Electronic Theses and Dissertations UC San Diego

Transcription

1 Electronic Theses and Dissertations UC San Diego Peer Reviewed Title: Trade and institutions Author: Mahakitsiri, Doungdao Acceptance Date: 2012 Series: UC San Diego Electronic Theses and Dissertations Degree: Ph. D., UC San Diego Permalink: Local Identifier: b Abstract: The first chapter of my dissertation analyses the relationship between trade and labor market outcomes. In particular, I provide evidence on one mechanism through which trade can affect wages via labor shortage. The second chapter studies the role of various institutions as a source of comparative advantage. I show empirically that each type of institutions recently advanced does not uniquely determine independent patterns of specialization and trade. The last chapter examines how free trade agreement and customs union impact trade at the extensive and intensive margins Copyright Information: All rights reserved unless otherwise indicated. Contact the author or original publisher for any necessary permissions. escholarship is not the copyright owner for deposited works. Learn more at escholarship provides open access, scholarly publishing services to the University of California and delivers a dynamic research platform to scholars worldwide.

2 UNIVERSITY OF CALIFORNIA, SAN DIEGO Trade and Institutions A dissertation submitted in partial satisfaction of the requirements for the degree Doctor of Philosophy in Economics by Doungdao Mahakitsiri Committee in charge: Professor Marc-Andreas Muendler, Chair Professor Thomas Baranga Professor Gordon Hanson Professor James E. Rauch Professor Krislert Samphantharak 2012

3 Copyright Doungdao Mahakitsiri, 2012 All rights reserved.

4 The dissertation of Doungdao Mahakitsiri is approved, and it is acceptable in quality and form for publication on microfilm and electronically: Chair University of California, San Diego 2012 iii

5 DEDICATION To my beloved parents, Prateep and Cholticha Mahakitsiri iv

6 TABLE OF CONTENTS Signature Page Dedication Table of Contents iii iv v List of Figures vii List of Tables Acknowledgements Vita Abstract of the Dissertation viii ix x xi Chapter 1 Trade, Wage Premia and Labor Shortages Introduction Data and variable description Sample and sources Thailand: trade policy and labor market Firm-Level Analysis Wage Premia Tariff Cuts and Wages Tariff Cuts and Vacancies to Employment Tariff Cuts, Vacancies to Employment and Wages Worker-level analysis Mincer wage regression with firm controls Mincer wage regression by tenure Mincer wage regression with skill mismatch Conclusion Chapter 2 Institutional Quality, Sectors Institutional Dependence, and Comparative Advantage Introduction Data and empirical specification Country-sector data Country-level data Sector-level data Empirical Specification Examining the Raw Data Estimation Results v

7 2.3 Robustness and Sensitivity Analysis Conclusions Chapter 3 The Effects of Free Trade Agreements and Customs Unions on the Extensive and Intensive Margins of Trade Introduction Gravity model and trade agreements Extensive and Intensive margins Free Trade Agreements and Customs Unions Data and empirical results Conclusion vi

8 LIST OF FIGURES Figure 2.1: Judicial quality and financial development Figure 2.2: Judicial quality and labor market flexibitliy Figure 2.3: Judicial quality and IP protection Figure 3.1: No. of PTAs. (Source: WTO) vii

9 LIST OF TABLES Table 1.1: Wages, Importers and Exporters Table 1.2: Wages by Occupation, Importers and Exporters Table 1.3: Characteristics, Importers and Exporters Table 1.4: Wages and Tariffs Table 1.5: Wages by occupational groups Table 1.6: Employment by occupational groups Table 1.7: Vacancies to Employment (V/E) and Tariffs Table 1.8: Wages, Vacancies to Employment (V/E) and Tariffs Table 1.9: Mincer regression for log worker hourly wages Table 1.10: Mincer regression for log worker hourly wages by tenure Table 1.11: Mincer regression for log worker hourly wages with skill mismatch Table 1.12: Descriptive Statistics Table 2.1: Raw data analysis Table 2.2: Baseline specification and joint institutional determinants Table 2.3: Separate Institutions on Alternative Samples Table 2.4: Joint specification on alternative samples Table 2.5: Pairwise Correlation of Country Characteristics Table 2.6: Alternative Industry Interactions Table 2.7: Contract intensity with alternative institutions and endowments 57 Table 2.8: Financial intensity with alternative institutions and endowments 58 Table 2.9: Sector volatility with alternative institutions and endowments. 59 Table 2.10: Innovation intensity with alternative institutions and endowments 60 Table 3.1: Data for year Table 3.2: Data for year Table 3.3: Data for year Table 3.4: Data for year Table 3.5: Data for years viii

10 ACKNOWLEDGEMENTS Iwishtothankmycommitteemembersfortheirvaluableadviceandprecious time. My utmost gratitude to Dr. Marc-Andreas Muendler, my committee chair, for his support, guidance, encouragement, and patience throughout the entire process. This dissertation would not have been possible without his help. Thank you God for answering my prayers and finally making it happen. ix

11 V 2003 B. S. in Economics Second Honour, Thammasat University, Bangkok, Thailand 2004 Master of Science in Policy Economics, University of Illinois, Urbana Champaign 2012 Ph. D. in Economics, University of California, San Diego x

12 ABSTRACT OF THE DISSERTATION Trade and Institutions by Doungdao Mahakitsiri Doctor of Philosophy in Economics University of California, San Diego, 2012 Professor Marc-Andreas Muendler, Chair The first chapter of my dissertation analyses the relationship between trade and labor market outcomes. In particular, I provide evidence on one mechanism through which trade can affect wages via labor shortage. The second chapter studies the role of various institutions as a source of comparative advantage. I show empirically that each type of institutions recently advanced does not uniquely determine independent patterns of specialization and trade. The last chapter examines how free trade agreement and customs union impact trade at the extensive and intensive margins. xi

13 Chapter 1 Trade, Wage Premia and Labor Shortages Recent trade theories with heterogeneous firms build upon labor market frictions and search to generate revenue sharing between firms and their employees and workforce adjustments following trade liberalization. However, little empirical attention has been paid to potential labor shortages. Using firm-level vacancy data from Thailand s manufacturing sector for , I construct the ratios of vacancies to employment to measure the extent of labor shortages across firms. I find that a cut in input tariffs raises not only wages at firms that use imported intermediates, but also their vacancies to employment ratios relative to firms that only source inputs locally. Importantly, firms with high vacancies to employment ratios pay higher wages, and even more so for importers of intermediate inputs. This evidence is consistent with the hypothesis that labor shortages constitute one empirically documentable mechanism by which importing firms pay higher wages: they search more intensively for workers, suffer more from hiring constraints, and hence increase their wage offers to raise adequately skilled employment following input tariff cuts. 1

14 2 1.1 Introduction Given concern over rising wage inequality observed in both developed and developing countries 1, the relationship between globalization and labor market outcomes has gained particular attention from both policy makers and academic researchers. While earlier literature has primarily focused on how international trade affects wages of workers with different levels of skill (between-group) 2, recent theoretical and empirical papers explore how trade impacts wages of workers with similar observed characteristics employed in heterogeneous firms (within-group). To explain the latter, various forms of labor market frictions that feature firmworkers rent sharing have been incorporated into the Melitz (2003) model. Hence, wages vary across firms because: different firms earn different profits and workers demand fair wages (Amiti and Davis (2011) and Egger and Kreickemeier (2009)); workers can shirk on the job and firms differ in their monitoring costs such that efficiency wages paid to deter shirking vary from firm to firm (Davis and Harrigan (2011)); or search and matching frictions, heterogeneity in worker ability, and generalized Nash-Bargaining over the surplus (Helpman, Itskhoki and Redding (2010)). So far, however, existing data sources have not allowed researchers to investigate the firm-level adjustment process and optimization under unemployment and potential skill mismatches directly. There is, however, little empirical documentation of the underlying mechanism. In this paper, I provide evidence on one specific underlying mechanism, labor shortage, through which trade can differentially affect wages of globalized firms relative to domestically-oriented firms. How this potential channel operates is consistent with the ideas behind existing theoretical models that combine heterogeneous firms and search frictions. to Melitz (2003), the opening of trade will induce within industry reallocations as more productive exporting firms, and arguably importing firms 3, expand and 1 See the survey by Goldberg and Pavcnik (2007). 2 See, for example, Lawrence and Slaughter (1993), Leamer (1996), and Feenstra and Hanson (1999, 2008). 3 Exporting and importing firms are larger, more productive, and pay higher wages, as found empirically by Bernard, Jensen, Redding and Schott (2007). Due

15 3 domestic firms shrink or exit 4. Yet, the expansion process is not immediate. It involves hiring new workers, which is costly and time consuming, and depends significantly on labor market conditions (Fajgelbaum (2011) 5 ). So as firms that wish to expand search more intensively for workers, competition in the labor market intensifies such that matching may remain incomplete and opening vacancies cannot be filled 6. It follows that the opening of trade raises the tightness of the labor market (as in Felbermayr, Prat and Schmerer (2011) and Helpman, Itskhoki and Redding (2010)), which in turn increases firms search costs 7 and hence wage payments necessary to induce the flow of people into job vacancies. To examine the empirical relevance of the labor shortage channel, I draw on arichfirm-leveldatasetfromthethailandproductivityandinvestmentclimate Survey (PICS) for the period 2003 to During this time, the Thai government unilaterally restructured its tariff regime to reverse an early policy of trade restrictions following the 1997 financial crisis. This led to tariff cuts in over digit HS tariff lines. A distinctive feature of the PICS dataset is that the survey s questionnaire asks about vacancies at each firm at year-end, making it possible to construct afirm-specificratioofthestockofvacanciestocurrentemployment( vacacy to employment ratio ). The PICS dataset also offers a small-scale link employeremployee data for a subset of 426 firms from years 2003 and , whereby 10 randomly sampled workers from each firm are interviewed face-to-face on a long 4 The selection effect of trade receives a vast empirical support, see, for example, Helpman (2006) and Bernard et al. (2007). 5 Fajgelbaum (2011) studied the the effect of capacity constraining labor market frictions on firm employment growth and export decisions. 6 The rate at which firms fill their vacancies is a decreasing function of labor market tightness in the theoretical model of Felbermayr, Prat and Schmerer (2011). 7 Following the standard Diamond-Mortensen-Pissarides approach, search cost (b) inthetheoretical model of Helpman, Itskhoki and Redding (2010) is assumed to be increasing in labor market tightness (x), whereby b = α 0 x α1. As a solution to the bargaining game, each worker receives a fraction of firm revenue per worker w(θ) = βγ r(θ), and due to firm s first-order conditions for the measure of workers sampled n(θ) = 1+βγ h(θ) βγ 1+βγ r(θ) h(θ) b 1, thus wage is in turn a function of search cost w(θ) =bn(θ) that is increasing in the tightness of the labor market. 8 Matched employer-employee datasets are more readily available for advanced countries, such as Denmark (Munch and Skaksen (2008)), Germany (Klein, Moser and Urban (2010) and Baumgarten (2010)), Sweden (Davidson, Heyman, Matusz, Sjöholm and Zhu (2011)), and for some developing countries, such as Brazil (Menezes-Filho, Muendler and Ramey (2008), Krishna, Poole and Senses (2011), and Helpman, Itskhoki, Muendler and Redding (2011)) and Mexico (Frías, Kaplan and Verhoogen (2009)).

16 4 list of socio-economic characteristics 9. This information enables me to conduct the analysis at the individual worker-level, in addition to the firm-level, to explicitly control for worker characteristics that could also affect wage outcomes. While theories emphasize the role played by exporting firms, the large and growing importance of trade in intermediates is now widely recognized. Goldberg, Khandelwal, Pavcnik and Topalova (2010) show that lower import tariffs leadindian firms that demand imported inputs to expand by launching new product varieties. Following Amiti and Davis (2011), I consider the separate role of intermediate input tariffs from that of final product tariffs, and discern importers from exporters to capture both cost and demand shocks to firms. The analysis at the firm level confirms previous study findings of wage premia at import-using firms, who benefit from lower input tariffs through cheaper and more accessible intermediate imports. Furthermore, the differential positive effects of a cut in input tariffs on wages between importers and non-importers are present for both skilled and unskilled workers (proxied by production and non-production workers respectively). Analogous to the wage responses, I find evidence that importing firms expand employment across all skill groups, suggesting that the results are not driven just by changes in skill composition (as in Verhoogen (2008)). Going beyond the wage analysis, I examine how the same tariff cuts affect labor shortage at the firm level. I find that import-using firms, who appear to try and expand their workforce, suffer more from unfilled vacancies relative to firms that do not import following input tariff cuts. This particular result parallels Holzner and Larch (2011) s model prediction that With trade the number of vacancies per firm is lower than in autarky for low-productivity firms, but higher for high productivity firms, because additional revenues from import and export allow high productivity firms to create more vacancies so as to increase their labor input in response to additional demand from abroad. More importantly, firms with hiring constraints, in particular importers, do indeed pay higher wages. This evidence is consistent with the hypothesis that labor shortage is an important 9 Other than general demographic questions, e.g., age, sex, education, past and current work related information, the PICS dataset also offers many more interesting variables, such as computer usage, bank account and credit card ownership, and online purchase transaction.

17 5 mechanism for higher wage premia at import-using firms, who search more intensively for workers, suffer more from hiring constraints, and hence offer higher wages arguably to attract additional employment. To test whether the documented labor shortage channel is robust to controlling for observed worker characteristics, I perform a Mincer regression of log worker wages using the small-scale linked employer-employee data from PICS that are available for a subsample of firms. At the worker level, I find consistent result of higher wage payments for the 10 randomly sampled workers employed by importing firms, which experience hiring constraints, relative to workers employed by firms that do not import inputs. Moreover, I test if there is a differential wage effect between newly hired and incumbent workers, hypothesizing that newly hired workers are better able to take advantage of a tighter labor market. By exploiting detailed information on worker tenure at the firm, I show that the importer wage premium is significant throughout and somewhat stronger for newly hired compared to incumbent workers. An intuitive and consistent explanation of this result may be that not only do incoming workers bargain more strongly, but existing workers ask for a wage renegotiation and command a pay raise possibly by threatening to leave. In addition, individual wage with controls for skill mismatches shows that the increases prevalence of vacancies is likely to be accompanied by a skill mismatch, i.e., the hiring of an under qualified worker into a skill-intensive position. Hence, an alternative interpretation for the evident wage premium is an excess wage paid conditional on what the worker would earn otherwise based on his qualification in a market environment without search frictions and mismatches. This study falls into a line of research that focuses on the firm component of wages. Recent empirical papers using matched employer-employee data provide evidence on the significance contribution of both wage premia and unobserved worker heterogeneity toward the overall firm wage component (see for example Schank, Schnabel and Wagner (2007), Munch and Skaksen (2008), Frías, Kaplan and Verhoogen (2009), Davidson, Heyman, Matusz, Sjöholm and Zhu (2011), Krishna, Poole and Senses (2011), and Baumgarten (2010)). Helpman, Itskhoki, Muendler and Redding (2011) develop and estimate a structural model that accounts for the

18 6 link between trade and wage dispersion between firms within sector-occupations, allowing for the firm effect to change over time and vary with export status. Unlike earlier studies that estimate the determinants of wage differences across firms, this paper seeks to identify the transmission mechanism of tariff shocks that give rise to wage premia, to understand and to provide evidence on the firm level adjustment of labor, and to explain why globalized firms pay higher wages to workers with similar observables compared to domestic firms following tariff cuts. The remainder of the paper is organized as follows. Section 1.2 describes the data and discusses Thai s trade policy and labor market. Section 1.3 presents the estimation strategy and results from the firm-level analyses of tariff cuts on wages, employment, and vacancies to employment ratio. In Section 1.4, I test for the wage responses at the individual worker-level, compare marginal wages to average wages, and examine the extent to which the quality of worker-job assignment, i.e., skill-mismatch influences wage outcomes. Section 1.5 concludes. 1.2 Data and variable description Sample and sources This paper uses firm level data from Thailand s manufacturing sector. The data come from the 2004 and 2007 rounds of the National Productivity and Investment Climate Survey (PICS), conducted by the Foundation of the Thailand Productivity Institute (FTPI) in collaboration with the World Bank and Thailand s Ministry of Industry (MOI). The survey samples the universe of registered firms following a stratified random sampling methodology under 3 criteria: sector, location, and size. In order for the World Bank to keep comparability across countries for which similar surveys are conducted 10, two manufacturing industries are selected in all medium and large economies. For this reason, the two manufacturing industries automatically selected for Thailand, a lower middle-income country, are food processing and garments. The 2007 round of the PICS covers 1,043 randomly sampled Thai firms, 10 The World Banks Enterprise Surveys (ES):

19 7 spanning 34 ISIC 4-digit industries for the period between A distinctive feature of the Thai PICS dataset is the availability of firm level data on vacancies at year-end, allowing for direct observation of firm-specific labor shortage that is particularly important for testing how tariff cuts will lead to change in labor market tightness, which in turn influences wage outcomes. Another special feature of this dataset is that it offers detailed information on 10 randomly sampled workers from each firm, making it a small-scale linked employer employee dataset that allows me to take into account both firm and worker heterogeneity in analyzing wage determination and changes. Each worker is interviewed face-to-face on a variety of quantitative and qualitative questions. For example, variables of interest include hourly wages of workers (in Thai Baht), working hours, age, gender, education, occupation, labor market experience, and tenure at the firm. In addition, the qualitative questionnaire covers workers opinions about their jobs. One of the questions asks According to you, what is the most appropriate level of education for the work you are doing?, which I exploit in the construction of the skill-mismatch variables. A worker is categorized as over qualified ( under qualified ) if his highest level of educational attainment 11 is higher (lower) than the level required for his current position 12. This worker-level information is, however, available only for year To complement this, I turn to the first round of the survey in which worker-level information for year 2003 is available on the same 426 firms that are interviewed in both rounds. As a result, the individual worker level analysis with controls for both firm and worker heterogeneity is carried out using these firms for year 2003 and There are 6 educational categories, namely Degree, Por Wor Sor (vocational), Por Wor Shor (upper secondary), Lower secondary, Primary, and Non (illiterate). 12 For example, if a person with a Por Wor Shor (upper secondary education) answers that he actually needs a Por Wor Sor (vocational education) to perform his skilled production job, then a worker is identified as under qualified. 13 The firm-level analysis using average hourly wage, instead of average wage, as the dependent variable is based on the reduced sample as used in the individual worker level analysis.

20 Thailand: trade policy and labor market Thai trade policy Tariffs arethemainbarrierstotradeinthailand,especiallyformanufacturing sector which accounts for over three quarters of total exports. Raw materials and intermediates are Thailand s number one import category, taking up almost half of total annual imports (Source: Thailand Ministry of Commerce). During the period under consideration, the Thai government restructured its tariff regime, commencing in June 2003, which led to tariff cuts in over a third of the tariff lines (The WTO (2007)). The tariff reductions reversed an early policy of trade restrictions following the 1997 financial crisis, which prompted the Thai government to discourage demand for imported goods, to increase tariff revenues, and to meet budget surplus agreed to the International Monetary Fund (IMF) as part of the bailout package requirements. The tariff restructuring between involved more than digit HS tariff lines (Kohpaiboon (2007)). Under the so-called three-rate program: import duties were cut to 1% on raw materials (from an average of 7%), (ii) to 5% on intermediate products (from an average of 12.1%) and (iii) to 10% on finished products (from an average of 20.3%) (The WTO (2007)). Importantly, as these changes were influenced by foreign policy ambitions of a newly elected government under the leadership of prime minister Thaksin Shinawatra (Sally (2007)), most of the tariff changes were not a result of broad-based domestic lobbying by firms and might considered to be unanticipated policy changes from the perspective of firms and workers (The WTO (2007)). To construct product tariffs, I map SITC 4-digit Rev.3 tariffs into firm s product code at ISIC 4-digit, using a correspondence table developed by Affendy et al. (2010) 14. To construct input tariffs, I use cost shares from Thailand s Input-Output table available through OECD STAN 15 to compute the weighted average tariffs onintermediate inputs faced by Thai importers. 14 The proxy is done using SITC Rev.2-SITC Rev.3 correspondence table and SITC Rev.3-ISIC Rev.3 correspondence table. Mapping matches more than 98 percent of commodities under SITC Rev.3. For the remaining commodities in which industrial category cannot be matched directly, the identification is done based on the closest code. 15

21 9 Thai labor market Although a large portion of labor force is employed in agricultural sector, the labor force structure is changing as more workers shift into manufacturing and services sectors. Within manufacturing, Thai firms report that their operations are still hindered by three major constraints: a heavy regulatory burden, deficient infrastructure especially outside Bangkok, and shortage of laborers (The World Bank (2008)). The same survey reports that it takes firms 5.2 weeks to find a suitable skilled production worker and 2.2 weeks to find an unskilled production worker. Out of 66 countries for which similar data are available, Thailand is relatively closer to the average country, which takes 3.8 weeks to fill skilled and 2 weeks to fill unskilled positions, compared to Indonesia that takes 2 weeks to fill skilled and 1 week to fill unskilled positions. So it is worth keeping in mind that Indonesia, arguably a flexible economy being analyzed in Amiti and Davis (2011), is quite distinct from Thailand in terms of labor market environment such that the origin of the wage premium could as well be differ. As an example, in the early 2006 Seagate, a US-based manufacturer of hard-disk drives, was reported to consider an investment in a new 40-billion Baht high-tech magnetic media facility in Thailand. In the end, Seagate chose to split the investment between two other locations: a substrate plant in Johor, Malaysia and an additional media facility in Singapore. Seagate cited the proximity of the two locations and its existing operation in Singapore as reasons for the choice, but many sources reported that limited labor availability in Thailand was a leading factor in this decision (Bangkok Post (2006)). 1.3 Firm-Level Analysis Wage Premia I begin by inspecting how wages vary by type of firm. Using domestically oriented firm as the omitted category, the result in column (1) of Table 1.1 shows that exporters that do not import and importers that do not export pay around

22 10 16 percent higher wages. Firms that both import and export, accounting about one third of the sample 16 and presumably the most productive group of firms, pay 32 percent higher wages. Additional controls for skill share (the ratio of nonproduction workers to total employment) and employment size in columns (2)-(4) leave wage premia at importing and exporting firms roughly unaffected. These wage differentials across firms with different levels of exposure to trade persist when the analyses are performed separately for production and non-production wages in Table 1.2. In addition, I find similar premia for employment, value added, total factor productivity, and capital stock at globalized firms as reported in Table 1.3, suggesting that importers and exporters are on average larger, more efficient, and more capital intensive compared to domestic firms. This evidence is similar to that from many other countries Tariff Cuts and Wages Iproceedbyexaminingtheimpactofinputandoutputtariff reductions on average wages at the firm level and by testing whether there are heterogeneous responses across firms with different import and export participation. The following equation is estimated using OLS with industry fixed effects and year fixed effects or interacted industry-year fixed effects, to control for unobserved industry-level heterogeneity, time variations, and industry specifics that might vary overtime. The standard errors are clustered at the industry level. ln (wage) i,j,t = α j + α t + α j,t + β 1 output tariff j,t + β 2 output tariff j,t EX i,j (1) + β 3 input tariff j,t + β 4 input tariff j,t IM i,j +EX i,j +IM i,j + Z i,j,t Γ + ε i,j,t 16 Although firms that export and import generally account for a small fraction of all firms in a standard firm-level dataset, a relatively high percentage of firms that do both activities in the PICS sample could be partly due to the dropping of firms with missing data or zero values for employment and wages, which tend to be firms that are smaller in size and do not participate in the global market.

23 11 Thedependent variableisthelogarithm ofaveragewageoffirmi in industry j at time t. Producttariff and input tariff denote the level of protections in industry j in which firm i operates on final goods and intermediate inputs respectively. EX is an indicator variable equal to one if firm i is an exporter and zero otherwise. IM indicates whether firm i imports any of its intermediate inputs. The vector Z includes firm-level controls, such as firm s productivity, employment and capital stock. To estimate the impact of tariff cuts on wages, I allow for separate effects of changes in incentive to export final products and changes in costs of imported inputs. I interact product tariffs withanexportdummytocapturethedifferential effect of a reduction in product tariffs between exporters and non-exporters. Similarly, I interact input tariffs with an import dummy to capture the differential effect of reducing input tariffs betweenimportersandnon-importers. Theoriesof rent sharing suggest that a fall in product tariffs willreducewagesofanyfirm due to increased import competition (β 1 > 0), but potentially increase wages of exporting firms relative to domestic firms (β 2 < 0) if exporting activity is favored by heightened foreign competition i.e., the general equilibrium effect outweighs the loss in profits due to the direct effect. So the overall wage impact of a decline in product tariffs on exporters is equal to β 1 + β 2. Likewise, theories of rent sharing suggest that lower input costs will raise profits and hence wages at any firm (β 3 < 0), and increase wages at firms that rely on imported inputs particularly strongly (β 4 < 0), with the overall effect on importers equal to β 3 + β 4. Alongside the benchmark regression on full sample, I also report the estimation results using log average hourly wage, instead of log average wage, as the dependent variable. The use of this alternative definition of wages serves as a robustness check that variation in wages across firms is not just reflecting change in over-time hours from existing workers. This additional analysis is based on the subsample of firms from years 2003 and 2006 for which data on hours worked are available. In the first column of Table 1.4, I find that a decrease in product tariffs has no significant impact on wages of exporting firms and firms serving only the domestic market. A potential explanation for this insignificant effect is that

24 12 our baseline specification (1), following Amiti and Davis (2008), captures only the contemporaneous effect of tariff changes. However, it might take some time for surge in import and competition to take effect on local producers, and possibly even longer time required for an adjustment of the real exchange rate to restore trade balance so as to create new profit opportunities for exporters. Next, I test for the effects of reducing input tariffs on wages at import-using firms and non-importers in column (2). The negative and significant coefficient on the interaction term between input tariff and firm s import status indicates that a fall in input tariffs raises wages at import using firms relative to firms that only source inputs locally. This finding is in line with the positive effect of lower import tariffs onimportingfirmsinindiafoundbygoldberg,khandelwal,pavcnikand Topalova (2010). They show that importers benefit from lower prices and more accessible imported intermediates, which lead them to expand by launching new product varieties 17. The same conclusions hold when I include both input and product tariffs in column (3). That is, I find a significant effect only for the interaction term between input tariff and an import dummy, suggesting that importing firms that experience a cut in their input costs pay higher wages relative to those that do not import inputs. All regressions include controls for firm productivity (TFP), capital stock and size. More productive firms and more capital-intensive firms pay higher wages. Conditional on productivity, larger firms pay lower wages, consistent with the later interpretation that firms with labor shortages pay wage premia (large unconstrained firms pay discount). The empirics in columns (5) and (6) from using log average hourly wage to account for possible changes in hours worked, which might also explain the negative relationship between input tariffs and importer interaction term and wages, confirm the results in the first part of the Table 1.4 using log average wage. These results are also robust to introducing industry-year fixed effects (columns (4) and (6)). In Table 1.4, I rerun equation (1) separately for production and non- production wages in order to test whether the differential impacts of input tariffs are 17 Goldberg, Khandelwal, Pavcnik and Topalova (2010) also find that lower input tariffsimprove firm performance in other dimensions, including output, TFP, and research and development (R&D) activity.

25 13 present for both types of workers. An increase in wages at importing firms seems to be largely shared across production and non-production, but non-production workers command somewhat stronger increase in their premium. These results show that wage premia at importers happen at all skill levels. As theory predicts that the selection effect of trade will trigger expansion at high productivity firms, I next test whether there is an employment growth that matches up with the pattern of wages, especially at importing firms where wage premia are evident. The results in Table 1.6 suggest that importing firms expand their employment opportunities by hiring both production and non-production workers. Thus, access to cheaper foreign inputs leads firms to expand by demanding more workers, and thereby potentially pushes up wages at importing firms following input tariff cuts Tariff Cuts and Vacancies to Employment Similar to other countries, I show in the previous section that firm level heterogeneity matters for how wages and employment respond to lower tariffs in Thailand. I now analyze how the same tariff cuts affect labor market conditions, particularly with respect to the costs of adjusting labor force across firms with different demands for workers. I repeat the wage analysis but replace the left hand side variable with firm s vacancies to employment ratio to investigate the extent to which employer suffers from labor shortage. I show in Table 1.7 that the result of vacancies to employment mirrors the results of wages and employment documented in the previous section. As before, I find no significant differential impact of reducing product tariffs on vacancies to employment between exporting and non-exporting firms. In column (2), the coefficient of the interaction term between vacancies to employment and importer status is significantly negative, consistent with the hypothesis that importers try to expand their workforce but experience labor shortages following a cut in input tariffs. Again, the same conclusions emerge when I include both product and input tariffs (column (3)) and introduce industry-year fixed effects (column (4)). This finding confirms Felbermayr, Prat and Schmerer

26 14 (2011) and Helpman, Itskhoki and Redding (2010) models predictions that the tightness of the labor market increases following the opening of trade. Interestingly, I find that the coefficient on the TFP term is negative and significant. This implies that highly productive firms typically face lower vacancies to employment ratios, consistent with their capacity to pay higher salaries out of their rents and to offer more generous compensation than low productivity ones. Therefore, this exercise shows that not only are wages affected in a heterogeneous way by tariff cuts, but so are firm-specific vacancies Tariff Cuts, Vacancies to Employment and Wages So far, I have documented that lower input tariffs lead to higher wages at import-using firms relative to firms that do not import inputs (section 3.2). I have also shown that these importers try to expand by increasing employment across all skill groups, but at the same time suffer more from unfilled vacancies relative to non-importers (section 3.3). So one possible way for importers to achieve their optimal size facing labor shortages is to raise their wage offers in order to induce the flow of workers into opening positions. To test this hypothesis, I reestimate equation (1) by including the vacancies to employment ratio and its interactions with import and export status as well as input and output tariff as additional explanatory variables. In doing so, I can test whether firms with higher vacancies to employment ratios, in particular importers where hiring constraints are severe, are also firms that are willing to pay higher wages to fill otherwise unfilled vacancies. The first two columns of Table 1.8 report the estimation results drawn on the full sample using average wages. The latter half shows the estimates when the same analysis is carried out using the reduced sample with average hourly wage as the dependent variable. A significantly positive coefficient estimate of the vacancies to employment ratio in column (1) suggests that firms with high vacancies to employment are also high wage firms. In the next column, I find that unfilled vacancies at exporters do not result in higher wage payments but unfilled vacancies at importers do. More specifically, a significantly positive coefficient on the interaction term between vacancies to employment and importer status suggests that im-

27 15 porting firms that have significantly higher incidence of labor shortages do indeed pay higher wages. These results are robust to using hourly wages (columns (3) and (4)) and interactive industry-year fixed effects (columns (2) and (4)). Therefore, consistent with the labor shortage mechanism, a joint examination of tariff cuts, vacancies to employment, and wages reveals that firms that find vacancies more costly, especially importers who try to expand their workforce but are restricted by a capacity constrained labor market, offer higher wages. 1.4 Worker-level analysis Mincer wage regression with firm controls Having provided some evidence on the empirical relevance of the labor shortage channel at the firm level, I now proceed onto the individual worker level to explicitly account for observable worker characteristics that could also affect vacancy and wage outcomes. I conduct the analysis using detailed information from the small-scale linked firm-workers data on a sub-sample of firms available for years 2003 and In Table 1.9, I perform a Mincer regression for log worker hourly wage to test whether the findings at the firm level still hold. I begin with the main specification (1) using the same set of firm controls plus worker observables. I find that the 10 randomly sampled workers employed at importing firms are paid higher hourly wages relative to workers employed at firms that rely only on domestic inputs. The estimates of the returns to worker characteristics are as expected. Highly educated and more experienced workers earn significantly higher hourly wages. Female workers are paid less relative to male workers, in line with the findings in other developing countries 18. Furthermore, there might be a stigma associated with past unemployment experience as suggested by a significantly negative coefficient on the ever unemployed variable. In addition to a full set of firm and worker characteristics, I include firm s vacancies to employment ratio into the Mincer log worker wage regression in col- 18 See The Global Gender Gap report, World Economic Forum for further detail.

28 16 umn (2). I find a significantly positive association between workers hourly wages and their firm s vacancies to employment ratio. This implies that workers employed at a firm that suffered more from labor shortage are being paid relatively higher hourly wages. A significantly positive coefficient on the vacancies to employment and importer status interaction term in the last column suggests that workers employed at an importing firm, which suffers more from hiring constraint, do receive significantly higher hourly wages compared to their peers with the same observable characteristics. These findings at the worker-level further confirm the importance of the labor shortage channel. A tightening of the labor market following tariff cuts pushes up wages at hiring-constrained firms so as to increase the attractiveness of and to fill the opening vacancies Mincer wage regression by tenure Inextillustratetheeffect of trade reform on hourly wages of workers by tenure at the firm. The idea is that unfilled vacancies should affect marginal wages for the last hires, not just average wages i.e., firm s most recently filled vacancies. I perform the analysis by restricting the sample to workers with tenure of one year or less and run the same mincer wage regression as in the previous section to measure the wage effect of marginal workers. I then repeat the same analysis drawing on the sample of workers with tenure of greater than one year to examine the effect on existing workers. Estimation results are reported in Table I find that the importer wage premium is significant throughout and somewhat stronger for newly employed compared to incumbent workers. Although the plain vacancies to employment term is not significant on its own, the interaction term between vacancies to employment and firm s import status is positive and significant, and again stronger for newly hired workers. A possible explanation for this is that new entrants have direct impact on wages at importing firms that severely suffered from hiring constraints following lower input tariffs. This is because, in a tightening labor market environment full of vacancies, potential workers know that their marginal product is quite high and hence bargain strongly. As a result, importing firms in need

29 17 of workers are forced to pay excess wages in order to increase the probability of filling their vacancies. Additionally, the significantly positive effect of the same interaction term for tenured workers suggests that they also ask for a wage renegotiation and command a pay raise possibly by threatening to leave. A higher wage offer attracts more workers from not only unemployed but also firms paying lower wages Fajgelbaum (2011). Therefore, this exercise shows that firms facing difficulties hiring pay higher wages to workers at all tenure levels, to keep incumbent workers from leaving for higher pay jobs as well as to attract new workers into vacant positions Mincer wage regression with skill mismatch In a tightening labor market environment in which expanding firms compete with one another for workers, firms that are desperate to fill vacancies may be willing to hire less qualified applicants to staff the positions rather than bearing the costs of unfilled vacancies, especially when the opportunity cost of not hiring is high. So in this last section, I explore how potential labor shortages affect the quality of matches, which in turn influences wage outcomes. In doing so, it can help strengthen our understanding of costly search: not only on what happen to wages when more productive firms try to expand on the extensive margin and there is an insufficient number of workers, but also what happen to the quality of matches when there is an inadequate quality of individuals available to firms. This is to control for the discrepancies between the qualities of workers which firms wish to employ for given vacancy characteristics and the actual job placements, and how that translate into wages. With the same firm and worker controls as before, I include two additional mismatch variables and their interactions with firm s vacancies to employment ratio on the right hand side of the mincer wage regression. An over-qualified and an under-qualified represent whether workers possess higher skills than they needed or lower skills than their jobs required as described in the data section. I find statistically significant effects on both variables in Table 1.11, however with opposite signs, negative for the over-qualified but positive for the under-qualified

30 18 workers. These results suggest that the over-qualified workers receive wage discounts because they are appointed to lower positions compared to their peers with the same skills such that their marginal products are lower. Conversely, if the under qualified workers fill more skill intensive jobs than their observable skills, then they command premiums compare to the same skill counterparts. The skill mismatch and firm s vacancy to employment interaction terms test whether or not the over (under) qualified workers who work at a hiring-constrained firm receive wage discounts (premiums). The interaction terms appear significant only at the under-qualified margin, implying that an alternative interpretation for the origin of the wage premia evident at importing firms might be as follows. In addition to paying higher wages because of labor shortages, firms may have also filled vacancies by hiring people who are actually under-qualified for jobs, which is usually unobserved otherwise to the researchers. As less qualified workers fill positions that have higher productivity than what they would have attained otherwise in a normal market environment, part of the wage premia documented, hence, is possibly an extra wage paid conditional on characteristics. The inclusion of the skill mismatches and their interactions with firm s vacancies to employment ratio slightly lowers the magnitude of the key coefficient of interest i.e., the interaction term between input tariff and import status. Therefore, in the presence of labor shortages, part of the differential positive effect of tariff cuts on wages found for importing firms could be explained by excess wages paid conditional on the qualifications of workers. 1.5 Conclusion Recent theories of firm heterogeneity and trade incorporate imperfect labor markets that feature firm-workers rent sharing to explain wage premia and differential wage responses of exporters, importers, and domestic firms following trade liberalization. There is, however, far less documentation of the mechanism involved. This paper sheds light on one particular channel, labor shortage, through which trade shocks are transmitted to wages. I show empirical support for this

31 19 mechanism with the firm-level vacancy data from Thailand s manufacturing sector which experienced unilateral home tariff reductions during Ifindthatlaborshortagematters,bothquantitativelyandqualitatively,in determining hiring and wage outcomes. A decline in input tariffs is associated with rising wages as well as vacancies to employment ratios at firms that use imported intermediates relative to firms that source inputs locally. Firms with high vacancies to employment ratios pay high wages, and in particular expanding importers of intermediate inputs pay even higher wages compared to non-importers. These findings provide a compelling empirical evidence in support of the insights from search theory emphasized in Helpman, Itskhoki and Redding (2010), Felbermayr, Prat and Schmerer (2011), and Holzner and Larch (2011). This paper s results also point to an additional gain from trade reform that could be realized in a capacity constraint-free labor market environment. In the presence of costly search, firms cannot immediately match with the needed skills, thereby restricting their capability to expand and inhibiting them from reaping the full benefits of lower input tariffs. Appropriate policy measures, such as educational and training policies, that would encourage more efficient matches and help reducing search costs should therefore be implemented in conjunction with liberalization policy.

32 20 Table 1.1: Wages, Importers and Exporters Dependent variable: log Wage (1) (2) (3) (4) Exporter without Imports 0.158*** 0.143*** 0.194*** 0.148*** (0.031) (0.031) (0.030) (0.030) Importer without Exports 0.167*** 0.145*** 0.181*** 0.138*** (0.043) (0.043) (0.041) (0.042) Exporter and Importer 0.324*** 0.296*** 0.350*** 0.290*** (0.026) (0.027) (0.025) (0.026) log Employment 0.029*** 0.054*** (0.009) (0.008) Skill share 1.764*** 1.443*** (0.097) (0.094) Year effects Yes Yes Yes Yes Industry effects Yes Yes Yes Yes Observations 3,700 3,700 3,700 3,700 R squared Notes: * significant at 10%; ** significant 5%;*** significant at 1%. Skill share is the ratio of non-production workers to total employment.

33 21 Table 1.2: Wages by Occupation, Importers and Exporters Dependent variable: log Wage Production Non-Production (1) (2) (3) (4) Exporter without Imports 0.163*** 0.133*** 0.251*** 0.235*** (0.037) (0.037) (0.042) (0.042) Importer without Exports 0.225*** 0.180*** 0.145** 0.121** (0.051) (0.051) (0.058) (0.058) Exporter and Importer 0.401*** 0.343*** 0.301*** 0.272*** (0.031) (0.032) (0.035) (0.036) log Employment 0.061*** 0.031*** (0.010) (0.012) Year effects Yes Yes Yes Yes Industry effects Yes Yes Yes Yes Observations 3,700 3,700 3,700 3,700 R squared Notes: * significant at 10%; ** significant 5%;*** significant at 1%.

34 22 Table 1.3: Characteristics, Importers and Exporters Dependent variable: log Labor log VA TFP log Capital (1) (2) (3) (4) Exporter without Imports 0.843*** 0.074* 0.369*** 1.379*** (0.050) (0.041) (0.044) (0.095) Importer without Exports 0.805*** 0.219*** 0.385*** 1.141*** (0.068) (0.057) (0.060) (0.129) Exporter and Importer 1.521*** 0.359*** 0.693*** 2.475*** (0.041) (0.034) (0.036) (0.078) Year effects Yes Yes Yes Yes Industry effects Yes Yes Yes Yes Observations 3,700 3,700 3,700 3,700 R squared Notes: * significant at 10%; ** significant 5%;*** significant at 1%. Total factor productivity (TFP) is computed using the superlative index number by Caves, Christensen and Diewert (1982).

35 23 Table 1.4: Wages and Tariffs Dependent variable: A. log Wage B. log Hourly Wage (1) (2) (3) (4) (5) (6) Output Tariff (0.317) (0.331) (0.525) Output Tariff*Exporter (0.152) (0.156) (0.158) (0.366) (0.364) Input Tariff (0.625) (0.660) (1.288) Input Tariff*Importer *** *** *** *** *** (0.128) (0.131) (0.132) (0.318) (0.319) Exporter 0.114*** 0.094*** 0.099*** 0.100*** (0.036) (0.021) (0.037) (0.037) (0.084) (0.085) Importer 0.057** 0.135*** 0.134*** 0.132*** 0.425*** 0.416*** (0.022) (0.036) (0.036) (0.037) (0.094) (0.095) TFP 0.402*** 0.402*** 0.402*** 0.403*** 0.343*** 0.337*** (0.014) (0.014) (0.014) (0.014) (0.035) (0.035) log Employment *** *** *** *** *** *** (0.012) (0.012) (0.012) (0.012) (0.031) (0.031) log Capital Stock 0.178*** 0.178*** 0.178*** 0.178*** 0.135*** 0.133*** (0.009) (0.009) (0.009) (0.009) (0.018) (0.018) Year effects Yes Yes Yes Yes Industry effects Yes Yes Yes Yes Industry-year effects Yes Yes Observations 3,682 3,682 3,682 3, Rsquared Source: A. Thai PICS data (full-sample), B. Thai PICS data (sub-sample), 2003&2006 Note: Standarderrorsareclusteredattheindustrylevel;*significantat10%;**significantat5%; *** significant at 1%.

36 24 Table 1.5: Wages by occupational groups Dependent variable: log Wage Production Non-Production (1) (2) (3) (4) Output Tariff (0.408) (0.533) Output Tariff*Exporter (0.202) (0.205) (0.264) (0.268) Input Tariff (0.884) (1.155) Input Tariff*Importer * * ** ** (0.176) (0.179) (0.230) (0.234) Exporter (0.047) (0.048) (0.062) (0.063) Importer 0.157*** 0.155*** 0.133** 0.132** (0.047) (0.047) (0.061) (0.062) TFP 0.405*** 0.405*** 0.238*** 0.239*** (0.013) (0.013) (0.017) (0.017) log Employment *** *** *** *** (0.013) (0.013) (0.017) (0.018) log Capital Stock 0.184*** 0.184*** 0.116*** 0.116*** (0.007) (0.007) (0.009) (0.009) Year effects Yes Yes Industry effects Yes Yes Industry-year effects Yes Yes Observations 3,682 3,682 3,682 3,682 R squared Source: Thai PICS data (full-sample), Firm-level observation. Note: Standard errors are clustered at the industry level; * significant at 10%; ** significant at 5%; *** significant at 1%.

37 25 Table 1.6: Employment by occupational groups Dependent variable: log Employment Production Non-Production (1) (2) (3) (4) Output Tariff (0.527) (0.470) Output Tariff*Exporter (0.261) (0.265) (0.233) (0.236) Input Tariff (1.141) (1.017) Input Tariff*Importer ** * ** ** (0.228) (0.231) (0.166) (0.168) Exporter 0.262*** 0.261*** 0.366*** 0.367*** (0.061) (0.062) (0.054) (0.055) Importer 0.364*** 0.363*** 0.335*** 0.335*** (0.060) (0.061) (0.053) (0.054) TFP 0.122*** 0.122*** 0.221*** 0.221*** (0.016) (0.017) (0.015) (0.015) log Capital Stock 0.257*** 0.257*** 0.234*** 0.234*** (0.007) (0.007) (0.007) (0.007) Year effects Yes Yes Industry effects Yes Yes Industry-year effects Yes Yes Observations 3,682 3,682 3,682 3,682 R squared Source: Thai PICS data (full-sample), Firm-level observation. Note: Standard errors are clustered at the industry level; * significant at 10%; ** significant at 5%; *** significant at 1%.

38 26 Table 1.7: Vacancies to Employment (V/E) and Tariffs Dependent variable: V/E (1) (2) (3) (4) Output Tariff (0.017) (0.016) Output Tariff*Exporter (0.008) (0.009) (0.009) Input Tariff (0.073) (0.070) Input Tariff*Importer ** ** ** (0.005) (0.005) (0.005) Exporter * (0.025) (0.008) (0.025) (0.025) Importer 0.015*** (0.005) (0.014) (0.015) (0.015) TFP *** *** *** *** (0.003) (0.003) (0.003) (0.003) log Employment *** ** *** ** (0.004) (0.004) (0.004) (0.004) log Capital Stock 0.007*** 0.007*** 0.007*** 0.007*** (0.002) (0.002) (0.002) (0.002) Year effects Yes Yes Industry effects Yes Yes Industry-year effects Yes Yes Observations 3,682 3,682 3,682 3,682 R squared Source: Thai PICS data (full-sample), Firm-level observation. Note: Standard errors are clustered at the industry level; * significant at 10%; ** significant at 5%; *** significant at 1%.

39 27 Table 1.8: Wages, Vacancies to Employment (V/E) and Tariffs Dependent variable: A. log Wage B. log Hourly Wage (1) (2) (3) (4) V/E 0.123** * (0.055) (0.193) (0.127) (0.379) V/E*Exporter (0.143) (0.353) V/E*Importer 0.491*** 0.796*** (0.167) (0.366) V/E*Output Tariff (0.664) (1.321) V/E*Input Tariff (0.576) (1.624) Output Tariff (0.363) (0.524) Output Tariff*Exporter (0.173) (0.177) (0.364) (0.364) Input Tariff (0.761) (1.289) Input Tariff*Importer ** ** *** *** (0.150) (0.152) (0.318) (0.320) Exporter 0.213*** 0.206*** (0.042) (0.044) (0.084) (0.089) Importer 0.183*** 0.213*** 0.420*** 0.496*** (0.040) (0.041) (0.094) (0.097) TFP 0.362*** 0.363*** 0.343*** 0.341*** (0.016) (0.016) (0.035) (0.035) log Employment *** *** *** *** (0.012) (0.012) (0.031) (0.031) log Capital Stock 0.178*** 0.178*** 0.135*** 0.133*** (0.009) (0.009) (0.018) (0.017) Year effects Yes Yes Industry effects Yes Yes Industry-year effects Yes Yes Observations 3,682 3, R squared Source: A. Thai PICS data (full-sample), B. Thai PICS data (sub-sample), 2003 and Firm-level observation. Note: Standard errors are clustered at the industry level; * significant at 10%; ** significant at 5%; *** significant at 1%.

40 28 Table 1.9: Mincer regression for log worker hourly wages Dependent variable: log Hourly Wage Worker-level (1) (2) (3) V/E 0.157*** 0.234* (0.051) (0.129) V/E*Exporter (0.115) V/E*Importer 0.424*** (0.115) V/E*Output Tariff (0.471) V/E*Input Tariff (0.502) Output Tariff*Exporter (0.106) (0.106) (0.106) Input Tariff*Importer ** ** ** (0.090) (0.090) (0.090) Exporter (0.025) (0.024) (0.024) Importer 0.106*** 0.098*** 0.092*** (0.026) (0.025) (0.025) TFP 0.044*** 0.043*** 0.044*** (0.007) (0.007) (0.007) log Employment (0.007) (0.007) (0.007) log Capital Stock 0.011*** 0.011*** 0.011*** (0.003) (0.003) (0.003) Schooling 0.091*** 0.091*** 0.091*** (0.002) (0.002) (0.002) Experience 0.035*** 0.035*** 0.035*** (0.002) (0.002) (0.002) Experience ** ** ** (0.000) (0.000) (0.000) Female *** *** *** (0.012) (0.012) (0.012) Ever unemployed *** *** *** (0.013) (0.013) (0.013) Occupation Yes Yes Yes Industry-year effects Yes Yes Yes Observations 7,128 7,128 7,128 R squared Source: Thai PICS data (sub-sample), 2003&2006. Worker-level obs. Note: Standard errors are clustered at the industry level; * significant at 10%;** significant at 5%; *** significant at 1%.

41 29 Table 1.10: Mincer regression for log worker hourly wages by tenure Dependent variable: log Wage Tenure <=1 (year) Tenure >1 (years) (1) (2) (3) (4) (5) (6) V/E *** 0.301** (0.126) (0.309) (0.055) (0.142) V/E*Exporter (0.300) (0.124) V/E*Importer 0.752** 0.377*** (0.310) (0.124) V/E*Output Tariff (0.488) (0.552) V/E*Input Tariff (0.552) (0.515) Output Tariff*Exporter (0.270) (0.271) (0.270) (0.115) (0.115) (0.115) Input Tariff*Importer * * * ** * * (0.236) (0.236) (0.236) (0.098) (0.098) (0.098) Exporter 0.159** 0.114* 0.112* (0.064) (0.061) (0.061) (0.028) (0.026) (0.026) Importer 0.141** 0.133** 0.132** 0.101*** 0.093*** 0.084*** (0.069) (0.066) (0.066) (0.029) (0.028) (0.027) TFP *** 0.048*** 0.048*** (0.017) (0.017) (0.017) (0.007) (0.007) (0.007) log Employment (0.019) (0.019) (0.019) (0.008) (0.008) (0.008) log Capital Stock *** 0.012*** 0.012*** (0.009) (0.009) (0.009) (0.004) (0.004) (0.004) Schooling 0.096*** 0.096*** 0.096*** 0.091*** 0.090*** 0.090*** (0.004) (0.004) (0.004) (0.002) (0.002) (0.002) Experience 0.032*** 0.032*** 0.032*** 0.035*** 0.035*** 0.035*** (0.007) (0.007) (0.007) (0.002) (0.002) (0.002) Experience ** ** ** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female *** *** *** (0.031) (0.031) (0.031) (0.013) (0.013) (0.013) Ever unemployed ** ** ** *** *** *** (0.031) (0.031) (0.032) (0.015) (0.015) (0.015) Occupation Yes Yes Yes Yes Yes Yes Industry-year effects Yes Yes Yes Yes Yes Yes Observations ,168 6,168 6,168 R squared Source: Thai PICS data (sub-sample) from years 2003 and Worker-level observation. Note: Standard errors are clustered at the industry level; * significant at 10%; ** significant at 5%; *** significant at 1%.

42 30 Table 1.11: Mincer regression for log worker hourly wages with skill mismatch Worker-level Dependent variable: log Hourly Wage (1) (2) (3) (4) Over-qualified *** *** (0.010) (0.010) V/E*Over-qualified (0.064) (0.066) Under-qualified 0.125*** 0.105*** (0.009) (0.009) V/E*Under-qualified 0.109* 0.123* (0.064) (0.065) V/E 0.234* 0.253** 0.305** 0.330** (0.129) (0.129) (0.129) (0.130) V/E*Exporter (0.115) (0.113) (0.114) (0.113) V/E*Importer 0.424*** 0.404*** 0.455*** 0.428*** (0.115) (0.113) (0.113) (0.112) V/E*Output Tariff (0.471) (0.464) (0.465) (0.460) V/E*Input Tariff (0.502) (0.494) (0.495) (0.490) Output Tariff*Exporter (0.106) (0.105) (0.105) (0.105) Input Tariff*Importer ** ** * * (0.090) (0.089) (0.089) (0.088) Exporter (0.024) (0.025) (0.025) (0.025) Importer 0.092*** 0.098*** 0.097*** 0.092*** (0.025) (0.026) (0.026) (0.026) TFP 0.044*** 0.044*** 0.042*** 0.043*** (0.007) (0.007) (0.007) (0.007) log Employment (0.007) (0.007) (0.007) (0.007) log Capital Stock 0.011*** 0.009*** 0.010*** 0.009*** (0.003) (0.003) (0.003) (0.003) Schooling 0.091*** 0.093*** 0.098*** 0.098*** (0.002) (0.002) (0.002) (0.002) Experience 0.035*** 0.033*** 0.034*** 0.032*** (0.002) (0.002) (0.002) (0.002) Experience ** * *** ** (0.000) (0.000) (0.000) (0.000) Female *** *** *** *** (0.012) (0.012) (0.012) (0.012) Ever unemployed *** *** *** *** (0.013) (0.013) (0.013) (0.013) Occupation Yes Yes Yes Yes Industry-year effects Yes Yes Yes Yes Observations 7,128 7,128 7,128 7,128 R squared Source: Thai PICS data (sub-sample) from years 2003 and Note: Standard errors are clustered at the industry level; * significant at 10%; ** significant at 5%; *** significant at 1%.

43 31 Table 1.12: Descriptive Statistics Mean A. Full Sample Firm Characteristics log Wage log Employment 4.71 log Capital Stock Vacancies per Employment Production Workers log Wage log Employment 4.53 Non-production Workers log Wage log Employment 2.63 No of Firms 925 No of Observations 3,700 B. Reduced Sample Firm Characteristics log Hourly Wage 3.54 log Employment 4.95 log Capital Stock Vacancies per Employment No of Firms 421 No of observations 842 Worker Characteristics log Hourly Wage 5.12 Years of Schooling 11 Years of Experiences 12 % of Female 58.2 % of Previously Unemployed 24.3 No of Workers 7,656 Source: A. Thai PICS data (full-sample), B. Thai PICS data (sub-sample), 2003 and Note: Production workers are skilled and unskilled blue collar. Non-production workers are professional and managerial, skilled and unskilled white collar.

44 Chapter 2 Institutional Quality, Sectors Institutional Dependence, and Comparative Advantage This paper examines the role of institutions as a new source of comparative advantage. In particular, I investigate the extent to which different pillars of institutions including contract enforcement, financial development, labor market flexibility, and intellectual property protection independently identify the pattern of specialization. I find support for the idea that countries with better institutional quality specialize in industries that are very dependent on institutional provisions. While jointly significant, I show empirically that each type of institutions does not uniquely determine independent sources of comparative advantage. The superiority of countries rule of law does not necessarily lead to specialization in only contract-intensive industries that use intensively inputs requiring relationship-specific investments but also: financially vulnerable sectors that depend more on external financing; high volatility industries that experience greater idiosyncratic shocks; and innovation intensive industries in which invention is crucial. Countries with better developed financial markets and more flexible labor laws capture larger world exports in contract-intensive industries, originally defined to be dependent only upon contracting institutions. 32

45 Introduction The prominent role of institutions has long been evident in the growth literature. Institutional differences are found to have a large influence on income per capita across countries, and even larger than the effects from geography or trade. However, it is only recently that institutional differences have been explicitly incorporated into the international trade model. A growing body of literature on institutions and trade points to the role of institutional environment as an important source of comparative advantage. The idea is that industries vary in the degrees of complexity such that some rely on institutions more than others. Also, countries differ in their institutional conditions and thus ability to provide the necessary services needed by different types of industries. Thereby, interacting countries institutional differences with sectors specific differences in institutional dependence could generate a new pattern of specialization and trade. The main prediction is hence countries with superior institutions specialize in industries that are very dependent on institutional provisions. So far, the literature has identified four different institutional framework as sources of comparative advantage, namely contract enforcement (Levchenko (2007), Nunn (2007), and Costinot (2009b)), financial market development (Manova (2008) and Beck (2003)), labor market flexibility (Cuñat and Melitz (2007)), and intellectual property (IP) protection (Vichyanond (2009)). Alternative to restricting attention to a particular type of institutions as standard in this literature, this paper takes into account all the institutional variables and conducts a joint analysis to verify their significance and compare their relative importance. More importantly, given that these institutions tend to be highly correlated, I examine whether earlier research has truly identified independent channels through which different pillars of institutions influence the pattern of trade. To do so, I perform a cross institutional-sector pair analysis by first interacting each country characteristic to all the industry characteristics and then the reverse, holding fix industry characteristic and varying it across various institutional features. The former tests whether countries that are superior in a given institution specialize in other nonoriginal pair industries, and the latter asks whether other country characteristics,

46 34 rather than the originally defined institutional-pair, cause countries to become specialize in a certain industry. In line with the empirical literature, I find support for the idea that a country s institutional quality matters for its production and trade. That is, countries with better institutional quality specialize in industries that are very dependent on institutional provisions. Although jointly significance, I observe that each institution does not play a separate role in determining the pattern of specialization. A comparative advantage in contract intensive industries that use intensively inputs requiring relationship-specific investments is not exclusive to a country s ability to enforce contracts. The superiority of countries rule of law also lead to specialization in financially vulnerable sectors that depend more on external financing, high volatility industries that experience greater idiosyncratic shocks, and innovation intensive industries in which invention is crucial. In testing for the institutional sources of comparative advantage, I use a factor content of trade approach developed by Romalis (2004) in which trade shares are explained by the interactions of factor intensity and relative factor abundance. He shows that countries capture larger US market shares in industries that use their abundant factors more intensively. Likewise, the mechanism through which institutions determine the structure of commodity trade can be captured in a similar vein via the interactions of countries institutional quality and sectors institutional dependence. When consider separately the effects of original institution-sector pairs, Ifindevidenceinsupportoftherolesplayedbycontractinginstitution,financial system, labor market regulation, and IP protection in influencing the pattern of specialization. When combining all the institution-sector characteristics in the same regression, I observe that they are jointly significance, however, their combined effects are still lower than those of the traditional H-O forces of comparative advantage. By varying institutional measures across industry measures, I find that countries with more developed financial markets specialize in not only financially intensive sectors but also specialize more strongly in contract-intensive industries. Countries with higher levels of IP protection specialize not only in innovation-

47 35 intensive industries, but again the effect is more pronounce when a measure of IP protection is interacted with an industry measure of contract intensity. By varying country characteristics and holding fix industry characteristic, I find that acountrywithbetter-developedfinancialmarketsandmore-flexiblelaborlaws captures larger shares of world exports in contract-intensive industries, whereas a countrys ability to enforce contract plays no significant role after other aspects of institutions are simultaneously accounted for. To empirically test for the relevance of institutional determinants of trade, earlier empirical work has constructed measures of industry-institutional intensity and interacted them with their respective institution to capture the benefits of country-industry matches. Levchenko (2007) and Nunn (2007) base their analysis on the incomplete contracts framework, assuming that some sectors require the use of customized inputs more than others. This, in turn, lead producers and suppliers into making relationship-specific investments. As a result, countries with better contract enforcements have less hold-up problems, thereby giving them comparative advantage in the production of goods that highly dependent on legal system to enforce contracts. Levchenko (2007) uses the Herfindahl index to proxy for the levels of industry-institutional dependence. His rationale is that industries that heavily rely on a few key inputs are more susceptible to the holdup problem. He finds that countries with superior institutional quality capture larger U.S. import shares in more institutionally intensive sectors. Nunn (2007) uses a newly constructed measure of input-relationship specificity, adopting Rauch (1999) classification to distinguish between generic inputs (those sold on an organized exchange in which substitutes are readily available on the open market) and inputs that require relationship-specific investments (whereby suppliers must make costly investments for final good producers). In line with Levchenko (2007), Nunn (2007) also finds that countries with better judicial quality export relatively more in industries for which relationship-specific investments are important. Manova (2008) emphasizes the role of financial market development, hypothesizing that the availability of outside capital is more important for industries in which necessary investments cannot be funded solely by internal cash flow.

48 36 As a consequence, entrepreneurs will find it easier to raise outside capital in industries that employ more tangible assets that can be used as a collateral. She shows that equity market liberalization increases exports disproportionately more in financially vulnerable sectors that use fewer collateralizable assets. Cuñat and Melitz (2007) focus their attention on labor market institutions. Their theoretical model, which links labor market regulations and sectors volatility, predicts that international differences in labor market flexibility affect how firms can adjust to idiosyncratic shocks. As a result, countries in which it is easier to hire workers in good times and lay them off in bad times when there is excess capacity should attract more industries that are subject to high variance shocks, and hence benefit most from being able to make frequent adjustment on their employment margins. They find evidence that countries with more flexible labor markets specialize in highly volatile sectors. Costinot (2009a) provides a unifying theory for explaining forces behind comparative advantage. He suggests that all of the aforementioned papers share the same fundamental in the log-supermodularity of countries institutional quality and sectors institutional dependence. This notion of log-supermodularity captures the idea that increasing one variable (the former) is relatively more important when the other variables are high (the latter). This paper, instead, looks at the relationship among different types of institutions. The survey by La Porta et al. (2008) suggests that diverse areas of laws and regulations vary systematically with legal traditions or origins. Common law origin countries are found to have better contract enforcemenr and greater security of property rights (Djankov et al. (2003)), more developed financial markets (La Porta et al. (1997, 1998)), and less stringent labor market regulations (Botero et al. (2004)). So a high correlation among different measures of institutional features is not surprising given evidence from Legal Origin theory, but it raises a concern that specific institutional measures could merely be proxies for one another such that the estimated importance of a particular institution for comparative advantage is not fully completed or might misinterpreted. The contribution of this paper, hence, is to shed some light on which institutions really do matter, whether the right institution-sector interac-

49 37 tions have been identified, and what are the consequences that a given institution might have on export performance of other industries whose characteristics do not necessarily resemble those originally identified in the literature. The roadmap for the paper is as follows. Section 2.2 provides an overview of the key variables, data sources, and empirical specification. Section 2.3 presents arawdataanalysisandreportstheestimationresults. Section2.4providesrobustness and sensitivity analysis. Section 2.5 concludes. 2.2 Data and empirical specification To investigate the effects of various institutional determinants on the patterns of specialization, this section overviews sources and definitions of countryspecific characteristics, industry-specific characteristics, and trade data used Country-sector data Data on export values at the 4-digit SITC Rev.2 industry level are from Feenstra et al. (2005) s World Trade Flows Database for the year I convert the original trade data into 3-digit ISIC level using Haveman s concordance tables 2 to match measures of sector-level institutional intensity. The lists of countries and industries are presented in table 11 and Country-level data The key country-level variables required for testing institutional sources of comparative advantage are measures of legal system, financial market development, labor market flexibility, and IP Protection. I use rule of law from the World Bank Governance Indicators (Kaufmann et al. (2004)) as a proxy for judicial quality (as in Nunn (2007) and Levchenko (2007)). This measure is a weighted average of a number of variables that measure individuals perceptions of the effectiveness and predictability of the judiciary, and the enforcement of contracts in each country 1 Iusedatafrom1997forcomparabilitywiththoseintheliterature. 2

50 38 between 1997 and As for a country s level of financial development, I use data from Bekaert et al. (2001) s Financial and Economic Development database, which measure the amount of credit extended by banks and other financial intermediaries to the private sector divided by GDP. This variable is used to capture the accessibility of firms to external funding sources. I use the rigidity of employment index from the World Bank s Doing Business database Botero et al. (2004) to measure labor market flexibility across countries. This employment index is the average of 3 sub-indices: the difficulty of hiring index, rigidity of hours index, and difficulty of redundancy index. I rescale the original index so that it is increasing in labor market flexibility. As for a measure of country s IP protection, I use the patent rights index from Ginarte and Park (1997). This index ranges from 0 (weakest) to 5 (strongest) and is the arithmetic sum of 5 indices, including extent of coverage, membership in international patent agreements, provisions for loss of protection, enforcement mechanisms, and duration of patents. Data on human capital and physical capital endowments are from Caselli (2005). As in Barro and Lee (2001), human capital endowments are proxied by human capital per worker defined as a function of the average years of schooling in the population over 25 years old. Physical capital endowments are generated from the perpetual inventory equation and divided by the number of workers to obtain capital stock per capita Sector-level data I use Nunn (2007) s measure of contract intensity 3 to capture the sensitivity of a given industry to the hold-up problems between producers and upstream suppliers. This measure is defined as the share of intermediate inputs that cannot be bought on organized ex-changes and are not reference-priced based on Rauch (1999) commodity classification 4. The measures of financial intensity and sec- 3 nunn 4 Top five least contract intensive industries are poultry processing, flour milling, petroleum refineries, wet corn milling, and aluminum sheet. Top five most contract intensive industries include photographic & photocopying equip. manuf., air & gas compressor manuf., analytical laboratory instr. Manuf., other engine equipment manuf., and other electronic component manuf.

51 39 tor volatility are constructed using data from Compustat s Industrial Annual file, which provide detail information on all publicly-traded firms in North America. Following Rajan and Zingales (1998) s methodology, the indicator of financial intensity or sector s reliance on outside finance is the ratio of capital expenditures minus cash flow from operations to capital expenditures for the median firm in each industry. Sector volatility is the average of the standard deviation of the annual growth rate of firm sales weighted by firm average employment over the period. The innovation intensity variable is calculated using data from STAN R&D Expenditure in Industry, OECD, weighted by value added. Data on skill intensity and capital intensity of production are from Braun (2003). Skill intensity is measured as the industry s mean wage over that of the whole manufacturing sector. Physical capital intensity corresponds to the industry s ratio of gross fixed capital formation to total value added Empirical Specification To test for the role of institutional determinants of trade, I exploit crosscountry variation in institutional quality and cross-industry variation in institutional dependence. Following the approach taken by Romalis (2004), whereby comparative advantage is identified from the interactions between country and industry characteristics, I augment his empirical specification to include institutional sources and estimate the following equation: ln X i,c = α i + α c + β z,q (z i Q c )+β f,f (findep i F indev c )+β v,l (vol i Flex c ) + β v,q (vol i Q c )+β IP (rd i IP c )+β k,k (k i K c )+β hh (h i H c )+ε ic (1) where X ic denotes the total value of exports in industry i from country c to all other countries in the world; z i is a measure of the importance of relationshipspecific investment in industry i, i.e., contract intensity; Q c is a measure of the judicial quality in country c or rule of law; findep i is a measure of sector i level of external capital dependence indicating financial intensity; F indev c is a measure of country c private credit availability which signify a country s level of financial

52 40 development; vol i is a measure of sector i sales volatility; Flex c is a measure of country c labor market flexibility; rd i is a measure of sector i innovation intensity; IP c is a measure of country c s level of IP protection; H c and K c denote country cendowmentsinskilledlaborandphysicalcapital,andh i and k i are the skill and capital intensities of production in industry i; α i and α c denote industry and country fixed effects respectively, and ε ic is the error term with i.i.d and zero mean. Using this specification to study factor content of trade, Romalis (2004) shows that countries capture larger US market shares in industries that use their abundant factors more intensively (β k,k > 0andβ h,h > 0). Similar arguments can be applied to institutional content of trade. The coefficients in front of all the interaction terms between country institutional measures and industry measures capture the extent to which the importance of a given institution varies from industry to industry depending on the levels of institutional intensity. A positive coefficient estimate of the interaction term between country s rule of law and industry s contract intensity (β zq > 0) indicates that countries with better rule of law export relatively more in contract intensive industries (Nunn (2007) and Levchenko (2007)). The intuition is that with better legal institutions, the probability that the contracts between input suppliers and producers of complex goods are enforced is higher. This in turn reduces under-investment in the relationshipspecific inputs and encourages more efficient production and export of final goods that are more susceptible to the hold-up problem. Similarly, β ff > 0impliesthat countries with greater degrees of financial liberalization export disproportionately more in financially vulnerable sectors that require more outside finance or employ fewer collaterizable assets (Manova (2008)). Likewise, β vl > 0suggeststhat countries with more flexible labor markets specialize in sectors with higher withinindustry dispersions of shocks Cuñat and Melitz (2007). A negative estimate of β vq term implies that countries with superior legal institutions specialize in less volatile sectors. Lastly, β IP > 0termmeansthatcountrieswithbetterprotection of intellectual property rights protection have higher export in industries that are more innovation intensive. All specifications are controlled for factor endowments forces of comparative advantage.

53 Examining the Raw Data Before estimating equation (1), I first check whether there is evidence in the raw data that various institutional features are important sources of comparative advantage. I divided sample into countries with good and poor institutional quality along the four dimensions: those above and below the sample median of rule of law, financial market development, labor market flexibility, and patent rights index. Similarly, I also group industries into those that are contract intensive, financial intensive, high volatility, and innovation intensive, and those that are not using the median level of the relevant industry-institutional intensity. The statistics in Table 2.1 show that there appears to be a pattern in which countries with good institutional quality have higher percentage of exports in industries that are institutionally intensive. For countries with poor quality of institutions, a greater percentage of exports are in industries that require less of the services provided by these institutions. I find that for good judiciary countries 81 percent of exports are in contract intensive industries, whereas for poor judiciary countries 73 percent of exports are in contract intensive industries. The percentage differences are more pronounce for the levels of IP protection. For strong IP protection countries 73 percent of exports are in innovation intensive industries, while for weak IP enforcement countries only 54 percent of exports are in innovation intensive industries Estimation Results Data on exports are available for 27 ISIC Rev.2 3-digit manufacturing industries and 90 countries 5. Table 2.2 tests for the four hypotheses of institutional sources of comparative advantage. To verify that the patterns of specialization identified in the previous work are also present in the current sample, Table 2.2 reports the OLS estimates of Eq. (1) where each of the institutional determinants is introduced in turn in 5 Due to 103 zero trade observations (some countries do not export in some sectors) and missing observations in some country-level institutional features, the number of observations in the regression is 1,887.

54 42 column (1)-(5). All specifications include country and industry fixed effects, with standardized beta coefficients 6 and robust standard errors reported. In column (1), I first examine the role of legal institutions in facilitating production and alleviating transaction costs when contracts are not perfectly enforced. I obtain apositiveandstatisticallysignificantcoefficient estimate on the judicial quality interaction term (β zq =0.138), confirming the importance of a country s quality of contract enforcement proposed by Nunn (2007) and Levchenko (2007). That is, countries with better rule of law have a comparative advantage in the production of final goods that rely heavily on a few key inputs or use intensively inputs requiring relationship-specific investments. Next, I examine the role of a country s level of financial development in column (2), as capture by the interaction between country measures of private credit availability and an industry measure of external capital dependence. The estimated β ff term is positive and statistically significant, a result that supports Manova (2008) s hypothesis that countries with better financial development export more in industries that rely heavily on external financing. In column (3), I estimate (1) with only the labor market flexibility interaction included. I find no evidence that countries with flexible labor markets export more in volatile sectors as argued by Cuñat and Melitz (2007). However, when I interact the rule of law with sector volatility, I find a negative and significant effect of β vq term in column (4). This is in line with Costinot (2009b) s model prediction that countries with good judicial quality are in a better position to specialize in less volatile goods. Column (5) reports the result of a regression of log exports on only IP protection interaction term. I see that β IP is positive and significant, suggesting that countries with higher levels of IP protection tend to specialize and export more in innovation intensive sector. I next re-estimate the specifications of columns (1)-(3) & (5) by also controlling for capital and skill endowment interactions reported in columns (6)-(9). 6 Standardized beta coefficient captures the change in standard deviation units of the dependent variable in response to a one standard deviation increase in the explanatory variable, and thus allows for direct comparison of the relative magnitude of different institution-sector interactions.

55 43 All the institutional interactions remain intact when the traditional H-O comparative advantage sources are taken into account. The results show that countries which are more skill abundant export more in skill-intensive industries. Similarly, countries that are more capital abundant export more in capital-intensive industries. Having provided some preliminary evidence on the relevance of institutional sources of comparative advantage, I proceed by including all the institutional determinants in the same regression in column (10) to jointly verify the significance of each individual institution and compare their relative importance. The coefficient estimates of all the institutional interactions are statistically significant and with the expected sign except for the rule of law interacted with sector volatility. Since beta coefficients are reported, the highest magnitude found at the judicial quality interaction can be interpreted as follows. Among the four pillars of institutions analyzed, the rule of law appears to be the strongest institutional source of comparative advantage. Unlike some empirical studies which find that institutional differences explain more of the world s pattern of trade than physical capital and skilled labor, e.g., Nunn (2007), and Levchenko (2007), I find that the combined effects of institutional forces are still lower than the combined effects of capital and skilled labor endowments. A one standard deviation increase in rule of law, financial development, IP protection altogether increases the log of country s export by 0.3 standard deviations, whereas a same increase in capital and labor interactions increases the log of export by 0.55 standard deviations. To ensure that the results are not just driven by large developed countries whose institutions are much better than developing ones, Table 2.3 and 4 present coefficient estimates when I run each institutional interaction one by one or altogether in one regression on the following subsamples of countries: OECD, Non-OECD, and all nations excluding African, American, Asian and Oceania, and European countries. While the effects of all the three pillars of institutions are in general positive and statistically significant as sample of countries changes, interesting variations of rule or law and financial development seem to be rooting from Non-OECD countries (top and middle sections of Table 2.3-column (3)), whereas the significance of IP enforcement appears to be coming from OECD coun-

56 44 tries (bottom section of Table 2.3-column (2)). These patterns are present when institution-sector interactions are included separately or jointly in Table 2.3 and 4 respectively. Hence, these results suggest that institutional differences as a source of comparative advantage is not restricted to only North-South trade due to better institutions in the North, but also valid for South-South as well as North-North trade. The importance of contract enforcement is not apparent for the group of OECD countries but for non-oecd countries. Conversely, the importance of IP protection is evident for the group of OECD countries but not for non-oecd countries. 2.3 Robustness and Sensitivity Analysis Although most of the mechanisms remain robust when all the institutional determinants are run in the same regression, this section further explores the issue concerning high correlation among different dimensions of institutions. In particular, I test whether each of the institutional features examined in the previous section is identifying an independent source of comparative advantage. Empirical evidence from Legal Origin Theory suggests that (i) domestic institutions are important for long-term economic development and (ii) developed countries have much better institutions than developing countries. The level and nature of development of institutions, in turn, can be traced back to how legal institutions were transplanted by different colonial powers. Legal system based on French civil law is associated with a heavier state-intervention, whereas British common law allows for a more market based orientation. The survey by La Porta et al. (2008) indicates that historical origin of acountry slawsishighlycorrelatedwithabroadrangeofitslegalrulesand regulations. La Porta et al. (1997, 1998) document that common law countries have stronger legal rules protecting investors than civil law countries, and since legal investor protections limit the expropriation by corporate insiders, thus the former tends to have more developed financial markets than the latter. Legal origin is also associated with contract enforcement (Djankov et al. (2003)), with common

57 45 law colonized countries having better contract enforcement and greater security of property rights than civil law colonized countries. Furthermore, Botero et al. (2004) show that common law origin countries generally have less stringent labor regulations relative to civil law origin countries. As a consequence, the potential effect that legal origin may have on a host of country characteristics plausibly causes different dimensions of domestic institutions to be highly correlated with one another. So if it is the case that the most contract intensive industries are also industries that most intensively require external finance, and because British legal origin countries have both good judicial quality and better financial development, then it is unclear whether which institutional source is really identifying the pattern of comparative advantage. Given the strong correlation among country characteristics, I test whether countries that are superior in a particular institutional environment specialize in other non-original pair sectors, and 2) whether other country characteristics, rather than the original institution-specific pair, cause countries to become specialize in the relevant industry. To do so, I first focus on a particular institution and interacted it with different industry measures and then the reverse, holding fix industry characteristic and varying it across various institutional measures. IfirstpresentpairwisecorrelationsofthefourinstitutionalfeaturesinTable 2.5. As shown, all of the institutional quality measures are positively correlated with one another. Countries with better rule of law tend to have more liberalized financial markets with correlation coefficient of While the rule of law s correlation with labor market flexibility is not as strong, its correlation with property rights protection is very high at The highest correlation is found between rule of law and GDP at So it is not surprising that the pattern in which the rule of law is correlated with other country characteristics is very similar to how GDP is also correlated with the same institutional variables. This raises the concern that if the rule of law is in fact merely a proxy for countries income, then the estimated importance of contract enforcement for comparative advantage is misinterpreted because high income countries might specialize in contract intensive industries for reasons unrelated to a country s ability to enforce contract. I show

58 46 the plotted relationship of judicial quality and financial development, labor market flexibility, and IP protection in Figure (1)-(3) respectively. The strong correlation between countries judicial quality and financial development is depicted in Figure (1) with a steep upward sloping fitted line. Switzerland, Singapore and the United Kingdom are among the countries that portray good judicial quality, developed financial market, flexible labor regulation and strong IP protection. To examine whether the right institution-sector pairs have been identified, I begin by interacting each country characteristic across different measures of industry characteristics in Table 2.6. The first column presents the interactions of rule of law with all the four measures of industry characteristics classified as contract intensity, financial intensity, sector volatility, and innovation intensity. I obtain a positive and significant effect only at the original institution-sector pair as shown in the first row of column (1). This provides a support for the hypothesis that countries with better contract enforcement have a comparative advantage in contract intensive industries and that the right interaction may have been correctly identified for country s rule of law. In column (2), however, I find positive and significant effects of financial development on both financially vulnerable and contract intensive industries. The estimates imply that financial development favors the specialization of not only sectors that are more dependent on external financing but also those that use intensively inputs requiring relationship specific investments. It is interesting to see that the effect of country s financial systems is even stronger for exports of contract intensive industries and not the originally identified financially intensive industries. A one standard deviation increase in country s level of financial development increases export in financially intensive industries by 0.03 standard deviations, while the same increase raises export of contract intensive sectors by 7 times as much or 0.22 standard deviations. One interpretation of this result is that a more readily available outside funds in more developed financial markets allows firms to finance vertical integration to get around the problem of weak contract enforcement. As such, countries with better financial development specialize more strongly in contract-intensive industries. Inextturntotheroleoflabormarketflexibilityincolumn(3). Even

59 47 though I do not find any significant effect from interacting labor market flexibility with sector volatility in the previous section, flexible labor regulations appear to be another important determinant for trade in contract intensive industries, as shown in column (3). Similarly, I find that country s IP protection is an important source of comparative advantage not only for sectors intensive in research and innovation but also contract intensive industries. In the last column, I interact all of the industry characteristics with GDP to see which industries do high income countries export. In column (5), I observe that rich countries specialize in contract intensive industries. Therefore, a particular institutional feature is important not only for own-industry pair trade but also for the performance of other non-original industry-specific characteristics. I proceed by doing the reverse, restricting attention to one particular industry characteristic and interacting it with alternative country characteristics in turn. I report a baseline specification with only original country-sector pair included in column (1) of Table 2.(7)-(10). In Table 2.7, I first test whether other dimensions of institutions, rather than judicial quality, drive country s comparative advantage in contract intensive industries. I do this by adding the interactions of contract intensity with financial development, labor market flexibility, and IP enforcement one by one in column (2) to (5) respectively. Column (2) shows that while the coefficient of contract intensity x rule of law remains positive and significant after contract intensity x financial development is inserted, the magnitude of the original interaction term falls by half, and that the stronger effect of institution-sector match is now seen at contract intensity x f inancial development. That is, the level of financial development is more important for exports of contract intensive industries compared the rule of law. A country s world market share of complex goods exports increases by almost two standard deviations for every one standard deviation increase in financial development. In column (3), labor market flexibility is also found to have important implication on contract intensive industries, although the effect is not as strong compared to the original institution-sector pair in the first row. A one standard deviation increase in labor market flexibility index increases country s market share in contract intensive goods by 0.14 standard

60 48 deviation. However, I find no role for country s IP protection or skill and capital endowments in influencing the pattern of specialization in contract intensive industries (column (4)-(5)). Overall, the estimated coefficient of z i Q c term remains positive and statistically significant throughout when alternative interactions are included one by one. However, when a full set of all the country-industry interaction terms is launched, the original rule of law interacted with contract intensity loses its significance to financial and labor market institutions. This imply that countries with better-developed financial markets and more-flexible labor laws capture larger world exports in contract-intensive industries. A plausible explanation is that good financial institutions and flexible labor markets may lessen the reliance of firms in contract dependent industries on legal institutions to enforce contracts, by enabling firms to finance vertical integration or adjusting employment needed to produce own customized inputs. To test the robustness further, I conduct a similar analysis by interacting rule of law (Q c ), labor market flexibility (Flex c ), and IP protection (IP c ), and skill and capital endowments (H c and K c )withameasureoffinancialintensityintable 2.8. In column (2), I find that when rule of law and financial intensity interaction is included, the estimated coefficient on the original financial institution-sector pair becomes insignificant. The result suggests that the level of financial development is less importance for a comparative advantage in financially intensive industries when countries exhibit good legal systems. A possible explanation is that it is not about financial instruments per se but more about lenders willingness to lend as they can rely on strong local courts in the event that the borrower defaults on loan payments. I also obtain a positive and significant coefficient estimate on F inancial Intensity x IP P rotection term, suggesting that a country s IP protection is an important trade determinant for financially vulnerable sectors. The original financial interaction term becomes insignificant when financial intensity is interacted with physical capital and skill labor, or when all the interaction terms are included in column (5)&(6) respectively. In Table 2.9, sector volatility is interacted with all the country characteristics. The only significant effect is observed at the interaction between sector

61 49 volatility x rule of law for all the alternative specifications. However, nothing is significant when the full set of country characteristics is run in one regression. Finally, Table 2.10 presents estimates when innovation intensity (IP c )isinteracted with different measures of institutional framework. Similar to Table 2.8, I find that when the cross interaction term between innovation intensity and rule of law is jointly examined, the original interaction pair rd i xip c )dropsoutofsignificance in column 2. A one standard deviation increase in rule of law increases exports in R&D intensive goods by 0.13 standard deviations. I see that financial development also matters for the specialization of innovation intensive industries, although the effect is weaker than the institutional strength of patent rights protection (column 3). I obtain a positive and significant estimate from interacting capital endowment with innovation intensity in column (4). When all of the country-industry interactions are included in the last column, I observe that rule of law is the most important institutional determinant for exports of innovation intensive industries. The main implication from this last section s exercise is that every time a rule of law enters into the equation, its institutional strength dominates the positive impact that a particular institution has on its original industry-specific pair or even renders some to become insignificant. This suggests that rule of law, compared to other types of institutions, has the strongest effect on the pattern of specialization of all institutionally intensive industries, however industry are defined except for contract intensity. Moreover, I document a reverse pattern in which a country s level of financial development is more important than the rule of law in explaining exports of contract intensive industries, and vice versa for financially vulnerable sectors that require more outside outside finance or employ fewer collateralizable assets. Hence, this paper contributes by pointing out that the complementarity between institutional quality and sectors institutional dependence is not limited to country-industry pair originally defined in the literature.

62 Conclusions Alternative to asking whether which institutions matter and to what industry characteristics, this paper examines whether different institutional features advanced recently uniquely identify independent sources of comparative advantage. Itakeintoaccountalloftheinstitutionalvariablesstudiedintheliterature including contract enforcement, financial development, labor market flexibility, and intellectual property protection and conduct a joint analysis to verify their significance and compare their quantitative importance. Given high correlation among different pillars of institutions such that specific measures of institutional quality might merely be proxies for one another, I also perform a cross institutional-sector inspection to test whether the right interaction has been identified. In line with the empirical literature, I find that the superiority of country s institutional environment influences its comparative advantage in sectors that use intensively the services provided by the domestic institutions. While jointly significance, each institution does not play a separate role in determining the pattern of specialization. A comparative advantage in contract intensive industries is not restricted to a country s ability to enforce contracts. When all country attributes are simultaneously considered, I observe that financial and labor market institutions matter more for exports of contract intensive industries than the originally defined contracting institutions. Hence, this paper sheds light on the consequences that a particular institution might have on export performance of other kinds of industries whose characteristics do not necessarily resemble those identified in the literature. Despite being relevant, the H-O counterpart effect for institutions is less obvious because of high correlation among different pillars of institutions. That is, the development in legal systems may imply stronger investor protections and hence better developed financial institutions. Therefore, well-functioning financial markets may substitute for contracting institutions by relaxing firms financial constraints, and thereby allowing them to vertically integrate to help reduce under-investment when contracts are not perfectly enforced. As a consequence, countries with good financial systems specialize not only in financially vulnerable sectors that are more

63 51 dependent on external financing, but also in contract intensive industries in which customized inputs are used in high proportion and thus more susceptible to the hold-up problem. Table 2.1: Raw data analysis Contract Intensive Not Contract Intensive Good Judicial 81% 19% Poor Judicial 73% 27% Financial Intensive Not Financial Intensive Developed Financial Market 73% 26% Underdeveloped Financial Market 71% 29% High volatility Low Volatility Flexible Labor Market 62% 38% Rigid Labor Market 55% 45% Innovation Intensive Not Innovation Intensive Good IP protectiont 73% 27% Poor IP protection 54% 46%

64 52 Table 2.2: Baseline specification and joint institutional determinants (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Rule of Law x Contract Intensity 0.138*** 0.268*** 0.207*** (0.037) (0.043) (0.046) Financial Development x Financial Intensity 0.076*** 0.079*** 0.033* (0.020) (0.019) (0.020) Labor Market Flexibility x Sector Volatility (0.045) Rule of Law x Sector Volatility *** * (0.030) (0.029) (0.029) IP Protection x Innovation Intensity 0.189*** 0.145*** 0.065** (0.023) (0.028) (0.030) Capital Endowment x Capital Intensity 0.393*** * 0.346*** (0.133) (0.112) (0.117) (0.118) (0.131) Skill Endowment x Skill Intensity 0.298*** 0.348*** 0.315*** 0.184** 0.244*** (0.075) (0.074) (0.078) (0.089) (0.085) Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Observations 1,887 1,887 1,887 1,887 1,887 1,887 1,887 1,887 1,887 1,887 R The dependent variable is the log of exports in industry i by country c to the rest of the world. Standardized beta coefficients are reported, with robust standard errors in brackets. *, ** and *** indicate significance at the 10, 5 and 1 percent levels.

65 53 Table 2.3: Separate Institutions on Alternative Samples (1) (2) (3) (4) (5) (6) (7) Full OECD Non- No No No Asia- No Samples OECD Africa America Oceania Europe Rule of Law x Contract Intensity 0.260*** *** 0.321*** 0.249*** 0.274*** 0.251*** (0.042) (0.127) (0.127) (0.065) (0.054) (0.050) (0.047) Capital Endowment x Capital Intensity 0.376*** *** 0.836*** 0.299** 0.336** 0.424*** (0.127) (0.612) (0.612) (0.182) (0.176) (0.135) (0.139) Skill Endowment x Skill Intensity 0.261*** 0.563*** *** 0.238*** 0.223*** 0.185** (0.071) (0.148) (0.148) (0.108) (0.087) (0.083) (0.078) Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effects Yes Yes Yes Yes Yes Yes Yes Observations 2, ,406 1,515 1,441 1,500 1,622 R Financial Development x Financial Intensity 0.077*** * 0.097*** 0.076*** 0.059*** 0.075*** (0.049) (0.025) (0.033) (0.018) (0.022) (0.023) (0.025) Capital Endowment x Capital Intensity ** 0.012** (0.150) (0.552) (0.173) (0.156) (0.114) (0.122) (0.137) Skill Endowment x Skill Intensity 0.327*** 0.572*** *** 0.315*** 0.290*** 0.236*** (0.082) (0.146) (0.111) (0.089) (0.083) (0.080) (0.085) Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effects Yes Yes Yes Yes Yes Yes Yes Observations 1, ,366 1,515 1,401 1,460 1,582 R IP enforcement x Innovation Intensity 0.142*** 0.441*** *** 0.164*** 0.134*** 0.118*** (0.028) (0.158) (0.047) (0.036) (0.032) (0.032) (0.035) Capital Endowment x Capital Intensity *** (0.117) (0.553) (0.178) (0.162) (0.122) (0.128) (0.144) Skill Endowment x Skill Intensity 0.185** 0.443*** *** * (0.088) (0.157) (0.124) (0.106) (0.102) (0.098) (0.102) Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effects Yes Yes Yes Yes Yes Yes Yes Observations 1, ,311 1,433 1,371 1,401 1,519 R The dependent variable is the log of exports in industry i by country c to the rest of the world. Standardized beta coefficients are reported, with robust standard errors in brackets. *, ** and *** indicate significance at the 10, 5, and 1 percent levels.

66 54 Table 2.4: Joint specification on alternative samples Full OECD Non- No No No Asia- No Samples OECD Africa America Oceania Europe Rule of Law x Contract Intensity 0.206*** *** 0.252*** 0.177*** 0.248*** 0.20 (0.046) (0.141) (0.068) (0.062) (0.055) (0.052) (0.053) Financial Development x Financial Intensity 0.036* *** 0.037* 0.006* (0.020) (0.025) (0.034) (0.019) (0.022) (0.024) (0.026) IP enforcement x Innovation Intensity 0.065** 0.441*** *** (0.030) (0.171) (0.045) (0.039) (0.034) (0.034) (0.035) Capital Endowment x Capital Intensity 0.344*** *** 0.745*** 0.247* 0.342** (0.131) (0.598) (0.190) (0.186) (0.138) (0.145) (0.155) Skill Endowment x Skill Intensity 0.251*** 0.447*** *** 0.227** 0.209** (0.084) (0.160) (0.122) (0.103) (0.097) (0.093) (0.098) Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effects Yes Yes Yes Yes Yes Yes Yes Observations 1, ,290 1,433 1,350 1,380 1,498 R The dependent variable is the log of exports in industry i by country c to the rest of the world. Standardized beta coefficients are reported, with robust standard errors in brackets. *, ** and *** indicate significance at the 10, 5, and 1 percent levels.

67 55 Table 2.5: Pairwise Correlation of Country Characteristics Rule of Financial Labor IP GDP Law Market Market Protection Rule of Law 1 Liberalisation Intensity * 1 Labor Market Flexibility * * 1 IP Protection * * * 1 GDP * * * * 1 Correlation coefficients are reported. * indicates significance at the 1 percent level.

68 56 Table 2.6: Alternative Industry Interactions Country Characterisitcs Industry Characteristics (1) (2) (3) (4) (5) Rule of Law Fin Dev Lab Flex IP GDP Contract Intensity x Column 0.209*** 0.217*** 0.254*** 0.105*** 0.464*** (0.049) (0.043) (0.058) (0.045) (0.118) Financial Intensity x Column * (0.030) (0.022) (0.046) (0.028) (0.099) Sector Volatility x Column (0.028) (0.022) (0.043) (0.027) (0.079) Innovation Intensity x Column *** (0.036) (0.023) (0.050) (0.034) (0.111) Capital Intensity x Capital Endow 0.331*** 0.359*** * 0.368*** (0.135) (0.133) (0.115) (0.132) (0.109) Skill Intensity x Skill Endow 0.257*** 0.329*** 0.317*** 0.251*** 0.253*** (0.090) (0.082) (0.079) (0.091) (0.078) Country Fixed Effects Yes Yes Yes Yes Yes Industry Fixed Effects Yes Yes Yes Yes Yes Observations 1,953 1,913 1,830 1,960 1,953 R The dependent variable is the log of exports in industry i by country c to the rest of the world. Standardized beta coefficients are reported, with robust standard errors in brackets. *, ** and *** indicate significance at the 10, 5, and 1 percent levels.

69 57 Table 2.7: Contract intensity with alternative institutions and endowments (1) (2) (3) (4) (5) (6) Contract Intensity x Rule of Law 0.271*** 0.121*** 0.231*** 0.282*** 0.175*** (0.042) (0.055) (0.045) (0.071) (0.084) (0.095) Contract Intensity x Fin Dev 0.175*** 0.160** (0.050) (0.056) Contract Intensity x Labor Flex 0.139*** 0.104*** (0.053) (0.056) Contract Intensity x IP Protection (0.067) (0.077) Contract Intensity x Capital Endow (0.387) (0.434) Contract Intensity x Skill Endow (0.505) (0.567) Capital Intensity x Capital Endow 0.416*** 0.479*** 0.402*** 0.415*** 0.474*** 0.469*** (0.132) (0.134) (0.132) (0.132) (0.144) (0.145) Skill Intensity x Skill Endow 0.248*** 0.231*** 0.258*** 0.248*** 0.234*** 0.239*** (0.074) (0.075) (0.075) (0.074) (0.077) (0.077) Country Fixed Effects Yes Yes Yes Yes Yes Yes Industry Fixed Effects Yes Yes Yes Yes Yes Yes R Observations 1,858 1,858 1,858 1,858 1,858 1,858 *, ** and *** indicate significance at the 10, 5, and 1 percent levels.

70 58 Table 2.8: Financial intensity with alternative institutions and endowments (1) (2) (3) (4) (5) (6) Financial Intensity x Rule of Law 0.094*** (0.041) (0.063) Financial Intensity x Fin Dev 0.079*** *** 0.053*** (0.019) (0.030) (0.022) (0.024) (0.030) (0.033) Financial Intensity xlabor Flex (0.044) (0.047) Financial Intensity x IP Protection 0.074*** (0.032) (0.045) Financial Intensity x Capital Endow (0.143) (0.155) Financial Intensity x Skill Endow (0.068) (0.082) Capital Intensity x Capital Endow (0.112) (0.112) (0.112) (0.111) (0.111) (0.111) Skill Intensity x Skill Endow 0.295*** 0.304*** 0.295*** 0.305*** 0.304*** 0.306*** (0.073) (0.073) (0.074) (0.073) (0.073) (0.073) Industry Fixed Effects Yes Yes Yes Yes Yes Yes R Observations 1,858 1,858 1,858 1,858 1,858 1,858 *, ** and *** indicate significance at the 10, 5, and 1 percent levels.

71 59 Table 2.9: Sector volatility with alternative institutions and endowments (1) (2) (3) (4) (5) (6) Sector Volatility x Rule of Law *** *** *** * (0.027) (0.042) (0.029) (0.047) (0.059) (0.066) Sector Volatility x Fin Dev (0.033) (0.035) Sector Volatility x Labor Flex (0.044) (0.048) Sector Volatility x IP Protection (0.045) (0.052) Sector Volatility x Capital Endow (0.135) (0.142) Sector Volatility x Skill Endow (0.075) (0.082) Capital Intensity x Capital Endow (0.116) (0.116) (0.116) (0.116) (0.116) (0.116) Skill Intensity x Skill Endow 0.262*** 0.263*** 0.262*** 0.261*** 0.262*** 0.262*** (0.077) (0.077) (0.077) (0.077) (0.077) (0.077) Country Fixed Effects Yes Yes Yes Yes Yes Yes Industry Fixed Effects Yes Yes Yes Yes Yes Yes R Observations 1,858 1,858 1,858 1,858 1,858 1,858 *, ** and *** indicate significance at the 10, 5, and 1 percent levels.

72 60 Table 2.10: Innovation intensity with alternative institutions and endowments (1) (2) (3) (4) (5) (6) Innovation Intensity x Rule of Law 0.130*** 0.119*** (0.043) (0.057) Innovation Intensity x Fin Dev 0.037* (0.022) (0.027) Innovation Intensity x Labor Flex (0.039) (0.043) Innovation Intensity x IP Protection 0.158*** *** 0.159*** 0.109*** (0.029) (0.041) (0.032) (0.029) (0.041) (0.043) Innovation Intensity x Capital Endow 0.274** (0.115) (0.137) Innovation Intensity x Skill Endow (0.073) (0.083) Capital Intensity x Capital Endow 0.222** 0.241** 0.234** 0.221** 0.230** 0.230* (0.119) (0.119) (0.119) (0.119) (0.122) (0.122) Skill Intensity x Skill Endow 0.147* * (0.090) (0.092) (0.093) (0.091) (0.101) (0.102) Country Fixed Effects Yes Yes Yes Yes Yes Yes Industry Fixed Effects Yes Yes Yes Yes Yes Yes R Observations 1,790 1,790 1,790 1,790 1,790 1,790 *, ** and *** indicate significance at the 10, 5, and 1 percent levels.

73 LIberalization Intensity LIberalization Intensity LIberalization Intensity Rule of Law Rule of Law Rule of Law Figure 2.1: Judicial quality and financial development Labor Market Flexibility Labor Market Flexibility Labor Market Flexibility Rule of Law Rule of Law Rule of Law Figure 2.2: Judicial quality and labor market flexibitliy

74 IP Protection IP Protection IP Protection Rule of Law Rule of Law Rule of Law Figure 2.3: Judicial quality and IP protection

75 Chapter 3 The Effects of Free Trade Agreements and Customs Unions on the Extensive and Intensive Margins of Trade Using Helpman, Melitz and Rubinstein (2008) s two-stage estimation procedure, this paper empirically estimates the effects of preferential trade agreements (PTAs), accounting for sample selection and firm heterogeneity bias, on the extensive and intensive margins of trade (i.e., the number of exporters and the trade volume per exporter respectively). I explicitly distinguish between free trade agreements (FTAs) and customs unions (CUs) to allow for their distinction in devices of integration, in particular the preferential rules of origin (RoO), to differentially affect trade at the two margins. Employing a large number of country pairs and worldwide PTAs, I find that CUs have a much larger trade effect than FTAs, especially at the extensive margin. Once the potential biases are addressed, I find that the effect of FTAs, previously found to be negative with the traditional OLS estimate, is positive and statistically significant. 63

76 Introduction The past two decades have seen a rapid increase in the number of preferential trade agreements (PTAs) signed around the world. As of 2012, there is a total of 511 PTAs notified to the GATT/WTO and 319 currently in force, with over 50 notifications occurred during alone (Figure 1). Given the worldwide proliferation (new) and expansion (existing) of PTAs, the analysis of their impact on trade flows seems especially relevant. Countries participating in PTAs are granted preferential access to members markets vis-a-vis non-members through reciprocal exchange of concessions on both tariffs andnon-tariffs trade barriers. While PTA is a widely used trade policy tool to bring closer economic integration, its real world consequences on members bilateral trade are less well understood. Earlier empirical research estimating the trade effect of PTA membership using the traditional gravity equations has delivered mixed findings. The discordant in empirical evidence across studies has generated a fair amount of follow-up research, trying to identify sources of econometric difficulties potentially responsible for inconsistent estimates and searching for best strategies to manage any biases to yield reliable estimates. The aim of this paper is to estimate the trade impact of PTAs by paying particular attention to the biases embodied in the standard gravity estimation, in particular the pervasiveness of zeros in bilateral trade data which presents a challenge to consistent estimation. I adopt Helpman, Melitz and Rubinstein (2008), henceforth HMR, two-stage estimation procedure that corrects for non-random selection of country pairs into exporting as well as firm heterogeneity bias. This method allows for a decomposition of the trade impact into their intensive and extensive margin components, which is crucial for understanding the welfare gains of PTAs owing to both scale (increase in the volume of trade of existing exporting firms) and selection effects (entering of new firms into export markets, hence more varieties). Also, this paper explicitly distinguishes between free trade agreements (FTAs) and customs unions (CUs) to allow the two regimes to differentially affect fixed and variable costs of exporting, and thus operate through the extensive and intensive margins differently. This is important because the aggregate

77 65 effect of PTAs on trade flows could be confounded as a result of pooling of FTA and CU with diverse legal instruments governing implementation, especially the preferential rules of origins (RoO). Using a large number of country-pairs and PTAs contracted worldwide, I find that CUs have a much larger trade impact than FTAs, in particular at the extensive margin. This finding is consistent with one interpretation that RoO feature of FTA, which is absent from CU and tends to be complex, restrictive, and vary greatly in details, could act as additional trade barriers and work to offset any preferential gains from tariff liberalization. On the extensive margin, the administrative burden of RoO could possibly limit market access by raising compliance costs involve in certifying origin, generally requires a two-step certification procedure from both public and private. On the intensive margin, RoO may potentially raise domestic production costs by constraining and distorting cumulation of inputs across members and non-members, and perhaps away from lowest price suppliers so as to qualify for a preferential treatment. Support for this adverse effect of RoO argument can be found in Anson et al. (2005a) and Estevadeordal and Suominen (2004), who show that RoO indeed have a negative impact on trade flows. Moreover, the results remain unaltered when membership to the European Union is identified using a separate dummy variable. Hence, a larger trade promoting effect of CUs compared to FTA is not merely driven by country pairs belonging to the EU who already trade a lot with one another, but more likely due to the overwhelming requirements of RoO feature of FTA that work to undo tariff preferences. Although initially used as an empirical tool to analyze bilateral trade flows by Tinbergen (1962) and later micro-founded 1 to provide a proper specification i.e., with controls for relative prices or multilateral resistance terms (Anderson and Wincoop (2003) and Feenstra (2003), ch5, p.35), the gravity estimation still suffers from biases if pairs of countries that do not trade with each other are ignored. Recognizing the importance of information contained in zeros, Helpman et al. (2008) develop a theoretically motivated gravity equation that features two- 1 See, for example, Anderson (1979), Helpman and Krugman (1985) and Anderson and Wincoop (2003)

78 66 stage estimation procedure. This approach is used to treat both sample-selection and firm heterogeneity in productivities, thereby capable of predicting positive as well as zero trade and enable the number of exporting firms to vary across destination countries as displayed in the data. Their two-system equation involves estimating the probability of a positive bilateral trade in the first stage Probit, i.e, the extensive margin. The predicted probability obtained from the first stage is then used to construct two variables that capture Heckman sample selection and firm heterogeneity bias. These correction terms are in turn included in the second stage to provide a consistent parameter estimate of the intensive margin effect. As conventional in the literature on PTA, a value of 1 is often assigned to asingledummyvariabletorepresentwhethercountrypairsbelongtoanykinds of preferential trading arrangement. In practice, FTAs and CUs differ greatly in terms of legal provisions and instruments used to govern their implementation. The RoO requirements, which are used to prevent tariff concessions on the subsequent exports of goods that enter the free trade zone through member with the minimum duty to other members, can be very costly. Their demanding requirements could potentially raise firms transaction costs and particularly strongly for firms in a country that signs many trade agreements with diverge RoO, for example Chile, Singapore, and Turkey. As a result, these extra costs of certification could adversely affect firms cost of doing business and may vary heterogenously across firms. Smaller and less efficient firms might find it more costly to deal with red-tape compliance due to their lacking of resources compared to larger, more productive firms. Hence, these extra administrative fixed costs of RoO in FTA might discriminate against smaller exporters and yield a smaller fraction of firms export compared to CU. Anson et al. (2005b) find that RoO-related administrative costs approximate to 2% of Mexico s preferential exports to the US market. Anson et al. (2005a) find average compliance costs are 6% in ad valorem equivalent, higher than 4% tariff preferences. After taking into account the extensive margin in the first stage Probit, I find that the effect of FTAs on trade volumes, previously found to be negative with the OLS estimates, is positive and statistically significant. This paper is not the first to explore the effects of PTAs on the two mar-

79 67 gins of trade. Helpman et al. (2008) find that PTA membership influences bilateral trade at the extensive margin but has no significant effect on the intensive margin of trade. Egger et al. (2011) using Poisson pseudo maximum likelihood estimation with endogenous PTA variable find the opposite a significant positive effect of PTAs on the intensive margin but no extensive margin effect. Nevertheless, neither of the two studies distinguishes between FTA and CU in their analyses. While Roy (2010) differentiates between the two regimes of PTAs using Baier and Bergstrand (2007) s panel data technique to account for time-invariant unobservables, he makes no distinction between the impact on the extensive and intensive margins of trade. He finds that CU members engage in significantly greater volumes of bilateral trade than FTA members. Similar to this paper, Baier et al. (2011) examine how various types of PTAs affect trade at the extensive and intensive margins but their focus is on goods margin of trade. They employ Hummels and Klenow (2005) decomposition methodology and panel data to examine goods extensive and intensive margins, and not country and firm as in this study. They find evidence that various types of PTAs statistically increase trade at both goods extensive and intensive margins, however provide no discussion on plausible economic reasoning underlying their results. The remainder of the paper is organized as follows. Section 3.2 gives a brief overview of the empirical literature on PTA in the gravity framework. Section 3.3 discusses sample selection and firm heterogeneity bias and reviews HMR s twostage estimation procedure. Section 3.4 discusses the differences between FTAs and CUs, particularly with respect to RoO. Section 3.5 describes datasets and provides estimation results. Section 3.6 concludes. 3.2 Gravity model and trade agreements For over half a century, the gravity equation has been a workhorse model used to estimate the impact of PTAs on trade flows. However, empirical evidence on the trade effect of PTAs has been mixed across studies. Aitken (1973), Abrams (1980), and Brada and Mndez (1985) found the European Community (EC) to have

80 68 an economically and statistically significant effect on trade flows among members, whereas Bergstrand (1985) and Frankel, Stein and Wei (1995) found insignificant effect. Frankel (1997) found positive significant effect from Mercosur, insignificant effect from the Andean Pact, and significant negative effect from the membership in the EC. Trefler (1993) and Lee and Swagel (2000) show that estimates of tradepolicy liberalization suffered from endogeneity bias. The formation of PTAs has been argued to fail an exogenous requirement. Countries generally select themselves into PTA and who to sign with rather than randomly assigned partners, thereby potentially leading to an endogenous trade agreement variable. As a result, a series of papers have used different econometric techniques trying to address the endogeneity issue. Baier and Bergstrand (2002) and Magee (2003) used instrumental variables (IV) techniques with cross sectional data. However, the quality of instruments used is questionable. Baier and Bergstrand (2002) employ capitallabor ratios, relative factor endowment differences with the rest of the world, and measures of remoteness, all of which are often included as explanatory variables in the gravity equation and have statistically significant effects on trade flows. Magee (2003) use an index of democracies, GDP similarities, intra-industry trade indices, trade surpluses, and relative factor endowment differences. Again, all the mentioned variables are clearly suspicious as they tend to correlated crosssectionally with unobservable determinants of trade flows. Baier and Bergstrand (2007) comment that IV estimation is not a reliable method for addressing the endogeneity bias of the PTA variable and experiment with the Heckman control function, however find that the method yield unstable findings due to the lack of suitable exclusion restrictions. They proposed using panel data with country-pair and country-and-time fixed effects to remove bias from time-invariant unobservables and control for multilateral resistance terms that vary over time respectively. They argued that panel techniques adjust well for endogeneity bias stemming from self-selection into trade agreements and that the effect of PTAs on trade flows is quintupled. Another line of research focuses on the issue of zeros trade in bilateral trade

81 69 data between country pairs that do not trade with each other. Silva and Tenreyro (2006) invoke Jensens inequality and propose using Poisson pseudo-maximumlikelihood (PPML). They add that the estimates would also be consistent in the presence of heteroskedasticity which likely to be inherent in trade data. Roy (2010) estimates the impact of trade agreements by combining panel fixed effects approach with PPML specification and find CUs to have a much larger impact than FTAs on aggregate trade. Nevertheless, the panel fixed effects address only the pairwise time-invariant unobservables. HMR extend the classical gravity model recognizing two sources of potential biases due to the omission of countries that do not trade with each other and the fraction of firms that export. They proposed two-stages estimation procedure, allowing for a decomposition of the trade impact into their intensive and extensive margin components. This paper finds HMR methodology highly relevant hypothesizing that RoO feature of FTA, which is absent from CU, may potentially raise fixed export costs and thus affect the two margins of trade differentially. 3.3 Extensive and Intensive margins The prevalence of zeros in bilateral trade data is a common feature but achallengingonetoconsistentestimationofgravityequations. Thestandard procedure often discards zero trade observations because the dependent variable is the log of trade flows and hence zero trade is dropped. Helpman et al. (2008) argues that zeros contain rather meaningful information underlying reasons that prevent positive trade. Therefore, ignoring zeros potentially lead to biased coefficients estimated with the standard OLS. Exploiting the presence of zeros, Helpman et al. (2008) develop a two-stage estimation procedure that allows for a decomposition of trade into the extensive and intensive margins. This section briefly outlines the approach toward estimating the effects of PTAs on the two trade margins. The full details of the HMR model and its derivation can be found in Helpman et al. (2008). HMR s version of the gravity estimation which allows for zero trade flows (selection) and controls for the fraction

82 70 of firms that export from j to i (firm heterogeneity) can be written in the log-linear form as: m ij = β 0 + λ j + χ i γd ij + ω ij + u ij (3.1) where m ij is exports from country j to country i, β 0 is a constant, λ j is a fixed effect of the exporting country, χ i is a fixed effect of the importing country, d ij is the bilateral trade costs, both natural and policy-based, between i and j, ω ij controls for the fraction of firms that export from j to i which could possibly be zero, and u ij is the error term, u ij N(0, σn). 2 Consistent estimation of equation (3.1) requires a consistent estimate of ω ij, which often not included, and u ij, the selection effect that could arise when country pairs with zero trade flows are excluded. Also, the ω ij term need not be symmetric, which makes it important to use separate country fixed effects for exporters and importers as proposed by Feenstra (2004). The selection of firms into export market is a function of firm zero profit condition, defined as a function of firms productivity, fixed and variable costs, trade barriers including trade agreements, and demand and elasticity of substitution across products. HMR derive a selection of firms into export markets with a latent variable z ij equation to be: z ij = γ 0 + ξ j + ζ i γd ij κφ ij + η ij (3.2) where z ij is the ratio of the export profits of the most efficient firm to the common fixed export cost for exporters from j to i and equal to zero when exports are zero. ζ j is an exporter fixed effects, ζ i is an importer fixed effect, φ ij is a country-pair fixed trade cost and η ij is an i.i.d. error term consisting of u ij and unmeasured fixed export costs. Since z ij is observed when trade is positive, the following Probit equation can be estimated: ρ ij = Φ(γ0 + ξj + ζi γd ij κφ ij + ηij) (3.3) where ρ ij is the probability of positive exports from j to i. Predicted ρ ij can be used to obtain predicted values of the latent variable, which is in turn used to

83 71 obtain consistent estimate of ω ij and u ij. HMR then showed that a transformation of the gravity equation (3.1) that will give consistent estimates is: m ij = β 0 + λ j + χ i γd ij +ln{exp[δ(ẑ ij + ˆ η ij)] 1} + β uη ˆ η ij)+e ij (3.4) where ẑij = Φ 1 (ˆρ ij )isthepredictedvalueofthelatentvariable,ˆ η ij = φ(ẑij)/φ(ẑ ij) is the inverse mills ratio i.e., the Heckman correction for sample selection that addresses for the biases generated by the unobserved country-pair level shocks u ij and η ij. ˆ z ij ẑij + ˆ η ij is a consistent estimate for E[zij., T ij = 1] and ˆ ω ij ln{exp[δ(ẑij + ˆ η ij)] 1} is a consistent estimate for E[ω ij., T ij =1],β uη corr(u ij, η ij )/σ u σ η and e ij is distributed i.i.d.. The two-stage procedure involves estimating Probit equation (3.3) to obtain the estimated probability of exporting and the effects of various trade barriers on the extensive margin of trade. In the second stage, equation (3.4) is estimated using nonlinear least squares to obtain the estimates of trade agreements on the intensive margin. These steps are carried out to addresses the biases generated by the unobserved country-pair level shocks u ij and η ij as well as the underlying unobserved firm-level heterogeneity. Without explicitly control for heterogeneity, the effects of trade barriers on firm level exports will be confounded with their effects on the number of firms that export, and without control for zero bilateral flows, estimates will be affected by selection bias. 3.4 Free Trade Agreements and Customs Unions In recent years, PTAs have become a major trade policy tool for almost all nations. PTA is an initiative whereby participating countries liberalize barriers to trade, both tariffs and non-tariffs, therefore allowing members to enjoy preferential market access vis-a-vis non members. Reciprocal PTA can be classified into two major categories, namely FTA and customs union 2. FTA is an arrangement in which member countries agree to eliminate tariffs, quotas, and preferences on most 2 Higher levels of economic integration: common market (CU plus free movement of factors of production) and economic union (CU plus monetary and fiscal policy coordination) are lumped into CU for the purpose of the analysis in this paper.

84 72 goods and services while mainitining own external tariffs. FTA can either be bilateral, usually between one large country (rich Northern countries) and smaller (poorer Southern) partner, or regional where one country sign with a regional body or association. CU is similar to FTA except that CU members adopt a common external tariff (CET), generally the average of the pre-existing tariff levels of member countries. Another important feature that separates between the two types of PTA is the extent of rules of origin requirement (RoO). Since CET is absent from FTA, RoO play an important role in preventing tariff concessions on the subsequent exports of goods that enter the free trade zone through member with the minimum duty to other members. Implementing and monitoring the system of RoOs can be challenged especially when countries belong to two or more FTAs with disparate RoO, thereby giving rise to the overlapping FTAs and complicated spaghetti bowl phenomenon. This expended capital can be very costly, especially for small exporters, e.g., cost of documenting products and verifying them prior to border crossings part of FTA arrangements. The study of Anson et al. (2005a) suggests that RoO undermine the effects of tariffs reductions due to an FTA. They use the restrictiveness RoO index and a measure of tariff preferences to explain Mexican exports to the US and show that close to zero market access is being provided by NAFTA. Hence, analyses which do not allow for the effects of FTAs and CUs to differ can fail to capture crucial aspect of trade policy, produce misleading conclusions, and confound policy decisions. FTAs could raise fixed costs of entering foreign markets from strict RoOs requirement, but at the same time decrease the variable trade costs from tariff reductions, such that the aggregate effect resulting from pooling the two trade margins and regime types of PTAs yield insignificant result. For this reason, the implementation of an FTA could have very different implications on the two margins of trade compared to CU.

85 Data and empirical results This paper uses trade data from the UN Comtrade database for every 10 years period from Samples of countries range from depending on the availability of the data in different years. Bilateral trade costs, including dummies for common border, language and whether countries ever had a colonial relationship, are from CEPII 3. I refine PTA membership into FTA and CU and each is assembled from notifications to the WTO 4 and supplemental sources from World Trade Law 5 and Tuck Trade Agreements Database 6. Data on excluded variables used in the first stage Probit estimation, including regulatory days and procedure and regulatory costs, are from The World Bank Doing Business Project. Tables present cross-section regression results for the years 1975, 1985, 1995, and Column (1)&(2) estimate a proper benchmark OLS specification that includes standard gravity variables and importing and exporting country fixed effects to control for multilateral resistance terms. The results show that country j exports more to country i when two countries are closer to each other, have a common land border, share the same language or colonial ties. I first examine the impact of PTAs using a single dummy variable in column (1). I find that PTAs have insignificant effect on aggregate bilateral trade flows for the 1975 and 1985 samples (Tables 3.1&3.2). This is in line with previous findings such as Bergstrand (1985) and Frankel et al. (1995). Next, I refine PTA into two separate dummies for FTA and CU from column (2) onward. I obtain a large positive coefficient estimate of CU but a strong negative coefficient estimate of FTA, both highly significant at the 1% level. This suggests that the insignificant effect of PTA found in column (1) may stem from combining FTA and CU with opposite effects. I proceed by taking explicit account of zero trade flows and estimate the trade impact of the two PTA regimes using the HMR s two stage procedure. Column (3) demonstrates a Probit regression of eq. (3), where the presence of trading relationship is a function of trade costs, database.html

86 74 both fixed and variable, as specified in the first column. Again, I control for both exporter and importer fixed effects. For the excluded variables, I use regulation costs of firm entry, as measured by the number of days, the number of legal procedures, and the relative cost as a percentage of GDP per capita to start up a business. These variables have been used and verified as satisfying the exclusion restriction, affecting fixed but not variable trade costs, in Helpman et al. (2008). I find that while CUs have a large and positive impact on the probability that country j export to i, FTAs play negative and insignificant role in the formation of trading relationship for both the years 1975 and 1985 in column (3) of Tables 3.1&3.2. This evidence provides a support for the hypothesis that, in practice, the ROO feature of FTA may work as additional trade barriers by raising fixed compliance costs and discouraging smaller, less efficient firms from entering into export markets. Turning to the second stage estimation, I use the first stage Probit estimates to construct two variables that account for non-random selection of countries into exporting and allow for firm heterogeneity in productivity. Column (4) reports estimation of eq.(4) using nonlinear least squares (NLS), with η and δ capturing the Heckman selection bias and firm heterogeneity respectively. After addressing for biases, the coefficient of the FTA dummy, which used to be negative with the standard OLS estimate in column (2), is now positive and statistically significant for both years 1975 and This suggests that by controlling for potentially negative influence of RoO on the fixed export costs, I now obtain a positive effect of FTA membership on export volume per firm. The intensive margin effect of FTA is about 0.55, implying that a typical FTA increases trade volumes of existing exporters by about 72 percent (compared to CU at 1.22 or approximately 237 percent in column (4)). In column (5), I relax the parameterization assumptions of Pareto distribution of firm heterogeneity and reestimate the intensive margin with polynomial terms of ˆ z ij. I find that the estimate of the intensive margin remains positive and significant with slightly higher magnitude, confirming that joint membership in FTA actually boosts export volume per firm after correcting for biases. For all specifications, trade promoting effect from countries belonging

87 75 to CU is much larger than FTA and joint tests reject the null equality of the coefficient on FTA and CU at the 1% level. IfurthertestwhetherthesuccessoftheEUisdrivingamuchlargertrade impact found for CUs compared to FTAs. To do so, I use a separate dummy variable for country pairs that belong to the EU so that CU dummy represents only country pairs belonging to CUs other than EU. I reestimate the two-equation system with a new set of trade agreement dummies, namely FTA, CU, and EU, in columns (6)-(8) of Tables I find a consistent result in which the coefficient estimates of CU are large positive and significantly greater than FTA, with the joint tests continue to reject the null of equality at the 1% level. Hence, countries belonging to CUs other than the EU engage in significantly greater trade at both the extensive and intensive margins. Although the probability that a pair of countries trade at all is positive and significant in the more recent years (1995, Table 3.3 & 2005, Table 3.4), CUs are still found to promote significantly greater trade than FTA at both margins. I also pool across all 4 years, , and use exporter-time and importer-time, instead of exporter and importer, fixed effects to capture time-varying exporter and importer specific and multilateral price terms in Table 3.5. Again, CUs have a much larger impact in all specifications as described earlier. Therefore, it is of great importance to consider both the extensive and intensive margins and allow for separate treatments of FTAs and CUs to capture the crucial aspects of trade policy decisions. 3.6 Conclusion This paper analyzes the impact of preferential trade agreements (PTAs) on trade flows in a setting with large country-pairs and worldwide PTAs, decomposing the impact into their extensive (number of exporting firm) and intensive (trade volume per firm) margin components and allowing the effects to differ between free trade agreements (FTAs) and customs unions (CUs). I address for biases embodied in the traditional gravity approach due to sample selection and firm level heterogeneity using HMR s two stage estimation procedure. I find that CUs

88 76 are more successful in promoting trade compared to FTAs, in particular at the extensive margin at which stiff rules of origin potentially limit market access due to high compliance costs. Especially in the early years, once the extensive margin component is explicitly taken into account, the coefficient estimate of FTA on the intensive margin, previously found to be negative with the standard OLS estimate, is positive and statistically significant. Moreover, the estimates and the ranking of FTAs and CUs are not sensitive to consider separately dummy for the EU. Hence, refinement of PTA into FTA and CU and decomposing the trade effect into their extensive and intensive margin component are significant for trade policy analyses. Although restrictive, rules of origin are tools needed to prevent transshipping of goods. On the other hand, RoO that are too restrictive could overturn the preferential tariff treatments conferred by FTA. As tariffs become lower, the benefit of further tariff reductions may be small, but the harmful effect of complex and divergence RoO could be large. Therefore, policymakers need to balance out the two so as to lower administrative costs and promote more efficient implementation. Figure 3.1: No. of PTAs. (Source: WTO)

89 77 Table 3.1: Data for year 1975 FTA & CU FTA, CU, and EU (1) (2) (3) (4) (5) (6) (7) (8) PTA OLS Extensive Intensive Intensive OLS Extensive Intensive (NLS) Distance ** ** ** ** ** ** ** ** (0.041) (0.041) (0.038) (0.051) (0.054) (0.041) (0.038) (0.053) Border ** 1.001** 0.848** ** 0.700** (0.181) (0.181) (0.176) (0.179) (0.181) (0.175) (0.176) (0.180) Language 0.631** 0.626** 0.592** ** 0.592** 0.193* (0.073) (0.073) (0.047) (0.077) (0.079) (0.073) (0.047) (0.078) Colonial ties 1.252** 1.258** ** 0.681** 1.204** ** (0.146) (0.149) (0.401) (0.146) (0.142) (0.146) (0.401) (0.142) PTA (0.202) FTA ** * 0.560** ** (0.188) (0.630) (0.180) (0.186) (0.188) (0.630) (0.191) CU 1.410** 1.746** 1.216** 1.366** 3.679** 1.746** 2.511** (0.419) (0.234) (0.208) (0.207) (0.255) (0.234) (0.284) EU ** * (0.244) (0.192) Regulatory Days & Procedures ** ** ** (0.093) (0.093) (0.062) (0.092) (0.062) Regulatory Costs (0.111) (0.111) (0.059) (0.111) (0.059) δ 1.292** (0.063) η ** 1.175** (0.086) (0.187) (0.188) z 3.730** 3.252** (0.492) (0.494) z ** ** (0.156) (0.156) z ** 0.031* (0.016) (0.016) Test FTA=CU [p<0.001] [p<0.001] [p<0.001] [p<0.001] [p<0.001] [p<0.001] [p<0.001] No. observations 9,776 9,776 20,160 9,776 9,776 9,776 20,130 9,776 R *** indicate significance at the 10, 5, and 1 percent levels. All regressions include importer and exporter fixed effects.

90 78 Table 3.2: Data for year 1985 FTA & CU FTA, CU, and EU (1) (2) (3) (4) (5) (6) (7) (8) PTA OLS Extensive Intensive Intensive OLS Extensive Intensive (NLS) Distance ** ** ** ** ** ** ** ** (0.041) (0.041) (0.038) (0.053) (0.055) (0.041) (0.038) (0.054) Border ** 0.974** 0.801** ** 0.673** (0.179) (0.178) (0.175) (0.178) (0.181) (0.174) (0.175) (0.183) Language 0.627** 0.625** 0.591** ** 0.591** 0.182* (0.073) (0.073) (0.046) (0.078) (0.080) (0.072) (0.046) (0.080) Colonial ties 1.268** 1.281** ** 0.730** 1.232** ** (0.146) (0.149) (0.400) (0.146) (0.141) (0.147) (0.400) (0.143) PTA (0.165) FTA ** * 0.399* ** (0.161) (0.347) (0.156) (0.161) (0.162) (0.347) (0.165) CU 0.968** 1.418** 0.931** 1.108** 2.929** 1.418** 2.016** (0.335) (0.219) (0.214) (0.215) (0.324) (0.219) (0.345) EU ** (0.272) Regulatory Days & Procedures ** ** ** (0.093) (0.092) (0.061) (0.092) (0.061) Regulatory Costs (0.112) (0.112) (0.058) (0.112) (0.058) δ 1.271** (0.065) η ** 1.169** (0.087) (0.188) (0.189) z 3.630** 3.234** (0.497) (0.499) z ** ** (0.157) (0.157) z * 0.030* (0.016) (0.016) Test FTA=CU [p<0.001] [p<0.001] [p<0.001] [p<0.001] [p<0.001] [p<0.001] [p<0.001] No. observations 9,866 9,866 20,449 9,866 9,866 9,866 20,407 9,824 R *** indicate significance at the 10, 5, and 1 percent levels. All regressions include importer and exporter fixed effects.

91 79 Table 3.3: Data for year 1995 FTA & CU FTA, CU, and EU (1) (2) (3) (4) (5) (6) (7) (8) PTA OLS Extensive Intensive Intensive OLS Extensive Intensive (NLS) Distance ** ** ** ** ** ** ** ** (0.032) (0.032) (0.026) (0.038) (0.040) (0.032) (0.026) (0.043) Border 0.805** 0.802** ** 0.915** 0.795** ** (0.143) (0.143) (0.144) (0.134) (0.136) (0.139) (0.144) (0.142) Language 0.800** 0.797** 0.557** 0.331** 0.347** 0.762** 0.557** 0.411** (0.062) (0.062) (0.044) (0.066) (0.065) (0.062) (0.044) (0.066) Colonial ties 1.048** 1.062** ** 1.277** 1.026** ** (0.132) (0.132) (0.221) (0.129) (0.125) (0.130) (0.221) (0.129) PTA 0.371** (0.095) FTA 0.252* 0.598* * (0.113) (0.238) (0.109) (0.109) (0.115) (0.238) (0.110) CU 0.545** 0.874** 0.728** 0.781** 1.438** 0.874** 1.094** (0.129) (0.198) (0.112) (0.112) (0.161) (0.198) (0.158) EU ** (0.141) Regulatory Days & Procedures (0.069) (0.069) (0.051) (0.069) (0.051) Regulatory Costs * * * (0.081) (0.081) (0.049) (0.080) (0.049) δ 0.958** (0.051) η 0.429** 1.354** 1.457** (0.070) (0.153) (0.153) z 4.005** 3.741** (0.384) (0.387) z ** ** (0.121) (0.122) z ** 0.069** (0.012) (0.012) Test FTA=CU [p=0.048] [p=0.367] [p<0.001] [p<0.001] [p<0.001] [p=0.3673] [p<0.001] No. observations 15,362 15,362 28,730 15,362 15,362 15,362 28,574 15,206 R *** indicate significance at the 10, 5, and 1 percent levels. All regressions include importer and exporter fixed effects.

92 80 Table 3.4: Data for year 2005 FTA & CU FTA, CU, and EU (1) (2) (3) (4) (5) (6) (7) (8) PTA OLS Extensive Intensive Intensive OLS Extensive Intensive (NLS) Distance ** ** ** ** ** ** ** ** (0.034) (0.034) (0.034) (0.043) (0.044) (0.034) (0.034) (0.047) Border 0.831** 0.829** ** 0.982** 0.823** ** (0.148) (0.149) (0.210) (0.145) (0.143) (0.148) (0.210) (0.145) Language 1.034** 1.033** 0.595** 0.407** 0.559** 1.027** 0.595** 0.555** (0.060) (0.060) (0.043) (0.066) (0.065) (0.060) (0.043) (0.066) Colonial ties 0.977** 0.980** ** 0.820** 0.948** ** (0.153) (0.154) (0.302) (0.145) (0.138) (0.154) (0.302) (0.140) PTA 0.745** (0.071) FTA 0.729** 0.488** 0.163* 0.396** 0.728** 0.488** 0.391** (0.078) (0.123) (0.083) (0.083) (0.078) (0.123) (0.084) CU 0.774** 0.945** 0.233* 0.591** 0.888** 0.945** 0.571** (0.097) (0.164) (0.094) (0.094) (0.101) (0.164) (0.100) EU * (0.154) Regulatory Days & Procedures (0.075) (0.075) (0.049) (0.075) (0.049) Regulatory Costs * * * (0.081) (0.081) (0.081) (0.046) δ 1.266** (0.053) η 0.528** 1.881** 1.874** (0.085) (0.181) (0.181) z 3.387** 3.404** (0.436) (0.439) z ** ** (0.136) (0.137) z * 0.028* (0.014) (0.014) Test FTA=CU [p=0.659] [p=0.017] [p=.040] [p=0.126] [p=0.017] [p=0.067] No. observations 18,826 18,826 27,543 18,826 18,826 18,826 27,462 18,745 R *** indicate significance at the 10, 5, and 1 percent levels. All regressions include importer and exporter fixed effects.

93 81 Table 3.5: Data for years FTA & CU FTA, CU, and EU (1) (2) (3) (4) (5) (6) (7) PTA OLS Extensive Intensive OLS Extensive Intensive Distance ** ** ** ** ** ** ** (0.027) (0.027) (0.019) (0.028) (0.026) (0.019) (0.021) Border 0.558** 0.557** ** 0.716** 0.557** ** 0.726** (0.129) (0.129) (0.104) (0.121) (0.125) (0.104) (0.082) Language 0.808** 0.803** 0.512** 0.523** 0.784** 0.512** 0.529** (0.048) (0.048) (0.026) (0.047) (0.047) (0.026) (0.035) Colonial ties 1.149** 1.163** ** 1.111** ** (0.125) (0.126) (0.164) (0.117) (0.123) (0.164) (0.073) PTA 0.708** (0.063) FTA 0.543** 0.264* 0.454** 0.525** 0.264* 0.442** (0.065) (0.111) (0.063) (0.066) (0.111) (0.055) CU 0.979** 0.997** 1.060** 1.479** 0.997** 1.319** (0.100) (0.112) (0.086) (0.096) (0.112) (0.069) EU ** (0.152) Regulatory Days & Procedures ** ** * (0.056) (0.056) (0.031) (0.056) (0.031) Regulatory Costs * * * (0.059) (0.059) (0.029) (0.059) (0.029) η 1.443** 1.482** (0.107) (0.089) z 2.435** 2.441** (0.264) (0.223) z ** ** (0.084) (0.071) z (0.009) (0.007) Test FTA=CU [p<0.001] [p<0.001] [p<0.001] [p<0.001] [p<0.001] [p<0.001] No. observations 53,830 53,830 96,882 53,830 53,830 96,573 53,521 R *** indicate significance at the 10, 5, and 1 percent levels. All regressions include importer-year and exporter-year fixed effects.

94 Bibliography Abrams, Richard K., International trade flows under flexible exchange rates, Economic Review, 1980, (Mar), Affendy, Arip M., Sim Yee Lau, and Satoru Madono, Commodity-industry classificationproxy: A correspondence table between SITC revision 2 and ISIC revision 3, MPRA Paper 27626, University Library of Munich, Germany March Aitken, Norman D., TheEffect of the EEC and EFTA on European Trade: A Temporal Cross-Section Analysis, The American Economic Review, 1973, 63 (5), pp Amiti, Mary and Donald R. Davis, Trade, Firms, and Wages: Theory and Evidence, The Review of Economic Studies, Anderson, James E., A Theoretical Foundation for the Gravity Equation, The American Economic Review, 1979,69 (1), pp and Eric van Wincoop, Gravity with Gravitas: A Solution to the Border Puzzle, The American Economic Review, 2003,93 (1), pp Anson, José, Olivier Cadot, Antoni Estevadeordal, Jaime de Melo, Akiko Suwa-Eisenmann, and Bolormaa Tumurchudur, Rules of Origin in North South Preferential Trading Arrangements with an Application to NAFTA, Review of International Economics, 2005, 13 (3), ,,,,, and, Rules of Origin in North South Preferential Trading Arrangements with an Application to NAFTA, Review of International Economics, 2005,13 (3), Baier, Scott L. and Jeffrey H. Bergstrand, On the endogeneity of international trade flows and free trade agreements, Technical Report, Journal of International Economics and, Do free trade agreements actually increase members international trade?, Journal of International Economics, 2007,71 (1),

95 83,, and Michael Feng, Economic Integration Agreements and the Margins of International Trade, Miemo, Barro, Robert J. and JongWha Lee, International data on educational attainment: updates and implications, Oxford Economic Papers, 2001,53 (3), Baumgarten, Daniel, Exporters and the Rise in Wage Inequality Evidence from German Linked Employer-Employee Data, Technical Report Beck, Thorsten, Financial Dependence and International Trade, Review of International Economics, 2003,11 (2), Bekaert, Geert, Campbell R. Harvey, and Christian Lundblad, Does Financial Liberalization Spur Growth?, Working Paper 8245, National Bureau of Economic Research April Bergstrand, Jeffrey H., The Gravity Equation in International Trade: Some Microeconomic Foundations and Empirical Evidence, The Review of Economics and Statistics, 1985,67 (3), pp Bernard, Andrew B., J. Bradford Jensen, Stephen J. Redding, and Peter K. Schott, Firms in International Trade, The Journal of Economic Perspectives, 2007,21 (3), pp Botero, Juan C., Simeon Djankov, Rafael La Porta, Florencio Lopez de Silanes, and Andrei Shleifer, The Regulation of Labor, The Quarterly Journal of Economics, 2004,119 (4), Brada, Josef C. and Jos A. Mndez, Economic Integration among Developed, Developing and Centrally Planned Economies: A Comparative Analysis, The Review of Economics and Statistics, 1985,67 (4), pp Braun, Matias, Financial Contractibility And Asset Hardness, University of California Los Angeles miemo, Caselli, Francesco, Chapter 9 Accounting for Cross-Country Income Differences, in Philippe Aghion and Steven N. Durlauf, eds., Philippe Aghion and Steven N. Durlauf, eds., Vol. 1, PartAofHandbook of Economic Growth, Elsevier, 2005, pp Caves, Douglas W., Laurits R. Christensen, and W. Erwin Diewert, The Economic Theory of Index Numbers and the Measurement of Input, Output, and Productivity, Econometrica, 1982,50 (6), pp Costinot, Arnaud, An Elementary Theory of Comparative Advantage, Econometrica, 2009,77 (4),

96 84, On the origins of comparative advantage, Journal of International Economics, 2009,77 (2), Cuñat, Alejandro and Marc J. Melitz, Volatility, Labor Market Flexibility, and the Pattern of Comparative Advantage, Working Paper 13062, National Bureau of Economic Research April Davidson, Carl, Fredrik Heyman, Steven Matusz, Fredrik Sjöholm, and Susan Chun Zhu, Globalization and Imperfect Labor Market Sorting, Technical Report 2011: Davis, Donald R. and James Harrigan, Good jobs, bad jobs, and trade liberalization, Journal of International Economics, 2011, 84 (1), Djankov, Simeon, Rafael La Porta, Florencio Lopez de Silanes, and Andrei Shleifer, Courts, Quarterly Journal of Economics, 2003, 118 (2), pp Egger, Hartmut and Udo Kreickemeier, Firm Heterogeneity And The Labor Market Effects Of Trade Liberalization, International Economic Review, 2009, 50 (1), Egger, Peter, Mario Larch, Kevin E. Staub, and Rainer Winkelmann, The Trade Effects of Endogenous Preferential Trade Agreements, American Economic Journal: Economic Policy, 2011,3 (3), Estevadeordal, Antoni and Kati Suominen, Paper presented at the IDB- CEPR-DELTA/INRA conference Rules of Origin in Regional Trade Agreements: Conceptual and Empirical Approaches, Washington, Feburary Fajgelbaum, Pablo D., Labor Market Frictions, Firm Growth and International Trade, Mimeo, Feenstra, Robert C., Advanced International Trade: Theory and Evidence, Princeton University Press, and Gordon H. Hanson, The Impact of Outsourcing and High-Technology Capital on Wages: Estimates For the United States, , The Quarterly Journal of Economics, 1999,114 (3), and, Global Production Sharing and Rising Inequality: A Survey of Trade and Wages, Blackwell Publishing Ltd,, Robert E. Lipsey, Haiyan Deng, Alyson C. Ma, and Hengyong Mo, World Trade Flows: , Working Paper 11040, National Bureau of Economic Research January 2005.

97 85 Felbermayr, Gabriel, Julien Prat, and Hans-Jorg Schmerer, Globalization and labor market outcomes: Wage bargaining, search frictions, and firm heterogeneity, Journal of Economic Theory, 2011, 146 (1), Frankel, Jeffrey, Regional Trading Blocs in The World Trading System, Washington, DC: Institute of Internationa Economic, 1997., Ernesto Stein, and Shang-Jin Wei, Trading blocs and the Americas: The natural, the unnatural, and the super-natural, Journal of Development Economics, 1995,47 (1), Frías, Judith A., David S. Kaplan, and Eric A. Verhoogen, Exports and Wage Premia: Evidence from Mexican Employer-Employee Data?, Technical Report Ginarte, Juan C. and Walter G. Park, Determinants of patent rights: A cross-national study, Research Policy, 1997,26 (3), Goldberg, Pinelopi Koujianou, Amit Kumar Khandelwal, Nina Pavcnik, and Petia Topalova, Imported Intermediate Inputs and Domestic Product Growth: Evidence from India, The Quarterly Journal of Economics, 2010,125 (4), and Nina Pavcnik, Distributional Effects of Globalization in Developing Countries, Journal of Economic Literature, 2007,45 (1), Helpman, Elhanan, Trade, FDI, and the Organization of Firms, Journal of Economic Literature, 2006,44 (3), pp and Paul R. Krugman, Market Structure and Foreign Trade, (Cambridge, MA: The MIT Press, 1985), 1985., Marc Melitz, and Yona Rubinstein, Estimating Trade Flows: Trading Partners and Trading Volumes, The Quarterly Journal of Economics, 2008, 123 (2), , Oleg Itskhoki, and Stephen Redding, Inequality and Unemployment in aglobaleconomy, Econometrica, 2010,78 (4), ,, Marc-Andreas Muendler, and Stephen Redding, Trade and Inequality: From Theory to Estimation, Mimeo, Holzner, Christian and Mario Larch, Capacity Constraining Labor Market Frictions in a Global Economy, SSRN elibrary, Hummels, David and Peter J. Klenow, The Variety and Quality of a Nation s Exports, The American Economic Review, 2005,95 (3), pp

98 86 Jan, 1903 Tinbergen, Shaping the world economy; suggestions for an international economic policy., Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi, Governance Matters III: Governance Indicators for 1996, 1998, 2000, and 2002, The World Bank Economic Review, 2004,18 (2), Klein, Michael W., Christoph Moser, and Dieter Urban, The Skill Structure of the Export Wage Premium: Evidence from German Manufacturing, Mimeo, Kohpaiboon, Archanun, Multination Enterprises and Industrial Transformation: Evidence from Thailand, Edward Elgar Publisher, Krishna, Pravin, Jennifer P Poole, and Mine Zeynep Senses, Wage Effects of Trade Reform with Endogenous Worker Mobility, Technical Report July Lawrence, Robert Z. and Matthew J. Slaughter, International Trade and American Wages in the 1980s: Giant Sucking Sound or Small Hiccup?, Brookings Papers on Economic Activity. Microeconomics, 1993,1993 (2), pp Leamer, Edward E, In Search of Stolper-Samuelson Effects on U.S. Wages, Technical Report 5427 January Lee, Jong-Wha and Phillip Lee Swagel, Trade Barriers And Trade Flows Across Countries And Industries, The Review of Economics and Statistics, 2000, 79 (3), Levchenko, Andrei A., Institutional Quality and International Trade, The Review of Economic Studies, 2007,74 (3), pp Magee, Christopher S., Endogenous Preferential Trade Agreements: An Empirical Analysis, The B.E. Journal of Economic Analysis & Policy, 2003, contributions.2 (1), 15. Manova, Kalina, Credit constraints, equity market liberalizations and international trade, Journal of International Economics, 2008,76 (1), Melitz, Marc J., The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity, Econometrica, 2003, 71 (6), Menezes-Filho, Naercio, Marc-Andreas Muendler, and Garey Ramey, The Structure of Worker Compensation in Brazil, with a Comparison to France and the United States, The Review of Economics and Statistics, 2008, 90 (2),

99 87 Munch, Jakob Roland and Jan Rose Skaksen, Human capital and wages in exporting firms, Journal of International Economics, 2008, 75 (2), Nunn, Nathan, Relationship-Specificity, Incomplete Contracts, and the Pattern of Trade, The Quarterly Journal of Economics, 2007,122 (2), Porta, Rafael La, Florencio Lopez de Silanes, and Andrei Shleifer, The Economic Consequences of Legal Origins, Journal of Economic Literature, Forthcoming, 2008., Florencio Lopez-De-Silanes, Andrei Shleifer, and Robert W. Vishny, Legal Determinants of External Finance, The Journal of Finance, 1997, 52 (3), pp , Florencio Lopez?de?Silanes, Andrei Shleifer, and Robert W. Vishny, Law and Finance, Journal of Political Economy, 1998, 106 (6), pp Rajan, Raghuram G. and Luigi Zingales, Financial Dependence and Growth, The American Economic Review, 1998,88 (3), pp Rauch, James E., Networks versus markets in international trade, Journal of International Economics, 1999,48 (1), Romalis, John, Factor Proportions and the Structure of Commodity Trade, The American Economic Review, 2004,94 (1), pp Roy, Jayjit, Do Customs Union Members Engage in More Bilateral Trade than Free-Trade Agreement Members?, Review of International Economics, 2010, 18 (4), Sally, Razeen, Thai Trade Policy: From Non-Discriminatory Liberalisation to FTAs, The World Economy, October 2007, 30 (10), Schank, Thorsten, Claus Schnabel, and Joachim Wagner, Do exporters really pay higher wages? First evidence from German linked employer-employee data, Journal of International Economics, 2007,72 (1), Silva, J. M. C. Santos and Silvana Tenreyro, The Log of Gravity, Review of Economics and Statistics, 2012/04/ , 88 (4), The World Bank, Investment Climate Assessment Update, Report No TH, 2008,June. The WTO, Trade Policy Review, Report by the Secretariat, Thailand, WTO WT/TPR/S/191, 2007.

100 88 Trefler, Daniel, Trade Liberalization and the Theory of Endogenous Protection: An Econometric Study of U.S. Import Policy, Journal of Political Economy, 1993, 101 (1), pp Verhoogen, Eric A., Trade, Quality Upgrading, and Wage Inequality in the Mexican Manufacturing Sector, The Quarterly Journal of Economics, 2008, 123 (2), Vichyanond, Jade, Intellectual Property Protection and Patterns of Trade, Working Papers 1201, Princeton University, Department of Economics, Center for Economic Policy Studies. November 2009.