cs′ is large enough to exceed the monopoly price which generates themost profit for automaker s. In that case, automaker will set the price at whichever is lower. Thatis,ps = min{cs′ , σσ − 1 · cs}.If seller s sets the monopoly price, then the following analysis will be the same as in Scenario 1. Ifcost differences are relatively small compared to automaker’s monopoly power, price will be set atthe cost of s′. Given ps, the price intermediary i charges to foreign buyer is:pi(s) =σσ − 1 · ps =σσ − 1 · cs′ .Demand received by intermediary i becomesqi(s) = Ai · c−σs′ · κM.The lowest cost automaker s makes profitpis(i) = (cs′ − cs) · qi(s) = (cs′ − cs) ·Ai · c−σs′ · κM.202.3. A model of automaker-intermediary matchesThere is a trade-off for automaker’s profit: on the one hand, automaker s would prefer its competitors′ to have a high cost so that the markup will be high; on the other hand, cs′ has to be low enough tobring enough demand. Profit is always lower in Bertrand competition (with the same intermediary)where ps = cs′ <σσ−1 · cs. Finally, the profit of intermediary i ispii(s) =1σ − 1 · cs′ · qi(s) =1σ − 1 · cs′ ·Ai · c−σs′ · κM.2.3.2 Pre-regulation: pricing and order allocationTiming of the pre-regulation game is as follows:1. Intermediaries draw idiosyncratic demand shifter Ai.2. Automakers offer prices ps to supply intermediaries (based on Bertrand competition).3. Intermediaries distribute orders to automakers.Before the regulation, Bertrand competition leads to a Nash equilibrium where each intermediarymatches with his lowest cost automaker. The lowest cost automaker may be within the region oroutside the region, depending on the cost of procurement across regions, τ . This pre-regulationmatching is efficient in the sense that intermediaries all source from their lowest cost producer(s).2.3.3 Post-regulation: pricing and order allocationThe introduction of registration requirements restricts the maximum number of matches to threefor each automaker and changes the sequence of decision making. The timing of post-regulationgame is as follows:1. Automakers observe previous demand shocks of intermediaries and form expectations abouttheir quality.2. Automakers list at most three intermediaries with the registry.3. Intermediaries draw demand shifters Ai as before.4. Automakers offer prices to supply intermediaries.5. Intermediaries distribute orders to automakers.27The post-regulation equilibrium concept is perfect Bayesian for both the listing and pricing game. Inequilibrium, each automaker lists three intermediaries and no lists contain common intermediaries.In other words, market division arises endogenously from the regulation. Automakers will not listthe same intermediary because if they do, they engage in Bertrand competition and automakerswith higher costs get nothing. It is subgame perfect in the sense that intermediaries will always27Intermediaries can only source from automakers who list them with the registry.212.3. A model of automaker-intermediary matchesturn to the lower cost automakers if they are listed by multiple ones. Backward induction predictsthat automakers will avoid listing the same intermediary.The pricing game is (trivially) part of a perfect Bayesian equilibrium. Specifically, this gameleads to monopoly pricing in the pricing stage since prices are set after matching by which timemarket has already been divided and intermediaries do not have alternative automakers to sourcefrom. Given an arbitrary partition of intermediaries, an automaker changing its list (to competewith a rival for a better intermediary) would be trading a monopoly profit for a Bertrand profit.If a Bertrand profit with a higher quality intermediary is larger than the current monopoly profitwith a lesser intermediary, the automaker with the lower cost would deviate and enter the weakerrival’s market to compete for the high quality intermediary.For some parameter values, the model generates the positive assortative matching equilibriumwhere the lowest cost automakers match with the intermediaries expected to generate the highestorders. It also yields a large number of equilibria where assortative matching does not obtain. Con-ditions under which assortative matching is supported are discussed below. Assortative matchingalways characterizes some of the equilibria and under certain parameters, it is the only equilibrium.Conditions for assortative matchingSuppose the current arbitrary market division is converse to positive assortative matching: au-tomaker s matches intermediary 2 with A2 and automaker s′ matches intermediary 1 with A1,where A1 > A2 and cs < cs′ . The question is whether automaker s wants to deviate from thecurrent monopoly situation with intermediary 2 to engage in Bertrand competition with s′ for thehigher quality intermediary 1. If automaker s stays with current market division, it will getpis(2) =1σ − 1 · c1−σs ·A2 · κ2M.If automaker s deviates to compete with s′ for A1, its profit would bepis(1) = (cs′ − cs) ·A1 · c−σs′ · κM.The first case to consider is when Bertrand competition posits a binding restriction, i.e., ps =cs′ <σσ−1 · cs. Non-sssortative matching is ruled out ifpis(1) > pis(2),i.e.,A1A2︸︷︷︸>1· (cs′ − cs)1σ−1 · cs︸ ︷︷ ︸<1· c−σs′( σσ−1 · cs)−σ︸ ︷︷ ︸>1> 1.This condition is satisfied if variation in quality is large and the second lowest cost gets closer tothe monopoly price.222.3. A model of automaker-intermediary matchesA second case occurs when Bertrand price is not binding, i.e., ps =σσ−1 ·cs < cs′ and automakers becomes better off to charges monopoly price. The profit of automaker s will always increase byswitching from intermediary 2 to higher quality intermediary 1.This discussion shows that non-assortative matching may not be equilibria whereas assortativematching is always an equilibrium. Also, we could get partial assortative matching if the aboveconditions are met by some pairs but not by the others.2.3.4 Post-regulation: welfare analysisAutomakers. The lower cost automakers favored by intermediaries pre-regulation will likely beworse off post-regulation because although they charge weakly higher prices, some lose intermedi-ary business due to the registration that limits relationships to three. But if the distribution ofintermediary quality is very skewed and those being dropped would not bring many orders anyway,their elimination will not have a large effect on the profit of the low cost sellers. Whether or notthe lower cost automakers become worse off depends on the trade-off between higher markup andfewer orders. Some high-cost automakers who did not receive business prior to the policy and nowserve intermediaries are better off.Intermediaries. Intermediaries of lowest qualities are not listed by any automaker. The regu-lation mechanically cut the number of intermediaries from N to 3M . The remaining intermediariesare worse off as well since they are faced with higher costs for two reasons: 1) all but three lose ac-cess to the low-cost automaker and 2) they face monopoly rather than Bertrand prices. Therefore,the total profits for intermediaries are decreasing post-regulation.Aggregate profit. Since the downstream intermediaries engage in monopolistic competition(MC) for foreign buyers and the upstream competition among automakers changes from Bertrandto MC, the integrated market structure for Chinese automobile exporters to compete for foreignbuyers is subject to double marginalization post-regulation (rather than the one layer markuppre-regulation). This acts to reduce aggregate profits.Let us imagine the special case of perfect competition among automakers pre-regulation (i.e.,zero cost variation). Buying from the lowest cost automaker with price set at the second lowestcost is the same as buying from any automaker (with second lowest cost c = cs′). Intermediariesare the only ones who mark up the prices to foreign buyers. Total profit piPre generated by eachautomaker-intermediary pair with cost c and quality A ispiPre(c) =1σ − 1 ·A · c1−σ · κ ·M.With market division post-regulation, automakers with cost c price at σσ−1 · c and total profit ofthe same automaker-intermediary pair becomespiPost(c) =[(σσ − 1)2− 1]·A · c1−σ · κ2 ·M.232.3. A model of automaker-intermediary matchesTherefore, the profit difference ∆pi is∆pi = piPost(c)− piPre(c) = A · c1−σ · κ ·M ·(ρκ− 1σ − 1)< 0where ρ = ( σσ−1)2 − 1.Next, we move on to consider the case where automaker costs are heterogeneous and they chargeps =σσ−1 · c′ where c′ ≥ c. The total profit of an automaker-intermediary pair becomespiPost(c′) = ρ ·A · (c′)1−σ · κ2 ·M.The ratio between post and pre regulation profits can be expressed aspiPost(c′)piPre(c)=( cc′)σ−1 · (σ − 1) · ρκ < 1.Thus, total profits for all the newly formed automaker-intermediary pairs are decreasing. For thealready existing partnerships including the lowest cost automakers, total profits stay the same ifmonopoly prices are lower than Bertrand prices (i.e., σσ−1 · cs < cs′) and decrease if Bertrand pricesare binding (i.e., cs′ <σσ−1 · cs).Inefficiencies also arise from non-assortative matching equilibrium where a high cost automakeris matched with a high quality intermediary who otherwise generates more profits with a lower costautomaker. In addition, intermediaries who are weeded out by the regulation no longer generateprofits. Aggregate profit decreases post-regulation.2.3.5 Bayesian updating processThe model allows uncertainty in intermediary demand shock (Ai) based on its quality µi, Bayesianupdating process is introduced to facilitate the formation of intermediary quality expectations. Tofully describe the updating process, a prior distribution of model parameters needs to be specified.Usually a conjugate prior will be chosen due to algebraic convenience, providing a closed-formexpression for the posterior. Given that intermediary qualities are assumed to follow a Poissondistribution with parameter µi, a natural prior choice is the Gamma distribution,µi ∼ Gamma(α0, β0), ∀i.Then the sampling model given µi can be specified asAti|µi ∼ Poisson(µi).Although automakers do not observe µi, they know these parameters are drawn from a Gammadistribution with hyper-parameters (α, β). Therefore, automakers could use realizations of previous242.4. Simulationdemand shocks (A1, A2, A3) to estimate (α, β) based on the assumption that Ati are i.i.d.28 andα = v −mβ = vm − 1where m = Mean(Apre), v = Var(Apre), and Apre = {A1, A2, A3}.29. At = {At1, At2, · · · , Atn} is usedto summarize observed demand shocks of intermediaries in each period. And t = 1, 2, 3 indicatespre-regulation periods while t = 4, 5, 6 stands for post-regulation years.Once the estimated priors (α, β) are obtained, automakers are able to update the expectedquality of each intermediary (µi) based on its previous realized demand shocks {A1i , A2i , A3i },E(µti|At−1i ) = (α+t−1∑t=1Ati)/(β + t− 1)where t = 4, 5, 6.Two assumptions worth mentioning are implicitly incorporated into the model. First, automak-ers are assumed to share the same expectation of intermediary qualities. Common expectationshelp simplify the model solution while preserving key features of the matches. Second, the modelassumes that realized demand shocks of intermediaries who were not listed by any automaker onthe registry cannot be observed. Automakers only use the last observed realized sales to formexpectations.2.4 SimulationIn this section, I simulate the model to see if it delivers results that are consistent with stylizedfacts of the industry described in Section 2.2. The simulation follows the order of steps shownin Section 2.3.2 and Section 2.3.3. Specifically, each intermediary sources from the lowest costautomaker to him pre-regulation.30In the post-regulation periods, automakers form common expectations about the orders thatintermediaries will deliver based on previously observed orders and the Bayesian updating process.If the cost of selling across regions τ is zero, the ranking of automakers is shared by all the inter-mediaries. Then the lowest-cost automaker that has not filled its registry lists the three availableintermediaries that are expected to bring the largest orders. Automakers do not list intermediariesthat already are chosen by lower-cost automakers because Bertrand competition implies that they28Although qualities are not directly observed by automakers, costs of automakers, general market conditions,matches among automaker and intermediaries and orders generated by intermediaries are common knowledge. Au-tomakers can perfectly infer intermediary qualities based on these revealed information at the end of each period.For simplicity, I model automakers to update Ai directly.29The equations used to estimate hyper-parameters (α, β) are derived from the predictive Negative-binomial dis-tribution of Ati. Details of the deduction can be found in Appendix A.3.30Due to the cost of selling across regions τ , intermediaries located in different regions may not share the samelowest cost automaker.252.4. Simulationwould not get the orders of this intermediary. Once the registry lists are established, intermedi-aries realize demand shifters, automakers offer prices to supply intermediaries, and final prices andprofits are obtained.If τ is large enough to make the lowest-cost automakers different across regions, the simulationprocedure is slightly adjusted to incorporate τ . Specifically, intermediaries ranking from high tolow based on their expected order sizes take turns to fill one of the three registration slots of hislowest cost automaker available. For example, if the first three intermediaries prefer the sameautomaker, that automaker is no longer available for other intermediaries to choose from. Theregistry is established once all the automaker slots are filled.The simulations assume assortative matching for the following reasons. First, assortative match-ing always characterizes some equilibria and with certain parameters it is the only equilibrium. Sec-ond, it is natural to consider assortative matching since efficient automakers are more motivated tospend a little effort to list people early and they would match with the best intermediaries. Finally,it is the best case scenario for government policy in terms of total surplus.I will begin with a brief discussion of parameter values chosen in the simulation and then moveon to present the simulation results.2.4.1 Choice of parametersTable 2.9 presents a list of parameters used in the simulation and the values chosen for each ofthem.Table 2.9: Description of model parametersParameter Description Value(α0, β0) parameters of Gamma distribution (2,1)M number of automakers 3, 3, 4aN number of intermediaries 16, 16, 18c0s automaker production cost c0s ∼ N(0.8, 0.01)τ cost of selling across regions τ = 2%× c0sσ demand elasticity 3ba It lists the number of automakers in each of the three regions.b Broda and Weinstein (2006) estimate the demand elasticity for Motor Cars andOther Motor Vehicles in the US to be 3. κM is normalized to 1.The parameters governing the Gamma distribution of intermediary qualities (µi) are chosen tomatch the distribution of intermediary order sizes which is granular and highly skewed. However,one difficulty is that the intermediary demand shock Ai (based on µi) can not be directly observed.Recall that the orders brought by intermediary i pre-regulation is qi(s) = Ai · c−σs · κM . Since inthe pre-regulation period, intermediaries tend to choose the same automaker to fill their orders,variations in cs is small and variations in Ai is required to explain variations in qi(s). Settingα0 = 2 and β0 = 1 generates the necessary skewness and variation in Ai as shown in the left panelof Figure 2.7. The right panel of Figure 2.7 shows the real distribution of intermediary order size262.4. Simulationfor comparison.31Figure 2.7: Intermediary demand shock (simulation) vs order size (data)The number of automakers and intermediaries are set to 10 and 50 respectively.32 They arerandomly assigned to 3 different regions. Each region receives roughly the same number of au-tomakers and intermediaries (shown in Table 2.9). Costs are generated from normal distributioncs ∼ N(0.8, 0.01). In order to address the effect of intermediary quality, the costs of automakers arescaled down. The cost of selling across regions is assumed to be 2% the cost of production. That is,τ = 2%× c0s. In the simulation, the pre-regulation same-region matches accounts for 36% while theshare of same region matches ranges from 33% to 43% post-regulation. Elasticity of substitution σis assumed to be 3 which represents a 50% monopoly markup.33 The product of demand elasticityand total automobile expenditure in the foreign country, κM , is normalized to 1. As mentionedpreviously, three periods before and three after the policy are simulated.2.4.2 Simulation resultsSimulation results corresponding to model predictions and stylized facts described in Section 2.2are presented in this subsection.First, the number of intermediaries with positive orders and their probability of exiting areexamined. Figure 2.8 presents the drop of intermediaries right after the policy. It is mainlydriven by the limited number of intermediaries allowed to be listed by each automaker on theregistration. The slightly increasing number of intermediaries post regulation results from the fact31Appendix A.3 provides a detailed explanation of how α0 and beta0 are associated with the skewness of thedistribution and the relative importance of priors in the Bayesian updating process.32Before registration is required for auto exporters, automakers and intermediaries do not identify themselves.This posits challenges for researchers to determine the ratio of automakers and intermediaries who would potentiallyhandle auto exports. The registration lists provide some hints and when combined with firm name orthography,several implications could be made to estimate the boundaries of this ratio. To make the paper succinct, discussionsregarding the ratio are presented in Appendix A.4.33σ = 3 is borrowed from Broda and Weinstein (2006) which estimates the elasticity of substitution for Motor Carsand Other Motor Vehicles in the US.272.4. Simulationthat automakers have more accurate information about intermediary qualities due to Bayesianupdating process. It enables automakers to list intermediaries who are more likely to bring orders.Figure 2.8: Number of intermediariesConsistent with the stylized facts, exiting intermediaries in the simulation have fewer automobileexport orders both before and after the regulation (Table 2.10). Intermediaries post regulation onaverage bring fewer orders than their pre-regulation counterparts. This is caused by increases inwholesaler prices. The last three columns of Table 2.10 indicate that intermediaries with largerexport sales are less likely to exit which is consistent to Fact 2 shown in section 2.2.2. However, inthe simulation the disadvantage of smaller intermediaries is not getting worse post-regulation.Table 2.11) provides information on the persistence of matches. The drop rate of automaker-intermediary relationships is as high as 57% according to the simulation which matches the observedscale of drop rate in data. The matches dissolve as automakers update priors. High level ofdissolution is mainly driven by the large number of small intermediaries who would bring ordersoccasionally and make it difficult for automakers to determine whom to put on the list.Next, simulation results on prices received by intermediaries are analyzed. Since intermediariescharge a constant markup when they sell to foreign buyers, changes in intermediary exportingprices should share the same pattern with wholesale prices. Among the 30 intermediaries out of50 listed in all periods, only 3 of them remain matched with their lowest cost automaker. Allthe other intermediaries will source from a less efficient automaker and prices will rise. Therefore,the first source of price increase comes from inefficiencies in matching. The second source of priceincrease comes from the structure of competition. Automakers shift from Bertrand competition pre-regulation to monopoly pricing post-regulation due to market division endogenously implementedby the regulation.Figure 2.9 shows that on average prices received by intermediaries increase post-regulation.Specifically, wholesale prices are about 65% (e0.6− 1=0.65) higher than before. Provided that costvariation is set small in the simulation, the margin of price increase is mainly driven by monopoly282.4. SimulationTable 2.10: Exiting vs surviving IntermediariesY=ln(q) Y=Prob(exit)(1) (2) (3) (4)OLS Hzd LP Hzd logit Hzd probitexit -0.582a(0.137)post -1.359a -0.313a -2.710a -1.522a(0.092) (0.060) (0.866) (0.431)exit×post 0.156(0.214)ln(q) -0.223a -1.562a -0.882a(0.050) (0.447) (0.241)post×ln(q) 0.129c 0.309 0.184(0.067) (0.670) (0.358)cons 0.580a 0.332a -0.785a -0.471a(0.095) (0.048) (0.221) (0.133)N 169 169 169 169R2 0.508 0.130Standard errors in parentheses. c p < 0.1, b p < 0.05, a p < 0.01.Intermediary-year level regression with intermediary clusters.Table 2.11: # of Automaker-Intermediaries pairst New Continue Dropped Total DropRate4 . . 17 30 0.575 17 13 13 30 0.436 13 17 . 30pricing with a 50% markup (σ = 3).34Welfare in terms of total industry profits.As discussed before, the lower cost automakers who receive orders pre-regulation will likely beworse off post-regulation because of lost orders since the registration limits the relationships to amaximum of three. But if the distribution of intermediary quality is very skewed and those beingdropped would not bring many orders anyway, dropping them will not have a large effect on theprofit of the lowest cost seller. The grey dash line in Figure 2.10 shows profits of the lower costautomakers who receive orders pre-regulation.35The pre-regulation profits are determined by thequality draws of intermediaries. The post-regulation profit of the lowest cost automaker actuallyincreased suggesting the effect of higher markups overcome the damage of lost orders.34The 2% search cost is not high enough to prevent intermediaries from sourcing from the most efficient seller evenif that seller locates in a different region. Since matches in pre-regulation periods do not change, there is no wholesaleprice variation for intermediaries then.35Under the assumed parameters in this simulation, only the most efficient automaker is used pre-regulation.292.4. SimulationFigure 2.9: Price change of intermediariesMarginal effects with 95% CIs. No price variation pre-regulation (t=1,2,3) due to single automaker.The solid black line in Figure 2.10 presents the total profit of all automakers.36 Some high-costautomakers who did not receive business prior to the policy now serve intermediaries. The largeincrease in total profits post-regulation shows that they are better off. In addition, the total profitincrease from period 4 to 5 can be attributed to the Bayesian updating process where automakershave better expectations of intermediaries and are more likely to list the ones who will bring orders.Figure 2.10: Profit of automakers36Since the lowest cost automaker is the only one matched with all intermediaries and its profit represents the totalprofit of automakers pre-regulation.302.4. SimulationFigure 2.11 displays the total profit of intermediaries and the aggregate profit of the autoexport industry. The regulation mechanically reduce the number of intermediaries together withthe profits. The remaining intermediaries are worse off as well due to higher wholesale prices asdiscussed above. The automobile industry as a whole becomes worse off post-regulation.Figure 2.11: Total profitParameter values used in this simulation: M=10, N=50,c0s ∼ N(0.8, 0.01), τ = 2%× c0s, and µi ∼ Gamma(2, 1).Welfare in the foreign country will fall for two reasons based on this model. First, foreign buyersconnected with intermediaries not being listed by any automaker will have to drop the order or findalternative sellers. Due to the regulation, intermediaries can only source from automakers who listthem on the registry. Foreign consumer surplus is lost due to unfulfilled orders. Second, even ifforeign buyers link with intermediaries on the registry, they are still worse off due to higher pricestransferred from the monopoly pricing of automakers. Foreign buyers may potentially gain in thelong run if the regulation pushes them to search and connect with direct exporting automakers.However, the interaction between direct and indirect exporting is assumed away in the currentmodel.Domestic consumers are hardly affected by this regulation. The large share of automobilesales in the Chinese market consists of passenger cars while the majority of China’s auto exportsis commercial vehicles such as buses and trucks. No direct interactions or spillovers are assumedbetween the domestic and foreign market in this paper. Therefore, export registration of automakersand intermediaries does not affect the production and sales in the domestic market.Alternative rationales for the regulationThe model associates the registration policy with aggregate welfare loss. The winners are a subsetof automakers. Therefore, the model can only provide a political economy explanation for the312.5. Conclusionpolicy—politically powerful automakers pushed the policy through despite the costs inflicted onsome automakers and all intermediaries.A search of Chinese news reports reveal a couple of other explanations for the policy. Onefrom the state media suggests the goal was to correct excess entry in the intermediated auto-exportindustry. Intuitively, if intermediaries incur fixed cost to enter the market and each additionalentrant will dilute the market share of existing firms without lowering prices appreciably, theremight be too many intermediaries. However, Dixit and Stiglitz (1977) show that with CES demand,the monopolistic competition equilibrium number of firms is the social optimum.37 The socialplanner problem in this paper is slightly different from the standard Dixit-Stiglitz setting since theorders brought by intermediaries benefit automakers without additional cost.38 If intermediariesare modeled as measureless firms who do not affect the price index and additional entry do notaffect the current profits of other intermediaries, the socially optimal number of intermediariesshould coincide with the free market equilibrium. Excess entry could be introduced into the modelby allowing firms to have market power and additional entry to generate business stealing effects.A number of papers using different utility functions characterize the conditions under which excessentry exist (e.g., Anderson and Renault, 1999; Gu and Wenzel, 2009; Dhingra and Morrow, 2012).The registration policy may reduce problems with after-sales services. First, it eliminatesmarginal intermediaries who arguably are more likely to offer inadequate after-sales service. Sec-ond, if the commercial dispute arises to government, a smaller number of intermediaries makes iteasier for government to track down the source of the problem.2.5 ConclusionThis paper takes advantage of the automaker-intermediary matching data to provide evidence onthe economics underlying the linkages among domestic producers and export intermediaries. Itshows that large automakers are matched with intermediaries who bring more foreign orders. Inaddition, large skewness in intermediary qualities and Bayesian updating of automaker beliefs giverise to the substantial churning in the sets of intermediaries registered by the automakers. Thesefindings are in line with studies examining seller-buyer matches in international markets.The matching data is made available because of a trade policy initiated to monitor China’sauto exports and potentially promote industry consolidation. This paper proposes a model todescribe how this registration policy facilitated market division as an equilibrium outcome and ledto inefficiency.Predictions from the model and results of the simulation show that this regulation benefitedautomakers especially those relatively less efficient ones who were not used by intermediaries before.Intermediaries became worse off. The survivors faced higher wholesale prices and the lowest qualityones were not listed by any automaker and exit the market. Inefficiencies of the policy mainly came37Lump sum subsidies which transfer consumer surplus to producers are not allowed.38Welfare of foreign consumers is not taken into consideration.322.5. Conclusionfrom endogenously risen market partition and double marginalization. Non-assortative matchingequilibria also generated inefficiencies since low cost automakers benefited more from high qualityintermediaries.Welfare consequences discussed in this paper are drawn in light of the model. The questionarises about why the government issued this regulation. A reading of media accounts does notreveal the rationale. Based on the model, a political economic argument can be constructed whereautomakers lobbied to reduce competition. Other potential rationales such as reducing fixed costsand extracting more foreign consumer surplus require additional modeling.Information on manufacturer-intermediary provides a unique opportunity to model matchingbetween firms. Future research is required to fully understand the Chinese government rationalefor the regulation.33Chapter 3Productivity, market penetration andallocation of sales3.1 IntroductionTrade literature has long been focusing on the difference between exporters and non-exporters.There is an ongoing debate over the causal relationship between firm performance and export status.Previous studies address this issues from a variety of perspectives emphasizing fixed costs of export,factor intensity, ownership, features of processing trade, and with the help of field experiment (seeMelitz, 2003; Dai et al., 2016; Defever and Rian˜o, 2017; Lu et al., 2010; Lu, 2010; Atkin et al.,2017). However, little has been done regarding the relationship between firm productivity andhow much firms export relative to sales in the domestic market. Based on information aboutChinese exporters, this paper provides new evidence on how firm productivity is associated withthe allocation of sales between foreign and domestic market.The empirical phenomenon documented in this research is rather surprising: The export todomestic sales ratio of Chinese firms is decreasing in firm productivity. This negative correlationbetween firm productivity and export intensity remains strong when we control firm ownership, fac-tor intensity, and rule out impacts of processing trade. This finding is in stark contrast with Melitz(2003) predictions where high productivity firms enter more markets and sales are proportional tofirm productivity conditional on market entry. Thus, sales ratio across markets are independentof firm productivity and high productivity firms are associated with high export intensity due tothe large number of markets they enter. Arkolakis (2010) and Eaton et al. (2011) build a slightlydifferent model with market penetration technology instead of fixed cost of entry that predicts thatfirm productivity is positively correlated with relative sales in the foreign market. Their predictionalso contradicts the empirical patterns we observe.To rationalize our empirical findings, we extend the Arkolakis (2010) model to allow for het-erogeneous marketing cost elasticities across countries. In this model, firms must incur marketingcosts to reach consumers. The returns to marketing is decreasing and the marketing cost elasticitygoverns the speed of deterioration. The model links firm productivity and sales through two mar-gins. The intensive margin, sales to existing consumers, increases proportionally across markets asfirm productivity grows. The extensive margin, the number of consumers reached, is governed bythe interaction of the marketing elasticity and firm productivity. In markets where the elasticity issmall, firms are able to easily approach consumers, making the role of productivity minimal. On343.1. Introductionthe contrary, when the market is difficult to penetrate and the marketing cost elasticity is high,productive firms have an advantage in reaching consumers. When the marketing cost elasticityis higher in China than abroad, productive firms have a relative advantage in selling to domesticconsumers resulting in a negative correlation between firm productivity and the foreign-domesticsales ratio.We conduct an exercise to support our market penetration explanation of the negative relation-ship between firm productivity and export intensity. Specifically, we hypothesize that if an industryrelies more on marketing, the negative correlation between firm productivity and export sales ra-tio should be more pronounced. We use advertisement expenditures over sales of the industry tocapture the relative importance of marketing. The interaction term between firm productivity andadvertising intensity is negative and significant in the export sales regressions which supports ourmarketing story.Finally, we compile evidence contradicting alternative explanations for a negative relationshipbetween export intensities and firm productivity. The first explanation is based on Melitz andOttaviano (2008) where markups are determined by the interaction between firm productivity andmarket size. More efficient firms charge higher markups but this is weakened in large marketsdue to increased competition. If the Chinese market is larger than the foreign market, the modelpredicts that export intensities and the ratio of home to foreign prices will decrease with firmproductivity. In contrast, our model predicts declining export intensities but constant markups.Using Hong Kong as a close substitute to the Chinese market, we examine export-to-domesticprice ratios and find evidence in favor of our model. The other alternative explanation concernsthe quality of products. Manova and Zhang (2012) find that firms vary in terms of the qualityof their products across destinations due to different inputs quality levels. They also predict thatmarkups are heterogeneous across firms. If this were true, then the price ratio between two exportdestinations should be correlated with firm productivity which is not supported by our empiricalresults.This paper is closely related to two strands of literature. The first consists of studies thatinvestigate the exporting performance of Chinese firms. Lu et al. (2010) find that, among foreignaffiliates, exporters are less productive than non-exporters in China. They argue that the fixedcost for foreign affiliates in foreign markets is lower than that in the Chinese market. Dai et al.(2016) argue that the low productivity of exporting firms is entirely driven by firms that engageonly in export processing—the activity of assembling tariff-exempted imported inputs into finalgoods for resale in foreign markets. These firms are less productive than non-exporters and causea decrease in the average productivity across all exporting firms. Lu (2010) uses factor intensityto explain the productivity difference between exporting firms and non-exporting firms. Whencountries differ in their factor endowment, sectors that are intensive in the locally abundant factorface higher competition in the domestic market than in foreign markets. Since China has a hugelabor supply, domestic rather than foreign markets select the most efficient firms in labor-intensiveindustries. Huang et al. (2015) also find that production in China from 1999 to 2007 became more353.2. Data and stylized facts of sales allocationcapital-intensive, while exporting became more labor-intensive. Defever and Rian˜o (2017) arguethat pure exporting firms are less productive than regular exporting firms, due to export subsidies.The second strand of literature investigates the exporting behavior of firms using data fromdeveloped countries. Bernard et al. (2003) use US data to study why exporting plants only exporta small fraction of their output. Crino` and Epifani (2012), using Italian data, find that Total FactorProductivity (TFP) is strongly negatively correlated with the export share to low-income desti-nations and uncorrelated with the export share to high-income destinations. In contrast to theirfindings, this paper finds that the negative correlation remains robust regardless of destinations.The rest of this paper is organized as follows. In Section 3.2, we introduce the two maindatasets used in this paper, presents the stylized facts as well as the empirical specifications aboutfirm productivity and sales ratio. In Section 3.3, we develop a theoretical model with heterogeneousmarketing cost elasticities to rationalize the empirical findings. In Section 3.4, we provide evidencebased on the importance of marketing across industries and check alternative explanations. Thelast section concludes this paper.3.2 Data and stylized facts of sales allocationIn this section, we first introduce the databases used in this research and describe the distributionof export intensity (i.e., exports over total sales) by ownership for Chinese exporters. Next, wedepict the correlation between firm productivity and export intensity with various controls. Inorder to make the empirical results directly comparable with model predictions, we move fromexport intensity to export-domestic sales ratio and include destination-specific controls.3.2.1 Data and summary statisticsIn this paper we combine two databases over the 2000-2006 period: The firm-level production datafrom Chinese Industrial Enterprises Database (henceforth Firm Survey) and the transaction-leveltrade data from Chinese Customs Export and Import Database (henceforth Custom Record).The first database is collected by China’s National Bureau of Statistics. It includes all state-owned enterprises (SOEs) and non-SOEs with annual sales above 5 million Chinese Yuan (about0.6 million US dollars39). The Firm Survey contains firm balance sheet information such as sales,employment, assets, intermediary inputs, export sales, and etc. Only manufacturing firms withpositive value-added, capital and sales remain in our sample. We also drop small firms with five orfewer employees or firms without valid postal codes. The number of firms in our sample ranges from134,775 in 2000 to 258,586 in 2006. Table 3.1 shows that around 30% of manufacturers included inour sample are exporters. Firms with foreign capital participation (i.e., joint ventures and foreign-owned enterprises) are more likely to export although they account for only 22% of all firms. Thesepatterns are in line with a number of papers (e.g., Dai et al., 2016; Lu et al., 2010).39The US Dollar to Chinese Yuan exchange rate during 2000-2006 was around 8.27.363.2. Data and stylized facts of sales allocationTable 3.1: Share of exporters by year or ownershipNumber of Firms Share of ExportersPanel A: by Year2000 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 OwnershipState-owned enterprises 523,540 18.21%Private-owned enterprises 517,403 20.99%Joint ventures 146,129 56.40%Foreign-owned enterprises 141,211 71.23%All Firms 1,328,283 29.13%Notes: This table summarizes the composition of firms and their export behaviorin Firm Survey Database. Only manufacturing firms with positive value-added,capital and sales remain in our sample. We also drop small firms with less thansix employees or firms without valid postal codes. Foreign-owned enterprises andjoint ventures categories include investors from Hong Kong, Macao, Taiwan andother foreign countries.In the second database, constructed by the Chinese Customs Office, trade transaction dataare collected for each 8-digit harmonized system (HS) product. It provides detailed informationon trade status (import or export), product quantity, trade value, origin and destination of eachtransaction, trade mode (ordinary or processing trade), firm associated with each transaction, firmlocation, and etc.While both databases contain firm identification numbers they follow different constructionrules. Therefore, we follow the matching algorithm proposed by Wang and Yu (2012) and use firmname, telephone number, name of the manager, and postal code to match firms by orthography.After merging the two databases we obtain 177, 396 firms, which accounts for 45.85% of self-reporting exporters40, and whose export sales comprise 54.54% of total self-reporting export value.Export intensity is defined as export divided by total sales. Table 3.2 shows that the distributionof export intensity is polarized. About 16% of exporters sell less than 10% of their output abroadand over 41% of exporters have an export intensity over 0.9. Over half of the wholly foreign-ownedenterprises (WFOEs) and joint ventures (JVs) export 90% of their output. Domestic firms exportless intensively compared with WFOEs and JVs. But there are still more than 30% of them shippingthe majority of their products to foreign markets. This phenomenon is in stark contrast to the40Bai et al. (2017b) pointed out that many firms export through trade intermediaries and that is why we observepositive export values from Firm Survey but could not trace them in the Custom Record.373.2. Data and stylized facts of sales allocationobservation in Bernard et al. (2003), who find that around two-thirds of US exporters sell less than10% of their output abroad.Table 3.2: Distribution of export intensity by ownershipExport Intensity Full Sample Domestic WFOE/JV(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%Notes: This table shows the distribution of export intensity byownership for Chinese exporters. Only manufacturing firms withpositive value-added, capital and sales remain in our sample. We alsodrop small firms with less than six employees or firms without validpostal codes. Foreign-owned enterprises and joint ventures categoriesinclude investors from Hong Kong, Macao, Taiwan and other foreigncountries.3.2.2 Firm productivity and export intensityWe use two ways to measure firm productivity: value-added per worker and TFP constructedfollowing the method in Levinsohn and Petrin (2003).41In order to show the correlation between firm productivity and export intensity, we first rankexporters by their productivity and divide them into 100 percentiles. Then, we calculate the averageexport intensity for each productivity percentile with 95% confidence intervals. Figure 3.1 presentsa negative correlation between firm productivity and export intensity.42 In this paper, we do notexamine the relationship between firm productivity and their export status. Instead, we emphasizethe effect of firm productivity on sales allocation between domestic and foreign markets given thata firm has already entered the foreign market(s). Thus, we drop all non-exporting firms and pureexporters (i.e., firms that export all of their output)43.The correlation between firm productivity and export intensity is further examined with in-dustry, ownership, and export mode controls respectively in order to check the robustness of the41In Appendix B.1, we discuss these two measurements of productivity and Figure B.1 shows that these twomeasurements are correlated with a coefficient of 0.76.42In Appendix B.2 we discuss the slightly hump-shaped correlation more carefully.43There are 111, 052 firms that are pure exporters. 14% of them are SOEs, 26% are private-owned, 21% are JVsand 39% are WFOEs. Defever and Rian˜o (2017) argue that the exporting behavior of pure exporters is different fromthat of regular exporting firms due to tax subsidies.383.2. Data and stylized facts of sales allocationFigure 3.1: Average export intensity and productivity percentile. Export Intensity0 20 40 60 80 100Value-added Per Worker Percentile. Export Intensity0 20 40 60 80 100TFP Percentile95% Confidence IntervalAverage Export IntensityNotes: This figure shows the correlation between firm productivity and export intensity. We excludenon-exporters and pure exporters (i.e., firms that export all of their outputs). We first rank firms by theirproductivity and divide them into 100 percentiles. Then, we calculate the average export intensity with95% CIs for each productivity percentile. The x-axis is the productivity percentile and the y-axis is theaverage export intensity within the corresponding percentile.pattern and rule out confounding factors.First, we investigate the correlation industry by industry to control the effect of factor intensityas a lurking variable. It is well known that China’s exports mainly come from labor intensiveindustries such as the manufacture of textile and electric machines. These labor intensive industriestake advantage of China’s cheap labor forces without much upgrade in production technology andusually consists of low productivity firms. Lu (2010) argues that capital-labor intensity differencesacross industries affect the correlation between firm productivity and export status. Figure 3.2demonstrates that even within the Manufacture of textile, higher productive firms have on averagelower export intensity.44Second, processing trade might be another lurking variable that dominates the negative corre-lation between productivity and export intensity. Table 3.3 reveals that over half of the Chineseexporters are engaged in processing trade and they export the majority of their output45. Firmsparticipating in processing trade are even less productive than non-exporters (Dai et al., 2016).Therefore, the large number of low productivity firms with assembling activities could potentiallydrive down the overall productivity of exporters who sell more abroad. In order to exclude thiseffect, we drop all firms engaged in processing trade and present the result in Figure 3.3. The robust44Exporters from the Manufacture of textile and Manufacture of electric machines and equipments industriesaccount for about 40% of China’s total export values during 2000-2006 period. Figures showing the correlations fora full set of 2-digit industries as well as a list of complete industry names are presented in the Appendix B.3.45This is one reason why export intensity of Chinese firms are unexpectedly high compared with that of US firms.393.2. Data and stylized facts of sales allocationFigure 3.2: Average export intensity and TFP percentile (Textile vs Electric machines)Notes: This figure shows the correlation between the average export intensity and productivity (TFP)percentile for two industries: The Manufacture of textile and Manufacture of electric machines andequipments. We only include firms that sell both in the domestic market and foreign markets.negative correlation between firm productivity and export intensity demonstrates that trade mode(or processing trade) does not explain the negative pattern.Table 3.3: Trade mode: ordinary vs processingFirm Number PercentBy Trade Mode:Ordinary Trade 85,503 48.20%Processing Trade 91,893 51.80%All 177,396 100%Notes: This table summarizes the merged data. Onlymanufacturing firms are included. We drop firmswhose value-added, capital, sales and export valuesare negative or zero. We also drop small firms withfive or fewer employees or without valid postal codes.Third, we investigate whether firm ownership has an impact on the relationship between pro-ductivity and export intensity. Table 3.2 shows that foreign invested firms (i.e., WFOEs and JVs)exhibits higher export intensity than domestic firms probably due to preferential tax credits andexport subsidies (Defever and Rian˜o, 2017). Lu et al. (2010) argue that among foreign affiliates inChina exporters are less productive than non-exporters. Whether more intensive exporters are oflower productivity is largely unexamined either within certain ownership or in general. We address403.2. Data and stylized facts of sales allocationFigure 3.3: Export intensity and productivity (excluding processing trade). Export Intensity0 20 40 60 80 100Value-added Per Worker Percentile. Export Intensity0 20 40 60 80 100TFP PercentileAverage Export Intensity 95% Confidence IntervalNotes: This figure shows the correlation between the average export intensity and productivity percentileexcluding processing trade. We only include firms that sell both in domestic and foreign market.the impact of firm ownership by showing the correlation of firm productivity and export intensityby four ownership groups in Figure 3.4. The negative correlation holds for each ownership type.Above all, we have demonstrated that export intensity is negatively correlated with firm produc-tivity among exporters from China. This pattern remains robust when we consider factor intensity,exclude processing trade, and control firm ownership.3.2.3 Firm productivity and sales ratio (firm level)We replace export intensity with sales ratio between foreign and domestic market mainly becausethe sales ratio is directly comparable with our model predictions described in Section 3.3. Sincethe total sales of a firm is the sum of export and domestic sales, the two measures (i.e., exportover total sales and export over domestic sales) are positively correlated. The negative correlationbetween firm productivity and sales ratio does not depend on which measure we use. Separatingexport from total sales in the denominator also allows us to examine the relative sales allocation offirms market by market.46The benchmark regression at the firm level is based on the following specification:ln(Yit) = α0 + α1 ln(φit) + Xit + FEjkt + ijkt, (3.1)46The issue of measurement error may arise if total sales, which is used to construct revenue TFP and appearsin the denominator of export intensity, includes an error term. Using value-add to construct TFP and replace totalsales with domestic sales in the denominator of the dependent variable help alleviate the problem. The sample offirms does not change when we replace export intensity with sales ratio since we only take firms who enter both thedomestic and foreign market into consideration.413.2. Data and stylized facts of sales allocationFigure 3.4: Average export intensity and TFP percentile by ownership. 20 40 60 80 100State-owned. 20 40 60 80 100Private-owned. 20 40 60 80 100Joint Venture. 20 40 60 80 100Foreign-ownedAverage Export IntensityTFP Percentile95% Confidence IntervalAverage Export IntensityNotes: This figure shows the correlation between the average export intensity and productivity (TFP)percentile by ownership. We only include firms that sell both in the domestic market and foreign markets.Foreign-owned enterprises or joint ventures include owners from Hong Kong, Macao, Taiwan and otherforeign countries.where Yit stands for foreign/domestic sales ratio of firm i in year t. φit indicates firm productivitymeasured by value added per worker as well as TFP. Xit includes a variety of control variables suchas firm size, capital/labor ratio, and firm ownership. We add province-industry-year fixed effectFEjkt to capture the unobserved trends of macro economic conditions. ijkt is the idiosyncraticerror.The regression results are presented in Table 3.4.47 Firm productivity, in terms of either value-added per worker (in the upper panel) or TFP (in the bottom panel), is shown to be negativelycorrelated with the sales allocation between foreign and domestic market. Specifically, a 10%increase in firm TFP leads to a 2.66% decrease in the foreign/domestic sales ratio conditional ontotal sales. In other words, when the productivity of a firm becomes higher, it sells relativelymore in the domestic market. The negative coefficients for capital-labor ratio indicate that labor47In Appendix B.1, we include Table B.1 to show that export sales increase with firm productivity. There is noirrationality with the Chinese firm export data.423.2. Data and stylized facts of sales allocationintensive firms export more than selling in the domestic market. This finding is consistent with thefact that China has comparative advantages in exporting labor-intensive products.Table 3.4: Firm productivity and sales ratio (export/domestic)Dependent Variable: ln(Export/Domestic Sales)(1) (2) (3) (4)ln(Labor Productivity) -0.202*** -0.143*** -0.168*** -0.155***(0.017) (0.014) (0.014) (0.018)ln(Capital/Labor Ratio) -0.133*** -0.173*** -0.169***(0.017) (0.017) (0.015)ln(Sale) -0.024(0.027)Constant 0.612*** 0.857*** 0.598*** 0.799***(0.065) (0.084) (0.086) (0.274)Ownership FE X XProvince-Industry-Year FE X X X XCluster By Industry X X X XObservations 275,872 275,872 275,872 275,872R-squared 0.348 0.351 0.368 0.369Dependent Variable: ln(Export/Domestic Sales)(1) (2) (3) (4)ln(TFP) -0.170*** -0.143*** -0.146*** -0.266***(0.021) (0.019) (0.018) (0.023)ln(Capital/Labor Ratio) -0.153*** -0.201*** -0.222***(0.016) (0.017) (0.016)ln(Sale) 0.130***(0.037)Constant 1.002*** 1.359*** 1.086*** 0.609**(0.142) (0.161) (0.163) (0.279)Ownership FE X XProvince-Industry-Year FE X X X XCluster By Industry X X X XObservations 275,872 275,872 275,872 275,872R-squared 0.348 0.351 0.369 0.369Notes: This table shows the correlation between export/domestic sales ratio andproductivity at the firm level.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.In order to eliminate the effect of low productivity assemblers who mainly engage in exporting,we drop all firms participating in processing trade.48 The first column of Table 3.5 shows the48Besides firms doing processing trade, another reason for the decreasing number of observations comes from themerging of two databases (i.e., the Firm Survey and Customs Records) which is required in order to identify theassemblers.433.2. Data and stylized facts of sales allocationregressions results without processing firms. Although slightly smaller in scale, the coefficient offirm productivity remain to be negative. Next, we examine the impact of firm ownership on thenegative correlation between productivity and sales allocation. Firm ownership dummy, Domesticequals to 1 if the firm is state-owned or private-owned and 0 otherwise, as well as its interactionwith productivity are included in the regression. The second column of Table 3.5 shows that highproductivity firms are associated with lower exports relative to sales in the domestic market, andthis effect is more pronounced among domestically owned firms. Coefficients presented in Table 3.5are based on the TFP productivity measure. Regression results with value-added per worker aresimilar and can be found in Appendix B.1.Table 3.5: Firm productivity and sales ratio (robustness)Dependent Var.: ln(Foreign/Domestic)(1) (2) (3)ln(TFP) -0.235*** -0.191*** -0.354***(0.032) (0.025) (0.040)ln(TFP)× Domestic -0.160***(0.038)ln(TFP)× Homogeneous 0.018(0.032)Homogeneous Dummy -0.031(0.234)ln(Capital/Labor Ratio) -0.260*** -0.225*** -0.229***(0.023) (0.015) (0.030)ln(Sale) -0.049 0.140*** 0.254***(0.038) (0.037) (0.049)Constant 2.440*** 1.100*** 0.242(0.300) (0.239) (0.401)Ownership FE X X XProvince-Industry-Year FE X X XCluster By Industry X X XExclude Processing Trade XObservations 69,691 275,872 80,147R-squared 0.474 0.370 0.420Notes: This table shows the correlation between foreign/domestic salesratio and productivity (TFP) at the firm level.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.We suspect that the sales allocation across markets may be associated with features of productvariety offered by different firms.49 A natural candidate of product feature is the elasticity ofsubstitution. We borrow the elasticity of substitution estimates from Soderbery (2015) to construct49Note that domestic sales at the product level is not available and only products being exported can be observed.443.2. Data and stylized facts of sales allocationthe elasticities at the firm level.50 Specifically, the firm level elasticity of substitution σi is definedas the average of σin weighted by the export share of each product n.σi =∑nExport value of product nTotal export value of firm i× σinThe higher the weighted elasticity, the lower the degree of differentiation. Firms are divided intotwo groups based on σi: homogeneous and differentiated goods exporters. We created a dummyvariable homogeneous which equals 1 if σi is larger than the median. We include homogeneousand its interaction with firm productivity into Equation 3.1.The results are shown in the last column of Table 3.5. The coefficient on firm productivity be-comes smaller when product differentiation variables are included. The interaction term is positivebut not significant to suggest any differential effect of homogeneous products.3.2.4 Firm productivity and sales ratio (firm-destination level)We further investigate the negative correlation between firm productivity and sales ratio withineach destination country. Firms of different productivity may export to different markets. Highproductivity firms might be more prone to enter high income countries with fewer sales while lowproductivity firms export a lot in low income markets. The composition of destination countriesmay contribute to the negative relationship based on the total export of firms.To control the influence of destination variations, we construct the sales ratio, Yibt, for eachdestination market b as followsYibt =sales in country bdomestic sales,and include the province-industry-country-year fixed effect, FEjkbt, into the specification.ln(Yibt) = α0 + α1 ln(φit) + Xit + FEjkbt + ijkbt (3.2)The number of destination markets at the firm level is also added as a control variable.The first column of Table 3.6 shows that the correlation between firm productivity and foreign-domestic sales ratio remains to be negative when destination country fixed effects are taken intoconsideration. In particular, a 10% increase in firm TFP leads to a 1.15% drop in the relative salesbetween foreign and domestic market. The number of countries firms enter does not affect therelative sales.To check the robustness of the result, we further divide destinations into high income (OECD)countries and Less Developed Countries (LDCs). The last two columns of Table 3.6 show thatcoefficients on firm TFP only change slightly at the third decimal with different country groups. Inaddition, we run regressions for the top ten export destinations separately and present the results50Soderbery (2015) uses the US trade data and estimates the elasticity of substitution for import goods at the HS8level.453.3. A model with heterogeneous marketing cost elasticitiesTable 3.6: Firm productivity and sales ratio: firm-destination levelDept Var.: ln(Foreign b/Domestic)All OECD LDCln(TFP) -0.115** -0.117*** -0.112*(0.045) (0.0438) (0.0594)ln(Capital/Labor Ratio) -0.205*** -0.211*** -0.194***(0.032) (0.0317) (0.0379)ln(Sale) -0.430*** -0.397*** -0.484***(0.048) (0.0437) (0.0669)ln(No. of markets) -0.033 -0.0860 0.0656(0.054) (0.0538) (0.0590)Constant 2.587*** 2.593*** 2.624***(0.385) (0.357) (0.512)Ownership FE X X XCountry-Province-Industry-Year FE X X XCluster By Industry X X XExclude Processing Trade X X XObservations 560,850 305,163 255,687R-squared 0.650 0.610 0.694Notes: This table shows the correlation between foreign/domestic sales ratio andproductivity at the firm-destination level.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.in Table B.10 of Appendix B.4.513.3 A model with heterogeneous marketing cost elasticitiesIn this section, we extend the Arkolakis (2010) model and allow marketing cost elasticity, how muchthe marginal cost of marketing increases with the number of consumers reached, to be heteroge-neous across destination countries. The model links firm productivity and sales by two margins.The intensive margin, sales to existing consumers, grows proportionally across markets as firmproductivity increases. The impact of productivity on the difference in extensive margins, numberof consumers reached, is small when the market is easy to penetrate while large if it is difficult toexpand the consumer base. We begin with a description of the marketing cost and model setup.Then we derive how marketing cost elasticities enter the sales ratio and how the ratio relates tofirm productivity.51Crino` and Epifani (2012), using data of Italian exporters, examine the relationship between firm productivityand export share to high-income countries from a different perspective. The dependent variable in their paper is theshare of export to OECD countries at the firm level and it captures the relative importance (or attractiveness) ofhigh-income countries. They focus on the export sales allocation among different type of countries while we focus onthe sales allocation between foreign and domestic market controlling the type of destination countries.463.3. A model with heterogeneous marketing cost elasticities3.3.1 Marketing costFollowing Arkolakis (2010), firms must incur marketing costs (e.g., sending out advertisements) toreach consumers in country b. nb ∈ [0, 1] captures the the fraction of consumers (in a market ofsize Lb) a firm aims to reach. The amount of labor required to reach these consumers becomesf(nb, Lb) =Lλb ·1−(1−nb)1−κb1−κb , if κb ∈ [0, 1) ∪ (1,+∞)−Lλb · ln(1− nb), if κb = 1(3.3)There are two parameters governing the cost of marketing. First, λ ∈ [0, 1] captures thecoverage of the marketing technology. When the coverage is narrow (λ = 1), each advertisementonly reaches one consumer, total marketing cost f(nb, Lb) increases with the market size Lb. Whenthe coverage is wide enough to include all consumers in a country (λ = 0), total marketing costbecomes independent of market size.The other parameter κb ∈ [0,+∞), the marketing cost elasticity, measures the degree of de-creasing returns to marketing. That is, within a market, the cost per consumer increases as thenumber of consumers already reached grows. A larger κb corresponds to a faster cost increasewith respect to the size of consumer base. When κb > 0, no firm can saturate the market dueto the surge of marketing cost for every additional consumer. When κb = 0, the marketing coststructure degenerates to the case in Melitz (2003) where firm either enters the market and sell toall consumers there (nb = 1) or stays out (nb = 0). Arkolakis (2010) assumes κb to be homogeneouswhile we allow it to vary across countries.523.3.2 Consumer demandA representative consumer in country b consumes a set of differentiated goods combined by CESutility with elasticity σ > 1 from country a. The goods are offered by a continuum of firms withheterogeneous productivity φ. Each firm is small and cannot affect the price index. The fractionof consumers in country b reached by a firm of type φ from country a is nab(φ). Then, the demandin b for commodity provided by firm φ in country a becomesqab(φ) = Lbpab(φ)−σP 1−σbyb , σ > 1 (3.4)where pab(φ) is the price charged by the firm, Pb indicates the price index, and yb is the total incomein country b which consists of wage level wb and aggregate profit of domestic firms pib.52We assume κb to be country-specific meaning that firms face the same level of difficulty in terms of marketpenetration in country b no matter where they come from. This parameter can be further relaxed to be pair-specificif US firms find it easier to accumulate consumers in Canada than exporters from China. Since we only have dataon Chinese exporters, we make κb country-specific for notation simplicity.473.3. A model with heterogeneous marketing cost elasticities3.3.3 Firm problemThe production technology used by firms is constant returns to scale and labor is the only factorof input. In order to produce q units of products, a firm has to hire qφ units of domestic labor.Suppose the wage in country a is wa. The production cost for a firm with productivity φ in countrya to produce q units of a product isC(φ, q) =waqφ(3.5)We assume the iceberg transportation cost between country a and country b is τab > 1 andτaa = 1.Given the structure of marketing cost (3.3), consumer demand (3.4), and production technology(3.5), the profit of a country a firm selling in country b, ispi(pab, nab;φ) = nabLbybp1−σabP 1−σb− nabLbybp−σab τabwaP 1−σb φ− Lλb1− (1− nab)1−κb1− κb (3.6)Given productivity φ, firms choose the optimal price pab and advertising intensity nab thatmaximize their profits (3.6). Then, we we havepab(φ) = σ˜τabwaφ, where σ˜ =σσ − 1 (3.7)nab(φ) = max{1− (φ∗abφ)σ−1κb , 0} , where (φ∗ab)σ−1 =Lλ−1bybσ(σ˜τabwa)1−σP 1−σb(3.8)φ∗ab is the threshold productivity for firms in country a export to country b, which is not affected bythe elasticity of marketing cost κb. Equation 3.8 indicates that firms of higher productivity reachmore consumers than less efficient ones (i.e., the extensive margin effect), especially in difficultmarkets and when consumer base is already sizable.Simulation results shown in Figure 3.5 provides a simple illustration for the effect of κb. Themarketing cost elasticity is assumed to be higher in the domestic country c, κc > κb.53 Whenκ is relatively low, as in foreign country b, firms selling there could quickly reach the majorityof consumers even with relatively low productivity. The more efficient firms do not exhibit anyadvantage along this margin. On the contrary, when κ is high, as in the domestic market c, theconsumer accumulation process becomes steady and the extensive margin growth is reliant on firmproductivity.Suppose the distribution of firm productivity is Pareto, with probability density function g(φ)and cumulative density function G(φ), as follows:g(φ) = θ(φ∗)θφθ+1, θ > σ − 153The parameters used in this simulation are the following: φ∗cc = 3, φ∗cb = 4,σ−1κc= 1 and σ−1κb= 4.483.3. A model with heterogeneous marketing cost elasticitiesFigure 3.5: Productivity and extensive margin effect n(φ)4 6 8 10 12 14 16 18 2000. Marginal Effect  DomesticForeignForeign/Domestic RatioNotes: This figure shows the simulation results under some given parameters. The extensive margineffects are 1 − (φ∗cbφ)σ−1κb and 1 − (φ∗ccφ)σ−1κc as shown in (3.8). In this simulation, we assume φ∗cc = 3,φ∗cb = 4,σ−1κc= 1 and σ−1κb= 4.G(φ) = 1− (φ∗)θφθ, φ ∈ [φ∗,+∞)where θ is the scale parameter of Pareto distribution and satisfies θ > σ− 1. Thus, the conditionaldistribution of the productivity of firms from country a exporting to country b isµ(φ) =θ(φ∗ab)θφθ+1, if φ ≥ φ∗ab0 , otherwise(3.9)Based on (3.4), (3.7) and (3.8), the export sales of a country a firm (with productivity φ) incountry b can be derived asr(φ) = n(φ)p(φ)q(φ) =σLλb (φφ∗ab)σ−1[1− (φ∗abφ )σ−1κb ] , if φ ≥ φ∗ab0, if φ < φ∗ab(3.10)Firm sales in Equation 3.10 can be decomposed into the intensive and extensive margins. Theextensive margin, [1 − (φ∗abφ )σ−1κb ], captures the fraction of consumers each firm could reach whilethe intensive margin, σLλb (φφ∗ab)σ−1, characterizes the average sales to each consumer reached bythe firm. As discussed above, the marketing cost elasticity κ only affects the extensive margin. Itgoverns the speed of consumer accumulation as productivity rises.493.3. A model with heterogeneous marketing cost elasticitiesIntegrating expression (3.10) across the pdf (3.9), we can obtain the average sales of firmsexporting from country a to country b as the following:r¯ab = σLλb [11− 1/θ˜ −11− 1/(θ˜κ˜b)] (3.11)whereθ˜ =θσ − 1 , κ˜b =κbκb − 1Pareto distribution implies that marketing costs are a constant share of a firm sales:m =θ − (σ − 1)θσProfits and wages can also be expressed as constant shares of income:pia = ηya, wa = (1− η)yawhere η = (σ − 1)/(θσ).3.3.4 Sales ratio between foreign and domestic marketThe sales ratio between destination country b and the domestic market c of Chinese exporters,γb(φ), is defined as:γb(φ) ≡ rcb(φ)rcc(φ)=σLλb(φφ∗cb)σ−1[1−(φ∗cbφ)σ−1κb]σLλc(φφ∗cc)σ−1[1−(φ∗ccφ)σ−1κc] , if φ ≥ φ∗cb0 , if φ < φ∗cb(3.12)Other than the relative market size, the foreign/domestic sales ratio of a firm depends on theproductivity threshold of exporting to country b (φ∗cb) and that for selling in the domestic market(φ∗cc). It also depends on the elasticity of marketing cost in both countries.If we examine the sales ratio along two margins, intensive and extensive, Equation 3.12 indicatesthat the intensive margins change proportionally to firm productivity while the extensive margins(in square parentheses) respond non-proportionally due to the impact of κ. In other words, condi-tional on market entry, the sales ratio to existing consumers does not change with firm productivitysince φ can be canceled out and only market size and productivity cutoffs play a role in determiningthe relative sales along the intensive margin. Due to the existence of κ and the functional form ofextensive margin effect, how changes in firm productivity are linked to the relative number of con-sumers obtained is not readily obvious. Proportion 1 below describes the conditions under whichhigh productivity is associated with low foreign/domestic sales ratio.5454To simplify the illustration, we assume that the threshold of productivity for selling in the Chinese market islower than that for selling in foreign markets, which means φ∗cc < φ∗cb. This assumption is reasonable since most503.3. A model with heterogeneous marketing cost elasticitiesProposition 1:55(a) If κb ≥ κc, then ∂ ln γb(φ)∂ lnφ > 0.(b) If κb < κc, then there exists a φ∗(> φ∗cb) which satisfies1κb(φ∗cbφ∗)σ−1κb1−(φ∗cbφ∗)σ−1κb= 1κc(φ∗ccφ∗)σ−1κc1−(φ∗ccφ∗)σ−1κc. Wehave ∂ ln γb(φ)∂ lnφ ≥ 0 for φ ∈ (φ∗cb, φ∗], and ∂ ln γb(φ)∂ lnφ < 0 for φ ∈ (φ∗,+∞).Thus, when κb ≥ κc, there is always a positive correlation between firm productivity andforeign/domestic sales ratio. When κ are homogeneous across markets, as assumed in Arkolakis(2010) and Eaton et al. (2011), higher productive firms sell more in foreign markets. The intuition isthat the initial consumer base in the domestic market is larger, the decreasing returns in marketingmakes it even harder to gain consumers at home than abroad as productivity increases.In Proportion 1 we show that when κb < κc, the correlation between firm productivity and salesratio is a hump-shaped curve with φ∗ being the turning point. Figure 3.5 presents the relationshipbetween sales along the extensive margin and firm productivity based on the simulation parametersintroduced before. We ignore the component from intensive margin because it does not change withfirm productivity. The extensive margin effect in the foreign market increases very quickly initiallybut stays almost constant later while the extensive margin effect in the domestic market is moresteady paced. As a result, when φ ∈ (φ∗cb, φ∗], the extensive margin effect in foreign markets islarger than that in the domestic market; when φ ∈ (φ∗,+∞], the extensive margin effect in foreignmarkets becomes smaller.The hump-shaped relationship predicted by the model corresponds to the stylized pattern weshow in Figure 3.1 with Chinese exporters. However, the squared productivity is not statisticallysignificant in regressions indicating that φ∗ and φ∗cb are close to each other in our sample.3.3.5 EstimationWe follow the procedure proposed by Arkolakis (2010) to derive the foreign-domestic sales ratio ofChinese exporters normalized by the average export sales to each country. Then, we estimate themarketing cost elasticities using this normalized sales ratio between foreign and domestic market.Firm productivity φ is assumed to follow a Pareto distribution which implies φφ∗cb= (1 −Prcb)−1/θ, where Prcb denotes the productivity percentile of a Chinese firm exporting to coun-try b. Combining this expression, Equation (3.11) and (3.12), we obtain the normalized sales ratioof a Chinese firm belonging to productivity percentile Prcb in market b asrcb/r¯cbrcc/r¯cc=11−1/θ˜ −11−1/(θ˜κ˜c)11−1/θ˜ −11−1/(θ˜κ˜b)(1− Prcb)−1θ˜ [1− (1− Prcb)1θ˜κb ](1− Prcc)−1θ˜ [1− (1− Prcc)1θ˜κc ]. (3.13)Chinese firms first sell in the domestic market and then enter foreign markets.55The proof of Proportion 1 is in the Appendix B.5.513.4. Evidence on the effect of marketing costThen, we apply the nonlinear least squares method to estimate θ˜, κc and κb for the top tendestination countries of Chinese exporters. Individual and average sales of Chinese exporters in eachdestination market as well as firm productivity percentiles can be drawn from data.56 We choose θ˜,κc and κb to minimize the squared difference between the left and right hand side of Equation 3.13.θ˜ is estimated to be 2 and a list of marketing cost elasticity estimates is presented in Table 3.7.57Consistent with the model assumption, marketing cost elasticity in China is estimated to be thehighest, i.e. κc = 2.7. This indicates that the Chinese market is the most difficult to penetrate andit is very costly to accumulate an additional customer. The estimated marketing cost elasticity forChina’s top ten export destinations ranges from 0.326 for Italy to 2.423 for Hong Kong, suggestingthat it is much faster to reach all the potential consumers in Italy than in Hong Kong. Themarketing cost elasticity in China is substantially larger than that of Arkolakis (2010) and Eatonet al. (2011). But the average elasticity across China’s top ten destination markets is 0.81, whichis reasonably similar to the calibrated value of 0.915 in Arkolakis (2010) and estimated value of 1.1in Eaton et al. (2011).Table 3.7: Estimates of marketing cost elasticitiesCoefficientκc 2.713κUSA 0.780κHKG 2.423κJPN 0.522κKOR 1.281κGER 0.492κGBR 0.406κAUS 0.443κCAN 0.364κTWN 1.067κITA 0.3263.4 Evidence on the effect of marketing costIn this section, we provide empirical evidence on the impact of marketing cost. Advertisementexpenditure is used to shed light on the relative importance of marketing activities across industries.Then, we investigate alternative theories that may provide the same empirical predictions and use56Firm percentiles are calculated according to their TFP measures and firms participating in processing trade areexcluded from the estimation.57Arkolakis (2010) uses a simple method of moments estimator to calibrate θ˜ and κ (which is denoted by β inArkolakis (2010) model). That is, the two parameters are exactly identified by equating the mean of normalizedexport intensity predicted by the model and in data for firms at the median percentile in each market. The calibratedθ˜ in Arkolakis (2010) is 1.65.523.4. Evidence on the effect of marketing costthe differential effects on price ratios to distinguish marketing cost channels from other alternativemechanisms.3.4.1 Advertisement expenditureDifferent industries feature different marketing strategies. For instance, firms in Manufacture ofwearing apparel may spend more on advertising in order to distinguish with other brands whilefirms manufacturing special equipment emphasize more on product innovation since their consumerbase is relatively stable and less responsive to advertisements. In light of the model proposed inSection 3.3, firm productivity affects the sales ratio by the difference in number of consumers reached(i.e., extensive margin). Therefore, if firms in an industry relies more on consumer accumulation,the negative correlation between firm productivity and sales ratio should be more pronounced.In order to capture the relative importance of marketing strategy, we use firm level informationto calculate the average advertisement expenditure of the industry. It is divided by the average salesto control industry size effects. The interaction of advertisement over sales and firm productivityis included in the specifications (3.1 and 3.2).58Table 3.8 presents the regression results at both firm and firm-destination levels. Firm produc-tivity, measured by TFP, is reaffirmed to be negatively correlated with sales ratio between foreignand domestic market. This negative correlation is more prominent in industries with higher ad-vertisement expenditures.59 Neither the sign nor the scale of the coefficients change much whenprocessing trade is excluded. Results from Table 3.8 supports the marketing cost channel that weproposed for the linkage between firm productivity and sales ratio.3.4.2 Alternative explanationsIn this paper, we approach the puzzling negative correlation between firm productivity and foreign-domestic sales ratio from the production side through market penetration technology. However,the disproportionate sales allocation across markets may also arise from the demand side. Forexample, Melitz and Ottaviano (2008) allows firms to charge variable markups and the degree ofmarket power depends on market size. Firms of higher productivity have larger sales than lowerproductive ones especially when the market is big. If Chinese market is larger than foreign market,more efficient firms would expand non-proportionally more at home.We construct a relative market size measure using the relative consumption60 in foreign overdomestic market. It varies across destination countries and different industries over time. Theinteraction term between firm productivity and relative market size is included in the regressions.58The advertisement over sales variable is captured by the province-industry-year fixed effects in the firm levelregressions and absorbed by the country-province-industry-year fixed effects in the firm-destination level regressions.59Similar results can be found in Appendix B.2 with value-added per worker being the productivity measure.60The consumption data come from the “Industrial Demand-Supply Balance Database (IDSB),” which is collectedby UNIDO. This database contains datasets based on the four-digit level of ISIC Revision 3 for each country andeach year. The apparent consumption in this database is calculated as the summation of domestic output and netimport. Since there are some missing values for domestic output, total imports or total exports, only half of theobservations can be used.533.4. Evidence on the effect of marketing costTable 3.8: Effect of advertisement expenditureDependent Variable: ln(Export/Domestic Sales)Firm level Firm-destination levelln(TFP) -0.240*** -0.207*** -0.212*** -0.101**(0.0224) (0.0325) (0.0352) (0.0458)ln(TFP)× Advertisement/Sales Ratio -11.28*** -12.27*** -8.402*** -8.036***(2.378) (4.092) (2.005) (2.828)ln(Capital/Labor Ratio) -0.222*** -0.260*** -0.184*** -0.205***(0.0157) (0.0229) (0.0225) (0.0319)ln(Sale) 0.130*** -0.0483 -0.188*** -0.429***(0.0366) (0.0377) (0.0517) (0.0479)ln(No. of markets) -0.0792 -0.0341(0.0495) (0.0541)Constant 0.588** 2.416*** 0.849** 2.597***(0.276) (0.299) (0.394) (0.385)Ownership FE X X X XProvince-Industry-Year FE X XCountry-Province-Industry-Year FE X XCluster By Industry X X X XExclude Processing Trade X XObservations 275,872 69,691 1,098,287 560,850R-squared 0.370 0.474 0.586 0.650Notes: This table shows the impact of advertising spending on the correlation between firmproductivity and foreign/domestic sales ratio.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.It is expected to be positive if market size has an impact on the negative correlation. In otherwords, as the foreign market is relatively small than Chinese market, firm productivity is negativelycorrelated with sales ratio. As the foreign market becomes comparable with the Chinese market,a positive interaction term would cancel out the negative effect, making productivity uncorrelatedwith relative sales ratio. If the foreign market is large enough to surpass the domestic market,the positive coefficient for the interaction term could overwhelm the negative correlation and makehigher productive firms export more.The first column of Table 3.9 presents a positive coefficient on the interaction suggesting animpact of market size. However, it becomes insignificant when we get rid of processing trade orcontrol firm-industry fixed effects (shown in the last two columns of Table 3.9).In addition, Melitz and Ottaviano (2008) also predicts a correlation between firm productivityand relative price ratio between foreign and domestic market. When domestic market is relativelylarge (and thus more competitive), more efficient firms would have weaker market power and there-fore charge a lower markup. As a result, firm productivity is positively correlated with the relativeprice ratio. In our model, as well as in Arkolakis (2010), firms enter monopolistic competition543.4. Evidence on the effect of marketing costTable 3.9: Effect of relative market sizeY=ln(Export/Domestic Sales)(1) (2) (3)ln(TFP) -0.244*** -0.100**(0.0388) (0.0504)ln(TFP)× Relative Market Size 0.0157** 0.00176 0.00639(0.00759) (0.00444) (0.00540)ln(Capital/Labor Ratio) -0.201*** -0.204*** -0.197***(0.0222) (0.0290) (0.0289)ln(Sale) -0.192*** -0.439*** -0.407***(0.0470) (0.0501) (0.0505)ln(No. of markets) -0.0341 -0.00499 -0.0219(0.0539) (0.0591) (0.0588)Constant 1.032*** 2.589*** 2.810***(0.342) (0.379) (0.243)Country-Province-Industry-Year FE X X XOwnership FE X X XCluster By Industry X X XExcluding Processing Trade X XIndustry Dummy × ln(TFP) XObservations 391,677 212,825 212,825R-squared 0.515 0.586 0.592Notes: This table shows the impact of market size on the correlation betweenexport/domestic sales ratio and productivity on the firm-destination level. Rel-ative market size varies at the industry-country-year level.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.with CES demand and charge constant markup. All the variations in relative sales come fromdifferences in quantity instead of price. Therefore, whether the negative correlation results fromdemand or production side can be resolved if we could compare the reactions of price and quantityratios respectively.One difficulty is that price information for products sold in the domestic market is not observed.As an alternative, we turn to Hong Kong (HK) which is close to mainland China both geographicallyand culturally. Hong Kong is treated as if it were a foreign country in the Customs records andwe assume selling to Hong Kong is comparable to selling in mainland China. In order to check thevalidity of this assumption, we replace domestic sales with sales to Hong Kong and replicate thebenchmark regressions with various controls. Trade data allows us to include additional controlsat the HS 8-digit product level.Table 3.10 shows that firm productivity remains negatively correlated with sales ratio betweenforeign and HK market. This pattern does not depend on the measurement of firm productivitynor the exclusion of processing trade. The scale of the effect becomes slightly smaller than that553.4. Evidence on the effect of marketing costTable 3.10: Firm productivity and sales ratio: Hong KongDependent Var.: ln(foreign/HK Sales)value-added per worker TFPln(Productivity) -0.0463 -0.0934** -0.0519** -0.0837**(0.0307) (0.0475) (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 -1.756 -4.027 -1.501 -3.764(1.390) (2.612) (1.379) (2.601)Product(HS 2-digit) FE X X X XCountry-Province-Industry-Year FE X X X XOwnership fixed effect X X X XExclude Processing Trade X XCluster By Industry X X X XObservations 441,946 110,086 441,946 110,086R-squared 0.360 0.516 0.360 0.516Notes: This table shows the correlation between sales ratio and productivity on the firm-destinationlevel. We calculate the export value on HS8 level for the same firm.The sales ratio is Export V alue to Country bExport V alue to Hong Kongat firm-product level. We include US, Japan, SouthKorea, Germany, UK, Canada, Italy, Australia and Taiwan. These markets are the top 10 destina-tions of Chinese exporting firms.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.in Table 3.6 (probably due to the additional product dimension) but still at a comparable level.61Given that Hong Kong provides a reasonably good substitute to the Chinese domestic market, weconstruct relative price ratios between selling in foreign country b and in Hong Kong as well asrelative quantity ratios.Table 3.11 shows the regression results. Price ratio turns out to be positively correlated withproductivity as predicted by the demand theory. However, the correlation is not statisticallysignificant. The coefficient on quantity ratio, on the other hand, is strongly negative suggestingthe relationship between firm productivity and sales ratio mainly comes from the differences inquantities rather than prices. These results lend support to the marketing cost theory proposed inthis paper and identifies production side effects from demand side mechanisms.Another potential explanation for our empirical findings relates to the quality of the products.Manova and Zhang (2012) find that more successful exporters use higher quality inputs to producehigher quality goods. Firms vary in terms of the quality of their products across destinations byusing inputs of different quality levels. Thus, firm sales will vary across markets due to the differingquality of its products. Manova and Zhang (2012) also argue that the markups are heterogeneousacross firms. Therefore, the price ratio between two export destinations should be correlated witha firm productivity. However, we find no such evidence in the Chinese data.61There is a 0.84% decrease in sales ratio with respect to 10% TFP growth compared with a 1.1% drop previouslywith domestic sales.563.5. ConclusionTable 3.11: Price ratio vs quantity ratio: Hong KongDependent Var.: ln(foreign/HK Sales)ln(Price Ratio) ln(Quantity Ratio)ln(TFP) 0.00706 -0.0921**(0.00845) (0.0369)ln(Capital/Labor Ratio) -0.00761 -0.0774**(0.00556) (0.0340)Constant 0.654** -4.398*(0.289) (2.660)Country-Province-Industry-Year FE X XOwnership FE X XProduct(HS 2-digit) FE X XExclude Processing Trade X XCluster By Industry X XObservations 109,743 109,743R-squared 0.431 0.516Notes: This table shows the correlation between price ratio, quantity ratio and pro-ductivity on the firm-destination level. We calculate the export prices (quantities)on HS 8-digit level for the same firm.The price (quantity) ratio is Export Price (Quantity) to Country bExport Price (Quantity) to Hong Kongat firm-productlevel. Here other countries (regions) include US, Japan, South Korea, Germany,UK, Canada, Italy, Australia and Taiwan. These countries (regions) are the top 10destinations of Chinese exporting firms.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.3.5 ConclusionThis paper examines the relationship between firm productivity and allocation of sales acrossmarkets. Using data on Chinese exporters, we establish the stylized fact that firm productivityis negatively correlated with sales ratio between foreign and domestic market. This empiricalpattern remains robust when we control firm ownership, capital intensity, and rule out the impactof processing trade at both firm and firm-destination level.This finding is in stark contrast with Melitz (2003) predictions where sales ratio across marketsare independent of firm productivity and high productivity firms are associated with high exportintensity due to the large number of markets they enter. The empirical pattern we observe isthe opposite to predictions by Arkolakis (2010) and Eaton et al. (2011) where market penetrationtechnology replaces the fixed cost of entry and firm productivity is positively correlated with relativesales in the foreign market.To rationalize our empirical findings, we extend the Arkolakis (2010) model to allow for het-erogeneous marketing cost elasticities across countries. The returns to marketing is decreasing andthe marketing cost elasticity governs the speed of deterioration. The model links firm productivityand non-proportional distribution of sales across markets through the extensive margin—number of573.5. Conclusionconsumer reached. When the marketing cost elasticity is higher in China than abroad, productivefirms have a relative advantage in selling to domestic consumers resulting in a negative correlationbetween firm productivity and the foreign-domestic sales ratio.Our market penetration explanation of the negative relationship between firm productivityand export intensity is supported by evidence on industry level reliance on marketing strategiesmeasured by advertisement expenditures. The model prediction on constant rather than variableprice ratio helps to distinguish our marketing theory with alternative explanations.Yet, more work is needed to investigate which factors determine local marketing cost elasticities.Future research aims to find a direct measure of marketing cost, estimate the marketing elasticityacross countries, and carry out counterfactuals for policies that help ease barriers to reaching localconsumers.58Chapter 4The legacy of 19th century treaties onthe current trade of Chinese cities4.1 IntroductionHistorical conditions can exert enduring influences on current economic relationships. A well-known example is the Acemoglu et al. (2000) finding that settler mortality in the colonial eraled to different institutional arrangements that explain current differences in prosperity.62 Headand Mayer (2014) report that on average the existence of a colonial relationship (past or present)between two countries increases bilateral trade by 150%.63 We contribute to the study of persistenteffects of colonization by taking advantage of two somewhat unique features of China. First, unlikemost of the rest of Asia, the country as a whole was not colonized but specific cities were convertedinto colonies by Western powers and Japan. Second, Chinese customs data records the city fromwhich exported goods originate. We use the Chinese experience to investigate whether treaty ports,forcibly established in the 19th century, influence contemporary international trading patterns ofChinese cities.Starting with Britain’s Treaty of Nanjing in 1842, a total of 14 foreign powers signed treatieswith 58 different Chinese cities over the span of 70 years. These arrangements include treaty ports,port concessions, and leased territories. Foreigners established schools, churches and introducedforeign laws and customs in settlements within the treaty cities. There was a large amount offoreign investment during the treaty port era. All these features of the period may have hadenduring impact on bilateral relationships between treaty countries and the host cities. That said,the treaty port era and current times are separated by seven decades, two devastating wars (withJapan from 1937 to 1945, followed by civil war until 1950) and the economic isolation imposedby the communist government. It is far from obvious how bilateral linkages established a centuryago might have survived such disruptions to facilitate trade today. Nevertheless, very long runhistorical influences have been found in the literature—Alesina et al. (2013) report that “societiesthat traditionally practiced plough agriculture today have less equal gender norms”—so we regardthe existence of persistent treaty effects as an open question.We use data on the trade of 335 Chinese cities with 212 trading partners for the 2000–2006 periodto test the hypothesis that treaty linkages exert long-lasting impacts on trade. Using a gravity62Nunn (2009) reviews a range of additional evidence for history effects on economic development.63The mean of 147 estimated coefficients from the literature is 0.92.594.1. Introductionframework including fixed effects for each city and country, we estimate that a past treaty raisesa city’s imports by 76% and exports by 40% with the specific signatory country. However, whenwe consider restricted samples of cities and countries and control for economic complementarityrelated to the propensity to be a host or recipient of treaty arrangements, we find that the bilateraltreaty effects lose their significance. There remains some evidence of a general increase in tradebetween cities and countries that were involved in any treaty arrangements in the period.The only paper of which we are aware that considers the legacy of the treaty port era inpromoting current trade is Keller et al. (2013) who focus on Shanghai, a treaty port establishedunder the Treaty of Nanjing. In part of their analysis, they fit a gravity model to trade betweenShanghai and 11 countries for the 1986 to 2009 period. They include current and historical FDI ascovariates (along with distance, GDP, and, in some specifications, bilateral linkage variables such ascommon language). They interpret the result that both FDI variables have positive and significanteffect as evidence that the treaty port era generated a legacy promoting modern trade.Recent papers investigate the economic development of treaty port cities in China over anextended period of time. Jia (2014) constructs a longitudinal data set of Chinese prefectures for10 different years over the 1776–2000 time period to examine population and GDP growth. Sheidentifies treaty port effects by comparing them to similar prefectures that did not include a treatyport and finds treaty ports enjoyed faster economic growth after the Open Door policy was enacted.Keller et al. (2011) use data from Chinese Maritime Customs (CMC) service to examine treaty porttrade over a similar long period of time. They chronicle many aspects of Chinese trade includinga rapid increase in the number of goods traded after the establishment of treaty ports and theimportant but steadily declining role of Hong Kong in entrepoˆt trade. Unlike our study, neitherpaper considers the relationship between current bilateral trade and historical bilateral treaties.Treaty ports in China can be considered a type of colonial expansion as they involved colonialpowers such as Britain, France, and Japan and occurred during the colonial period. A number ofstudies document persistent economic effects of colonization. Engerman and Sokoloff (2002) andFeyrer and Sacerdote (2009) document the influence of European colonization on the economicperformance of the colonies. Head et al. (2010) find that once colonial ties are severed, trade erodessteadily over the course of 30–40 years (a generation) but continues to be higher than trade betweencountries without a colonial history. The persistent effects of colonial histories may be related tocommon institutions. For example, Lo´pez de Silanes et al. (1998) emphasize that countries withdifferent legal systems, i.e., based on British common law or Roman civil law, offer different investorprotection which affect their financial development. Colonialism created common legal systems, afeature that the gravity literature has established as a source of greater bilateral trade. Our analysisextends research on the effects of colonialization.Our paper also relates to the literature showing effects of history-based trust and distrust oneconomic relations. In their study of 17 European countries, Guiso et al. (2009) find that lowerbilateral trust lead to less trade as well as less direct investment between two countries. Glick andTaylor (2010) study the effects of war on bilateral trade with data extending back to 1870 and604.2. Treaty Ports, Concessions, and Leased Territoriesprovide evidence for large and persistent impacts of war on trade and global economic welfare. Cheet al. (2013) investigate the long-run effect of Japanese invasion on China’s contemporary tradewith Japan. Using the civilian casualty rate across 28 Chinese provinces as the key explanatoryvariable, they find that a higher casualty rate is associated with lower Japanese trade and foreigndirect investment. In our study, the treaties signed by Chinese under foreign pressure might havegenerated trade-reducing resentment or trade-enhancing familiarity.Finally, our analysis contributes to studies examining the multilateral effects of bilateral link-ages. An obvious type of multilateral effect of a bilateral linkage is trade diversion. A number ofstudies (e.g., Krishna, 1998; Ornelas, 2005) provide evidence on trade divergence for countries out-side FTAs. On a more positive note, Saggi and Yildiz (2011) develop a theoretical model predictingthat bilateral agreements can be stepping stones to multilateral liberalization. Multilateral effectsof dollarization are identified in Lin and Ye (2010) who finds that dollarization encourages bilateraltrade between dollar-using countries and the U.S. as well as multilateral trade among dollar-zonecountries. Recent work by Morales et al. (2014) models interdependence of export markets via aconcept they refer to as “extended gravity.” The idea is that “export entry requires a costly adap-tation process: some firms are better prepared than others to export to certain countries becausethese firms have previously served similar markets and have therefore already completed part ofthe costly adaptation process.” If the cities that signed trade agreements are similar to each other,then it seems likely that the incremental sunk costs of entering the first such city will be higherthan for subsequent entry. This will lead to patterns in which a given country tends to have highexports to all the formerly treated cities. Our study investigates multilateral effects of bilateraltreaties by considering whether treaty port experience increases trade between countries and hostcities even if they are not directly linked by a bilateral treaty.The next section provides details on the timing and characteristics of the treaties we evaluate.Section 4.3 describes the trade data and explains how we specify the treaty indicator variables. Insection 4.4, we describe the empirical framework and report and interpret the results. The finalsection summarizes and discusses the implications of our analysis for research on the effects ofhistorical variables.4.2 Treaty Ports, Concessions, and Leased TerritoriesWe consider three types of agreements that provided specific foreign countries with special accessto China: treaty ports, concessions, and leased territories. The most common arrangement wasthe establishment of a treaty port. 77 treaty ports were opened in 58 cities with six foreigncountries. Inside some of the treaty ports, areas of land called concessions were created where foreignmerchants could reside and establish businesses. This provided access to China for countries otherthan the six that established treaty ports. Finally, four cities (including Hong Kong) were occupiedby five foreign countries as leased territories. We provide further detail on each arrangement below.Before the 19th century, foreigners who came to China had to do business on Chinese terms.614.2. Treaty Ports, Concessions, and Leased TerritoriesThe Qing government was willing to grant favours to the foreign merchants and let them trade,but only under strict regulations.64 As early as 1757, foreign merchants were restricted to the portof Guangzhou and found themselves bound by imperial decrees. Doing business through merchantsyndicates called Hong is an example.65 Foreign merchants were not allowed to deal directly withlocal people nor hire Chinese servants by themselves. Even their stay in Guangzhou were limitedto the trading periods and restricted to a small area along the shoreline. Foreign merchants facinghigh prices and administration fees began to fight against this so called “Canton System” and thustriggered the opium war and opening of treaty ports.Starting from the 1842 Treaty of Nanjing, Western powers imposed a number of asymmetrictreaties on China. As a result, China was forced to pay large amounts of reparations, open upports for foreign merchants to conduct business, make concessions for foreign residency, and leaseterritories to foreign countries (e.g., Hong Kong was leased to Great Britain). In the following 60years, treaty ports were opened in more than 50 cities. Note that several treaty ports may locatein different areas within the same city. Some of the treaty ports, such as Ulan Bator (the capitalof Mongolia), are no longer territories of China now. Though called “ports,” many of them werenot located along the coast or rivers but along the border with the Soviet Union.Concessions (or settlements) are areas of land inside treaty cities designated for homes andbusinesses of foreign residents. The British concession in Shanghai opened in 1845 was the firstforeign concession in modern China. In the following 60 years, 14 countries established almost 30concessions in 12 treaty cities in China. Most of them were owned by one foreign country whichhad administrative power over both economic and political issues within its territory. Therefore,concessions are also referred to as a “state within a state.” Public concessions in Shanghai andXiamen were jointly held by several countries.After the 1895 Sino-Japanese War, foreign powers competed to divide up China by means ofleased territories. The UK, France, Germany, Russia, and Japan held territories in four citiesin China. In contrast to treaty ports where China retained territorial control, leased territoriesusually allowed foreign powers complete sovereignty. The leased territories were “rented” by foreigncountries mainly for strategic and military purposes. They were much larger than concessions andmore likely to include adjacent water areas. Governor-generals were assigned to practice foreignlegislations in their “colonies”.In the appendix, Table C.1 identifies the salient features of the different types of treaty arrange-ments and Table C.2 matches the 14 treaty countries to 55 host cities.66 The primary arrangementwas a treaty port but more countries (14 versus 6) were involved in concessions. All arrange-ments allowed foreigners to live and work. Broadly speaking, the extent of political and economicautonomy was greatest in leased territories and most limited in a treaty port arrangement.64The Qing Dynasty (1644–1911) is the last imperial dynasty of China.65The term Hong derives from a Chinese word meaning company. A Hong was a union of authorized merchantswith monopoly power over business with foreign traders. Firms belonging to Hong were licensed by the governmentand paid large sums of fees for their positions. They collected fees and duties from foreign merchants, transmittedgovernment decrees and supervised foreign traders’ business activities.66The 77 treaty ports map into a smaller number of modern Chinese cities.624.2. Treaty Ports, Concessions, and Leased TerritoriesThe proposition that treaty port linkages influence current trade requires that there 1) existstrade-promoting and relationship-specific capital and 2) that this capital has been maintained overmany generations. We discuss possible forms this capital could take below.One obvious avenue for trade promotion is the creation of port infrastructure. Facilities toassist loading and warehousing were introduced. Railways or roads connecting treaty ports andscattered production areas were developed. With the establishment of the Imperial MaritimeCustoms Administration in 1854, lighthouse and beacons were set up along the coast and largerrivers. However, this infrastructure, to the extent it persists today, is available to all tradingpartners and unlikely the source of bilateral trade promotion. In the empirical work, we will usecity-fixed effects and thereby capture general trade promotion associated with port infrastructure.Direct investment between recipient country j and host city i would be another source of trade-promoting capital. Being the only locations foreign businesses were permitted in China, host citiesattracted many foreign firms, especially in banking and transportation industries. In 1844, 11 firmsfrom Britain and the U.S. established headquarters in Shanghai. The number of foreign companiesincreased to more than 120 in the following ten years (Zhang, 1993). Li et al. (1981) reports thatthe number of foreign firms grew from 343 to 579 from 1872 to 1892. Meanwhile, the number offoreign merchants expanded from 3,673 to 9,945.Some of these trade facilitating institutions managed to retain their business in China such asthe success of Hongkong and Shanghai Banking Corporation (HSBC), originally a British company.Established in 1865 in the treaty port of Shanghai, HSBC remained business in China except for1941–1945 period when all foreign invested banks were forced to leave Chinese market by Japan.In April 1955, HSBC handed over this office to the communist government and its activities werecontinued in rented premises. Today, HSBC has the largest service network among foreign banks inChina.67 In spite of all the disruptions to the economy of China, many foreign firms such as HSBCcontinued to operate. The development of trade-facilitating foreign service industries potentiallyprovides a legacy that still benefits the bilateral trade today between treaty partners. One pieceof supporting evidence for an FDI legacy affecting trade today is the finding in Keller et al. (2013)that countries with more FDI in Shanghai in 1921 have more trade with the city today.Finally, the opening of treaty ports allowed host cities to get access to new technologies andproducts that could exert permanent influence on their trade structure. In particular, treaty citesmight have developed manufacturing industries to suit the needs of their treaty partner. Forexample, Nield (2010) documented that a group of French engineers came to Fuzhou (one of thefirst five treaty ports opened due to the Treaty of Nanjing) in 1866 to launch shipyards, arsenals andspecial navy schools. These activities were supported by scholar generals in Qing government underthe “Westernization Movement” which mainly took place in treaty ports. Although the movementfailed at last due to the rigid feudal system, these attempts to learn foreign technologies broaden theeyes of Chinese businessmen and laid industrial foundations for further trade to be built on. After67See Guo (1992) “Main activities of HSBC in China” in “Concessions of foreign powers in China” (in Chinese)for a detailed description of business activities of the HSBC in modern China.634.2. Treaty Ports, Concessions, and Leased Territories1895, some industries appeared in treaty ports which was meant to bring in machinery, petroleum,and transport equipment that China lacked. By improving the match between foreign demand andlocal supply, trade structure evolved to encourage trade between host cities and their industrializedtreaty partners.68Relationship-specific human capital also plays an important role in promoting bilateral trade.69Transactions require the matching of buyers and sellers and familiarity with the business practicesof each party is essential. From the point of view of a host city, there may be country-specificknowledge needed to conduct business effectively with a foreign country. This knowledge maybe about culture and business practices or related to knowledge of particular legal institutions.French and the U.S. treaties allowed foreigners in China to be governed by the law of their owncountry instead of Chinese law, thereby exposing the treaty cities to foreign business practices andinstitutions. Likewise, there may be specific knowledge required to conduct business in a particularhost city and treaty countries gained access to this knowledge.Trust and reputation are also important. Foreign powers established municipal authorities,schools, and provided public services such as street cleaning and police in treaty ports. The treatyport era brought a large number of foreign missionaries who built schools to teach math, mod-ern science, and languages. By 1860, there were 90 Catholic and 50 Christian primary schoolsestablished in the first five port cities (Gu, 1981).70 Moreover, Chinese and foreigners were notseparated since many Chinese compradors (or middlemen), rich merchants, or even refugees alsolived in foreign concessions (Murphey, 1975). While the individuals in the treaty era are no longeraround, the positive “reputation” could persist and still influence trade today. As pointed out byJia (2014), human capital and social norms seem to be more important than geography and tangibleinstitutions. On the other hand, with its vast area and insular nature, China was too self-containedto be strongly affected by Western traders outside treaty ports. The spread of Western ideas andtechniques were deterred by both long distance and psychological resistance of Chinese people.Therefore, trade facilitating human capital established during the treaty port era may have beenconfined to the cities with treaty linkages and did not extend to non-treaty cities.Persistent relationship-specific capital provides a mechanism for historical treaty linkages toexert influence on current bilateral trade. However, whether this capital still persists is uncertaingiven the interlude of war and economic isolation between the treaty port era and today. Inthe years after the establishment of treaty linkages, China suffered from constant war and socialturmoil.71 Foreign investments were either evacuated from China or handed over to the communist68At the same time, the endowment of host cities also affect the selection of treaty ports. The British, for instance,aimed those places where large quantities of tea and silk could be obtained. Xiamen, Fuzhou and Ningbo were chosenas treaty ports by the British since they located nearer than Guangdong to the tea-producing areas (Murphey, 1975).However, we need not to worry about any selection effect on bilateral trade today since tea and silk no longer featureChina’s present trade structure.69Evidence of the influence of relationship-specific human capital is found in the trade and immigration literatureinitiated by Gould (1994) and Head and Ries (1998). Immigrants are associated with more trade with their countriesof origin, perhaps because they lower transactions costs.70The first five treaty ports opened by Treaty of Nanking are Guangzhou, Xiamen, Fuzhou, Ningbo, and Shanghai.71For example, during the Second Sino-Japanese war (1937–1945), China’s foreign trade was primarily controlled644.3. Data descriptionTable 4.1: Top trading partnersExport from ChinaCity Country Value (billion USD)Shenzhen U.S. 25.56Shanghai U.S. 23.47Dongguan U.S. 14.96Shanghai Japan 14.02Suzhou U.S. 12.99Import to ChinaCity Country Value (billion USD)Shanghai Japan 19.36Shanghai U.S. 12.61Shenzhen Japan 11.62Shenzhen Korea 9.89Suzhou Korea 9.11party. Foreign trade was carried out under the planned economy. In the remainder of the paper,we confront the hypothesis of persistent treaty linkage effects with the data.4.3 Data descriptionWe combine two datasets. First, China’s current trade data are drawn from the database con-structed by China’s General Administration of Customs. Disaggregated monthly transaction leveldata are collected for each HS-8 digit product from 2000 to 2006. The number of observations eachmonth ranges from about 78,000 in January 2000 to over 230,000 in December 2006. The datasetprovides detailed information on trade status (import or export), quantity, trade value, origin anddestination of each transaction, transportation mode, firm associated with each transaction, firmlocation, ownership (domestic, state owned, or foreign). Since our variable of interest—treatyports—does not vary over time over our period of study, we simply aggregate the monthly flowsover the 2000–2006 period to produce a cross-sectional dataset.72 We observe a higher incidence ofpositive export flows than import flows: 37,045 city-country pairs are linked by exports and 18,055pairs take part in importing.The top five trading partners in terms of Chinese cities’ export and import value are presentedin Table 4.1.73 Exports from Shenzhen to the U.S. and imports from Japan to Shanghai lead theby Japan. From 1946 to 1948, the two parties of China were engaged in a civil war and the U.S. dominated China’sforeign trade. The sovereignty of foreign trade was finally given back with the establishment of the People’s Republicof China when trade was mainly centralized or strictly regulated within the planned economic regime.72Alternatively, we could have aggregated by year and clustered standard errors at the country-city level to accountfor correlated errors. Since trade between countries and cities might be volatile, we decided summing over the yearswas better to smooth the data.73Note that Hong Kong, Macao and Taiwan (HMT) are not included in the China’s customs data and are thereforeexcluded from our ranking here.654.3. Data descriptionbilateral trade activities in China. While Shanghai was home to concessions to the US, UK andFrance, Shenzhen was a small city in the treaty era.74These examples point to the importance ofcity and country fixed effects to capture size differences.Data and references used to construct treaty linkages are primarily collected from the historybook “Treaty Ports and Concessions in Modern China” (in Chinese) by Zhang (1993). The sourceidentifies 77 treaty ports with the partner country(ies), year established, and location. Anothersource is Zhang (1993) who documents concessions and leased territories.75 Treaty linkages used inthis paper were reorganized based on Zhang (1993) by adjusting cities more than a hundred yearsago to their current locations and municipal cities.76Table 4.2: Number of city linkages, by type and recipient countryPort & Port &Port Concession Lease Concession Lease TotalUnited Kingdom 16 0 0 7 1 24Japan 12 5 2 3 0 22France 7 4 0 1 1 13Russia 7 2 0 0 1 10United States 2 3 0 0 0 5Germany 0 3 0 0 1 4Austria 0 1 0 0 0 1Belgium 0 1 0 0 0 1Denmark 0 1 0 0 0 1Italy 0 1 0 0 0 1Netherlands 0 1 0 0 0 1Norway 0 1 0 0 0 1Spain 0 1 0 0 0 1Sweden 0 1 0 0 0 1Total 44 25 2 11 4 86As previously discussed, recipient countries established links to Chinese cities via treaty ports,concessions, and leases. In some cases, a recipient country had established multiple arrangementsover time (e.g., Great Britain had a treaty port in Hong Kong as well as a subsequent lease).Table 4.2 displays the number of city links for each of the 14 recipient countries. A full accountingof the arrangements appears in the appendix. Great Britain was linked to 24 cities. It establishedtreaty ports (only) in 16 cities, had a treaty port as well as a concession in 7 cities, along withits port treaty and lease in Hong Kong. Japan had arrangements with 22 cities, followed by74After being designated as a Special Economic Zone in 1980, its population soared from about 300 thousand toover 10 million.75We also cross validated treaty linkages information documented in this book with other sources such as Li (2012)and Wikipedia.76Hong Kong and treaty ports in Taiwan are excluded from our sample since they’re considered foreign in China’scustom statistics. But they do deserve a careful examination. Some cities used to be a part of the Qing empire nowbelong to Mongolia or Russia and therefore dropped from our sample. Further, with the process of urbanization,different cities in the 19th century now consist of the same city are recorded only once in our list.664.4. Specification and resultsFrance and Russia who were linked to 13 and 10 cities, respectively. Seven countries—Austria,Belgium, Denmark, Italy, Netherlands, and Norway—only had a concession in the public concessionin Xiamen.Figure 4.1: Geographic distribution of host citiesFigure 4.1 provides a map of the cities that host treaty ports, concessions, and leases. Weidentify treaty ports with black dots, concessions with blue squares, and leased territories with redtriangles. All the other cities included in our sample are represented with grey dots. We observethat, compared with the distribution of all the cities in China, treaty ports were concentrated incoastal cities and along the Yangzi River. Some located along the “Silk Road” as well as near thenorthern and western border of China. Aside from Tianjin in the North, concession cities appearalong the Yangzi River or the Southeast coast. Leased territories were chosen in four coastal cities.Three of them were close to the capital city Beijing while the fourth (Hong Kong) guards the SouthChina Sea.4.4 Specification and resultsA straightforward way to test the effect of treaty linkages on current trade is to estimate a gravity-type, bilateral trade equation using year, country, and city fixed effects:ln(Xij) = βBTLij + δ ln(Distij) + EXi + IMj + ij (4.1)674.4. Specification and resultswhere Xij represents trade between city i and country j. We aggregate trade over the 2000–2006period and consider exports (from cities) and imports (to cities) separately. The bilateral treatylinkage dummy BTLij equals 1 if city i and country j are linked by a treaty. Distij is the geodesicdistance between city i and country j. EXi and IMj are city and country fixed effects while ij isthe idiosyncratic error. We cluster standard errors at the city level. All observations correspond tocity-country dyads.There is considerable heterogeneity in the types of linkages between host cities and recipientcountries. As shown in Table 4.2, linkages can be in the form of treaty ports, concessions, or leasedterritories and there are many cases of multiple links. We experiment with different ways of definingBTLij . First, we define five variables that correspond to the five different arrangements displayedin Table 4.2: port only (P); concession only (C); lease only (L); port and concession (PC); portand lease (PL). Estimating different effects for each of the five arrangements provides flexibility butposes a challenge for identification due to limited variation. We also define two aggregate forms ofBTL. The first is Tmaxij , a binary variable equals to the maximum value across P, C, L, PC, andPL. That is, Tmaxij turns on so long as there is any type of BTL between city i and country j.We also calculate Tsumij as the sum of the P+PC , C+PC, and L+ PL dummies. It therefore cantake the values of 0 (no linkage), 1 (for P, C, and L cities), or 2 ( for PC and PL cities) .The results of the different linkage specifications appear in Table 4.3. The first three columnsshow results for city imports and the second three exports. Columns (1), (2), (4) and (5) show thatTmax and Tsum enter significantly. However, the significance of the linkage variables is higher forimports (1% significance) than for exports (10% significance). Based on R2 and root-mean squarederror, we observe both specifications fit the data equally. Columns (3) and (6) reveal generallypositive effects of each of the five types of arrangements. Significant effects are exhibited for ports(10% ), concessions (1%) and ports&concessions (1% ) in the case of imports, whereas in the exportregressions concessions (10%) and port&lease (5%) are significant.Table 4.3 indicates that historic treaty linkages are associated with more current trade. Sincethe specifications include country and city fixed effects, a linkage leads to more trade than what isobserved on average for that country and for that city. Specifying linkages as either Tmax or Tsumprovide equivalent fit to the data. We use Tmax hereafter because it is slightly easier to interpret.Exponentiating gives the trade multiplier for having had some type of BTL. Columns (1) and (4)suggest that a BTL of any kind raises imports and exports of a city by 76% and 40% respectively.These magnitudes are in line with coefficients estimated for current regional trade agreements.Averaging 108 estimates that use origin and destination fixed effects Head and Mayer (2014) findan average RTA effect of exp(0.36) − 1 =43%. The remarkable aspect of our results is that theagreements considered here have been inoperative since World War 2.In the initial specification, we include trade between all countries and cities. To explore robust-ness of the results to different samples, we consider two sub-samples: (1) All cities and only the 14countries that had some type of treaty linkage (which we refer to as “recipient” countries) and (2)All countries but only the 55 cities with some linkage (“host” cities). As before, we report results684.4. Specification and resultsTable 4.3: Bilateral treaty linkage effects(1) (2) (3) (4) (5) (6)City Imports City ExportsTmax 0.564a 0.334c(0.174) (0.181)Tsum 0.423a 0.255c(0.135) (0.131)Port (P) 0.431c 0.141(0.244) (0.197)Concession (C) 0.832a 0.664c(0.294) (0.363)Lease (L) 0.264 0.367(0.584) (0.578)P&C 0.771a 0.320(0.291) (0.221)P&L -0.164 0.344b(0.603) (0.142)ln Dist -1.614a -1.614a -1.615a -1.204a -1.205a -1.205a(0.099) (0.099) (0.099) (0.085) (0.085) (0.085)Number of obs. 18055 18055 18055 37045 37045 37045R2 0.490 0.490 0.490 0.672 0.672 0.672rmse 2.490 2.490 2.490 1.588 1.588 1.588Standard errors in parentheses. Significance: c p <0.1, b p <0.05, a p <0.01694.4. Specification and resultsseparately for imports (columns 1 and 2) and exports (3 and 4). Table 4.4 shows that the resultschange dramatically: the coefficients on Tmax shrink and become statistically insignificant in bothsub-samples. History’s shadow has all but disappeared.Table 4.4: Restricted sample(1) (2) (3) (4)Imports ExportsRecipients Hosts Recipients HostsBilateral treaty linkage (Tmaxij) 0.135 -0.086 0.025 0.114(0.161) (0.202) (0.126) (0.176)ln Distij -0.941a -2.077a -0.069 -1.371a(0.263) (0.175) (0.258) (0.185)Number of obs. 3903 4686 4072 7668R2 0.454 0.648 0.636 0.764rmse 1.543 2.335 1.244 1.472Standard errors in parentheses: Significance: c p <0.1, b p <0.05, a p <0.01The contrasting results in Table 4.3 and Table 4.4 can be reconciled via a data generatingprocess in which recipient countries trade more with any host city, regardless of whether they hada linkage with the host city. To see this, let Hi be a dummy variable indicating that city i hasbeen host to one or more treaties and Rj indicates that country j has received treaty privilegesfrom at least one Chinese city. Neither dummy can be identified, of course, in a model with i and jfixed effects. However, we can estimate their interaction, HiRj . The interacted variable is a binaryvariable identifying a pairing of any host city and any recipient country, not just those who werebilaterally linked by an actual treaty. The augmented specification isln(Xij) = βBTLij + γHiRj + ln(Distij) + EXi + IMj + ij . (4.2)The parameter γ (where the Greek g is a mnemonic for “group” or “generalized” effect) representsthe additional trade between any host city when exporting to any recipient country. Meanwhile βis the incremental effect of a bilateral treaty link. For a country with a bilateral treaty link, thetotal trade effect is γ+β. In equation 4.1, the estimate βˆ captures both the bilateral and the groupeffects. When we confine the sample to recipient countries only, the Rj = 1 so HiRj becomes ani-specific term that is absorbed by the city fixed effect. Similarly, when we confine the sample tohost cities only, the Hi = 1 for all observations so the HiRj term becomes j-specific and is fullycaptured by the country fixed effect.If the HiRj term belongs in the specification, the estimates of β listed in Table 4.4 are unbiasedwhereas estimates shown in Table 4.3 are upwardly biased because the fixed effects fail to absorbthe HiRj term. The results in Table 4.4 indicate that β is small and not significantly differentfrom zero. Receiving favourable treatment is associated with higher trade with all host cities andhaving a direct link has a negligible additional effect. The lesson we take from these results is thatthe sample matters even if importer and exporter fixed effects are included in the specification.704.4. Specification and resultsUnobserved bilateral linkages lead to bias. Unfortunately, in cases such as this where the variableof interest is historical does not vary over the time frame of the estimation, it is impossible toinclude bilateral fixed effects.What is the source of these group effects that appear to expand trade between recipient countriesand host cities even in the absence of a direct treaty link? Port infrastructure can be ruled outbecause, if it expanded trade generally, it would be absorbed in the city fixed effect. A possibleexplanation is that host city industrial structure changed in a manner that is conducive to tradewith recipient countries but not non-recipient countries. This may be explained by the formerbeing higher income than the latter. Alternatively, the experience gained through participation ina treaty arrangement might have created knowledge useful for trade between host cities and recipientcountries (group benefits). Figure 4.2 illustrates the coexistence of bilateral treaty linkages (redsolid lines) and their 3rd party effects on countries and cities involved in port arrangements (bluedash lines). In the diagram, country F has a treaty with city C that enables trade with host citiesA and B (but not non-host city D).Figure 4.2: Bilateral and group linkagesNote: Bilateral linkages between pairs of host cities and recipient countries determined in treaties are denotedby red solid lines while group linkages are represented by blue dash lines. Note that city D and country Gare non-host city and non-recipient country respectively.One potential source of group benefits is the Unilateral Most Favored Nation clause. As statedin the treaty with UK in 1843, “additional privileges” granted to foreign countries were to be“extended to and enjoyed by British subjects”. Under this condition, treaty ports opened by onecountry was not a private property but shared with any other country who obtained the privilegeof UMFN. From 1843 to 1896, at least 6 countries obtained UMFN from the Qing government.77Even for countries without UMFN, business activities were largely welcomed in foreign concessionsowned by other countries. Zhang (1993) provides evidence of British concessions where merchantsfrom other foreign countries were allowed and welcomed to operate. While preferential access underUMFN is not relevant for trade today, the knowledge and experience developed during trade under77Please refer to Li (2012)714.4. Specification and resultsUMFN may persist.Another potential mechanism for the group effect is offered by Morales et al. (2014). Theydevelop a theoretical framework to analyze the dynamic entry decision of exporters. Firms incurexport entry costs in each country, but entry costs become lower if they have exported to a similarcountry previously. The formulation of export entry costs reflect a costly adaptation process suchas product modifications to suit local tastes or regulations. Therefore, firms are more likely toenter markets similar to their previous export destinations. Bilateral treaty linkages facilitate theestablishment of trade relationships between treaty partners. The extended gravity effect allowsother host cities and recipient countries that share many similarities with these treaty partners tobenefit since entry costs in these markets are reduced.Thus far, we have only compiled indirect evidence that trade is higher between any recipientcountry and any host city (γ > 0). We can explicitly measure γ by estimating equation (4.2) usingthe full sample. The estimation results are shown in Table 4.5. Columns (2) and (4) present resultsfor imports and exports for the group linkage variable HiRj . It is large and statistically significant(1% level). Host city imports from recipient countries are 73% higher and their exports to recipientcountries 48% higher than city-country pairs without group linkages. Bilateral treaty linkages losetheir significance.Table 4.5: Bilateral and multilateral effects(1) (2) (3) (4)Imports ExportsTreaty link (Tmax) 0.564a 0.167 0.334c 0.020(0.174) (0.167) (0.181) (0.173)Host w/ Recipient (HR) 0.546a 0.390a(0.144) (0.103)ln Dist -1.614a -1.612a -1.204a -1.204a(0.099) (0.099) (0.085) (0.086)N 18055 18055 37045 37045R2 0.490 0.490 0.672 0.672rmse 2.490 2.488 1.588 1.588Standard errors in parentheses. c p <0.1, b p <0.05, a p <0.01We have observed that treaty linkages confer group benefits between recipient countries and hostcities. The presence of a direct bilateral link appears to have no significant impact on bilateral trade.Therefore, trade promotion mechanisms through pair-specific investment or knowledge legacies thatcannot be applied to other similar host cities or recipient countries are ruled out. For example, therole of relationship-specific human capital might be weakened since trust and business networks aremore likely to be confined within national borders or local areas.In the discussion above, we offer complementary industrial structure and knowledge as expla-nations for group effects. These sources of group benefits may or may not be “caused” by historicaltreaty arrangements. To further push on the results, we conduct falsification exercises. Our idea is724.4. Specification and resultsto match 14 placebo recipients to the 14 true ones and 55 placebo cities to the 55 true ones.Table 4.6: Probit prediction of treaty recipients and hosts(1) (2)Countries Citiesln GDP 1.023a 0.120(0.224) (0.109)ln GDP per capita 0.808b 0.264b(0.380) (0.128)ln Dist -0.002(0.515)Coastal 0.619a(0.223)Observations 185 332Standard errors in parenthesesc p <0.1, b p <0.05, a p <0.01In order to identify the placebo entities, we estimate two probit regressions, one for countriesand the other for hosts, where the dependent binary variable is coded as one for real recipients andhosts and zero otherwise. For both regressions, we use GDP and per capital GDP (both logged) assize and income variables. In the country regression, we add (log) distance from China and we adda binary variable indicating coastal for in the city regressions. We have data for 185 countries and332 cities. Table 4.6 displays the results. Column (1) reveals that both GDP and per capita GDPare significant but distance is not. The latter result is likely because recipient countries Russia andJapan are close to China whereas the remaining 12 recipients are distant. Per capita GDP and alocation on the coast matter determine the likelihood of being a host city.We identify the placebo countries and cities by generating predicted probabilities based on theprobit regressions and choosing countries and cities with the highest probabilities among those thatdid not have actual treaties. The probits do well in predicting the real recipients and links: 13of the 14 recipients are among the 17 countries with the highest predicted probability (the 14thcountry is Russia and ranks 23rd). Among the 50 cities with highest predicted probabilities, 21 areactual host cities. The top four placebo countries (in terms of predicted probabilities) are Canada,Australia, Switzerland, and Korea. A complete list of the ranking of placebo and true countriesaccording the probit predictions appears in the Appendix.We construct two sets of group variables incorporating our placebos and add them to thespecification. HiRpj equals one when trade is between a true host and a placebo recipient. Hpi Rjequals one when trade is between a true recipient and a placebo host. Table 4.7 reports results forimports and exports. The first set of results repeats the specification without placebo variables forcomparison and columns (2) and (5) show results with the placebos included. We observe that theplacebo group effects are positive and significant in all cases. Their magnitude is somewhat smallerthan the “true” group effects. The difference between the true and placebo effects is borderlinesignificant in some cases. In columns (3) and (6), we add a variable measuring the “propensity”734.4. Specification and resultsTable 4.7: Placebos and propensitiesCity Imports City Exports(1) (2) (3) (4) (5) (6)True Placebos Propensities True Placebos Propensitiestreaty (Tmax) 0.167 0.144 -0.031 0.020 0.015 -0.099(0.167) (0.167) (0.209) (0.173) (0.173) (0.194)True HR 0.546a 0.942a 0.334b 0.390a 0.531a 0.237b(0.144) (0.170) (0.152) (0.103) (0.114) (0.110)HRp 0.720a 0.324a(0.178) (0.098)HpR 0.812a 0.426a(0.136) (0.118)P(H)× P(R) 3.682a 2.309a(0.582) (0.552)ln(distance) -1.612a -1.604a -1.596a -1.204a -1.202a -1.195a(0.099) (0.098) (0.096) (0.086) (0.086) (0.084)N 18055 18055 18055 37045 37045 37045R2 0.490 0.493 0.492 0.672 0.672 0.672rmse 2.488 2.483 2.484 1.588 1.587 1.586Standard errors in parenthesesc p <0.1, b p <0.05, a p <0.01for having a group effect. P(Hi) × P(Rj) is the product of the probit propensities for city i andcountry j.78 Columns (3) and (6) show the results when we replace the placebo group variableswith the propensity variable. The results reveal that this propensity has a significant positive effecton trade. Also, once we control for propensity, the true group effect, HiRj , is now significant atthe 5% level for both imports and exports.The results of our falsification exercise reveals trade complementarity between levels of devel-opment of trading partners. The complementarity is evident when we match the true host cities tohigh-income placebo countries and observe a high level of bilateral trade. Likewise, true recipientstrade more with developed placebo hosts. This complementarity also appears when we introducea variable calculated as the product of the predicted probabilities, probabilities mainly reflectingper capita GDP. This complementarity may be viewed as a type of Linder effect.Controlling for the economic complementarity of trade partners is crucial in our analysis asrelatively developed cities and foreign countries established 19th century treaty linkages. Thecomplementarity could be captured and the bias avoided in a specification with bilateral fixedeffects. However, in cases such as ours where there is no temporal variation in the key bilateralvariable of interest, bilateral fixed effects are infeasible. Hence, researchers need to be especiallycareful about the choice of sample and specification. One option is to combine the treated pairs witha similar sets of control pairs using some type of matching scheme. Another option is to introduce78This product is coded as zero for the countries and cities for whom we do not have data for the probit regressions.They are very small entities.744.5. Conclusiona bilateral variable such as the product of the partner-country per capita GDPs. Interestingly,in standard gravity models, country incomes enter multiplicatively and are therefore absorbed bycountry fixed effects in the linear in logs transformation. Thus, the inclusion of a GDP productterm requires a non-standard functional form such as the product of propensities that we haveintroduced in this paper.4.5 ConclusionThis paper examines the effect of treaty linkages established between Chinese cities and foreigncountries during the 19th century on China’s current bilateral trade. We hypothesize that historicallinkages may exert contemporary effects by some form of persistent capital that lowers trade costs.Using data on the trade of Chinese cities, we find initial evidence that treaties are associated withsubstantially higher trade today. However, once we add additional controls allowing for highertrade among groups of trading entities, the bilateral linkage effect disappears.We believe these empirical results provide a useful lesson for gravity estimation of policy effects.The sample of “control” trading partners matters. We find strong results for bilateral linkage effectswhen we use all countries and cities in the data set, but these results disappear in restricted samples.Essentially, even though we employ country and city fixed effects in all specifications, they cannotcapture unobserved bilateral effects that may be correlated with the variable of interest. In our case,we determine that the unobserved bilateral influence was related to a complementarity between thelevels of development of trading partners resulting in higher trade.While we do not find that a historical treaty link between a Chinese city and foreign country isassociated with more trade today, there is some evidence of group effects: Trade is higher among thegroup of countries and cities that were involved in treaty arrangements. We propose two possibleexplanations for the observed group effects. First, participation in a treaty arrangement changedthe industrial structure of cities party to a treaty in a manner that continues to facilitate economicexchange today. Second, the experience gained through participation in a treaty arrangementcreated knowledge that has passed down through generations. Theoretical underpinnings of groupeffects are developed in recent research of Morales et al. (2014) modeling the interdependence ofexport markets. Our results are consistent with the proposition that bilateral linkages promotemultilateral trade by generating group effects.75Chapter 5ConclusionThis thesis is a collection of three essays on how Chinese firms link with foreign markets. The firstessay, Chapter 2, takes advantage of automaker-intermediary matching data to provide evidenceon the economics underlying the linkages among domestic producers and export intermediaries.On the registry, larger automakers are matched with intermediaries who bring more foreign orders.Substantial churning is observed in the sets of intermediaries registered by the automakers. Thesefindings are in line with studies examining seller-buyer matches in international markets. Thisessay proposes a model of export through intermediaries to describe how this registration policyfacilitated market division as an equilibrium outcome and generated inefficiency. Non-assortativematching equilibria also led to inefficiencies since low cost automakers benefited more from highquality intermediaries.This paper also discusses the welfare consequences in the context of the model. Automakersbenefit from this regulation mainly due to the gaining of market power while intermediaries becomeworse off. The question is more complex regarding the rationales that motived the Chinese govern-ment to issue this regulation. Potential justifications such as reducing fixed costs and extractingmore foreign consumer surplus require additional modeling in future research.In Chapter 3, a surprising empirical finding is documented among Chinese exporters: Firmproductivity and the foreign to domestic sales ratio are negatively correlated. This pattern remainsrobust to ownership, factor intensity, and processing trade controls. Existing trade theories (e.g.,Melitz, 2003) predict constant sales ratios across markets. Firms of higher productivity enter morecountries and therefore sell more in the foreign market as a whole. Such predictions contradictthe empirical findings established in this research. In order to rationalize the negative correla-tion between firm productivity and the export-domestic sales ratio, heterogeneous marketing costelasticities are introduced into the Arkolakis (2010) model. A higher marketing cost elasticity do-mestically gives rise to a faster sales expansion in the home market as firm productivity grows.Further, the model predicts negative correlation between firm productivity and the quantity ratiowhile the price ratio between foreign and domestic market is not affected. This prediction helpsto distinguish the marketing cost heterogeneity channel proposed in this research from alternativeexplanations. Empirically, advertisement expenditures are used to capture the effect of marketingcost variation across industries. Future research aims to find a direct measure of marketing cost,estimate the marketing elasticity across countries, and carry out counterfactuals for policies thathelp ease barriers to reaching local consumers.The last essay, in Chapter 4, examines the effect of treaty linkages established between Chinese76Chapter 5. Conclusioncities and foreign countries during the 19th century on China’s current bilateral trade. Initial evi-dence shows that treaties are associated with higher bilateral trade today. However, this bilaterallinkage effect disappears once we restrict the control groups or add propensity controls. Ratherthan bilateral, the impact of treaty is shown to be multilateral. That is, trade is higher amongthe group of countries and cities that were involved in treaty arrangements. Two possible expla-nations for the observed group effects include complementary development in industry structuresand knowledge of business practices passed down through generations. These results are consis-tent with the proposition that bilateral linkages promote multilateral trade by generating groupeffects. Empirical results from this research shed light on the gravity estimation of policy effects.Specifically, country and city fixed effects cannot capture unobserved bilateral effects that may becorrelated with the policy variable.77BibliographyAcemoglu, D., Johnson, S., and Robinson, J. A. (2000). The colonial origins of comparativedevelopment: An empirical investigation.Ahn, J., Khandelwal, A. K., and Wei, S.-J. (2011). The role of intermediaries in facilitating trade.Journal of International Economics, 84(1):73–85.Akerman, A. et al. (2010). A theory on the role of wholesalers in international trade based oneconomies of scope. Technical report, Stockholm University, Department of Economics.Alesina, A., Giuliano, P., and Nunn, N. (2013). On the origins of gender roles: Women and theplough. The Quarterly Journal of Economics, 128(2):469–530.Anderson, S. P. and Renault, R. (1999). Pricing, product diversity, and search costs: A bertrand-chamberlin-diamond model. The RAND Journal of Economics, pages 719–735.Arkolakis, C. (2010). Market penetration costs and the new consumers margin in internationaltrade. Journal of Political Economy, 118(6):1151–1199.Atkin, D., Khandelwal, A. K., and Osman, A. (2017). Exporting and firm performance: Evidencefrom a randomized experiment. The Quarterly Journal of Economics, 132(2):551–615.Bai, X., Krishna, K., and Ma, H. (2017a). How you export matters: Export mode, learning andproductivity in china. Journal of International Economics, 104:122–137.Bai, X., Krishna, K., and Ma, H. (2017b). How you export matters: Export mode, learning andproductivity in china. Journal of International Economics, 104:122–137.Bernard, A. B., Eaton, J., Jensen, J. B., and Kortum, S. (2003). Plants and productivity ininternational trade. The American Economic Review, 93(4):1268–1290.Bernard, A. B., Grazzi, M., and Tomasi, C. (2011). Intermediaries in international trade: Directversus indirect modes of export. Technical report, National Bureau of Economic Research.Bernard, A. B., Moxnes, A., and Ulltveit-Moe, K. H. (2014). Two-sided heterogeneity and trade.Technical report, National Bureau of Economic Research.Blum, B. S., Claro, S., and Horstmann, I. (2009). Intermediation and the nature of trade costs:Theory and evidence.78BibliographyBlum, B. S., Claro, S., and Horstmann, I. (2010). Facts and figures on intermediated trade. TheAmerican Economic Review, 100(2):419–423.Broda, C. and Weinstein, D. E. (2006). Globalization and the gains from variety. Quarterly Journalof Economics, 121(2).Che, Y., Du, J., Lu, Y., and Tao, Z. (2013). Once an enemy, forever an enemy? the long-runimpact of the japanese invasion of china from 1937 to 1945 on trade and investment. Technicalreport, unpublished manuscript.Chen, B. and Li, Z. (2014). An anatomy of intermediaries in china’s export market. China EconomicJournal, 7(2):187–213.Crino`, R. and Epifani, P. (2012). Productivity, quality and export behaviour. The EconomicJournal, 122(565):1206–1243.Crozet, M., Lalanne, G., and Poncet, S. (2013). Wholesalers in international trade. EuropeanEconomic Review, 58:1–17.Dai, M., Maitra, M., and Yu, M. (2016). Unexceptional exporter performance in China? The roleof processing trade. Journal of Development Economics, 121:177–189.Defever, F. and Rian˜o, A. (2017). Subsidies with export share requirements in china. Journal ofDevelopment Economics, 126:33–51.Dhingra, S. and Morrow, J. (2012). Monopolistic competition and optimum product diversity underfirm heterogeneity. London School of Economics, mimeograph.Dixit, A. K. and Stiglitz, J. E. (1977). Monopolistic competition and optimum product diversity.The American Economic Review, 67(3):297–308.Eaton, J., Kortum, S., and Kramarz, F. (2011). An anatomy of international trade: Evidence fromfrench firms. Econometrica, 79(5):1453–1498.Eaton, J., Kortum, S. S., and Sotelo, S. (2012). International trade: Linking micro and macro.Technical report, National bureau of economic research.Engerman, S. L. and Sokoloff, K. L. (2002). Factor endowments, inequality, and paths of develop-ment among new world economics. Technical report, National Bureau of Economic Research.Eslava, M., Tybout, J., Jinkins, D., Krizan, C., Eaton, J., et al. (2015). A search and learning modelof export dynamics. In 2015 Meeting Papers, number 1535. Society for Economic Dynamics.Feyrer, J. and Sacerdote, B. (2009). Colonialism and modern income: islands as natural experi-ments. The Review of Economics and Statistics, 91(2):245–262.79BibliographyGanapati, S. (2016). The modern wholesaler: Global sourcing, domestic distribution, and scaleeconomies. Technical report, Yale University.Gelman, A., Carlin, J. B., Stern, H. S., and Rubin, D. B. (2014). Bayesian data analysis, volume 2.Chapman & Hall/CRC Boca Raton, FL, USA.Glick, R. and Taylor, A. M. (2010). Collateral damage: Trade disruption and the economic impactof war. The Review of Economics and Statistics, 92(1):102–127.Gould, D. M. (1994). Immigrant links to the home country: empirical implications for us bilateraltrade flows. The Review of Economics and Statistics, pages 302–316.Gu, C. (1981). Missionary and modern China. Shanghai People’s Press, Shanghai.Gu, Y. and Wenzel, T. (2009). A note on the excess entry theorem in spatial models with elasticdemand. International Journal of Industrial Organization, 27(5):567–571.Guiso, L., Sapienza, P., and Zingales, L. (2009). Cultural biases in economic exchange? TheQuarterly Journal of Economics, 124(3):1095–1131.Guo, T. (1992). Main activities of HSBC in China, pages 85–93. Wenshi press.Head, K., Jing, R., and Ries, J. (2016). Import sourcing of chinese cities: Order versus randomness.Journal of International Economics, forthcoming.Head, K. and Mayer, T. (2014). Gravity Equations: Workhorse, Toolkit, and Cookbook, volume 4,pages 131–195. Elsevier.Head, K., Mayer, T., and Ries, J. (2010). The erosion of colonial trade linkages after independence.Journal of International Economics, 81(1):1–14.Head, K. and Ries, J. (1998). Immigration and trade creation: econometric evidence from canada.Canadian journal of economics, pages 47–62.Head, K., Ries, J., Sun, X., and Hong, J. (2015). The legacy of nineteenth century treaties on thecurrent trade of chinese cities. Asia-Pacific Journal of Accounting & Economics, 22(3):251–270.Hu, W.-M., Xiao, J., and Zhou, X. (2014). Collusion or competition? interfirm relationships in thechinese auto industry. The Journal of Industrial Economics, 62(1):1–40.Huang, H., Ju, J., and Yue, V. Z. (2015). A unified model of structural adjustments and interna-tional trade: Theory and evidence from china. Unpublished Manuscript.Jia, R. (2014). The legacies of forced freedom: China’s treaty ports. Review of Economics andStatistics, 96(4):596–608.80BibliographyKeller, W., Li, B., and Shiue, C. H. (2011). China’s foreign trade: Perspectives from the past 150years. The World Economy, 34(6):853–892.Keller, W., Li, B., and Shiue, C. H. (2013). Shanghai’s trade, china’s growth: Continuity, recovery,and change since the opium wars. IMF Economic Review, 61(2):336–378.Krishna, P. (1998). Regionalism and multilateralism: A political economy approach. The QuarterlyJournal of Economics, 113(1):227–251.Levinsohn, J. and Petrin, A. (2003). Estimating production functions using inputs to control forunobservables. The Review of Economic Studies, 70(2):317–341.Li, B. (2012). Treaties of Modern China. Shanghai Dictionary Press, Shanghai.Li, K., Xia, X., and Gu, R. (1981). Brief Analysis of China’s Foreign Trade History. Foreign TradePress, Beijing.Lin, S. and Ye, H. (2010). Dollarization does promote trade. Journal of International Money andFinance, 29(6):1124–1130.Lo´pez de Silanes, F., La Porta, R., Shleifer, A., and Vishny, R. (1998). Law and finance. Journalof political economy, 106:1113–1155.Lu, D. (2010). Exceptional exporter performance? evidence from chinese manufacturing 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.Manova, K. and Zhang, Z. (2012). Export prices aross firms and destinations. The QuarterlyJournal of Economics, 127:379–436.Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate industryproductivity. Econometrica, 71(6):1695–1725.Melitz, M. J. and Ottaviano, G. I. P. (2008). Market size, trade, and productivity. The Review ofEconomic Studies, 75(1):295–316.Midler, P. (2010). Poorly made in China: an insider’s account of the China production game. JohnWiley & Sons.Morales, E., Sheu, G., and Zahler, A. (2014). Extended gravity. mimeo.Murphey, R. (1975). China Meets the West: the Treaty Ports. Collier Macmillan Publishers, NewYork London.81BibliographyNield, R. (2010). The China Coast: Trade And The First Treaty Ports. Joint Publishing, HongKong.Nunn, N. (2009). The importance of history for economic development. Annu. Rev. Econ., 1(1):65–92.Ornelas, E. (2005). Trade creating free trade areas and the undermining of multilateralism. Euro-pean Economic Review, 49(7):1717–1735.Saggi, K. and Yildiz, H. M. (2011). Bilateral trade agreements and the feasibility of multilateralfree trade. Review of International Economics, 19(2):356–373.Soderbery, A. (2015). Estimating import supply and demand elasticities: Analysis and implications.Journal of International Economics, 96(1):1–17.Tybout, J., Jinkins, D., Xu, D. Y., Eaton, J., et al. (2016). Two-sided search in internationalmarkets. In 2016 Meeting Papers, number 973. Society for Economic Dynamics.Wang, Z. and Yu, Z. (2012). Trading partners, traded products and firm performances of china’sexporter-importers: Does processing trade make a difference? The World Economy, 35(12):1795–1824.Zhang, H. (1993). Treaty ports and concessions in modern China. Tianjin People’s Press, Tianjin.82Appendix AAppendix for Chapter 2A.1 Registration policyIn 2007 version of the registration policy, requirements for automakers include (1) makers shouldbe enlisted in “Automakers and Auto Products Bulletin” published by National Development andReform Commission; (2) hold effective product certificates (CCC certificate); (3) have export com-patible after-sales maintenance ability and sales network in major markets. For export intermedi-aries: (1) should obtain authorization from automakers and only export vehicles from authorizedmakers; (2) automakers and listed intermediaries should share the legal responsibility of productquality and after-sales services. Note that firms located in export processing zones and export 100%of their products are not restricted by this policy.79A.2 Comparing staying and exiting intermediariesIn this section, a full range of aspects are compared between staying and exiting intermediaries.Those who are dropped from the market are of smaller size, sell at lower prices, have fewer marketaccess, handle fewer product varieties and less specialized in the auto industry. Intermediariespost-regulation are in general better than their pre-regulation counterparts in every aspect listedabove. However, exitors post-regulation are even smaller in size and market coverage while slightlyinferior in terms of product variety handled and knowledge of the auto industry (indicated by theshare of auto exports).Table A.3 illustrates similar comparisons from a different perspective. It shows the probabilityof intermediary exit is negatively correlated with its export size and smaller intermediaries are morelikely to exit post-regulation. The predictions on exit decisions that correspond with prices andforeign market coverages do not change much with the regulation. The effect of product varietysuggests that intermediaries with smaller number of products are more likely to exit. However,this factor becomes less important post-regulation. Similar pattern is shown for the share of autoexports intermediaries handle.79That’s why the empirical analysis of this paper only include ordinary exports.83A.2. Comparing staying and exiting intermediariesTable A.1: Exiting vs staying Intermediaries(1) (2) (3) (4)ln(ExpVal) ln(q) ln(AveP) ln(MedP)exit -1.876a -1.417a -0.526a -0.452a(0.113) (0.111) (0.068) (0.068)post 1.717a 0.567a 1.148a 1.150a(0.119) (0.121) (0.070) (0.072)exit×post -0.717a -0.663a -0.069 -0.063(0.151) (0.142) (0.095) (0.096)N 4197 4197 4197 4197R2 0.290 0.160 0.174 0.165Standard errors in parentheses. c p <0.1, b p <0.05, a p <0.01Intermediary-year level regression with intermediary clusters.Table A.2: Exiting vs staying Intermediaries (cont.)(1) (2) (3) (4) (5) (6)ln(nMkt) nMkt ln(nProd) nProd autoShr HS87Shrexit -0.556a -1.050a -0.458a -0.692a -0.107a -0.175a(0.050) (0.095) (0.038) (0.069) (0.016) (0.018)post 0.243a 0.397a 0.152a 0.139a 0.233a 0.222a(0.060) (0.096) (0.039) (0.053) (0.018) (0.020)exit×post -0.272a -0.423a -0.096b -0.042 -0.083a -0.063b(0.063) (0.129) (0.044) (0.063) (0.023) (0.026)N 4197 4197 4197 4197 4197 4197R2 0.150 0.155 0.140 0.149Standard errors in parentheses. c p <0.1, b p <0.05, a p <0.01Intermediary-year level regression with intermediary clusters.84A.2. Comparing staying and exiting intermediariesTable A.3: Discrete hazard model(1) (2) (3) (4) (5) (6)LP Probit Logit LP Probit Logitpost -0.086 -0.029 0.007 0.068 0.289 0.541(0.095) (0.326) (0.564) (0.104) (0.345) (0.591)ln(q) -0.090a -0.263a -0.436a -0.040a -0.109a -0.178a(0.004) (0.015) (0.026) (0.007) (0.021) (0.035)post×ln(q) -0.002 -0.053b -0.099b -0.023b -0.099a -0.175a(0.006) (0.023) (0.040) (0.010) (0.031) (0.053)ln(P ) -0.088a -0.256a -0.428a -0.057a -0.168a -0.279a(0.006) (0.021) (0.035) (0.008) (0.023) (0.039)post×ln(P ) 0.012 0.022 0.035 -0.007 -0.023 -0.040(0.010) (0.033) (0.056) (0.011) (0.036) (0.061)ln(nmkt) -0.078a -0.332a -0.578a(0.016) (0.055) (0.092)post×ln(nmkt) 0.019 -0.073 -0.213(0.020) (0.088) (0.146)ln(nprod) -0.126a -0.454a -0.758a(0.021) (0.067) (0.112)post×ln(nprod) 0.097a 0.375a 0.659a(0.029) (0.099) (0.166)HS87Shr -0.115a -0.338a -0.556a(0.032) (0.092) (0.152)post× Shr 0.057 0.231c 0.391c(0.045) (0.131) (0.218)cons 1.519a 2.952a 4.931a 1.243a 2.184a 3.625a(0.057) (0.200) (0.343) (0.067) (0.214) (0.362)N 4197 4197 4197 4197 4197 4197R2 0.217 0.240Standard errors in parentheses. c p <0.1, b p <0.05, a p <0.01Intermediary-year level regressions with intermediary clusters.85A.3. Notes for Bayesian statistics and quality distributionA.3 Notes for Bayesian statistics and quality distributionThe following bullet points are drawn from Gelman et al. (2014). They explain a general rule ofBayesian updating process and provide intuitions of the parameters (e.g., effective sample size).• Data: Q = {q1, q2, · · · , qn}• Sampling model: qi|µ ∼ Poisson(µ), where qi are i.i.d. and µ is the population parameterwhich is unknown and random.• Prior: µ ∼ Gamma(α, β) and prior mean E(µ) = α/β• Posterior: µ|Q ∼ Gamma(α+∑ qi, β + n) and posterior meanE(µ|Q) = (α+∑qi)/(β + n) =αβββ + n+∑qinnβ + nNote that the posterior mean is a weighted average of prior mean and sample mean. And therelative weight is determined by β (effective sample size) and n.• Predictive distribution: q ∼ Neg-bin(α, 1/(β + 1)). Also, E(q) = α/β and Var(q) =α(β + 1)/β. These are the moments we use to generate priors in the simulation.Based on the above analysis, two justifications can be made for the rather arbitrary choice ofα0 = 2 and β0 = 1 which govern the distribution of intermediary quality draws.First, as demonstrated in the stylized fact 2 that the majority of intermediaries receive zeroor only a small amount of auto export orders. The average size of intermediary demand shock isassociated with the mean of µ which is determined by α0/β0.The second benchmark rests on the weight allocated between priors and data. In Bayesianupdating process, the expectation of µi conditional on observed Ai is a weighted average of priormean and sample mean. In reality, firms usually have little information on the prior distribution.Therefore, more weight is assigned to the realization of intermediary demand shocks in previousperiods (i.e., sample mean). This will be achieved by choosing a small β0 which is the effectivesample size compared to the observations used for belief updating.Together, they provide a very skewed distribution of µi and the Poisson Ai (Figure A.1).A.4 Intermediary/automaker ratioSuggestive evidence on the number of intermediaries per automaker before policy.Figure 2.1 indicates that intermediary-automaker ratio in 2006 was iAv/mAvDir = 746/142 =5.2 where automakers and intermediaries available are identified by both orthography and post-policy list.80 And Figure A.2 suggests that the ratio between direct exporting automakers and those80How to identify intermediaries before policy? The cleanest way is to use the registration list. However, interme-diaries shaken out by the policy can not be found on the list. Therefore, orthography provides additional information86A.4. Intermediary/automaker ratioFigure A.1: µi ∼ Gamma(2, 1) and Ai ∼ Poisson(µi)who list at least one intermediary with the registration is mRgDir = 1.05mSlot. In other words,about 82% of automakers on the registration use intermediaries, mSlot = 0.82mRg. Then, whatwe’re interested in is the number of intermediaries available in 2006 and the number of automakerswho potentially need intermediaries.iAvmSlot=iAvmRgDir/1.05=iAvmAvDir/1.05= 1.05× 5.2 = 5.46The above method relies on the credibility of iAv and mAvDir which are calculated partiallybased on orthography. We could use another more restrictive measure by predicting the numberof automakers and intermediaries who would be on the registration list in 2006 based on iAv andmAvDir. Figure A.3 shows that on average the number of auto-exporters identified are larger thanthose registered on the list. That is, mAvDir = 1.38mRg and iAv = 2.24iRg. Then,iRgmSlot=iRg0.82mRg=iAv/2.240.82mAvDir/1.38=5.2× 1.382.24× 0.82 = 3.9It is restrictive in the sense that the number of automakers on the list is more likely to cover theuniverse of automakers who intends to export than intermediaries.to identify intermediaries. We worry that orthography is not precise and might generate errors. In oder to make thenumbers comparable pre and post policy, orthography is also used to identify intermediaries beyond the list afterpolicy. Also since we’re interested in the number of intermediaries per automaker, I use the same method to identifyautomakers. In sum, the number of automakers and intermediaries available in Figure 2.1 are identified by bothregistration list and orthography.87A.4. Intermediary/automaker ratioFigure A.2: Direct export automaker/Automakers who list intermediaries) (mDir/mSlot)Figure A.3: Number of auto-exporters: available vs registration88Appendix BAppendix for Chapter 3B.1 Correlation between value-added per worker and TFPFigure B.1 shows the correlation between two productivity measurements. The x-axis is ln(TFP)and the y-axis is ln(value-added per worker). The correlation between two measurements is 0.76.Figure B.1: The Correlation between Value-added Per Worker and TFPNotes: This figure shows the correlation between two productivity measurements. The x-axis is ln(TFP)and the y-axis is ln(value-added per worker). The correlation between two measurements is 0.76.The following tables provide regression results with value-adder per worker as the measure offirm productivity.B.2 Productivity and sales ratio: quadratic regressionFigure 3.1 shows that the correlation between firm productivity and export intensity is an hump-shaped curve. However, the positive correlation before the turning point is not significant. Thus,we use a linear regression model to describe the correlation. In this section we use a quadraticregression to address the hump-shaped correlation. The regression can be written as follows:ln(Export/Domestic Sales Ratioijkt) = η0 + η1 ln(Pijkt) + η2 ln(Pijkt)2+ other controls+ µjkt + ijkt(B.1)89B.3. Industry list and firm shareTable B.1: Firm productivity and export sales(1) (2) (3) (4) (5) (6)LaborProd K/L Ownership TFP K/L Ownershipln(LabProd) 0.426a 0.397a 0.401a(0.011) (0.010) (0.010)TFP 0.809a 0.796a 0.791a(0.013) (0.012) (0.012)KL 0.065a 0.032a 0.085a 0.060a(0.010) (0.010) (0.008) (0.009)Ownership FE X XProvince-Industry-Year FE X X X X X XCluster By Industry X X X X X XN 386924 386924 386924 386924 386924 386924R2 0.307 0.309 0.325 0.474 0.477 0.489Notes: ln(LabProd) stands for labor productivity measured by value-added per worker.Standard errors in parentheses. c p<0.1, b p<0.05, a p<0.01.The results are shown in Tables B.7 and B.8. We find that the coefficient of ln(Productivity)2is always negative and significant. But the coefficient of ln(Productivity) is only significant whenwe use a firm’s TFP as our measurement and the regression is at the firm level. Thus, a linearregression is a better way to address the correlation between the firms’ export/domestic sales ratioand their productivity.B.3 Industry list and firm shareThere are 30 manufacturing industries at two-digit level. The name list of these industries is asfollows and the number in the bracket is the proportion of firms in that industry.Figures B.2 and B.3 show the correlation between the average export intensity and productivitypercentile by industries. The industries are on 2-digits level. From the first industry to the last one,the capital/Labor ratio of industry increases. The only two industries, which are abnormal, are“Recycling of Waste and Scrap ”(industry 9) and “Manufacture of Tobacco Products” (industry29).B.4 Firm productivity and sales ratio: top 10 destinationsIn Regression 3.2 we only control the export destination fixed effect. Next, we run regressions foreach export destination. We rank the destinations according to the number of Chinese exportingfirms that sell products in that country. We choose the top ten export destinations: the US, HongKong, Japan, South Korea, Germany, the UK, Canada, Australia, Taiwan, and Italy. The resultsare given in Table B.10. We find that the negative correlation remains robust for the top ten most90B.4. Firm productivity and sales ratio: top 10 destinationsTable B.2: Firm productivity and sales ratioDependent Variable: ln(Export/Domestic Sales)(1) (2) (3)ln(Value-added per worker) -0.065** -0.146*** -0.272***(0.030) (0.026) (0.033)ln(Value-added per worker)× Domestic -0.019(0.032)ln(Value-added per worker)× Homogeneous 0.080**(0.033)Homogeneous Dummy -0.221(0.152)ln(Capital/Labor Ratio) -0.228*** -0.170*** -0.150***(0.021) (0.015) (0.028)ln(Sale) -0.211*** -0.024 0.063*(0.028) 0.027) (0.036)Constant 2.659*** 0.833*** 0.575(0.298) (0.269) (0.396)Ownership FE X X XProvince-Industry-Year FE X X XCluster By Industry X X XExclude Processing Trade XObservations 69,691 275,872 80,147R-squared 0.472 0.369 0.419Notes: This table shows the correlation between export/domestic sales ratio and productivity (value-added per worker) on the firm level.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.91B.4. Firm productivity and sales ratio: top 10 destinationsTable B.3: Firm productivity and sales ratio (value-added/worker): firm-destination levelDependent Variable: ln(Export/Domestic Sales)All OECD LDCln(Productivity) 0.027 0.00133 0.0686(0.045) (0.0399) (0.0606)ln(Capital/Labor Ratio) -0.202*** -0.202*** -0.201***(0.031) (0.0290) (0.0396)ln(Sale) -0.527*** -0.488*** -0.589***(0.037) (0.0337) (0.0470)ln(No. of markets) -0.029 -0.0826 0.0720(0.055) (0.0539) (0.0597)Constant 2.697*** 2.708*** 2.712***(0.386) (0.362) (0.504)Ownership FE X X XCountry-Province-Industry-Year FE X X XCluster By Industry X X XExclude Processing Trade X X XObservations 560,850 305,163 255,687R-squared 0.649 0.610 0.694Notes: This table shows the correlation between export/domestic sales ratio and productivityon the firm-destination level.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.92B.4. Firm productivity and sales ratio: top 10 destinationsTable B.4: Effect of advertisement expenditure: value-add/workerDependent Variable: ln(Export/Domestic Sales)(1) (2) (3) (4)ln(Productivity) -0.141*** -0.0438 -0.0688** 0.0420(0.0181) (0.0312) (0.0313) (0.0462)ln(Productivity)× Advertisement/Sales Ratio -7.037*** -10.72*** -7.241*** -9.202***(1.759) (3.353) (1.986) (3.141)ln(Capital/Labor Ratio) -0.170*** -0.229*** -0.149*** -0.202***(0.0154) (0.0209) (0.0230) (0.0307)ln(Sale) -0.0231 -0.208*** -0.338*** -0.526***(0.0268) (0.0277) (0.0384) (0.0366)ln(No. of markets) -0.0773 -0.0297(0.0496) (0.0544)Constant 0.789*** 2.645*** 1.039*** 2.701***(0.272) (0.297) (0.390) (0.385)Ownership FE X X X XProvince-Industry-Year FE X XCountry-Province-Industry-Year FE X XCluster By Industry X X X XExclude Processing Trade X XObservations 275,872 69,691 1,098,287 560,850R-squared 0.369 0.473 0.585 0.649Notes: This table shows the impact of advertising spending on the correlation between export/domesticsales ratio and productivity.The first two columns are at the firm level. The last two columns are at the firm-destination level.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.93B.4. Firm productivity and sales ratio: top 10 destinationsTable B.5: Effect of relative market size (value-added per worker)Dependent Variable: ln(Export/Domestic Sales)(1) (2) (3)ln(Productivity) -0.100*** 0.0215(0.0303) (0.0451)ln(Productivity)× Relative Market Size -0.00702* -0.00939** -0.00348(0.00388) (0.00430) (0.00438)ln(Capital/Labor Ratio) -0.160*** -0.199*** -0.183***(0.0222) (0.0278) (0.0280)ln(Sale) -0.337*** -0.519*** -0.505***(0.0337) (0.0390) (0.0392)ln(No. of markets) -0.0346 -0.00168 -0.0132(0.0538) (0.0592) (0.0587)Constant 1.220*** 2.677*** 2.562***(0.331) (0.384) (0.322)Country-Province-Industry-Year FE X X XOwnership FE X X XCluster By Industry X X XExcluding Processing Trade X XIndustry Dummy × ln(Productivity) XObservations 391,677 212,825 212,825R-squared 0.515 0.586 0.591Notes: This table shows the impact of market size on the correlation between export/domestic salesratio and productivity on the firm-destination level.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.94B.4. Firm productivity and sales ratio: top 10 destinationsTable B.6: Firm productivity and sales ratio: value-added/worker (HK)ln(Price Ratio) ln(Quantity Ratio)(1) (2)ln(Productivity) -0.00731 -0.0867*(0.0104) (0.0514)ln(Capital/Labor Ratio) -0.00483 -0.0616*(0.00608) (0.0360)Constant 0.720** -4.735*(0.288) (2.669)Country-Province-Industry-Year FE X XOwnership FE X XProduct(HS 2-digit) FE X XExclude Processing Trade X XCluster By Industry X XObservations 109,743 109,743R-squared 0.431 0.515Notes: This table shows the correlation between price ratio, quantity ratio and pro-ductivity on the firm-destination level. We calculate the export prices (quantities)on HS08 level for the same firm.The price (quantity) ratio is Export Price (Quantity) to Country bExport Price (Quantity) to Hong Kongat firm-productlevel. Here other countries (regions) include US, Japan, South Korea, Germany,UK, Canada, Italy, Australia and Taiwan. These countries (regions) are the top 10destinations of Chinese exporting firms.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.95B.4. Firm productivity and sales ratio: top 10 destinationsTable B.7: Quadratic 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/Labor 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 0.474Notes: This table shows the correlation between export/domestic sales ratio and pro-ductivity at the firm level.We use value-added per worker to measure a firm’s productivity in the first two and useTFP in last two columns.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.96B.4. Firm productivity and sales ratio: top 10 destinationsTable B.8: Quadratic 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/Labor 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(No. of markets) -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 0.65081Notes: This table shows the correlation between export/domestic sales ratio and produc-tivity at the firm-destination level.We use value-added per worker to measure a firm’s productivity in the first two and useTFP in last two columns.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.97B.4. Firm productivity and sales ratio: top 10 destinationsTable B.9: List of industriesCIC Industry name (share of firms)17 Manufacture of Textile (12.22%)18 Manufacture of Wearing Apparel, Footwear and Headwear (8.01%)35 Universal Equipments Manufacturing (7.67%)39 Manufacture of Electric Machines and Equipments (6.82%)26 Chemical Raw Materials and Manufacture of Other Basic Chemical Raw Ma-terials (6.50%)40 Manufacture of Telecommunication Equipments, Computers and Other Elec-tric Equipments (5.94%)34 Manufacture of Metal Products (5.62%)31 Manufacture of Non-metal Products (4.54%)30 Manufacture of Plastic Products (4.53%)37 Manufacture of Transportation Equipments (4.31%)36 Manufacture of Special Equipments (4.03%)19 Manufacture of Leather, Fur Apparel, Feather and Products (3.40%)42 Manufacture of Arts and Crafts and Other Manufacturing (2.91%)24 Manufacture of Cultural, Educational and Sporting Products (2.38%)14 Manufacture of Food Products (2.09%)41 Manufacture of Instruments and Appliances, Culture-related and Office Ma-chinery (2.02%)27 Manufacture of Pharmaceuticals (2.02%)20 Manufacture of Wood and Articles of Wood, Bamboo, Bine, Palm Fibre, Strawand Grass (1.74%)22 Manufacture of Pulp, Paper, Paperboard and Articles of Paper and Paper-board (1.47%)21 Furniture Manufacturing (1.38%)29 Manufacture of Rubber Products (1.35%)13 Processing Industry of Agricultural and Subsidiary Food (1.30%)33 Manufacture and Casting of Non-ferrous Metals (1.19%)32 Manufacture and Casting of Ferrous Metals (1.02%)15 Manufacture of Drinking Products (0.87%)23 Printing and Reproduction of Recorded Media (0.76%)25 Processing of Crude Oil, Coking and Nuclear Fuel (0.42%)28 Manufacture of Chemical Fibres (0.41%)16 Manufacture of Tobacco Products (0.11%)43 Recycling of Waste and Scrap (0.02%)Manufacture of Textile has the most firms and is among the top 3 industries in terms of export value.Manufacture of Electric Machines and Equipments follows Textile and Wearing Apparel to be the thirdlargest industry in terms of number of firms. It is among the top 5 in export value and its ranking isrising in recent years.98B.4. Firm productivity and sales ratio: top 10 destinationsFigure B.2: Average export intensity and productivity (value-added per worker). 20 40 60 80 100Leather. 20 40 60 80 100Wearing Apparel. 20 40 60 80 100Crafts. 20 40 60 80 100Cutural Products. 20 40 60 80 100Furniture. 20 40 60 80 100Office Machinery. 20 40 60 80 100Metal. 20 40 60 80 100Textile0. 20 40 60 80 100Recycling. 20 40 60 80 100Universal Equipmentsl. 20 40 60 80 100Special Equipments. 20 40 60 80 100Wood. 20 40 60 80 100Electric. 20 40 60 80 100Plastic0. 20 40 60 80 100Rubber. 20 40 60 80 100Agricultural0. 20 40 60 80 100Food. 20 40 60 80 100Non-metal0. 20 40 60 80 100Printing. 20 40 60 80 100Telecommunication. 20 40 60 80 100Transportation. 20 40 60 80 100Pharmaceuticals0.2.4.60 20 40 60 80 100Drinking0. 20 40 60 80 100Paper. 20 40 60 80 100Raw Materials. 20 40 60 80 100Non-ferrous Metal0.2.4.60 20 40 60 80 100Ferrous Metal0. 20 40 60 80 100Fiber-.50.511.50 20 40 60 80 100Tobacco0. 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 productivity (value-added per worker) percentile by industries. We only include firms that sell both in the domestic marketand foreign markets. The industries are on 2-digits level. From the first industry to the last one, thecapital/Labor ratio of industry increases. The only two industries, which are abnormal, are “Recyclingof Waste and Scrap ”(industry 9) and “Manufacture of Tobacco Products” (industry 29).99B.4. Firm productivity and sales ratio: top 10 destinationsFigure B.3: Average export intensity and productivity (TFP). 20 40 60 80 100Leather. 20 40 60 80 100Wearing Apparel. 20 40 60 80 100Crafts. 20 40 60 80 100Cutural Products. 20 40 60 80 100Furniture. 20 40 60 80 100Office Machinery. 20 40 60 80 100Metal. 20 40 60 80 100Textile0. 20 40 60 80 100Recycling. 20 40 60 80 100Universal Equipments. 20 40 60 80 100Special Equipments. 20 40 60 80 100Wood. 20 40 60 80 100Electric. 20 40 60 80 100Plastic. 20 40 60 80 100Rubber0. 20 40 60 80 100Agricultural0. 20 40 60 80 100Food. 20 40 60 80 100Non-metal0. 20 40 60 80 100Printing. 20 40 60 80 100Telecommunication0.2.4.60 20 40 60 80 100Transportation0.2.4.60 20 40 60 80 100Pharmaceuticals0. 20 40 60 80 100Drinking. 20 40 60 80 100Paper. 20 40 60 80 100Raw Materials. 20 40 60 80 100Non-ferrous Metal0. 20 40 60 80 100Ferrous Metal0. 20 40 60 80 100Fibres-.50.510 20 40 60 80 100Tabacco0. 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 productivity (TFP)percentile by industries. We only include firms that sell both in the domestic market and foreign markets.The industries are on 2-digits level. From the first industry to the last one, the capital/Labor ratio ofindustry increases. The only two industries, which are abnormal, are “Recycling of Waste and Scrap”(industry 9) and “Manufacture of Tobacco Products” (industry 29).100B.4. Firm productivity and sales ratio: top 10 destinationssignificant destinations.Table B.10: Firm productivity and sales ratio: top 10Dependent 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 XNotes: This table shows the correlation between export/domestic sales ratio andproductivity for the top ten destinations.The productivity in the first two columns is value-added per worker. The productivityin last two columns is TFP.Standard errors in parentheses. **Significant at 5%; ***significant at 1%.USA United States; HKG Hong Kong; JPN Japan; KOR South Korea; GER Ger-many; GBR: United Kingdom; AUS Australia; CAN Canada; TWN Taiwan; ITAItaly.101B.5. Proof of Proposition 1B.5 Proof of Proposition 1Proof: By (3.12), we haveln γ(φ) = λ[ln(Lb)− ln(Lc)] + (σ − 1)[ln(φ∗cc)− ln(φ∗cb)]+ ln[1− (φ∗cbφ)σ−1κcb]− ln[1− (φ∗ccφ)σ−1κcc].Therefore,∂ ln γ(φ)∂ lnφ= φ σ−1κcb 1φ(φ∗cbφ )σ−1κcb1− (φ∗cbφ )σ−1κcb−σ−1κcc1φ(φ∗ccφ )σ−1κcc1− (φ∗ccφ )σ−1κcc= (σ − 1) 1κcb(φ∗cbφ )σ−1κcb1− (φ∗cbφ )σ−1κcb− 1κcc(φ∗ccφ )σ−1κcc1− (φ∗ccφ )σ−1κcc .In the following, we denoteΩ ≡ 1κcb(φ∗cbφ )σ−1κcb1− (φ∗cbφ )σ−1κcb− 1κcc(φ∗ccφ )σ−1κcc1− (φ∗ccφ )σ−1κcc,and consider the three cases as below.Case 1: κcb = κcc = κ. In this case,Ω =1κ (φ∗cbφ )σ−1κ1− (φ∗cbφ )σ−1κ−(φ∗ccφ )σ−1κ1− (φ∗ccφ )σ−1κ .By φ ≥ φ∗cb > φ∗cc, we have 1 ≥φ∗cbφ >φ∗ccφ . Therefore,(φ∗cbφ )σ−1κ1− (φ∗cbφ )σ−1κ>(φ∗ccφ )σ−1κ1− (φ∗ccφ )σ−1κ,which implies that Ω > 0. Thus, we obtain ∂ ln γ(φ)∂ ln(φ) > 0.Case 2: κcb > κcc. We first show the following two inequalities (B.2) and (B.3).1κcb(φ∗cbφ )σ−1κcb1− (φ∗cbφ )σ−1κcb>1κcc(φ∗cbφ )σ−1κcc1− (φ∗cbφ )σ−1κcc, (B.2)and1κcc(φ∗cbφ )σ−1κcc1− (φ∗cbφ )σ−1κcc>1κcc(φ∗ccφ )σ−1κcc1− (φ∗ccφ )σ−1κcc. (B.3)102B.5. Proof of Proposition 1Then using (B.2) and (B.3), we obtain Ω > 0, and thus ∂ ln γ(φ)∂ ln(φ) > 0.To show (B.3), note that φ ≥ φ∗cb > φ∗cc, and thus 1 ≥φ∗cbφ >φ∗ccφ . Therefore1κcc(φ∗cbφ )σ−1κcc1− (φ∗cbφ )σ−1κcc>1κcc(φ∗ccφ )σ−1κcc1− (φ∗ccφ )σ−1κcc.This implies (B.3). To show (B.2), note that 1κxσ−1κ1−xσ−1κis an increasing function of κ for any givenx ∈ (0, 1). Then (B.2) holds because κcb > κcc and 1 ≥ φ∗cbφ > 0.Combing Case 1 and Case 2, we obtain the desired result of part (a).Case 3: κcb < κcc. In this case, notice that when φ→ φ∗cb,(φ∗cbφ )σ−1κcb1− (φ∗cbφ )σ−1κcb→∞, and 0 <(φ∗ccφ )σ−1κcc1− (φ∗ccφ )σ−1κcc< +∞.Thus Ω > 0. In addition, we know that both(φ∗cbφ)σ−1κcb1−(φ∗cbφ)σ−1κcband(φ∗ccφ)σ−1κcc1−(φ∗ccφ)σ−1κccare decreasing with respectto φ. We can prove that there is an unique φ∗, such that 1κcb(φ∗cbφ∗ )σ−1κcb1−(φ∗cbφ∗ )σ−1κcb= 1κcc(φ∗ccφ∗ )σ−1κcc1−(φ∗ccφ∗ )σ−1κcc, Ω > 0 forφ ∈ (φ∗cb, φ∗), and Ω < 0 for φ ∈ (φ∗,+∞). Thus we obtain desired result of part (b). 103Appendix CAppendix for Chapter 4104Appendix C. Appendix for Chapter 4Table C.1: Description of treaty arrangementsTreaty port Treaty concession Leased territoryNumber of cities 55 12 4Number of recipients 6 14 5Time frame 1842–1924 1845–1902 1898–1914Expatriate presence andcommercial freedomMerchants were al-lowed to trade withanyone and rent res-idences from localpeople or build theirown houses in “settle-ments” granted by thegovernment. Nominalland rents paid to thegovernment.∗Concessions are usuallyseveral square kilome-ters large. Governmentfrom recipient countriesrented the area fromChina and then subletparts of the land toforeign merchants ormissionaries.Merchantsfrom countries otherthan the recipient ofthe concession were alsowelcomed to operate inthe concession.†They could be severalhundred square kilome-ters large and usuallycontained adjacent wa-ter areas. Military oc-cupation without rentor duties.Governance:Commercial law, taxes,and dutiesTariffs were set togetherby the government andforeign recipients. Lo-cal Chinese officialsowned the right tocollect duties and exe-cute law enforcement.Consuls from recipi-ents could intervene ifforeign residents wereinvolved in criminalactivities and disputes.Municipal councils andboards of directors con-sisting of consuls andelected merchants fromthe recipient countrywere in charge of lo-cal administration (fis-cal regulation, taxation,policing, infrastructure,etc.)Military occupation andfull governance (civillaw, taxation, police,etc.) Recipients ofleased territory possessjudicial authority overall cases no matter thecitizenship of the de-fendant, whereas Chi-nese could be tried un-der Chinese law in con-cessions.Commercial and resi-dential property devel-opmentForeigners were allowedto rent or build res-idences, consulates,business offices, banks,churches, warehouses,schools and hospitals,renovate road, port andother infrastructurewithin the grantedarea.Same as Treat Ports. Recipients were allowedto build railways andextract natural re-sources in addition tothose rights obtainedby Treaty Ports.∗: Foreigners were not allowed to stay overnight in cities without treaty ports.†: For example, more than 1300 Japanese citizen (working for 20 Japanese firms) lived in the Britishconcession in Hankou in 1905. (p. 80 Zhang1993)105Appendix C. Appendix for Chapter 4Table C.2: Treaty linkagesRC Port Year RC Port Year RC Concession YearUK Shanghai 1842 Japan Suzhou 1896 U.S. Xiamenp 1902Guangzhou 1842 Hangzhou 1896 France Shanghai 1849Ningbo 1843 Jingzhou 1896 Guangzhou 1861Fuzhou 1844 Dandong 1903 Tianjin 1861Xiamen 1844 Hulunbeier 1905 Hankou 1896Shantou 1860 Shenyang 1905 Xiamen 1902Tianjin 1861 Dandong 1905 Japan Hangzhou 1896Yingkou 1861 Liaoyang 1905 Suzhou 1897Zhenjiang 1861 Tieling 1905 Tianjin 1898Jiujiang 1861 Changchun 1905 Hankou 1898Yantai 1861 Jilin 1905 Jingzhou 1898Wuhan 1861 Yanbian 1905 Xiamen 1899Wenzhou 1876 Haerbin 1905 Fuzhou 1899Wuhu 1876 Qiqihaer 1905 Chongqing 1901Yichang 1876 Mudanjiang 1905 Xiamen 1902Beihai 1876 Heihe 1905 Germany Tianjin 1895Chongqing 1890 Yanbian 1909 Hankou 1895Rikaze 1893 Germany Qingdao 1898 Xiamen 1902Foshan 1897 Russia Tacheng 1851 Russia Hankou 1896Wuzhou 1897 Kashi 1860 Tianjin 1900Baoshan 1897 Shizuishan 1881 Denmark Xiamenp 1902Weihai 1898 Wulumuqi 1881 Austria Xiamenp 1902Changsha 1902 Tulufan 1881 Italy Xiamenp 1902Jiangmen 1902 Hami 1881 Norway Xiamenp 1902Rikaze 1906 Changji 1881 Belgium Xiamenp 1902Ali 1906 Dalian 1898 Sweden Xiamenp 1902U.S. Shenyang 1903 Netherland Xiamenp 1902Dandong 1903 Concession Spain Xiamenp 1902France Shantou 1860 UK Shanghaip 1845Tianjin 1861 Xiamen 1852 Leased TerritoryYantai 1861 Guangzhou 1861 UK Weihai 1898Hainan 1876 Tianjin 1861 HongKong 1898Chongzuo 1887 Hankou 1861 France Zhanjiang 1898Honghe 1887 Jiujiang 1861 Japan Dalian 1905Qiannan 1895 Zhenjiang 1861 Qingdao 1914Hekou 1895 Xiamenp 1902 Germany Qingdao 1898Nanjing 1899 U.S. Shanghaip 1848 Russia Dalian 1898Zhanjiang 1899 Tianjin 1861RC stands for recipient country. pindicates public concession jointly owned by several recipientcountries. If one city appears many times in the table above, a different treaty port under that citywas opened each time it shows up. All the city names displayed are their current names.106Appendix C. Appendix for Chapter 4Table C.3: Probit regression predictions for countriesCountry Predicted Prob TrueR/PlaceboUnited States 0.9999 TrueJapan 0.9965 TrueGermany 0.9749 TrueUnited Kingdom 0.9599 TrueFrance 0.9516 TrueItaly 0.9124 TrueCanada 0.8308 PlaceboSpain 0.7491 TrueNetherlands 0.6877 TrueAustralia 0.6728 PlaceboSwitzerland 0.5803 PlaceboNorway 0.5056 TrueKorea 0.5016 PlaceboSweden 0.4835 TrueBelgium 0.4505 TrueDenmark 0.3953 TrueAustria 0.3914 TrueMexico 0.3095 PlaceboIreland 0.2756 PlaceboFinland 0.2270 PlaceboBrazil 0.1742 PlaceboGreece 0.1147 PlaceboRussia 0.1036 TrueSaudi Arabia 0.0930 PlaceboSingapore 0.0797 PlaceboPortugal 0.0790 PlaceboIsrael 0.0572 PlaceboUnited Arab Emirates 0.0540 PlaceboNote: The second column presents predicted probabilityof being a Recipient country. The last column indicateswhether the country is truly recipient country or placeboone.107


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items