UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Essays on international trade and labor in Indonesia Liang, Yawen 2016

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

Item Metadata

Download

Media
24-ubc_2016_september_liang_yawen.pdf [ 2.36MB ]
Metadata
JSON: 24-1.0308074.json
JSON-LD: 24-1.0308074-ld.json
RDF/XML (Pretty): 24-1.0308074-rdf.xml
RDF/JSON: 24-1.0308074-rdf.json
Turtle: 24-1.0308074-turtle.txt
N-Triples: 24-1.0308074-rdf-ntriples.txt
Original Record: 24-1.0308074-source.json
Full Text
24-1.0308074-fulltext.txt
Citation
24-1.0308074.ris

Full Text

Essays on International Trade andLabor in IndonesiabyYawen LiangB.A., Dongbei University of Finance and Economics, 2009B.A., The University of Western Ontario, 2009M.A., The University of British Columbia, 2010A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Economics)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)August 2016c© Yawen Liang 2016AbstractThis dissertation studies the impact of international trade on the Indonesia labor market.The first chapter investigates how task trading induce workers to change jobs. To un-derstand the link between international trade and workers’ occupation choices, I propose ageneral equilibrium model with heterogeneous workers self-selecting into different tasks ac-cording to their skill-specific comparative advantage, individual task-specific abilities andtask prices. Task outsourcing from foreign countries acts as a demand shock that influencesworkers’ occupation decisions through changing task prices. The model predicts that occu-pational employments shifts with foreign task demand. I use Indonesian data to estimatethis effect. The main finding is that during the post-opening period (2002-2006), growth inmining goods demanded by foreign countries induced workers to move into manual jobs.The second chapter uses the Indonesia plant level data to examine how importing interme-diate goods affects the demand for highly educated workers within and across production andnon-production occupations categories. We estimate a model of importing and skill-biasedtechnological change in which selection into importing arises due to unobservable heteroge-nous returns from importing. Both instrumental variable regression and marginal treatmenteffect estimates confirm that importing has substantially increased the relative demand foreducated workers within each occupation. In contrast, we do not consistently estimate asignificant impact of importing on the relative demand for non-production workers.The last chapter examines the relationship between trading dynamics of plants and theaggregate skill demand at different margins (reallocation of workers across plants versus theskill composition changes within plants). We find that plants that switched from domesticto trade grew in employment shares. This growth was skill biased for plants that startedimporting, but not for plants that started exporting. Consequently, the growth in size andincrease in the skill intensities of the plants that switched from non-importing to importingincrease in the aggregate demand for skilled workers. Plants that stopped importing orexporting laid off workers, more unskilled workers are involved in this reduction. The plantsthat continue trading grew in size, and the growth was not bias toward workers of any skilltype. Given that always-trading plants are most skill intensive, their growth increase theaggregate skill demand.iiPrefaceChapter 2 Does Importing Intermediates Increase the Demand for Skilled Workers? andChapter 3 Reallocation and Composition Effects of Trade on the Demand for Skilled Workersare joint works with Professor Hiroyuki Kasahara and Professor Joel Rodrigue. I was involvedthroughout each stage of the research: collecting and preparing data, designing model andempirical method, carrying out estimation, organizing and presenting results, and writingseveral subsections of the manuscript.iiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix1 Does Trade Liberalization Induce Occupational Movement? . . . . . . . 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Baseline Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.2.1 Model Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.2.2 Equilibrium in Closed Economy . . . . . . . . . . . . . . . . . . . . . 61.2.3 Equilibrium with Trade of Tasks . . . . . . . . . . . . . . . . . . . . . 71.2.4 Solution to the Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . 71.2.5 Empirical Implications . . . . . . . . . . . . . . . . . . . . . . . . . . 91.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111.3.1 Data Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111.3.2 Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131.3.3 Local Trade Shocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161.4.1 OLS Regressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161.4.2 Instrumental Variable Regressions . . . . . . . . . . . . . . . . . . . . 201.5 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231.5.1 Endogeneity of Supply Shocks . . . . . . . . . . . . . . . . . . . . . . 231.5.2 Include Agricultural Sector . . . . . . . . . . . . . . . . . . . . . . . . 281.6 Counterfactual Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33ivTable of Contents2 Does Importing Intermediates Increase the Demand for Skilled Workers? 352.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352.2 A Simple Model of Importing, Selection and SBTC . . . . . . . . . . . . . . 392.2.1 Selection, SBTC and Skill Demand . . . . . . . . . . . . . . . . . . . 412.2.2 The Marginal Treatment Effect . . . . . . . . . . . . . . . . . . . . . 412.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432.3.1 Data Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432.3.2 Importing and Worker Education . . . . . . . . . . . . . . . . . . . . 442.3.3 Variable Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482.3.4 Instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522.4.1 Benchmark IV Findings . . . . . . . . . . . . . . . . . . . . . . . . . . 522.5 Marginal Treatment Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 642.5.1 Propensity Scores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 642.5.2 Treatment Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652.5.3 Policy Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 702.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 713 Reallocation and Composition Effects of Trade on the Demand for SkilledWorkers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 743.2.1 Data Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 743.2.2 Measure of Skill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 743.3 Entry and Exit of Plants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773.4 Impact of Trade on Skill Demand . . . . . . . . . . . . . . . . . . . . . . . . 793.4.1 Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 793.4.2 Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 813.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92AppendicesA Appendix for Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100A.1 Solve for the Baseline Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 100A.2 Matching ISCO68 with 1990 US Census Occupational Classification . . . . . 103A.3 Task Groups of Occupations . . . . . . . . . . . . . . . . . . . . . . . . . . . 104vTable of ContentsA.4 Measure the Distance to Port (Airport) . . . . . . . . . . . . . . . . . . . . . 104A.5 First Stage Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107A.6 Model with Agricultural Sector . . . . . . . . . . . . . . . . . . . . . . . . . . 107A.6.1 Model Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107A.6.2 Equilibrium with Trade of Tasks . . . . . . . . . . . . . . . . . . . . . 107A.6.3 Solution to the Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . 109A.6.4 Empirical Implications . . . . . . . . . . . . . . . . . . . . . . . . . . 111B Appendix for Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116B.1 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116B.1.1 Manufacturing Plant Data . . . . . . . . . . . . . . . . . . . . . . . . 116B.1.2 Regional Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118B.1.3 Industrial Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118B.2 Estimating MTE and Treatment Effects . . . . . . . . . . . . . . . . . . . . . 121B.3 Estimating Hicks-Neutral Productivity . . . . . . . . . . . . . . . . . . . . . 125B.4 First Differences, IV and Bias . . . . . . . . . . . . . . . . . . . . . . . . . . 131B.5 Capital-Skill Complementarity . . . . . . . . . . . . . . . . . . . . . . . . . . 132B.6 Investigating Differences with Amiti and Cameron (2012) . . . . . . . . . . . 133B.7 Additional Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137viList of Tables1.1 Percentage of Workers, by Migration Status within Five Years . . . . . . . . . 31.2 Means and standard errors of task measures by gender and education . . . . 151.3 Percentage of Workers in Each Task Group by Industry . . . . . . . . . . . . 161.4 OLS Estimates of the Relationships Between Relative Occupation Shares andLabor Supply and Foreign Task Demand Shocks, 1990-1996 . . . . . . . . . . 211.5 OLS Estimates of the Relationships Between Relative Occupation Shares andLabor Supply and Foreign Task Demand Shocks, 2002-2006 . . . . . . . . . . 221.6 IV Estimates of the Relationships Between Relative Occupation Shares andLabor Supply and Foreign Task Demand Shocks, 1990-1996 . . . . . . . . . . 241.7 IV Estimates of the Relationships Between Relative Occupation Shares andLabor Supply and Foreign Task Demand Shocks, 2002-2006 . . . . . . . . . . 251.8 OLS Estimates of the Relationships Between Relative Occupation Shares andLabor Supply and Foreign Task Demand Shocks, 2002-2006 (workers olderthan 18) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261.9 IV Estimates of the Relationships Between Relative Occupation Shares andLabor Supply and Foreign Task Demand Shocks, 2002-2006 (workers olderthan 18) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271.10 IV Estimates of the Relationships Between Relative Occupation Shares andLabor Supply and Foreign Task Demand Shocks, 2002-2006 (including agri-cultural sector) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.1 Importing and Skill Intensity 2006, full sample . . . . . . . . . . . . . . . . . 462.2 A Decomposition of Plant-Level Skill Growth by Import Status . . . . . . . . 482.3 Definitions of the Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492.4 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492.5 Skill Demand Equation Across Occupations . . . . . . . . . . . . . . . . . . . 542.6 Skill Demand Equation for All Workers in Levels . . . . . . . . . . . . . . . . 582.7 Robustness Checks: The Skill Demand Equation in Differences . . . . . . . . 592.8 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 622.9 Importing and Standardized Production . . . . . . . . . . . . . . . . . . . . . 63viiList of Tables2.10 Treatment Effects of Importing on Skill Demand . . . . . . . . . . . . . . . . 682.11 Robustness Check: Treatment Effects of Importing on Skill Demand for Pro-duction Workers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693.1 Decompose by occupation group . . . . . . . . . . . . . . . . . . . . . . . . . 763.2 Decompose Skill Share Changes, by Production Dynamics of Plants . . . . . 783.3 Trade and Skill Intensities of Plants in 2006 . . . . . . . . . . . . . . . . . . . 833.4 Decompose Skill Share Changes, by Importing Dynamics of Plants . . . . . . 843.5 Deompose Skill Share Changes, by Exporting Dynamics of Plants . . . . . . . 853.6 total employment and trade status, first differenced . . . . . . . . . . . . . . . 863.7 skilled and unskilled employment (high-school+) and trade status, first differ-enced . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873.8 skilled and unskilled employment (college+) and trade status, first differenced 883.9 skilled worker share and trade status, first differenced . . . . . . . . . . . . . 90A.1 Occupation Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113A.1 Occupation Groups (Cont.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114A.2 First Stage Regressions Correspond to IV regressions in Table 1.7 . . . . . . . 115B.1 Estimates of Skill Demand Equation . . . . . . . . . . . . . . . . . . . . . . . 123B.2 Bandwidth Choices by Cross-validation . . . . . . . . . . . . . . . . . . . . . 125B.3 Import Decision Model using Logit for the Sample of Production Workers . . 127B.4 Investigating Differences with Amiti and Cameron (2012) . . . . . . . . . . . 134B.5 A Decomposition of Plant-Level Skill Growth by Import Status . . . . . . . . 135B.6 Robustness Checks: Dropping Capital, R&D, and Training . . . . . . . . . . 136B.7 Robustness Checks: Skill Threshold Definitions . . . . . . . . . . . . . . . . . 137B.8 First Stage Results: Import Status . . . . . . . . . . . . . . . . . . . . . . . . 138B.9 First Stage Results: Export Status . . . . . . . . . . . . . . . . . . . . . . . . 139B.10 First Stage Results: Import Status, Large Instrument Set . . . . . . . . . . . 140B.11 Robustness Check: Skill Supply Control . . . . . . . . . . . . . . . . . . . . . 141B.12 Robustness Check: Large Instrument Set . . . . . . . . . . . . . . . . . . . . 142B.13 Robustness Check: Import Intensity . . . . . . . . . . . . . . . . . . . . . . . 143B.14 Robustness Check: TFP Measurement . . . . . . . . . . . . . . . . . . . . . . 144B.15 Robustness Check: Instrumenting Export Status . . . . . . . . . . . . . . . . 145B.16 Robustness Check: Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . 146B.17 Robustness Check: Capital-Skill Complementarity . . . . . . . . . . . . . . . 147B.18 Importing and Standardized Production . . . . . . . . . . . . . . . . . . . . . 148B.19 Exporting, Initial Skill-Levels, and SBTC . . . . . . . . . . . . . . . . . . . . 149viiiList of Figures1.1 Net Export in Indonesia During 1990-2006 . . . . . . . . . . . . . . . . . . . . 41.2 Initial Foreign Shares times Percentage Change in Regional Task Shocks (100×(1− δω)Yˆ ∗rω): 1990-1996 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171.2 Initial Foreign Shares times Percentage Change in Regional Task Shocks (100×(1− δω)Yˆ ∗rω): 1990-1996 (Cont.) . . . . . . . . . . . . . . . . . . . . . . . . . . 181.3 Initial Foreign Shares times Percentage Change in Regional Task Shocks (100×(1− δω)Yˆ ∗rω): 2002-2006 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181.3 Initial Foreign Shares times Percentage Change in Regional Task Shocks (100×(1− δω)Yˆ ∗rω): 2002-2006 (Cont.) . . . . . . . . . . . . . . . . . . . . . . . . . . 191.4 Changes in Occupational Employment Shares in Indonesia (2002-2006) . . . . 322.1 Importing and Skill Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . 402.2 Support of Estimated Propensity Scores . . . . . . . . . . . . . . . . . . . . . 642.3 Estimated MTE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 662.4 Estimated Weights for ATE, TT, TUT, MPRTEs, and PRTE (Dependentvariable: Lps/(Lps + Lpu)) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673.1 Trend of Skilled Worker Share, by Different Skill Measurements . . . . . . . . 75A.1 Process of Measuring Transportation Cost . . . . . . . . . . . . . . . . . . . . 106B.1 Trend of the Average Input and Output Tariff, 1996-2006 . . . . . . . . . . . 119B.2 Change in Tariffs, 1991-2000, Relative to 1991 Level . . . . . . . . . . . . . . 120B.3 Import Airshare and Weight Instruments Across Industries . . . . . . . . . . 121B.4 Estimated MTE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126ixChapter 1Does Trade Liberalization InduceOccupational Movement? ATask-based Analysis UsingIndonesia Data1.1 IntroductionThe effects of trade opening on the employment and wage distributions in developing coun-tries are extensively studied. Researchers found that trade liberalization increases the demandof skilled workers, and provide various explanations to explain the findings. Possible expla-nations behind the link between trade and demand of workers with different skills includecapital inflows (Feenstra and Hanson (1997)), quality upgrading (Verhoogen (2008)), importcompetition (Feenstra and Hanson (1996)), skill biased technology change (Pavcnik (2003),Kasahara et al. (2015)), and trade protection (Goldberg and Pavcnik (2005)). Goldberg andPavcnik (2007) provide a review of studies about globalization and labor market outcomesin developing countries. Despite the different approaches, this literature shares the viewthat workers are different in skills. However, an increasing number of studies emphasize therole of task content of jobs in explaining the relationship between international trade andlabor market outcomes.1 The shift in focus was motivated by the finding that the growthof employment in developed countries is non-monotonic in their workers’ skill levels duringthe 1990s and early 2000s.2 Specifically, the expansion of overall employment in the afore-mentioned period can be decomposed into three components: a rising demand for workersin high-skilled occupations, a shrinking demand for workers in medium-skilled occupations,and, surprisingly, an increase in the demand for workers in low-skilled occupations. This“polarized” pattern cannot be explained by traditional labor models, in which workers are1The “task trade” terminology was introduced by Grossman and Rossi-Hansberg (2008). Task trade studyincludes, though not exclusively, Grossman and Rossi-Hansberg (2008), Baldwin and Robert-Nicoud (2014),Blinder (2006), Blinder (2009), Hanson et al. (2005). Acemoglu and Autor (2011) and Firpo et al. (2011)emphasize the importance of task content of jobs in shaping labor markets.2See Autor and Dorn (2013) for the US, and Goos et al. (2014) for the EU countries.11.1. Introductiononly distinguished by skills but not by the jobs they do. One leading explanation focuseson the task content of work, arguing that the shrinking demand for medium-skilled workersis due to intensive offshoring, because what they typically do—the routine tasks—can becompleted more cheaply, by workers in lower wage countries.3 Because tasks are typicallydefined at the occupation level, trade opening does not simply change the relative demand ofworkers with low or high skills; rather, it influences the demand for workers doing differentjobs. In the context of developing countries, although there are many papers studying the re-lationship between globalization and labor maket outcomes, the effect of trade liberalizationon occupational choices is still under-examined relative to its developed-country counterpart.This paper contributes to the literature by examining the impact that the exposure to foreignmarkets has on the occupational employment in a specific developing country, Indonesia. Tomy knowledge, no other study to date focused on this question.Indonesia is suitable for this study for several reasons. First, it is one of the most unskilledlabor-abundant countries in the world. Unlike the middle-income countries such as Braziland Mexico, there is less concern about Indonesia opening to both more developed andless developed countries. Second, the data I use cover the period 1996-2006, over whichIndonesia experienced extensive trade reform. Indonesia became a member of the WorldTrade Organization (WTO) in late 1995. Following the WTO agreement, it was committedto liberalizing trade in the following 10 years.4 Finally, labor mobility in Indonesia is relativelylow, because it is an archipelago comprising many islands and its dialects vary widely acrossthe country. Table 1.1 presents distribution of the 5-year migration status of working agepopulation in Indonesia in 1995, 2000 and 2005. Across the 31 provinces and more than 220consistently defined regions, over 93 percent population stayed in the same major and minoradministrative units. The variations in trade opening across regions, together with the lowlabor mobility can be used to empirically estimate the effects of international trade.In order to understand the link between international trade and workers’ occupationchoices, I extend the framework developed by Burstein et al. (2015) to allow for internationaltrade. In this general equilibrium model, heterogeneous workers self-select to performingdifferent tasks according to their group-specific comparative advantage (e.g. male workersare more physically adept than female workers), individual task abilities (e.g. IQ) and thetask prices. International trade is in the spirit of Heckscher-Ohlin model in the sense thatthe trade of goods is essentially the trade of tasks being performed to produce the goods. Asa result, trade opening influences workers’ occupation choices through changing the relative3Using the US data, Ebenstein et al. (2014) find that occupations that are exposed to trade or offshoringto low wage countries are associated with wage declines. This result is consistent with the interpretation thatworkers in low income locations are substitutes for U.S. workers, because their competition exerts a downwardpressure on wages in the U.S.4Most trade expansion happened after the economy stabilized after the Asian crisis.21.1. IntroductionTable 1.1: Percentage of Workers, by Migration Status within Five YearsYear 1995 2000 2005same province, same region 93.64 93.73 95.02same province, different region 3.61 3.19 2.57different province 2.69 3.07 2.35abroad 0.07 0.01 0.06Data source: Indonesian Census, provided by IPUMS international.Sample restricted to working age population.prices of tasks. The model predicts that an economy with a positive shock of foreign demandfor a task motivates more workers to perform that task. I classify occupations into manual,routine and cognitive according to their predominant task.5Regions of Indonesia are treated as small open economies. National-wide trade open-ing disproportionately affect locations with different industry compositions. I exploit theregional variation in the foreign task demand shocks and employment dynamics to identifythe effect of trade on occupational employment. I find that during the period of 2002-2006,the employment distribution of workers across cognitive, routine and manual occupations aresignificantly correlated with the foreign demand of these tasks. Workers move away fromroutine-intensive jobs in regions that exhibit large increases in the net export of manual- orcognitive-intensive goods; conversely, more workers move to routine occupations in regionsthat have large increases in net export of routine-intensive goods.6 However, these effectswere insignificant for the 1990-1996 period, when Indonesia was relatively closed to trade.7After estimating the relationship between occupation shares and foreign task demand, Iperform counterfactual analyses to understand how trade liberalization might have influencedthe national employment distribution across occupations in Indonesia. Overall, internationaltrade induced workers to move from routine-intensive and cognitive-intensive jobs to manual-intensive jobs. This pattern is mainly caused by the large growth of the net export of mininggoods, which use manual workers intensively in its production process. As shown in Figure1.1, mining goods account for more more than half of the Indonesia trade expansion.The rest of the paper is organized as follows. Section 1.2 introduces the model which5Rather than grouping the occupations by earnings, I employ task groups. I use this grouping because thetask-earning correspondences are different in developed countries and Indonesia (e.g. routine-intensive jobsare middle wage jobs in the US, but they are highest-paid occupations in Indonesia). Task grouping is morecomparable with the polarization studies about the US, especially in the context of international trade, wheretrade directly influences workers doing different jobs, rather than workers earning different wages.6There is little shift of workers between manual and cognitive jobs. This is because the skill requirementsof these jobs are more divergent.7The estimations base on tariff changes in the two period. The insignificant effect in 1990-1996 is a result ofweak relationship between tariff and trading values, as non-tariff barriers play a significant role in determininginternational trade during that period.31.2. Baseline ModelFigure 1.1: Net Export in Indonesia During 1990-2006Source: United Nations Commodity Trade, arranged by World Integrated Trade Solution (WITS). The In-donesia labor force survey data only allows me to study the occupational employment effect of trade in theperiod before the first dashed line and after the second dashed line.links international trade with workers’ occupation choices. Section 1.3 describes the dataand descriptive statistics. Section 1.4 empirically tests the model implications. Section 1.5tests the robustness of the results. Section 1.6 performs the counterfactual analyses andsection 1.7 concludes.1.2 Baseline ModelIn this section I introduce the baseline model and derive its empirical implications. The basicsetting of this model follows Burstein et al. (2015). I extend their model to allow for tradein tasks. In order to derive a reduced form empirical specification from this model, I limitthe analysis to three tasks. The three-task setting provides a clear and intuitive conclusionabout the relationship between task trading and the occupational employment. Moreover,this model is general enough to accommodate the discussion of employment distribution andthe comparison with related studies on other developed countries8.8For example, Ebenstein et al. (2014) and Firpo et al. (2011).41.2. Baseline Model1.2.1 Model SettingsAt each time t, a single final good Yt is produced by a combination of tasks indexed by ω withω ∈ {1, 2, ...,Ω},9 according to a constant elasticity of substitution (CES) production functionYt =(Ω∑ω=1µ1ρωYρ−1ρωt) ρρ−1, (1.1)where ρ > 0 is the elasticity of substitution across tasks, Yωt ≥ 0 is the output of task ωand µω is the task-augmenting productivity that is assumed to be constant over time. Thereis a continuum of workers indexed by i, each supplies one unit of labor in one of the tasksinelastically. The set of workers is partitioned into skill groups indexed by j = 1, 2, . . . , J ,according to observable characteristics10, and the mass of workers in each group at time t isLjt.A worker i in group j performing task ω generates task outputY iωt = Tjωiωt.That is, a worker’s productivity in ω has a component Tjω determined by the group j to whichhe belongs, and an unobserved task-specific ability iωt. The two components are independentacross workers and between each other for a single worker. The first component Tjω capturesthe comparative advantage of workers in different skill groups across tasks.11 Group j has acomparative advantage (relative to j′) in task ω (relative to ω′) if Tjω/Tjω′ > Tj′ω/Tj′ω′ . Forexample, if the productivity of the high skill group (H) relative to the low skill group (L) issuch that THω > TLω, THω′ > TLω′ and THω/THω′ > TLω/TLω′ , then the high-skilled workersare better at both ω and ω′ tasks, but the high skill group has a larger relative productivity ofω to ω′. The second component iωt captures individual-specific skills in the Ω tasks. Assumeat every time t, that each worker in skill group j draws a vector of task-specific abilitiesit = (i1t, i2t, ..., iΩt) from a multivariate Fre´chet distribution,G(; j) = exp− Ω∑ω=1− θ˜j1−vjω1−vjParameter θ˜j > 1 is the j-specific dispersion of productivities across tasks (dispersion de-creases with it). Parameter vj ∈ [0, 1] captures the correlation of the worker’s productivity9In the empirical sections, ω belongs to the set cognitive, routine, manual. This method of grouping followsAcemoglu and Autor (2011), which is commonly used in polarization studies.10In the empirical sections, these groups are determined by education-gender cells.11Here I implicitly assume that the comparative advantages of skill groups over tasks do not change overtime.51.2. Baseline Modelacross tasks. To simplify the notation, define θj ≡ θ˜j/(1− vj). The Fre´chet distribution as-sumption is made for analytical tractability, as it provides simple expressions for employmentshares. Moreover, under the Fre´chet distribution assumption, the average wage within groupj can be expressed as a CES aggregate of task prices and productivities. Let Pωt be the priceof task ω at time t, a worker i in group j chooses his occupation by maxω{PωtTωjiωt}.1.2.2 Equilibrium in Closed EconomyTask Supply. The assumption that task-specific abilities follow a multivariate Fre´chetdistribution implies that the probability of a worker randomly sampled from skill group jchoosing to work in task ω ispiωjt =(Pωt · Tωj)θj∑Ωω′=1(Pω′t · Tω′j)θj . (1.2)Invoking the law of large numbers, piωjt also measures the fraction of workers in group jthat perform task ω. Equation (1.2) indicates that task sorting depends on PωtTωj , which isthe earnings that someone from skill group j with the mean task-specific ability obtains byworking in ω, relative to the power mean of the earnings for the group over all tasks. Thesensitivity of the employment fraction to price changes or productivity changes increases inθj . When workers’ task-specific skills are less dispersed, or the task skills are more correlated,employment shares are more responsive to the changes in rewards.The average wage of workers in skill group j who work in task ω, as denoted by Wωjt,is the product of task price Pωt, group-specific productivity Tωj and the average individual-specific productivity of all workers in group j that perform task ω. The Fre´chet distributionassumption implies that Wωjt = Wjt withWjt = γj[Ω∑ω=1(Pωt · Tωj)θj] 1θj, (1.3)where γj ≡ Γ(1− 1θj(1−vj))and Γ(·) is the Gamma function.The supply of task ω equals to the total cost of producing it:PωtYωt =J∑j=1WjtLjtpiωjt. (1.4)Task Demand. According to the final good production, the demand of task ω can bederived directly by equating its marginal product to its price. Normalizing the price of the61.2. Baseline Modelfinal good to be one, the task demand becomesYωt = µωYt · P−ρωt (1.5)Equilibrium. To close the model, in any period t, the final goods market clears, whichimplies that the following equation holds:Yt =J∑j=1WjtLjt. (1.6)The equilibrium of this economy is characterized by ΩJ+2Ω+J+1 endogenous variables{Pωt, Yωt, piωjt, Yt,Wjt : ω ∈ {1, 2, ...,Ω}, j ∈ {1, 2, ..., J}} determined by task supply (1.4)where piωjt and Wjt are the results of a Roy-type model described in (1.2) and (1.3), the taskdemand in the closed economy (1.5) and final good market clearing condition (1.6).1.2.3 Equilibrium with Trade of TasksWhen tasks are traded, the task demand becomes the sum of domestic and foreign demand.Foreign demand Y ∗ωt is treated as exogenous due to trade opening and lower task tradingcosts. The demand function with task trading isYωt = µωYt · P−ρωt + Y ∗ωt. (1.7)The equilibrium with task trading is summarized by ΩJ + 2Ω + J + 1 endogenous variables,{Pωt, Yωt, piωjt, Yt,Wjt : ω ∈ {1, 2, ...,Ω}, j ∈ {1, 2, ..., J}}, which are determined by tasksupply (1.4), where piωjt and Wjt are the results of the Roy-type model described in (1.2)and (1.3), the task demand in open economy (1.7) and final good market clearing condition(1.6).1.2.4 Solution to the EquilibriumTreating each region r as a small economy with zero labor mobility, I add subscript r tothe equilibrium conditions described in Sections 2.2 and 2.3. To analyse the impact of tasktrading on occupational employment, I log differentiate the ΩJ + 2Ω + J + 1 equilibriumequations. Define xˆ ≡ ∆x/x = d ln(x). The equilibrium with task trading can be describedby the following linear system71.2. Baseline Modelpˆirωj = θj(Pˆrωt −∑ω′pirω′jt0Pˆrω′) , (1.8)Wˆrj =∑ω′pirω′jt0Pˆrω′ , (1.9)Yˆrω =J∑j′=1φrj′ωt0(Wˆrj′ + Lˆrj′ + pˆirωj′)− Pˆrω, (1.10)Yˆrω = δrωt0(Yˆr − ρPˆrω) + (1− δrωt0)Yˆ ∗rω , (1.11)Yˆr =J∑j′=1ξrj′t0(Wˆrj′ + Lˆrj′), (1.12)where at each time t for region r, φrωjt ≡ WrjtLrjtpirωjt∑Jj′=1Wrj′tLrj′tpirωj′tis the share of group j’searnings in task ω, δrωt ≡ 1 − Y∗rωtYrωtis the share of the local consumption of task ω, andξrjt ≡ WrjtLrjt∑Jj′=1Wrj′tLrj′tis the share of group j’s earnings in total earnings.Substitute (1.8) (1.9) and (1.12) into the change in log of task supply (1.10) and taskdemand (1.11), and equate the resulting expressions of the supply change and demand changeto obtain a linear system of task price changes:1− ρδrωt0 −∑jφrjωt0θj Pˆrω +∑j(δrωt0ξrjt0 − φrjωt0(1− θj))∑ω′pirjω′t0Pˆrω′=∑j(φrjωt0 − δrωt0ξrjt0)Lˆrj − (1− δrωt0)Yˆ ∗rω for ω ∈ {1, 2, ...,Ω}.The solution of this system expresses {Pˆrω}ω∈{1,2,...,Ω} in terms of beginning-of-period shares{δrωt0 , ξrjt0 , φrjωt0}ω∈{1,2,...,Ω},j∈{1,2,...,J} and exogenous shocks {Lˆrj , Yˆ ∗rω}ω∈{1,2,...,Ω},j∈{1,2,...,J}}.Equation (1.8) implies that, for any skill group j, pˆirωj− pˆirω′j = θj(Pˆrω− Pˆrω′). Assumingthat the comparative advantage of skill groups do not change over time, employment sharechange differences are purely driven by the task price change differences. As derived inAppendix A.1, when there are two or three tasks, the solution to this equilibrium systemsatisfies the following propositions.Proposition 1 If an economy satisfies the equilibrium conditions as described in Section2.3 and ρ 6= 1, a positive demand shock in task ω increases the the relative price of ωto ω′, along with ω’s relative employment share. That is, d(Pˆrωj − Pˆrω′j)/dYˆ ∗rω > 0,d(Pˆrωj − Pˆrω′j)/dYˆ ∗rω′ < 0, d(pˆirωj − pˆirω′j)/dYˆ ∗rω > 0 and d(pˆirωj − pˆirω′j)/dYˆ ∗rω′ < 0. The81.2. Baseline Modelinfluence of a third task demand shock can be positive or negative: d(Pˆrωj − Pˆrω′j)/dYˆ ∗rω′′ ≷0 and d(pˆirωj − pˆirω′j)/dYˆ ∗rω′′ ≷ 0 if ω′′ /∈ {ω, ω′}.Proposition 2 The magnitude of the effect of demand shock change (Yˆ ∗rω) on the relativeprice changes(Pˆrωj − Pˆrω′j)and employment changes(pˆirωj − pˆirω′j)are increasing with thebeginning-of-period foreign consumption share of ω and decreasing with the foreign consump-tion shares of ω′; That is, d2(Pˆrωj − Pˆrω′j)/dYˆ ∗rωd(1−σrωt0) > 0, d2(Pˆrωj − Pˆrω′j)/dYˆ ∗rωd(1−σrωt0) < 0, d2(pˆirωj − pˆirω′j)/dYˆ ∗rωd(1−σrωt0) > 0 and d2(pˆirωj − pˆirω′j)/dYˆ ∗rωd(1−σrω′t0) <0.1.2.5 Empirical ImplicationsIn this subsection, I derive the empirical implications of the model discussed in the previoussubsection. A direct implication of the model is on the relationship between internationaltrade and relative task prices, as described by proposition 1 and proposition 2. However, theobserved wages of workers reflects not only the prices of the tasks they perform, but alsotheir observed skill and unobserved task related abilities (W iωjt = PωtTωjiωt). As both taskprices and composition of workers respond to demand shocks (e.g. foreign demand changes),the model does not provide clear implication on the occupational wages. This paper focuson employment effect due to the limitation on acquiring task prices.12I assume that the task-specific productivity distribution parameter θj is the same acrossskill groups, i.e. θj = θj′ = θ for all j, j′, and that the comparative advantage of groups andthe general equilibrium scaling factor are the same across regions. Categorizing the tasks intothree groups, cognitive (C), routine (R) and manual (M), the change in relative employmentof the routine task to cognitive task satisfies the following equation:pˆirRj−pˆirCj =J∑j=1αL,R−Crjt0 Lˆrj+βC,R−Crt0(1−δrCt0)Yˆ ∗rC+βR,R−Crt0 (1−δrRt0)Yˆ ∗rR+βM,R−Crt0 (1−δrMt0)Yˆ ∗rM ,(1.13)where βR,R−Crt0 is positive, βω,R−Crt0is negative, and the sign of the demand shock on the thirdtask M , βM,R−Crt0 , is ambiguous. Because regions in Indonesia are considered as individualsmall open economies, each region reaches its own equilibrium. As a result, the coefficientsare region specific, with the magnitudes depending on the initial status of the regions, in-cluding initial task trade shares ( {1− δrωt0}ω∈{C,R,M}), and initial task employment shares({pirω′jt0}ω∈{C,R,M}).13 I use interaction terms to capture these initial-status effects. The12Gottschalk et al. (2015) derive the trend of task prices through a bounding exercise. Their methodprovides a possible way of analyzing price effect of trade. As it requires adjustment of the baseline model andthe equilibrium conditions, I leave this analysis to future studies to keep the integrity of this paper.13See appendix A.1 for specifications of the coefficients in terms of the model parameters.91.2. Baseline Modelregression specification is thus:pˆirRj − pˆirCj =J∑j=1αL,R−Cj Lˆrj +∑ω∈{C,R,M}βω,R−C(1− δrωt0)Yˆ ∗rω (1.14)+J∑j=1αXL,R−cj∏ω′∈{C,R,M}(1− δrω′t0)Lˆrj+∑ω∈{C,R,M}βδ×ω,R−C∏ω′∈{C,R,M}(1− δrω′t0)Yˆ ∗rω+∑ω∈{C,R,M}βpi×ω,R−C∏ω′ 6=ωpirω′t0 Yˆ∗rω + urj .The specification of the relative employment of routine to manual task is analogous.There are three main challenges in implementing these empirical specifications. First,there are no data directly measuring the amount of tasks. Trade or production data measuresthe value of “goods” being exchanged or produced, but the tasks underlying their productionare conceptual and thus cannot be directly measured. Therefore, I use the following indirectmeasure. Under the assumption that workers’ task-specific productivities are the same acrossindustries, (i.e. an accountant working in a shoe production firm is equally productive if heworks in a coal mining firm) for every industry k, I use the local occupation shares withinthat industry to measure its task intensities (Lrωkt0/Lrkt0). Industrial output multiplied bythis local task intensity is the corresponding task being used.14 Foreign demand of the goodis measured by net export NXk.Second, regional trade data are not available in all industries and they suffer from endo-geneity problem in the sense that local trade changes depend on its own employment changes.In order to measure local shocks, I adopt a technique introduced by Bartik (1991). The ideaof this approach is to use national trade changes in disaggregated industries, interact themwith initial employment shares in those industries at the regional level, and then aggregateup. A region’s initial industrial employment share is used as a prediction of the effect of anational shock on the region.15 This method is widely used to identify the effects of tradeusing regional variations within countries (e.g., Topalova (2007) and Autor et al. (2013)). Toaddress the endogeneity problem, I instrument the industrial trade shocks by changes in theaverage world export tariff weighted by bilateral export values. Combining the task mea-surement within industries and the Bartik formulated regional shocks, the foreign demand of14Local task intensities are used because regions can use different technologies to produce goods.15One issue with this measurement is that there are no industries in the model, and as such this approxi-mation lacks an internally consistent microfoundation.101.3. Datatask ω in region r can be expressed asY ∗rω =∑kLrkt0Lrt0Lrωkt0Lrkt0NXkt1 =∑kLrωkt0Lrt0NXkt1 . (1.15)The change in logarithm of Y ∗rω is therefore equal to:Yˆ ∗rω = ln[∑kLrωkt0Lrt0NXkt1/∑kLrωkt0Lrt0NXkt0](1.16)and the instrumental variable for Yˆ ∗rω iŝTARIFF rω = ln[∑kLrωkt0Lrt0TARIFFkt1/∑kLrωkt0Lrt0TARIFFkt0]. (1.17)Third, the effects of a foreign task shock depend on the region’s beginning-of-period sharesof foreign usage for all tasks ({1− δrω}). Apply (1.15) to getδrω =Y ∗rωYrω=∑k Lrωkt0Y∗k∑k Lrωkt0Yk.Again, the regional variation of the share of task ω being used domestically,δrω, comes fromthe local industrial compositions.1.3 Data1.3.1 Data SourceMy primary data source is the Indonesia labor force surveys (SAKERNAS) conducted be-tween 1990 and 2006. The data between 1999 and 2001 cannot be used, as the locationinformation of individual workers was not reported. The 1997 and 1998 data are excluded asoutliers, because employment, wage and trade values are extremified by the Asian financialcrisis. To get the changes in fundamental variables over the longest possible periods beforeand after 2000, I mainly use the data in 1990, 1996, 2002 and 2006. The period covered inthe analyses coincides with the time frame in which job polarization is observed in developedcountries. The period 1990-1996 and 2002-2006 are pre- and post- liberalization intervals asIndonesia joint WTO after 1996. The surveys cover individual household members aged 10or older nationalwide,16, and document the individuals’ personal characteristics (age, gender,education attainment) employment status, working industries and occupations. This indi-16Diplomatic Corps households, households that are in the specific enumeration area and specific householdsin the regular enumeration area are not sampled.111.3. Datavidual information can be aggregated to region according to the location information fromthe surveyed households. The industries where individuals are employed are recorded bythe 2-digit KLUI classification system in the 1990-1996 data and the 3-digit KBLI classi-fication system in the 2002-2006 data. The KLUI and KBLI classifications originate frominternational classifications ISIC rev.2 and ISIC rev.3, and can be easily matched. The corre-spondence between ISIC rev.2 and ISIC rev.3 is used to get consistent industry classifications.The occupations are coded according to the 1968 version of ISCO (2-digit for 1990-1996 dataand 3-digit for 2002-2006 data). The occupations need to be matched with the US occupa-tion classifications to append a set of standardized job descriptors for each occupation (e.g.,routineness). The details of the matching process are provided in Appendix A.2. In additionto the demographic and working characteristics, wages were also reported in the surveys af-ter 1994, and thus are used to construct the income shares of each skill group within eachoccupation category.The international trade data are from the UN Comtrade Database, which records yearlyimport and export values by industry. Using this commodity trade data, I restrict my analysisto task trading embedded in goods. Although analysing the effects of service trade is ofinterest because a big share of task trade is in the form of services, it is beyond the scope ofthis paper due to data limitations. The tariff data are from the UNCTAD Trade AnalysisInformation System (TRAINS). Both trade and tariff data are organized by World IntegratedTrade Solution (WITS), which provides industrial information classified by 2-digit (up to 4-digit) ISIC rev.3 codes.The input-output table for Indonesia provides the value of output consumed domesticallyand exported by industries. These data, together with the industrial task intensity measuredby employment and the local industrial employment, allow me to generate local task exportingshares (1− δrωt0 in the model). However, the input-output table is only available after 1995,so I compute the industrial foreign usage shares by combining the net export values fromthe UN Comtrade data with the production values from the UN Industrial DevelopmentOrganization (UNIDO) for manufacturing industries.Finally, the transportation costs to regions’ nearest ports and airports are use as additionalIVs. As I do not observe transport costs to the port directly, I construct the measure oftransport costs for each region as follows. To incorporate geographical information, I firstdivide Indonesia into cells of one kilometer squared and assign a value between 1 and 10 toeach cell, where “10” is the highest cost (steepness of slope, sea vs. land). Then, I use ArcGISto find the least cumulative-cost path between any plant and its nearest port. Finally, themeasure of transport cost is obtained from the least cumulative-cost after dividing it by thesample standard deviation. The details related to the construction of this measurement areprovided in Appendix A.4.121.3. Data1.3.2 TasksThis study aims at providing evidence that are relevant to the polarization studies. I dividethe occupations into groups that are commonly used in the polarization literature: cognitive(occupations with a high intensity of abstract thinking tasks or those involving interpersonalcommunication requiring cognitive thinkings, such as, scientists and managers); routine (oc-cupations that require workers to repeatedly perform cognitive or manual works, such as,clerks and assembly workers); and manual (jobs that require finger dexterity or physical ca-pability, but that usually require little complicated thinking, such as, service workers and busdrivers).I use the job descriptors created by Acemoglu and Autor (2011) to categorize occupations.Their measurements are based on the Occupational Information Network (O*NET) data, inwhich each highly disaggregated US job is assigned a list of task scores. As there are numeroustask numbers under each occupation, it is subjective to decide which ones best represent agiven task constructs.17 Using the O*NET data, labor economists have also constructedother task content indexes. For example, Firpo et al. (2011) construct five task measuresincluding: information content, routine, the importance of face-to-face contact, the need foron-site work, and the importance of decision making; Goos et al. (2014) categorize O*NETvariables into cognitive, routine and manual following a method similar as Autor and Dorn(2013) and Acemoglu and Autor (2011), with some differences in the selection of O*NETvariables. An exception is Blinder and Krueger (2013), in which “offshorability” is measuredthrough a survey approach. In the methods that base on the O*NET data, the choices oftask variables creates substantial overlap, and the measures constructed by Acemoglu andAutor (2011) are widely adopted for consistency across studies.18The task content of occupations created by Autor and Acemoglu contains measures ofcognitive, routine and manual tasks with two sub-measures for cognitive and routine tasks. Imatch the US and Indonesia occupations and standardize the scores to have mean zero andcross-occupation standard deviation one across the 75 consistently coded occupations, using2006 occupational employment as weights.19 After standardization, a task score that anoccupation has reflects how intensive that task is for that occupation, relative to the averageintensiveness of that task across all occupations in Indonesia.2017Details about the task score computations, together with the choice of O*NET variables, are provided inAcemoglu and Autor (2011).18The measures of “offshorability” (or “non-offshorability”) emphasize the importance of face-to-face con-tact and presenting at specific locations. These measures are less relevant to this study, due to the limitedinformation on trade in service, and the importance of on-site in goods production.19The 2006 employment levels are computed using the Indonesia Labor Force Survey data. Using employ-ment weights in other years to standardize the scores does not change the list of occupations under eachgroup.20Acemoglu and Autor (2011) standardized their original O*NET scores according to US employment.131.3. DataOne concern of using this task measurement is that the same occupation may includedifferent tasks in different countries. For example, a textile maker in US may work withmachines and is defined as a routine worker, but a textile maker in Indonesia may sew byhands and is defined as a manual worker. Since this paper focus on international trade, it isimportant to use consistent task measures across countries. The results should be interpretedas whether trade liberalization in Indonesia leads workers to move toward “foreign-demanded”jobs.21Table 1.2 displays the task measures by education and gender using the 1990 survey data.It is clear that the task measures are non-monotonic in education levels, especially for femaleworkers. These non-monotonic patterns suggests that the traditional education-based divisionof workers does not suit the analysis when tasks are directly traded or when some specific tasksare needed to produce traded goods. Using these task scores, I categorize the occupationsinto cognitive, routine and manual by their predominant task usage. Appendix A.3 lists theoccupations under each category, ordered by their corresponding task intensities.1.3.3 Local Trade ShocksThe local task trade shocks are computed according to equation (1.16). The task intensitiesof the 37 industries are measured by the employment shares of the workers performing thetasks in the industries. Table 1.3 lists the task intensities by seven aggregated industriesin the two beginning-of-period years. The trade data are mostly in agriculture, mining andmanufacturing, where goods are produced. Trade opening in manufacturing is expected to bea positive shock on routine task demand, as it uses routine workers intensively. For example,the net export of electronic parts is considered to represent Indonesia’s value added to theproducts, and the task content of this job is mostly routine.As described by equation (1.16), local shocks are computed by the changes in log of netexports, where regional net exports are obtained by interacting national value with localindustry shares and aggregating up. Figures 1.2 and 1.3 show the distribution of foreigntask demand shocks across regions in Indonesian during 1990-1996 and 2002-2006.22 Thevariations is largest in Java island, where more than 56 percent of Indonesian populationlive. Notice that during 2002-2006, cognitive task shocks were most concentrated on Javaisland, routine shocks are more equal across islands, and manual task shocks are highest onKalimantan island. This is consistent with the industry distributions in the country.21If measures of task content of jobs based on occupations in developing countries are available, it is straightforward to investigate the substitution of tasks across countries. However, there is little information aboutthe occupation tasks in developing countries.22There were some changes of administrative division in some provinces, which include Kalimantan Timur,Kalimantan Barat, Riau and Sulaweisi Tengah. I only keep the regions that are consistently defined over timeto compute local changes. Also, provinces Maluku Utara, Irian Jaya Barat and Papua are excluded from theLabor force survey, and thus from my analysis.141.3. DataTable 1.2: Means and standard errors of task measures by gender and educationPanel A: Male WorkersAll ≤Jr.SchoolJr. HighSch.High Sch. College &Univ.NON-ROUTINE COGNITIVEnon-routine cognitive analytic -0.004 -0.228 -0.247 -0.019 0.931(0.005) (0.008) (0.009) (0.009) (0.016)non-routine cognitive interpersonal -0.033 -0.160 -0.243 -0.086 0.730(0.006) (0.009) (0.011) (0.009) (0.018)ROUTINEroutine cognitive 0.068 -0.058 0.043 0.238 -0.073(0.006) (0.008) (0.011) (0.010) (0.020)routine manual 0.094 0.436 0.367 0.024 -0.920(0.006) (0.009) (0.011) (0.010) (0.012)NON-ROUTINE MANUALphysical 0.204 0.618 0.564 0.099 -1.017(0.006) (0.009) (0.011) (0.010) (0.013)interpersonal -0.118 -0.448 -0.372 -0.042 0.827(0.006) (0.009) (0.011) (0.009) (0.013)Panel B: Female WorkersNON-ROUTINE COGNITIVEnon-routine cognitive analytic 0.020 -0.605 -0.561 0.195 1.115(0.011) (0.013) (0.019) (0.018) (0.018)non-routine cognitive interpersonal 0.090 -0.276 -0.405 0.109 0.960(0.010) (0.011) (0.018) (0.018) (0.019)ROUTINEroutine cognitive -0.137 -0.335 -0.138 0.157 -0.240(0.010) (0.014) (0.021) (0.019) (0.024)routine manual -0.199 0.194 0.338 -0.294 -1.041(0.010) (0.017) (0.023) (0.018) (0.012)NON-ROUTINE MANUALphysical -0.467 -0.097 -0.014 -0.549 -1.226(0.008) (0.009) (0.013) (0.014) (0.012)interpersonal 0.243 -0.167 -0.284 0.316 1.130(0.010) (0.015) (0.021) (0.017) (0.013)Source: 2006 Indonesia Labor Force Survey. Task measures are constructed by Autor and Acemogluwith construction procedure described in Autor and Acemoglu (2011). The measures are based onO*NET, which are created for US occupations. I match the tasks scores with Indonesia occupationsand standardized the scores to have mean zero and cross-occupation standard deviation oneaccording to Indonesia employment.151.4. Empirical ResultsTable 1.3: Percentage of Workers in Each Task Group by Industry1990 2002Cognitive Routine Manual Cognitive Routine ManualMining 0.0409 0.1271 0.8320 0.0838 0.2182 0.6980(0.0290) (0.0689) (0.0952) (0.0874) (0.1190) (0.1348)Manufacture 0.0381 0.8618 0.1001 0.0567 0.8742 0.0691(0.0230) (0.0903) (0.0780) (0.0298) (0.0839) (0.0747)Construction 0.1050 0.2883 0.6067 0.0582 0.3412 0.6007(0.0674) (0.3085) (0.3748) (0.0186) (0.3893) (0.4032)Wholesale & Retail 0.5023 0.3431 0.1546 0.4377 0.3932 0.1691(0.2836) (0.1971) (0.0869) (0.2096) (0.2673) (0.0746)Transportation & Communication 0.0378 0.3282 0.6340 0.0582 0.3887 0.5532(0.0251) (0.2438) (0.2619) (0.0396) (0.2701) (0.3073)Social and Individual Service 0.2814 0.3952 0.3234 0.3299 0.3791 0.2910(0.2209) (0.2397) (0.3064) (0.2887) (0.2968) (0.3556)Source: 1990 & 2002 Indonesia Labor Force Survey. The original 37 industries are aggregated to 7main groups according to the aggregated industry classification codes.1.4 Empirical ResultsPropositions 1 and 2 predict the effects of demand shocks {Yˆ ∗rω}ω∈{C,R,M} and supply shocks{Lˆrj}j∈{1,2,...,J} on the changes in relative employment {pˆirR − pˆirω}ω∈{C,M}. This sectionpresents the regression results from estimating equation (1.14) to quantify the effects of tradeopening on task employment.A worker is defined as “skilled” if his/her education attainment is high school or aboveand “unskilled” otherwise. This definition places the threshold for high education at a lowerlevel than the studies using data from developed countries, but it is natural given the lowerlevel of education attainment of the country. Indonesian high school graduates are exposedto greater logical and quantitative training (Hendayana et al. (2010)). The skill groups jare defined by gender-skill cells (2 × 2). All of the regressions include non-farming workersbetween 15 to 65 years of age who are not self-employed or unemployed. Employment ismeasured by the number of workers. The relative employment of routine over cognitive androutine over manual, and changes in the two time periods are run separately. All of thechanges are in log.1.4.1 OLS RegressionsI first run the regressions without taking the possible endogeneity problem of trade and supplyshocks into consideration. Table 1.4 and Table 1.5 present the results of these baselineregressions for the 1990-1996 and 2002-2006 periods respectively. The first three columns161.4. Empirical ResultsFigure 1.2: Initial Foreign Shares times Percentage Change in Regional TaskShocks (100× (1− δω)Yˆ ∗rω): 1990-1996(a) Cognitive Task Shock(b) Routine Task Shock171.4. Empirical ResultsFigure 1.2: Initial Foreign Shares times Percentage Change in Regional TaskShocks (100× (1− δω)Yˆ ∗rω): 1990-1996 (Cont.)(c) Manual Task ShockFigure 1.3: Initial Foreign Shares times Percentage Change in Regional TaskShocks (100× (1− δω)Yˆ ∗rω): 2002-2006(a) Cognitive Task Shock181.4. Empirical ResultsFigure 1.3: Initial Foreign Shares times Percentage Change in Regional TaskShocks (100× (1− δω)Yˆ ∗rω): 2002-2006 (Cont.)(b) Routine Task Shock(c) Manual Task Shock191.4. Empirical Resultsuse the log change of the relative employment of routine to cognitive occupation as thedependent variable and the last three columns use the relative employment of routine tomanual occupation. Columns (1) and (4) are the regressions ignoring the possibility that theimpacts of supply shocks depend on foreign usage shares (1− δ), and not controlling for theeffect of the initial share of the third factor.23 In columns (2)-(3) and (5)-(6), I add themgradually.The effects of the foreign task shocks only appear to be significant in the 2002-2006 periodwith signs consistent with the predictions of proposition 1. Focusing on this period, it is clearthat among the three types of shocks, the routine task demand shock measured by trade havesignificant effects on routine-to-manual employment. The impact of cognitive task net exporton routine-to-cognitive employment change is negative, but this effect becomes insignificantonce supply shocks are interacted with initial foreign shares. If a region has all of its routineproduction serving foreign countries, a 1 percent increase in the foreign demand of cognitivetask will decrease the relative employment of routine to cognitive occupations by 0 to 11percent; a 1 percent increase in the foreign demand of routine tasks will increase the relativeemployment of routine to manual occupations by 5 percent. These effects vary with thebeginning-of-period net export shares (1− δ). According to proposition 2, the magnitude ofthe effect of each task shock depends on its initial foreign usage share of all other tasks, butI find this result only holds when the dependent variable is routine to cognitive employmentchange. Columns (2)-(3) together with (5)-(6) suggest that labor moves from cognitive ormanual jobs to routine jobs when foreign countries demand more routine tasks, and labormoves from routine to manual tasks when there is more demand for manual tasks. However,workers do not “move up”: an increase in demand for cognitive tasks does not move laborfrom routine to cognitive job.One puzzle is the insignificant effects of all demand shocks in the period 1990-1996. Thisis mainly caused by the low rate of openness in 1990. As suggested by equation (1.13), thechanges in foreign demand shock of each task ω ∈ {C,R,M} is augmented with the initialshare of a task being used by foreign countries (1−δω). For all the three tasks, the shares aredoubled from 1990 to 2002. On average, more than 99 percent of cognitive and manual tasksand more than 93 percent of routine tasks to be used domestically in 1990. The nationwidecloseness leads to small variations of the openness-augmented task shock measures.1.4.2 Instrumental Variable RegressionsForeign demand of tasks are partly determined by domestic task prices, which are endoge-nously determined with employment in different occupations. As a result, any factors that23The solution to this model suggests that the size of the third task in each region matters for the estimation.Appendix A.1 provides the coefficients in terms of model parameters.201.4. Empirical ResultsTable 1.4: OLS Estimates of the Relationships Between Relative OccupationShares and Labor Supply and Foreign Task Demand Shocks, 1990-1996(1) (2) (3) (4) (5) (6)pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiM pˆiR − pˆiM pˆiR − pˆiM(1− δC)Yˆ ∗C 0.052 0.019 0.020 0.065** 0.133** 0.130**(0.037) (0.059) (0.060) (0.030) (0.052) (0.051)(1− δR)Yˆ ∗R -0.023* -0.020 -0.020 -0.017 -0.036* -0.035*(0.010) (0.017) (0.017) (0.014) (0.020) (0.020)(1− δM )Yˆ ∗M -0.006 -0.023 -0.021 0.062* 0.089** 0.087**(0.022) (0.039) (0.040) (0.031) (0.040) (0.039)(1− δR)(1− δM )× (1− δC)Yˆ ∗C 0.001 0.003 0.003 0.001 0.000 0.000(0.003) (0.002) (0.002) (0.002) (0.002) (0.002)(1− δC)(1− δM )× (1− δR)Yˆ ∗R -0.007 -0.012*** -0.012*** 0.001 -0.002 -0.003(0.004) (0.002) (0.002) (0.004) (0.004) (0.004)(1− δC)(1− δR)× (1− δM )Yˆ ∗M 0.005 0.008** 0.008** -0.002 0.001 0.002(0.003) (0.002) (0.003) (0.003) (0.003) (0.003)Lˆ1 (male-skilled) -0.047 -0.383 -0.376 0.168 0.388 0.367(0.138) (0.246) (0.250) (0.122) (0.245) (0.250)Lˆ2 (male-unskilled) 0.134 0.044 0.036 -0.200* -0.275 -0.262(0.101) (0.142) (0.149) (0.106) (0.140) (0.141)Lˆ3 (female-skilled) 0.126 0.501** 0.502** -0.032 -0.264 -0.253(0.111) (0.202) (0.202) (0.100) (0.184) (0.185)Lˆ4 (female-unskilled) 0.047 0.141 0.140 0.083 0.239* 0.240*(0.070) (0.113) (0.114) (0.074) (0.118) (0.117)Constant -0.196 -0.260 -0.222 -0.290 -0.194 -0.098(0.273) (0.282) (0.349) (0.269) (0.271) (0.286)Lˆ× (1− δ) No Yes Yes No Yes Yes3rd Task Share No No Yes No No YesProv. FE Yes Yes Yes Yes Yes YesGroup FE Yes Yes Yes Yes Yes YesObservations 480 480 480 518 518 518R-squared 0.138 0.184 0.184 0.135 0.166 0.166Std. errors are clustered at region level*** p<0.01, ** p<0.05, * p<0.1211.4. Empirical ResultsTable 1.5: OLS Estimates of the Relationships Between Relative OccupationShares and Labor Supply and Foreign Task Demand Shocks, 2002-2006(1) (2) (3) (4) (5) (6)pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiM pˆiR − pˆiM pˆiR − pˆiM(1− δC)Yˆ ∗C -0.119*** -0.063 -0.066 -0.019 0.036 0.039(0.040) (0.063) (0.060) (0.040) (0.059) (0.061)(1− δR)Yˆ ∗R 0.040 0.045* 0.046* 0.023 0.050** 0.050**(0.027) (0.027) (0.027) (0.025) (0.025) (0.025)(1− δM )Yˆ ∗M 0.005 -0.002 -0.005 -0.004 -0.124*** -0.125***(0.009) (0.035) (0.034) (0.015) (0.042) (0.042)(1− δR)(1− δM )× (1− δC)Yˆ ∗C 0.002 0.002 0.002 -0.002 -0.003 -0.003(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)(1− δC)(1− δM )× (1− δR)Yˆ ∗R -0.004* -0.004* -0.005** 0.005*** 0.004* 0.004*(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)(1− δC)(1− δR)× (1− δM )Yˆ ∗M 0.003* 0.002 0.003 -0.004** -0.001 -0.001(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)Lˆ1 (male-skilled) -0.146 0.180 0.208 -0.190 0.502* 0.512*(0.165) (0.359) (0.359) (0.157) (0.288) (0.292)Lˆ2 (male-unskilled) 0.042 -0.127 -0.226 -0.313** -0.404 -0.410(0.130) (0.220) (0.228) (0.124) (0.268) (0.270)Lˆ3 (female-skilled) 0.086 0.154 0.142 0.277*** -0.011 -0.013(0.121) (0.270) (0.270) (0.099) (0.232) (0.232)Lˆ4 (female-unskilled) 0.031 0.006 0.088 0.244*** 0.271* 0.273*(0.083) (0.179) (0.180) (0.089) (0.157) (0.157)Constant -1.389*** -1.598*** -1.138** -0.088 -0.322 -0.371(0.301) (0.379) (0.459) (0.159) (0.222) (0.274)Lˆ× (1− δ) No Yes Yes No Yes Yes3rd Task Share No No Yes No No YesProv. FE Yes Yes Yes Yes Yes YesGroup FE Yes Yes Yes Yes Yes YesObservations 505 505 505 463 463 463R-squared 0.163 0.195 0.203 0.143 0.181 0.182Std. errors are clustered at region level*** p<0.01, ** p<0.05, * p<0.1221.5. Robustness Checksinfluence the domestic task prices will change both occupational employment and foreigndemand of tasks, leading to an endogeneity problem. To address this problem, I use theset of tariff changes described by (1.17) as instruments. The industrial tariffs are averagesof the tariffs that foreign countries charge on Indonesian exports, weighted by the exportvalues. Tariff charged by foreign countries are either set to fulfill international agreements orto accommodate foreign economic environment, making it not likely to be directly correlatedwith domestic occupational employment.Table 1.6 and table 1.7 present the IV estimations for the periods 1990-1996 and 2002-2006. The first stage regressions suggest that the local task tariffs are significantly negativelycorrelated with the local task trade in 2002-2006, but the relationship is weak in 1990-1996.24 The weak relationship between tariff and trade values during the pre-opening periodis probably caused by non-tariff barriers that prevents trade. Due to this weak IV problem,I focus on the results in 2002-2006.Columns (1) to (3) in Table 1.7 use the log change of the relative employment of routineto cognitive occupation as dependent variable and columns (4) to (6) use the relative em-ployment of routine to manual occupation as dependent variable. Controls of the interactionbetween foreign demand shocks and initial conditions are gradually added from column (1)to column (3), and from column (4) to column (6). The results suggest that foreign demandof routine task is positively related with the relative employment of routine jobs to cognitivejobs, and routine jobs to manual jobs. The impact of foreign cognitive (manual) task demandon routine-to-cognitive (routine-to-manual) employment change is negative. If all tasks wereinitially produced for foreign countries (δω = 0 ∀ω{C,R,M}), 1 percent increase in theforeign routine task increases the routine-to-cognitive employment by 12-15 percent, and in-creases the routine-to-manual employment by 25-28 percent. An increase in foreign demandof cognitive task and manual task reduces the routine-to-cognitive employment and routine-to-manual employment, respectively. Foreign demand of the third tasks (manual for columns(1)-(3), and cognitive for columns (4-6)) do not seem to influence the relative employment ofthe other two.1.5 Robustness Checks1.5.1 Endogeneity of Supply ShocksAs task prices change, workers may change their education decisions accordingly. If this isthe case, the estimates are biased since Lˆω is endogenous. In order to address this problem, Irestrict the sample to be workers aged over 18. Since high school decisions are already made24The first stage results are provided in table A.2231.5. Robustness ChecksTable 1.6: IV Estimates of the Relationships Between Relative Occupation Sharesand Labor Supply and Foreign Task Demand Shocks, 1990-1996(1) (2) (3) (4) (5) (6)pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiM pˆiR − pˆiM pˆiR − pˆiM(1− δC)Yˆ ∗C 0.016 0.515 0.527 0.360 0.783* 0.807*(0.461) (0.324) (0.348) (0.274) (0.424) (0.476)(1− δR)Yˆ ∗R -0.274* -0.081 -0.084 -0.147* -0.043 -0.041(0.145) (0.058) (0.057) (0.084) (0.055) (0.056)(1− δM )Yˆ ∗M -0.057** -0.034 -0.027 0.051 0.085** 0.089**(0.025) (0.038) (0.041) (0.034) (0.037) (0.039)(1− δR)(1− δM )× (1− δC)Yˆ ∗C 0.002 0.002 0.004 0.008 0.011 0.011(0.011) (0.007) (0.007) (0.006) (0.007) (0.008)(1− δC)(1− δM )× (1− δR)Yˆ ∗R -0.018* -0.012*** -0.013*** -0.009 -0.011* -0.011*(0.010) (0.004) (0.004) (0.006) (0.006) (0.006)(1− δC)(1− δR)× (1− δM )Yˆ ∗M 0.018* 0.011** 0.011** 0.001 0.002 0.002(0.009) (0.005) (0.005) (0.005) (0.006) (0.006)Lˆ1 (male-skilled) -0.077 -0.498* -0.471* 0.238 0.291 0.330(0.234) (0.267) (0.272) (0.149) (0.279) (0.270)Lˆ2 (male-unskilled) 0.281 0.122 0.085 -0.222* -0.263 -0.289(0.212) (0.171) (0.199) (0.125) (0.190) (0.205)Lˆ3 (female-skilled) 0.257 0.532** 0.547** -0.027 -0.333 -0.362(0.219) (0.252) (0.247) (0.130) (0.225) (0.228)Lˆ4 (female-unskilled) -0.212 0.167 0.164 -0.025 0.180 0.173(0.213) (0.130) (0.135) (0.101) (0.154) (0.160)Constant 0.268 -0.174 0.010 -0.066 -0.359 -0.540(0.343) (0.316) (0.324) (0.264) (0.335) (0.521)Lˆ× (1− δ) No Yes Yes No Yes Yes3rd Task Share No No Yes No No YesProv. FE Yes Yes Yes Yes Yes YesGroup FE Yes Yes Yes Yes Yes YesObservations 471 471 471 510 510 510Std. errors are clustered at region level*** p<0.01, ** p<0.05, * p<0.1241.5. Robustness ChecksTable 1.7: IV Estimates of the Relationships Between Relative Occupation Sharesand Labor Supply and Foreign Task Demand Shocks, 2002-2006(1) (2) (3) (4) (5) (6)pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiM pˆiR − pˆiM pˆiR − pˆiM(1− δC)Yˆ ∗C -0.261** -0.250 -0.264* -0.081 -0.045 -0.025(0.105) (0.165) (0.140) (0.070) (0.153) (0.150)(1− δR)Yˆ ∗R 0.122* 0.123 0.154* 0.216** 0.295*** 0.280***(0.070) (0.088) (0.085) (0.087) (0.109) (0.103)(1− δM )Yˆ ∗M -0.021 0.042 -0.013 -0.033 -0.367** -0.336**(0.027) (0.238) (0.213) (0.055) (0.180) (0.156)(1− δR)(1− δM )× (1− δC)Yˆ ∗C -0.001 -0.010 -0.005 -0.012 -0.012 -0.010(0.015) (0.017) (0.015) (0.012) (0.017) (0.015)(1− δC)(1− δM )× (1− δR)Yˆ ∗R 0.000 0.010 0.004 0.012 0.012 0.010(0.012) (0.017) (0.015) (0.011) (0.017) (0.015)(1− δC)(1− δR)× (1− δC)Yˆ ∗M 0.004 0.004 0.005 0.000 0.003 0.002(0.006) (0.006) (0.005) (0.005) (0.005) (0.005)Lˆ1 (male-skilled) -0.263 0.131 0.200 -0.340** 0.700** 0.725**(0.173) (0.394) (0.390) (0.158) (0.350) (0.349)Lˆ2 (male-unskilled) -0.005 -0.328 -0.388 -0.481*** -0.518 -0.519*(0.162) (0.278) (0.264) (0.169) (0.316) (0.309)Lˆ3 (female-skilled) 0.133 0.144 0.128 0.352*** -0.164 -0.157(0.121) (0.274) (0.269) (0.100) (0.261) (0.253)Lˆ4 (female-unskilled) 0.073 0.187 0.242 0.340** 0.403** 0.394**(0.120) (0.215) (0.204) (0.133) (0.201) (0.192)Constant -1.489*** -1.619*** -1.271** -0.380 -0.696* -0.852**(0.307) (0.471) (0.513) (0.273) (0.359) (0.423)Lˆ× (1− δ) No Yes Yes No Yes Yes3rd Task Share No No Yes No No YesProv. FE Yes Yes Yes Yes Yes YesGroup FE Yes Yes Yes Yes Yes YesObservations 493 493 493 452 452 452Std. errors are clustered at region level*** p<0.01, ** p<0.05, * p<0.1251.5. Robustness Checksat a younger age, this older age group are less likely to change their education decisions whentask prices change.The results for 1990-1996 period are still insignificant and thus not provided. Table 1.8and Table 1.9 present the OLS and IV estimation results the restricted sample. Comparedwith table 1.7, the results are not qualitatively different except that routine demand fromforeign do not move workers from cognitive tasks to routine tasks in a significant fashion.Table 1.8: OLS Estimates of the Relationships Between Relative OccupationShares and Labor Supply and Foreign Task Demand Shocks, 2002-2006 (workersolder than 18)(1) (2) (3) (4) (5) (6)pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiM pˆiR − pˆiM pˆiR − pˆiM(1− δC)Yˆ ∗C -0.060 -0.019 -0.025 -0.026 -0.017 -0.015(0.045) (0.046) (0.044) (0.033) (0.043) (0.044)(1− δR)Yˆ ∗R 0.026 0.035 0.036 0.027 0.051** 0.052**(0.034) (0.032) (0.033) (0.023) (0.024) (0.024)(1− δM )Yˆ ∗M 0.006 0.033 0.023 -0.007 -0.100*** -0.101***(0.009) (0.036) (0.033) (0.017) (0.037) (0.037)(1− δR)(1− δM )× (1− δC)Yˆ ∗C 0.000 0.001 0.001 -0.003 -0.003 -0.003(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)(1− δC)(1− δM )× (1− δR)Yˆ ∗R -0.002 -0.003 -0.004 0.006*** 0.005* 0.005*(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)(1− δC)(1− δR)× (1− δM )Yˆ ∗M 0.003 0.001 0.002 -0.003 -0.001 -0.001(0.002) (0.002) (0.002) (0.002) (0.003) (0.003)Lˆ1 (male-skilled) -0.182 0.073 0.089 -0.240 0.350 0.359(0.169) (0.365) (0.362) (0.159) (0.295) (0.301)Lˆ2 (male-unskilled) 0.007 -0.108 -0.195 -0.358*** -0.388 -0.394(0.129) (0.227) (0.228) (0.127) (0.276) (0.277)Lˆ3 (female-skilled) 0.068 0.244 0.227 0.312*** 0.008 0.005(0.123) (0.264) (0.264) (0.099) (0.210) (0.210)Lˆ4 (female-unskilled) 0.052 0.013 0.082 0.242*** 0.333** 0.334**(0.081) (0.163) (0.160) (0.090) (0.154) (0.154)Constant -1.331*** -1.614*** -1.121** 0.158 -0.201 -0.245(0.293) (0.438) (0.511) (0.214) (0.231) (0.285)Lˆ× (1− δ) No Yes Yes No Yes Yes3rd Task Share No No Yes No No YesProv. FE Yes Yes Yes Yes Yes YesGroup FE Yes Yes Yes Yes Yes YesObservations 495 495 495 451 451 451R-squared 0.153 0.194 0.204 0.165 0.206 0.206Std. errors are clustered at region level*** p<0.01, ** p<0.05, * p<0.1261.5. Robustness ChecksTable 1.9: IV Estimates of the Relationships Between Relative Occupation Sharesand Labor Supply and Foreign Task Demand Shocks, 2002-2006 (workers olderthan 18)(1) (2) (3) (4) (5) (6)pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiM pˆiR − pˆiM pˆiR − pˆiM(1− δC)Yˆ ∗C -0.154 -0.219 -0.229* -0.123** -0.101 -0.080(0.136) (0.144) (0.126) (0.059) (0.089) (0.088)(1− δR)Yˆ ∗R 0.089 0.076 0.107 0.165** 0.271*** 0.251***(0.081) (0.084) (0.081) (0.082) (0.098) (0.095)(1− δM )Yˆ ∗M -0.024 0.081 0.057 -0.049 -0.383*** -0.337**(0.030) (0.166) (0.160) (0.064) (0.141) (0.131)(1− δR)(1− δM )× (1− δC)Yˆ ∗C -0.013 -0.003 -0.000 -0.005 -0.004 -0.002(0.016) (0.012) (0.010) (0.010) (0.013) (0.012)(1− δC)(1− δM )× (1− δR)Yˆ ∗R 0.010 0.006 0.002 0.004 0.004 0.001(0.013) (0.012) (0.011) (0.008) (0.012) (0.011)(1− δC)(1− δR)× (1− δM )Yˆ ∗M 0.007 -0.003 -0.001 0.001 0.004 0.003(0.007) (0.004) (0.004) (0.005) (0.006) (0.005)Lˆ1 (male-skilled) -0.324* 0.000 0.080 -0.338** 0.712** 0.720**(0.187) (0.402) (0.392) (0.158) (0.361) (0.367)Lˆ2 (male-unskilled) -0.074 -0.281 -0.379 -0.460*** -0.620* -0.603*(0.177) (0.258) (0.251) (0.172) (0.328) (0.322)Lˆ3 (female-skilled) 0.172 0.259 0.249 0.398*** -0.061 -0.065(0.125) (0.271) (0.271) (0.112) (0.249) (0.239)Lˆ4 (female-unskilled) 0.047 0.208 0.273 0.300** 0.512** 0.489**(0.120) (0.208) (0.198) (0.128) (0.219) (0.211)Constant -1.388*** -1.637*** -1.257** -0.062 -0.580 -0.748(0.316) (0.570) (0.609) (0.282) (0.448) (0.500)Lˆ× (1− δ) No Yes Yes No Yes Yes3rd Task Share No No Yes No No YesProv. FE Yes Yes Yes Yes Yes YesGroup FE Yes Yes Yes Yes Yes YesObservations 487 487 487 443 443 443Std. errors are clustered at region level*** p<0.01, ** p<0.05, * p<0.1271.5. Robustness Checks1.5.2 Include Agricultural SectorTo address concerns about the omission of agricultural workers and goods, I extend the modelto allow for the production of agricultural goods and occupations. Specifically, let there bea measure one of workers with homogeneous preferences. At any time t, the representativeworker’s optimization problem ismax{Y Nt ,Y At }α ln(Y Nt ) + (1− α) ln(Y At ) (1.18)s.t PAt YAt + YNt ≤ Ytwhere Y Nt is the consumption of non-agricultural goods and YAt is the consumption of agri-cultural goods, PAt is the relative price of agricultural goods, and the non-agricultural goodis taken to be the numeraire. The production of non-agricultural goods is produced by non-agricultural tasks ΩN = {1, 2, . . . ,Ω} according to a CES production function as describedby (1.1). The agricultural good is simply produced by agricultural task A. Domestic taskdemands are derived by equating marginal utility with marginal cost and the task supplyis same as the baseline model. Suppose there are three non-agricultural tasks: cognitive,routine and manual (ΩN = {C,R,M}), and treat each region r as a small open economy, theempirical specification derived from the model is25pˆirRj − pˆirCj =J∑j=1αL,R−Cj Lˆrj +∑ω∈ΩNβω,R−C(1− δrωt0)Yˆ ∗rω + βA,R−C Yˆ ∗rA (1.19)+J∑j=1αXL,R−cj∏ω′∈ΩN(1− δrω′t0)Lˆrj+∑ω∈{ΩN ,A}βδXω,R−C∏ω′∈ΩN(1− δrω′t0)Yˆ ∗rω+∑ω∈{ΩN ,A}βpiXω,R−C∏ω′ 6=ωpirω′t0 Yˆ∗rω + urj .Compared with the specification derived from the baseline model (1.14), there are extraterms that capture the impact of foreign demand for agricultural task. Directly running theregression (1.14) without these terms may suffer from omitted variable problem. However,as long as the instrumental variable are uncorrelated with the shock of foreign demand foragricultural task, the two stage least square results are still valid. To verify this, I re-runthe regressions according to (1.19) and use the same instrumental variables for the net ex-ports of non-agricultural tasks. As discussed in section 4.2, the instruments are weak in the25See Appendix A.6 for model details. Analogous expression holds for pˆirR − pˆirM .281.6. Counterfactual Analysispre-opening period 1990-1996, probably because there are trade barriers other than tariff,so I focus on the post-opening period in which tariff changes are significantly negativelycorrelated with net export changes. The regression results displayed in table 1.10 are con-sistent with those in table 1.7 where agricultural sector were not considered, except that theimpact of task trading on the relative employment of routine over cognitive task switchesfrom weakly significant to insignificant. Again, the impact of foreign task demand on therelative employment is more significant for routine-to-manual employment. One explanationof this result is that the trade-induced technology change substitute routine workers butcomplements cognitive workers. If regions more exposed to trade of routine-intensive goodsget more of such technological growth, the effect of routine-task trade as a demand shockand its technology effect will cancel out, leaving the relationship between trade and relativeemployment ambiguous.1.6 Counterfactual AnalysisThis section uses the occupation effect of trade identified in section 1.4 to perform counter-factual analysis. The analysis aims to addressing the question: How did trade liberalizationin 2002-2006 influence the national employment distribution across occupations in Indonesia?First, I compute the effect of trade on the occupational employment distribution for eachregion (r) and gender-skill cell (j). Define the relative employments as piRCrjt ≡ LRrjt/LCrjt andpiRMrjt ≡ LRrjt/LMrjt. Then we have the employment shares of cognitive, routine and manualoccupations to bepiCrjt ≡ LCrjt/Lrjt =piRMrjtpiRCrjt + piRMrjt + piRCrjt piRMrjt,piRrjt ≡ LRrjt/Lrjt =piRCrjt piRMrjtpiRCrjt + piRMrjt + piRCrjt piRMrjt,andpiMrjt ≡ LMrjt/Lrjt =piRCrjtpiRCrjt + piRMrjt + piRCrjt piRMrjt.For the period 2002-2006, I compute the predicted growth of the relative employment̂∆ ln(piRCrj)and̂∆ ln(piRMrj)after running the regressions specified in columns (3) and (6) in table 1.7.The predicted employment shares of the three occupations in 2006 are thenĈrj06 =R̂Mrj06R̂Crj06 +R̂Mrj06 +R̂Crj06R̂Mrj06,291.6. Counterfactual AnalysisTable 1.10: IV Estimates of the Relationships Between Relative OccupationShares and Labor Supply and Foreign Task Demand Shocks, 2002-2006 (includingagricultural sector)(1) (2) (3) (4) (5) (6)pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiC pˆiR − pˆiM pˆiR − pˆiM pˆiR − pˆiM(1− δC)Yˆ ∗C -0.346 -0.198 -0.213 -0.057 -0.051 -0.044(0.251) (0.189) (0.175) (0.077) (0.127) (0.127)(1− δR)Yˆ ∗R 0.134 0.130 0.156* 0.203** 0.267** 0.248**(0.110) (0.083) (0.083) (0.086) (0.107) (0.099)(1− δM )Yˆ ∗M -0.026 -0.137 -0.161 -0.033 -0.423* -0.366*(0.035) (0.218) (0.217) (0.054) (0.224) (0.201)(1− δR)(1− δM )× (1− δC)Yˆ ∗C 0.010 -0.018 -0.014 -0.008 -0.007 -0.006(0.040) (0.020) (0.020) (0.010) (0.008) (0.008)(1− δC)(1− δM )× (1− δR)Yˆ ∗R -0.007 0.001 -0.001 0.022* -0.002 -0.000(0.024) (0.009) (0.009) (0.012) (0.012) (0.011)(1− δC)(1− δR)× (1− δM )Yˆ ∗M -0.002 0.016 0.015 -0.008 0.007 0.005(0.020) (0.015) (0.014) (0.007) (0.009) (0.008)Yˆ ∗A -0.241 0.385 0.331 0.251 0.148 0.137(0.985) (0.515) (0.488) (0.336) (0.302) (0.279)(1− δC)(1− δR)(1− δM )× Yˆ ∗A -0.000 0.001 0.001 -0.001** 0.001 0.000(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)Lˆ1 (male-skilled) -0.221 0.362 0.402 -0.277* 0.680* 0.658*(0.288) (0.422) (0.423) (0.155) (0.372) (0.364)Lˆ2 (male-unskilled) 0.036 -0.485 -0.541* -0.476*** -0.530* -0.523*(0.169) (0.303) (0.294) (0.171) (0.306) (0.297)Lˆ3 (female-skilled) 0.110 -0.082 -0.074 0.320*** -0.203 -0.176(0.220) (0.308) (0.303) (0.105) (0.262) (0.251)Lˆ4 (female-unskilled) 0.107 0.201 0.272 0.395*** 0.342* 0.344*(0.228) (0.211) (0.212) (0.135) (0.205) (0.197)Constant -1.103 -2.329** -1.928* -0.935 -0.986* -1.034*(1.806) (1.000) (1.022) (0.643) (0.566) (0.555)Observations 495 495 495 454 454 454Prov. FE Yes Yes Yes Yes Yes YesGroup FE Yes Yes Yes Yes Yes YesdlnxXdelta No Yes Yes No Yes YesShare of others No No Yes No No YesStd. errors are clustered at region level*** p<0.01, ** p<0.05, * p<0.1301.6. Counterfactual AnalysisR̂rj06 =R̂Crj06R̂Mrj06R̂Crj06 +R̂Mrj06 +R̂Crj06R̂Mrj06,andM̂rj06 =R̂Crj06R̂Crj06 +R̂Mrj06 +R̂Crj06R̂Mrj06,where R̂ωrj06 =(1 +̂∆ ln(piRωrj))piRωrj96 for ω ∈ {C,M}. The predicted employment of anoccupation ω, L̂ωrj06 =ω̂rj06Lrj06, can be computed with observed trade shocks L̂ωrj06(Y∗06 =Y ∗06), or zero trade shocks L̂ωrj06(Y∗06 = Y∗96).Then, I aggregate up the predicted employment with and without trade to get the oc-cupational employment distribution changes caused by trade at the national level. For eachoccupation ω, the trade induced occupation share change is:∑r∑j L̂ωrj06(Y∗06 = Y∗06)∑r∑j Lrj06−∑r∑j L̂ωrj06(Y∗06 = Y∗96)∑r∑j Lrj06. (1.20)During 2002-2006, Indonesia’s trade with China, which joined the WTO in 2002, accountsfor more than 60% of the growth in Indonesia’s trade with other developing countries. Thesignificant growth of trade with China plays an important role in shaping the overall effectof trade on Indonesia’s distribution of occupational employment. Figuring out the effectof trading with China helps us understand Indonesia’s labor market response to the newlyforged ties with a growing large open economy. Moreover, the result can be compared withother studies that investigate the occupation effect of developed countries trading with largedeveloping countries. The occupational employment caused by trading with China is com-puted in the same way as described by equation (1.20), with the counterfactual value of the2006 trading with China set to its 1996 level.Figure 1.4 presents the overall employment changes of the three occupation groups to-gether with the estimated changes caused by trade expansion, and by trade with China alone,in the 2002-2006 period in Indonesia. Overall, international trade induced workers to movefrom routine-intensive and cognitive-intensive jobs to manual-intensive jobs. This is patternis mainly caused by the large growth of the net export of mining goods, which uses manualworkers intensively in the production process. Figure 1.4 also presents the changes in theshare of occupational employment attributable to the trade with China alone. Similar tothe developed countries, trading with China reduced the share of workers performing rou-tine tasks, which are intensively used in the production of manufacturing goods, but this“polarization” effect is more toward the manual jobs.2626Autor et al. (2013) found the negative effect of trading with China on the share of manufacturing workers311.6. Counterfactual AnalysisFigure 1.4: Changes in Occupational Employment Shares in Indonesia (2002-2006)Source: Indonesia labor force survey. The data include all non-farming workers aged 15-65, excluding thosewho were unemployed or self-employed. Occupations are grouped into cognitive, routine and manual accordingto A.3.The residual employment share changes include the effects of domestic demand changesand the trade effects that are not captured in this model. Possible domestic demand changesthat contribute to the residual part include industry reallocation (expansion of wholesale andretail sector, social service sector), increase in GDP per capita (demand goods with higherquality, or increase service goods demand), and technology change (less manual workers areneed to produce the same goods). Since the regression analysis in section 1.4 is based on thevariations across regions, it captures the employment differences of individual regions causedby different trade shocks. The model, as well as the empirical exercises, treat regions asindependent small open economies. As a result, other nation-wide general equilibrium effectsof trade opening are omitted. Another occupational effect of global economic integration thatis counted as residual is trade in services. While service activities were conventionally viewedas nontradable, the literature on service trade is expanding.27 With little information aboutservice trade in Indonesia, I restrict the foreign task demand shocks to be those induced byin the U.S.27For example, Amiti and Wei (2006) find that service trade has significant positive effect on productivityin US; Jensen et al. (2005) measure the tradability of service goods and occupations, and find higher workerdisplacement rate in tradable services using US data.321.7. Conclusiontrade in goods.1.7 ConclusionThe studies on how trade influences labor market dynamics have come a long way. In thepast, many researchers investigated the skill upgrading effect of trade opening in develop-ing countries. The researchers found an increasing demand for skilled workers following anopening up to trade. Although the papers offer different reasons for such a phenomenon, acommon thread through this literature is its focus on trade in goods and the classification ofworkers into the skilled versus the unskilled. However, since the mid-2000s, an expanding listof economists have argued that the nature of international trade has been changing. Ratherthan exchanging goods, the values of goods are now being added in many locations, a mode ofproduction known as the “trade in tasks”. One important consequence of this change is thattrade liberalization directly influence the demand of workers doing different jobs, rather thanthe demand of workers with different skills. There are some studies about the relationshipbetween trade opening and occupations of workers in developed countries, my paper looks atthe effect task trading on the occupational decisions of workers from a developing country’sprospect.To delineate the relationship between task trading and occupation choices, I proposea general equilibrium model with a Roy-type assignment component. The workers self-select into tasks according to their comparative advantage, individual task abilities and taskprices. Task outsourcing from foreign countries acts as a demand shock that changes workers’occupation decisions through task prices. Occupations are classified into three categoriesaccording to their predominant tasks: cognitive, routine and manual. The model has clearempirical implications for the effects of foreign task demand shocks on local labor markets.In particular, the model suggests that regions experiencing a larger increase in the foreigndemand of a task will have more workers performing that task, and the size of this effect isrelated to the region’s initial level of openness and occupational employments.I use the Indonesian data from the 1990-2006 period to test the model’s empirical im-plications. I identify the effects of task trading using regional variations in the changes ofoccupation shares and in the exposure to various task trades. The main findings are thatduring the pre-opening period, trade had insignificant effects on the occupational choices;however, in the post opening period, task trading led workers to switch jobs according tothe demands from foreign countries. In Indonesia, after trade opening, the growing demandfor manual-task-intensive mining goods induces workers to move to manual jobs, especiallyfrom routine-task intensive occupations. Trading with China reduced the share of workersin routine-intensive occupations. Meanwhile, since other foreign countries’ demand for man-331.7. Conclusionufacturing goods did not expand much, workers move away from routine jobs into the moredemanded manual jobs, and a small proportion of them move into cognitive jobs. Thesefindings are consistent with the main predictions of my model, and they are in line with thestudies on developed countries that show that task offshoring induce workers to change jobs.34Chapter 2Does Importing IntermediatesIncrease the Demand for SkilledWorkers? Plant-level Evidencefrom Indonesia2.1 IntroductionWorkhorse models of international trade almost universally suggest that increased integrationinto international markets will encourage resources to be reallocated towards workers, firms,or industries in which the country has a comparative advantage. In developing countries,for example, trade liberalization is often supported by the argument that trade will expandin labor-intensive industries which, in turn, are predicted to increase the relative demandand wages for unskilled labor. Surprisingly, in many contexts, exactly the opposite has beenfound. Numerous studies confirm that among developing countries, trade liberalization hasincreased the relative plant-level demand for skilled labour (Sanchez-Paramo and Schady(2003); Goldberg and Pavcnik (2007)) and, likewise, has caused the skill premium to rise(Harrison and Hanson (1999), Gindling and Robbins (2001), Attanasio et al. (2004)). Despitethese stark trends, the underlying cause of the increased demand for skilled workers, thecontribution from trade, and the implications for income inequality remain key, unresolvedissues (Goldberg and Pavcnik (2005)).28This paper contributes to this literature by examining whether starting to import for-eign materials has an impact on the demand for highly educated workers along Indonesianmanufacturing plants between 1996 and 2006. The idea that importing may affect firm or-28Our work is likewise related to studies of trade, employment and wages (Trefler (2004); Gonsaga et al.(2006); Bernard et al. (2007); Egger and Kreikemeier (2009); Davis and Harrigan (2011); Felbermayr et al.(2011); Amiti and Davis (2012)), studies of trade, wages and the demand for skilled workers (Bernard andJensen (1997b); Yeaple (2005); Verhoogen (2008); Fr´ıas et al. (2009); Chor (2010); Helpman et al. (2010);Bustos (2011); Cos¸ar (2013); Vannoorenberghe (2011)), and studies of trade, wages and skill-biased technolog-ical change (Feenstra and Hanson (1999); Matsuyama (2007); Costinot and Vogel (2010); Bloom et al. (2011);Burstein and Vogel (2012); Burstein et al. (2013);Parro (2013)).352.1. Introductionganization or productivity is neither new or controversial. Rather, it is widely reported thatusing foreign intermediate goods in production often requires the plant-level adoption of moresophisticated technology, inducing skill-biased technological change (SBTC, hereafter).29 Theadoption of foreign technology, and thus importing in a developing country, is likely to in-duce further structural changes within individual manufacturing plants. In fact, there isa rich literature indicating that the reallocation of workers is strongly related to changesin the demand for skilled labour within firms, rather than across industries (Berman et al.(1994); Bernard and Jensen (1997b); and Biscourp and Kramarz (2007)). We extend thisline of research by relating changes in the relative use of educated workers within and acrossoccupations to observable decisions to import intermediate materials at the plant-level.Our data are exceptionally well suited to this objective. Typically, researchers have usedvariation in occupation categories, such as non-production or white-collar workers, to con-struct a proxy for skilled labor (Bernard and Jensen (1997a); Harrison and Hanson (1999);Pavcnik (2003); Biscourp and Kramarz (2007).30 Likewise, Amiti and Cameron (2012)in-vestigate the impact of trade liberalization on the wages of production workers relative tonon-production workers. They find that falling input tariffs has caused the wage of non-production workers to fall relative to the wage of production workers within Indonesianmanufacturing firms that import their intermediate inputs. A major advantage of this pa-per’s study is that it is able to capture a much more precise measure of skill at the plant-levelin a large, developing economy. Specifically, our panel data record the education-level ofevery worker in every Indonesian manufacturing plant with at least 20 employees. Moreover,29This is particularly true when it is imported from industrialized nations for which there is substantialevidence of skill-biased technological change. Doms et al. (1997) provide evidence that the adoption of newfactory automation technologies lead to skill upgrading. Further, recent literature on trade and heterogeneousfirms suggests that importing foreign intermediate goods increases productivity. See Muendler (2004), Amitiand Konings (2007), Kasahara and Lapham (2013), Halpern et al. (2015), and Kugler and Verhoogen (2009)among others. There is also significant evidence that skill-biased technological change can increase the skill-premium even in developing countries (e.g., Kijima (2006)). Burstein et al. (2013) provide an alternativemodel whereby importing directly induces skill-biased technological change.30Important exceptions are Bustos (2011) and Koren and Csillag (2011). Using a panel of Argentinean man-ufacturing firms with the detailed information on worker’s education level Bustos (2011) finds that exportingincreases the demand for skilled labor, while our results suggest that importing, rather than exporting, is moreimportant for skill upgrading. Using Hungarian linked employer-employee data, Koren and Csillag (2011) findthat the wage gap between workers with a high school diploma and those with primary schooling is largeramong workers operating imported machines than among workers operating domestic machines. Similarly,a number of studies use linked data firm and employee data to establish a number of related findings. Inparticular, Frazer (2013) examines the effect of importing on Rwandan manufacturing wages, Hummels et al.(2014) characterizes the relationship between offshoring and wages across skilled and unskilled Danish work-ers, Martins and Opromolla (2009) investigates the impact of importing on Portuguese manufacturing wages,while Krishna et al. (2014) and Filho and Muendler (2007) study how trade reform affects Brazilian wagesand worker displacement, respectively. Ebenstein et al. (2014) examine the impact of offspring on US wagesusing CPS. While these studies offer insight into the effect of international trade on the workers’ wages ordisplacement, we examine the effect of trade on the relative employment of skilled workers to unskilled workersat the plant-level.362.1. Introductionwe are able to distinguish the distribution of worker education across non-production andproduction workers within each plant. Our data dose not, however, record wages by workereducation and, we cannot study the impact of importing on the education wage premium.In the last two decades, Indonesia has experienced a large increase in the supply ofeducated workers. In fact, using the balanced panel of manufacturing plants, we find thatthe plant-level average share of educated workers—defined as the workers with a highschooldiploma—increased by 14.5 percentage points between 1996 and 2006. When we decomposethe overall increase in the share of educated workers into the increase within occupationcategories and the increase due to reallocation between occupation categories, we find thatthe skill upgrading within production and within non-production workers account for morethan 95 percent of the overall increase in the share of educated workers; the reallocationfrom production to non-production workers account for less than 5 percent. Since little ofthe skill-upgrading at the plant-level can be explained by the reallocation of workers acrossoccupations, existing studies that focus on the relative demand for non-production workersto production workers provide limited insight on how importing affects the overall demandfor educated workers. This paper contributes to the literature by investigating the impact ofimporting on the demand for educated workers within occupation categories.Quantifying the impact of importing on the demand for educated workers requires over-coming a number of key empirical challenges. First, we are particularly concerned that thedemand for skill and the decision to import are endogenously determined. We develop anumber of detailed instruments to capture exogenous variation in plant-level import ship-ping costs. We exploit this variation to identify robust IV estimates of the causal impact ofimporting on the demand for skilled labour. Our IV results consistently indicate that mostwithin-firm education-based skill-upgrading occurs within occupations. Moreover, traditionalmeasures of skill upgrading in existing studies, such as the fraction of non-production workers,tend to understate the degree of skill upgrading induced by importing.Second, we are also concerned that the unobservable impact of trade on the demand foreducated workers will vary substantially across heterogeneous plants. For instance, importingforeign intermediate goods may provide plants with an incentive to hire more educated work-ers, but the degree of skill-upgrading may depend crucially on the plant’s existing, potentiallyunobserved, heterogeneous ability to implement foreign technology. When the effect of im-porting on the demand for skill varies across plants, there is no single “effect” of importingon skill demand. Furthermore, we expect plants with greater ability to adopt technology willself-select into importing because these plants gain more from importing. This “selection ongains” complicates estimation in general, and even a standard first differenced estimator isinvalid as an estimator for the average treatment effect because this source of the bias cannotbe differenced out.372.1. IntroductionBy applying the treatment effect framework developed by Heckman and Vytlacil (2005)Heckman and Vytlacil (2007a) and Heckman and Vytlacil (2007b), we estimate the MarginalTreatment Effect (MTE, hereafter) curve as well as various summary measures of the impactof importing on the relative demand for skilled labor in the Indonesian manufacturing sector,such as the average effect among all plants (the average treatment effect; the ATE, hereafter),the average effect among importers (the treatment effect on the treated; the TT, hereafter),and the average effect among non-importers (the treatment effect on the untreated; the TUT,hereafter).The estimated MTE curve is well above zero and downward sloping, where the latterfeature provides direct evidence that the impact of importing on the demand for educatedworkers varies across plants (i.e., the coefficient is random) and plants that receive a largeridiosyncratic gain from importing are more likely to start importing. The estimates of theATE, the TT, and the TUT of importing on the demand for educated workers are significantlypositive. Furthermore, the TT is consistently estimated to be substantially larger than theATE, which, in turn, is estimated to be larger than the TUT. This suggests that, on average,the effect of importing among plants that have chosen to import is substantially larger thanthat among plants that have chosen not to import in our sample. These findings are not justof academic interest, but imply that while importing may have had an important impact onthe demand for educated workers among plants that were induced to import in our sample,it is unclear that further policy change will greatly affect the demand for educated workersamong new importers.In the presence of heterogenous effects, the instrumental variable (IV) estimator identifiesthe Local Average Treatment Effect (LATE), which is the average effect of importing amongplants induced to change their import status by an instrument (Imbens and Angrist (1994)).However, the plants that are induced to start importing by an instrument may be differentfrom the plants that would have been induced to start importing by a policy change. Weuse the framework of Carneiro et al. (2010) to study the average impact of further policychanges on the demand for educated workers among the set of plants induced to import bythe change in policy. The results suggest that further policy changes that promote importingwould have increased the demand for educated workers among the plants induced to startimporting, but these changes are smaller than that implied by the TT or our IV estimates.The next section describes our empirical model and the nature of selection. Section 3describes our data set and documents the relationship between importing and plant-levelskill-intensity. Section 4 describes the empirical results. The last section concludes.382.2. A Simple Model of Importing, Selection and SBTC2.2 A Simple Model of Importing, Selection and SBTCConsider a simple two-country model of importing where home (Indonesian) firms considerwhether or not to import from abroad. Consumers have CES preferences and the marketstructure is monopolistic competition. A home firm producing variety ω faces home demandq(ω) = B(Zd)p(ω)−η where q is quantity demanded, p is the output price, η is the elasticityof substitution, and Zd is a vector of observed variables that serve as a demand shifter.31Firms hire capital, skilled labor, and unskilled labor on competitive factor markets andcombine them with intermediate materials - purchased domestically or imported - to produceoutput according to the production functionf(K,M,Ls, Lu, A, ϕ) = ϕKαkMαm{[ALs](σ−1)/σ + L(σ−1)/σu }αlσ/(σ−1), (2.1)where K is capital, M is total intermediate materials, Ls is the number of skilled workers,Lu is the number of unskilled workers, σ > 1 is elasticity of substitution between skilled andunskilled workers, ϕ is a Hicks neutral productivity term, and A is a skilled labor augmentingtechnology term. For notational simplicity we abstract from differences across occupations,though allowing differences across production and non-production workers is straightforward.To consider the potential impact of importing on the relative demand for skilled workers,we allow foreign imported inputs to affect the level of skilled labor augmenting technology aslnA(X,D, β˜) = Dβ˜ +Xγ˜′ + U˜ , (2.2)where D is a dummy variable for the use of imported inputs, β˜ is a firm-specific parameterthat captures the effect of importing on skill-biased technology A, X is a vector of observables(to be specified later), and U˜ is a skill-biased technology shock. The skill biased technologyterm A depends on X and the import decision D as in (2.2) but we assume that importingdoes not affect the Hicks neutral technology level ϕ.32For simplicity, we assume constant returns to scale technology where the plant’s marginalcost is determined byc(A,ϕ) = minLs,Lu,K,MWsLs+WuLu+WkK+WmM subject to f(Ls, Lu,K,M,A, ϕ) ≥ 1 (2.3)31We abstract from the possibility of exporting for the moment, though it will be straightforward to extendour framework to allow for it. We can also directly extend the model to allow for unobserved demand/costshifters.32It is also possible to allow importing to improve Hicks-Neutral productivity or product quality. Weabstract from the former possibility only for expositional simplicity, while the latter consideration leads to asimilar estimating equation. Although we investigate different mechanisms which drive skill upgrading throughimports, import-induced skill-biased technological change may operate through either productivity or productquality.392.2. A Simple Model of Importing, Selection and SBTCβ˜∗ϕ(a) Marginal Importer’s β˜∗S1 − S0ϕ(b) Marg. Importer’s S1 −S0E[S1 − S0|X = x, UD = p]p(c) MTEFigure 2.1: Importing and Skill Demandand (Ws, Wu, Wk, Wm) is a vector of factor prices. If the plant chooses to import, it incurs afixed import cost fm(Zc) where Zc contains a vector of observed variables that determine thefixed cost. Then the plant’s net profit function is pi(A,Zd, Zc, ϕ,D) = r(A,Zd, ϕ)−Dfm(Zc),where r(A,Zd, ϕ) = maxqpq − c(A,ϕ)q. A firm will import whenever the net profit fromimporting is greater than the net profit achieved using domestic materials alone,∆pi(X,Zd, Zc, ϕ, β˜) := pi(A(X, 1, β˜), Zd, Zc, ϕ, 1)− pi(A(X, 0, β˜), Zd, Zc, ϕ, 0) ≥ 0. (2.4)Note that ∆pi(X,Zd, Zc, ϕ, β˜) is strictly increasing in β˜ and ϕ.Suppose we allow β˜ to vary across plants asβ˜ =¯˜β + ˜,where¯˜β is the mean of β˜ and ˜ is the plant-specific return to importing. Then, for each valueof X, Zd, Zc, and ϕ, it is straightforward to determine a threshold value of β˜ that inducesfirms to start importing by the condition ∆pi(X,Zd, Zc, ϕ, β˜∗) = 0. This threshold value β˜∗depends on X, Zd, Zc, and ϕ. Naturally, firms with low initial productivity ϕ will requirelarger values of β˜ to justify importing.33Under the assumption of heterogeneous returns to importing, we can illustrate the se-lection mechanism by considering the locus of β˜’s for the marginal importer. Figure 1(a)demonstrates that this locus is strictly decreasing in initial Hicks-neutral productivity, ϕ,while fixing the value of X, Zd, and Zc. Firms with low Hicks-neutral productivity (i.e., lowvalue of ϕ) choose to import only if they receive high idiosyncratic returns from importing(i.e., high value of β˜).33This argument is analogous to that in Lileeva and Trefler (2010) which studies the heterogeneous returnto exporting on Hicks-neutral productivity.402.2. A Simple Model of Importing, Selection and SBTC2.2.1 Selection, SBTC and Skill DemandConsider the first order conditions from cost minimization problem (2.3). Denote the log ofthe demand for skilled workers relative to unskilled workers by SD = ln(Ls/Lu) forD ∈ {0, 1},where the subscript D indicates its dependence on import status. Given market wages, therelative demand for skilled workers is determined by equating the ratio of the marginalproduct of skilled and unskilled workers to the ratio of their wages asSD = (σ − 1) lnA− σ ln(Ws/Wu)= Dβ +Xγ′ + U= D(β¯ + ) +Xγ′ + U, (2.5)where (β, β¯, , γ, U) are ( β˜σ−1 ,¯˜βσ−1 ,˜σ−1 ,γ˜σ−1 ,U˜σ−1) and X subsumes ln(Ws/Wu). In ourcontext, importing is an endogenous decision because the import decision D and skill-biasedtechnology shock U are likely to be correlated. Moreover, as indicated in (4), plants with agreater ability to adopt skilled-biased technology (i.e., high value of β) will self-select intoimporting because they will achieve larger productivity gains from importing. Therefore, βand D are also correlated.Because of this positive sorting on the gain from importing, we would expect that thechange in skill demand will be greater among plants that choose to import relative to non-importers should they have started importing. As illustrated in Figure 1(b), and similarto Figure 1(a), the effect of importing on demand for skill for the marginal importer is adecreasing function of Hicks-neutral productivity ϕ.2.2.2 The Marginal Treatment EffectWe are interested in identifying the impact of importing on the demand for skilled labor, β,which may vary across plants. Imbens and Angrist (1994) show that, under certain conditions,using a single dummy instrument, an IV estimator identifies the local average treatment effect(LATE), or the average value of β among plants who are induced to change their importchoices by the instrument. When multiple dummy instruments are used, an IV estimatoridentifies a weighted average of the instrument-specific LATEs. Therefore, an IV estimatorprovides an estimate of an interpretable quantity even when the effect of importing on thedemand for skilled workers is heterogenous across plants, although the LATE is generallydifferent from the average value of β.To evaluate the heterogeneous impact of importing on the demand for skill, we use theframework developed by Heckman and Vytlacil (1999) Heckman and Vytlacil (2005) Heckmanand Vytlacil (2007a) and Heckman and Vytlacil (2007b) as follows. The relative demand for412.2. A Simple Model of Importing, Selection and SBTCskilled labor SD for D = 0 or 1 can be written asS1 = µ1(X) + U1 and S0 = µ0(X) + U0, (2.6)respectively, where, allowing for the average value of β to depend on X in (2.5), µ1(X) ≡E[S1|X] = β¯(X) + Xγ′ and µ0(X) ≡ E[S1|X] = Xγ′ while U1 =  + U and U0 = U . Theimpact of importing on the demand for skilled workers depends on the plant-specific abilityto adopt foreign technology embedded in imports since S1 − S0 = β¯(X) + U1 − U0.To derive an empirical specification for the decision to import, let Z ≡ (X,Zd, Zc, ϕ)and write (2.4) as ∆pi(Z, β˜) ≡ ∆pi(X,Zd, Zc, ϕ, β˜). Define the latent variable, D∗, asD∗ = ∆pi(Z, β˜) = µD(Z) − V , where µD(Z) = E[∆pi(Z, β˜)|Z] is a deterministic functionof observable variables Z while V = ∆pi(Z, β˜)− µD(Z) is a mean-zero unobserved stochasticcomponent. Then, we have a latent variable model of importing:D∗ = µD(Z)− V, D ={0 if D∗ < 01 if D∗ ≥ 0. (2.7)A plant imports if D∗ ≥ 0; it does not import otherwise. We assume that the distribution ofV , denoted by FV , is continuous and strictly increasing and, furthermore, that (U0, U1, V ) isstatistically independent of Z given X and ϕ.34Let P (Z) denote the probability of importing conditional on Z so that P (Z) ≡ Prob(µD(Z) >V ) = FV (µD(Z)). P (Z) is called the propensity score. Define UD ≡ FV (V ), and the randomvariable UD is uniformly distributed on [0, 1] by construction. Because V is the unobservedcomponent of the net benefit of importing, UD represents the quantiles of the unobserved netbenefit from importing. Then the import decision (2.7) is alternatively written as D = 1 ifP (Z) ≥ UD and D = 0 otherwise.We define the marginal treatment effect (MTE) as∆MTE(x, p) = E[S1 − S0|X = x, UD = p] = β¯(x) + E[U1 − U0|X = x, UD = p]. (2.8)This is the average effect of importing on skilled demand for plants with X = x and UD = p.Because a plant is indifferent between importing and not importing when P (Z) = UD, theMTE captures the mean impact from importing on the demand for skilled labor among plantswith X = x and P (Z) = p when the realization of UD is such that the plant is just indifferentbetween importing and not importing.Estimating the MTE for each value of UD = p within the support of P (Z), we are ableto construct the empirical counterpart to the locus of returns as outlined in Figure 1(c).34The latter is implied by the independence and monotonicity assumptions of Imbens and Angrist (1994)as shown by Vytlacil (2002).422.3. DataCompared to Figures 1(a) or (b), the x-axis is measured in import probabilities rather thanproductivity in Figure 1(c) since we allow firms to differ in many dimensions rather than onlyproductivity. Propensity scores are a natural metric to summarize those observed differencesin a single dimension. When firms self-select into importing based on their unobservedbenefits from importing, we expect the MTE curve to be downward sloping because, if firmschoose to import even if their observed characteristics suggest that they were not likely toimport (i.e., the low value of P (Z) = p), the unobserved component of net benefit fromimporting must be high (i.e., the high value of E[U1 − U0|X = x, UD = p] in (2.8)). Incontrast, we expect the MTE curve to be flat in the absence of self-selection based on theunobserved benefit from importing.Further, as described by Heckman and Vytlacil (2005) Heckman and Vytlacil (2007a)and Heckman and Vytlacil (2007b), the MTE also allows us to compute all the conventionaltreatment parameters, such as the ATE, the TT, and the TUT, as weighted averages ofthe MTE, each computed with a different weighting function. Details for each of thesecalculations can be found in Appendix B.2.2.3 Data2.3.1 Data SourcesOur primary source of data is the Indonesian manufacturing survey between 1995 and 2007,where we mainly use the data recorded in the census years 1996 and 2006 because, in thesetwo years, the Indonesian manufacturing survey records the distribution of academic achieve-ment in two distinct occupation categories (non-production and production) in each plant.Specifically, in each plant we observe the number of workers with primary, secondary andpost-secondary education. We construct relative skill measures that are directly based on theworkers’ education levels for each occupation category.The survey covers all manufacturing plants with at least 20 employees.35 In the 2006data set, 93 percent of plants are also single-plant firms. The data set captures a wide set ofplant-level characteristics which we use to study the nature of plant-level heterogeneity. Inparticular, the survey records all expenditures on imported intermediate materials. It alsoincludes plant-level input and output variables, such as total revenues, capital stock, domesticmaterials, and other plant-level information including the percentage of sales from exports,the percentage of ownership held by foreign investors, total plant-level expenses on researchand development (R&D), and total plant-level expenditures on worker training. Appendix35A limitation of this paper is that in low-income countries a large share of firms have few employees (seeMcCaig and Pavcnik (2014) and McCaig and Pavcnik (2015) for examples). However, these are likely alsofirms that do not directly import or export and typically lack a skilled labor force.432.3. DataA.1 provides a detailed description of our variable construction.To control for regional labor market conditions we augment the manufacturing surveywith the Indonesian household survey. The Indonesian household survey covers a nationallyrepresentative sample of households and documents key labor force information includinggender, age, location, educational attainment and labor force experience. We use the house-hold survey to develop a measure of the skill premium in each location and year. We areunable to use the manufacturing survey data for this task since it does not provide a measureof wages by education level.2.3.2 Importing and Worker EducationDescriptive StatisticsPanel A of Table 2.1 documents plant-level differences in employment across six education-based (highest attainment) categories: less than primary school, primary school, junior highschool, high school, college graduates and post-graduate educated workers. The top panelcompares the percentage of plant-level employment across importing and non-importingplants in 2006. We find that importing plants, on average, hire fewer workers in each edu-cational category below high school and more workers with high-school diplomas, college de-grees, or post-graduate training. For example, 61 percent of workers in importing plants haveat least a high school degree, while only 36 percent of workers in non-importing plants havea high school degree or better. In columns (8)-(10), “Training/Worker” and “R&D/Worker”report the average per worker expenditures on training and research and development (R&D),respectively, in thousands of 1983 Indonesian rupiahs while “Non-Prod./All Workers” reportsthe percentage of non-production workers in total employment in each plant. We find thatthe expenditures on training workers or investing in R&D among importers is more than dou-ble what is spent by non-importers on average. Likewise, importers tend to have a relativelylarge number of non-production workers in their plants.Panel A of Table 2.1 also compiles similar statistics for exporting plants, non-exportingplants, and foreign-owned plants.36 We observe a number of stark patterns: foreign plantstend to employ more educated workers than domestic plants while exporting plants appearskill-intensive when compared to their non-exporting counterparts. Nonetheless, within eachgroup we continue to find that importing plants hire a greater percentage of educated workers.The last row of Panel A documents the distribution of skilled labor for plants which did notimport in 1996, where the reduction in sample size is driven by the fact that only a fractionof 2006 firms exist in 1996. The skill differences across firms which never import and thosewhich start importing demonstrate very similar patterns to the full sample even though the36We classify a plant as foreign plant when at least 10 percent of its equity is held by foreign investors.442.3. Datasample is much smaller.Panel B of Table 1 documents the percentage of workers in each educational categorywithin production or non-production workers. For production workers, importing plants arefound to systematically hire more workers with education levels above high school. Whilethis remains true for non-production workers, it is much less stark. Instead, we observe thatimporting plants tend to hire a substantially greater share of college-educated non-productionworkers.Although importers always appear to be more skill-intensive on average within each oc-cupation category, the mechanism that drives the correlation between importing and skill-intensity may differ between production and non-production workers. While the use of im-ported materials might induce the adoption of new production processes which in turn re-quires hiring more skilled production workers, importing might require substantial increasesin the number of non-production workers for trade related activities such as dealing with cus-toms agents or arranging shipments from foreign countries. Given the potential for differencesacross occupation categories, we analyze the impact of importing on the demand for educatedworkers within the production occupation separately from that within the non-productionoccupation.Among production workers, many Indonesian plants do not hire any workers with collegeor post-graduate training. As a result, defining a skilled worker as a “college graduate” inour sample of production workers would eliminate a significant number of plants that arewholly composed of workers without college education. On the other hand, using a high-school education threshold would potentially obscure a key margin on which firms upgradeemployee skill in response to importing among non-production workers because the differencein non-production hiring between importers and non-importers is clearest at college level, asdocumented in Table 1. For these reasons, we choose to define a skilled worker as one withat least a high school degree for production workers, and as one with at least a college degreefor non-production workers.Decomposing Changes in Plant-Level Skill SharesTo better characterize the development of the Indonesian labor market, we examine theimportance of the reallocation of labor from the production to non-production occupation(a “between” component) relative to the education upgrading within each occupation (a“within” component). Specifically, we decompose the change in the overall share of educated452.3. DataTable 2.1: Importing and Skill Intensity 2006, full samplePanel A: All WorkersHighest Degree Completed/Fraction Training R&D Non-Prod.Obs.No Primary Primary Jr. High High College Grad. Worker Worker All WorkerAll PlantsImporters0.015 0.071 0.302 0.538 0.073 0.0006 70.9 73.8 0.1845,512(0.075) (0.171) (0.218) (0.237) (0.093) (0.004) (724.4) (1037.1) (0.151)Non-Imp. 0.059 0.275 0.306 0.323 0.036 0.0003 23.2 17.8 0.135 23,952(0.151) (0.302) (0.248) (0.295) (0.078) (0.005) (570.0) (294.4) (0.163)Exporting PlantsImporters 0.007 0.069 0.222 0.609 0.091 0.0011 150.7 158.2 0.184 1,519(0.044) (0.124) (0.208) (0.235) (0.103) (0.0065) (1,310.4) (1,826.3) (0.159)Non-Imp. 0.030 0.190 0.293 0.437 0.050 0.0003 65.0 46.5 0.150 3,690(0.102) (0.239) (0.227) (0.292) (0.075) (0.0034) (1,141.0) (613.4) (0.160)Non-Exporting PlantsImporters 0.018 0.072 0.333 0.511 0.066 0.0004 40.6 41.6 0.184 3,993(0.084) (0.186) (0.214) (0.232) (0.088) (0.0031) (260.8) (461.2) (0.148)Non-Imp. 0.065 0.291 0.309 0.302 0.033 0.0002 15.6 12.6 0.132 20,262(0.158) (0.309) (0.252) (0.291) (0.078) (0.0056) (383.0) (183.9) (0.163)Foreign-Owned PlantsImporters 0.008 0.070 0.170 0.651 0.099 0.0015 176.4 360.0 0.196 303(0.045) (0.111) (0.183) (0.238) (0.108) (0.0054) (744.1) (3,726.2) (0.177)Non-Imp. 0.023 0.130 0.208 0.555 0.083 0.0007 59.9 111.5 0.178 376(0.086) (0.185) (0.205) (0.294) (0.108) (0.0038) (337.6) (843.4) (0.158)Initial Non-ImportersImporters 0.038 0.114 0.273 0.503 0.072 0.0008 47.2 59.9 0.191 659(0.127) (0.193) (0.212) (0.259) (0.088) (0.004) (297.2) (387.5) (0.163)Non-Imp. 0.050 0.224 0.325 0.365 0.036 0.0002 16.6 14.7 0.154 7,465(0.134) (0.280) (0.244) (0.283) (0.064) (0.003) (241.9) (212.7) (0.159)Panel B: Production vs. Non-Production WorkersProduction Workers Non-Production WorkersNo Primary Primary Jr. High High College+ No Primary Primary Jr. High High College+All PlantsImporters 0.016 0.078 0.328 0.544 0.035 0.002 0.018 0.168 0.566 0.245(0.077) (0.182) (0.234) (0.264) (0.071) (0.037) (0.092) (0.204) (0.236) (0.232)Non-Imp. 0.061 0.290 0.324 0.309 0.017 0.017 0.085 0.193 0.534 0.172(0.156) (0.314) (0.267) (0.315) (0.065) (0.107) (0.232) (0.288) (0.352) (0.257)Exporting PlantsImporters 0.008 0.081 0.240 0.627 0.044 0.003 0.024 0.104 0.529 0.340(0.046) (0.144) (0.225) (0.266) (0.084) (0.026) (0.086) (0.157) (0.254) (0.271)Non-Imp. 0.030 0.206 0.314 0.429 0.021 0.013 0.053 0.133 0.543 0.258(0.104) (0.256) (0.247) (0.320) (0.060) (0.091) (0.166) (0.223) (0.322) (0.292)Non-Exporting PlantsImporters 0.018 0.077 0.361 0.513 0.031 0.002 0.016 0.194 0.581 0.207(0.086) (0.195) (0.229) (0.256) (0.066) (0.040) (0.094) (0.215) (0.227) (0.203)Non-Imp. 0.067 0.305 0.326 0.287 0.016 0.018 0.091 0.206 0.532 0.154(0.163) (0.322) (0.270) (0.309) (0.066) (0.110) (0.244) (0.298) (0.358) (0.245)Foreign-Owned PlantsImporters 0.009 0.084 0.181 0.686 0.041 0.004 0.023 0.074 0.498 0.401(0.044) (0.134) (0.200) (0.269) (0.085) (0.039) (0.075) (0.136) (0.276) (0.299)Non-Imp. 0.023 0.145 0.229 0.564 0.038 0.012 0.034 0.102 0.506 0.346(0.085) (0.208) (0.227) (0.336) (0.100) (0.081) (0.116) (0.198) (0.334) (0.330)Initial Non-ImportersImporters 0.039 0.126 0.297 0.507 0.031 0.010 0.035 0.166 0.540 0.250(0.129) (0.207) (0.227) (0.286) (0.064) (0.088) (0.131) (0.226) (0.258) (0.235)Non-Imp. 0.052 0.243 0.348 0.343 0.014 0.011 0.056 0.201 0.570 0.162(0.141) (0.298) (0.264) (0.305) (0.047) (0.082) (0.181) (0.277) (0.320) (0.229)Notes: Standard deviations are in parentheses. The first column indicates current import status, where“importers” denotes plantsthat import and “non-importers” captures plants that do not import in the current year. The first panel pools all plants in all years.The second and third panel split the sample by export status, while the fourth and fifth panels split the sample by the country ofownership. Specifically, foreign-owned plants are defined as those plants where at least 10% of equity is held by foreign investorswhile domestic plants are defined as plants for which more than 90% of equity is held by domestic investors.462.3. Dataworkers for each plant as∆LsL= ∆LpsLp× LpL︸ ︷︷ ︸within prod.+ ∆LnsLn× LnL︸ ︷︷ ︸within non-prod.+(LpsLp− LnsLn)×∆LpL︸ ︷︷ ︸between, (2.9)where ∆ (Ls/L) = (Ls/(Ls + Lu))06 − (Ls/(Ls + Lu))96 is the change in the overall share ofeducated workers between 1996 and 2006, ∆(Ljs/Lj)=(Ljs/(Ljs + Lju))06−(Ljs/(Ljs + Lju))96is the change in the share of educated workers within occupation j, ∆ (Lp/L) =(Lp/(Lp + Ln))06− (Lp/(Lp + Ln))96 is the change in the share of workers in the productionoccupation, and (Lj/(Lp + Ln)) ={(Lj/(Lp + Ln))96+(Lj/(Lp + Ln))06}/2 and(Ljs/(Ljs + Lju))={(Ljs/(Ljs + Lju))96+(Ljs/(Ljs + Lju))06}/2 are the average of the cor-responding share variables. The superscripts “p” and “n” represent production and non-production occupations, while the subscript “s” and “u” represent skilled and unskilledworkers, respectively. Likewise, the subscripts “96” and “06” indicate whether a variableis measured in 1996 or 2006.Table 2.2 reports the average value of each of the three decomposition components acrossplants. In the second column, defining educated workers as those workers with a highschooldiploma, the overall share of educated workers increased by 14.5 percentage points from 0.322to 0.467 between 1996 and 2006, and this increase is mostly explained by the skill upgradingwithin occupations.37 In fact, skill upgrading within production workers and non-productionworkers, respectively account for 86 percent and 10 percent of the overall change in the shareof skilled workers while the reallocation of workers from the production to non-productionoccupation contributes less than 5 percent. The third and fourth columns compare the plantsthat never imported (“non-switchers”) with those that started importing (“switchers”) andshow that skill upgrading is higher at all margins for switchers except within non-productionworkers. The overall differences in skill share growth between switchers and non-switchers islargely driven by the differences in skill share growth within production workers.The fifth to eighth columns repeat the same decomposition exercise, but define skilledworkers as those workers with an educational attainment of no less than college. In the fifthcolumn, the overall share of college educated workers increased by 1.75 percentage pointsfrom 0.325 to 0.500 between 1996 and 2006, and the “within non-production” term in thedecomposition accounts for more than 60 percent of the change in the share of college educatedworkers. We similarly find that the differences in skill share growth within non-productionworkers is largest determinant of the difference between the college worker share growthacross switchers and non-switchers.37Table B.5 in Appendix reports details. Appealing to census data, we also find that there is similar growthin the supply of skilled (highschool or college) manufacturing workers.472.3. DataTable 2.2: A Decomposition of Plant-Level Skill Growth by Import StatusDef. of Skilled Workers Highschool+ College+Initial Non-importers Initial Non-importersSample All non- All non-switchers switchers switchers switchers∆(Ls/L) 0.1446 0.1636 0.1425 0.0175 0.0270 0.0144within prod. 0.1248 0.1409 0.1235 0.0058 0.0067 0.0048within non-prod. 0.0137 0.0113 0.0128 0.0106 0.0147 0.0085between 0.0060 0.0114 0.0062 0.0011 0.0056 0.0011No. Obs 10,537 658 7,464 10,537 658 7,464a. Source: Indonesia Manufacturing Survey in 1996 and 2006.b. Skilled workers are defined as workers with education no less than highschool in the second to fourthcolumns and workers with no less than college in the fifth to last columns. Plants with no productionworkers in 1996 or 2006 are excluded (only three observations). Plants with no non-production workerin either period are treated as having zero within-non-production changes, and the mean value ofskill share in non-production sector (Lns /Ln) is computed using the period when the number of non-production workers is positive. Plants with no non-production workers in both 1990 and 2006 simplyhave a zero within non-production component and zero between component.2.3.3 Variable DefinitionsAll outcome variables and most explanatory variables are measured in 2006. The lagged valueof outcome variables are also included in the set of explanatory variables so that our sampleconsists of plants that are present in both the 1996 and 2006 data sets. The definitions ofvariables and their descriptive statistics are reported in Tables 2.3 and 2.4.We consider eight different outcome variables. Our first two measures, ln (Lps/Lpu)06 andln (Lns /Lnu)06, directly capture the number of skilled workers within each occupation categoryin 2006, where Ljs and Lju are the number of skilled workers and unskilled workers, respec-tively, employed in occupation j ∈ {p, n}. We define a skilled production worker as onewith at least a highschool diploma and a skilled non-production worker as one with a collegedegree. A non-trivial number of plants that do not hire any skilled workers are droppedfrom our sample when we use the log of the ratio of skilled workers to unskilled workers asan outcome variable. Because this omission may generate selection bias, we also considerthe outcome variables that measure the fraction of skilled workers’ in each occupation cat-egory, denoted by (Lps/(Lps + Lpu))06 and (Lns /(Lns + Lnu))06. To examine the total relativedemand for educated workers, the fifth and sixth outcome variables aggregate skilled workersacross occupations where ln (Ls/Lu)06 ≡ ln (Lns + Lps/Lnu + Lpu)06 and (Ls/(Ls + Lu))06 ≡((Lps + Lns )/(Lps + Lpu + Lns + Lnu))06. We keep the definition of skill, a highschool diploma orcollege degree, consistent across occupations in these measures and present both results in allof our regressions. Finally, our last outcome variable considers the log ratio of non-productionworkers to production workers, ln (Ln/Lp))06 ≡ ln ((Lns + Lnu)/(Lps + Lpu))06, or the fractionof non-production workers, (Ln/(Ln + Lp))06 ≡ ((Lnu + Lns )/(Lps + Lpu + Lns + Lnu))06, whichare often used as measures of skill intensity in the existing literature.482.3. DataTable 2.3: Definitions of the VariablesVar. DefinitionS (1) The log of the ratio of skilled workers to unskilled workers in 2006 in occupation j ∈ {p, n},ln(Ljs/Lju)06, across both occupations, ln (Ls/Lu)06, or the log of the ratio of non-productionworkers to production workers in 2006, ln (Ln/Lp)06. (2) The fraction of skilled workers in 2006in occupation j ∈ {p, n}, (Ljs/(Ljs + Lju))06, across both occupations (highschool or college),ln (Ls/Lu)06, and the fraction of non-production workers in 2006, (Ln/(Ln + Lp))06.D Equal to one if plant imports materials from abroad in 2006; zero otherwise.X Export dummy, capital stock, Hicks-neutral productivity, a foreign ownership dummy, a dummyfor positive R&D expenditures, a dummy for positive training expenditures, the log of theratio of skilled workers’ wages to unskilled workers’ wages in 1996 and in 2006 in each region,local changes in the supply of skilled labor, the 1996 value of the outcome variables (denotedby replacing “06” with “96”), a dummy for no hiring of skilled workers or unskilled workers inoccupation j ∈ {p, n} in 1996 denoted by djs,96 := 1(Ljs = 0) or dju,96 := 1(Lju = 0), TFP constructedby the Levinsohn and Petrin method, 3-digit ISIC industry dummies, and province dummies.Z\X Transport costs to the nearest port, the fraction of Indonesian imports shipped by air in industry j,the average weight of Indonesian imports in industry j, a change in output and input tariff ratesat 5-digit ISIC level between 1996 and 2001.Notes: A skilled worker is defined as a worker with high school eduction and an unskilled worker is definedas a worker without high school education. Occupation categories “p” and “n” denote production workersand non-production workers, respectively. All variables are measured in 2006 unless stated otherwise.Table 2.4: Descriptive StatisticsD = 0 D = 1 D = 0 D = 1Explanatory Variable(a) Mean S.D. Mean S.D. Explanatory Variable Mean S.D. Mean S.D.TC 0.874 0.924 0.631 0.814 log(Ws/Wu)high06 0.435 0.187 0.449 0.152Air 0.087 0.068 0.114 0.090 log(Ws/Wu)coll06 0.571 0.271 0.555 0.218Export 0.175 0.380 0.434 0.497 log(Ws/Wu)high96 0.472 0.146 0.459 0.140Capital 13.570 1.808 15.086 2.034 log(Ws/Wu)coll06 0.580 0.243 0.586 0.264Hicks-neutral ϕ 5.229 0.582 5.538 0.636 ln(Lps/Lpu)high96 -0.735 1.359 -0.389 1.502Foreign 0.017 0.130 0.081 0.273 ln(Lns /Lnu)coll96 -0.622 1.095 -0.930 1.173R&D 0.057 0.231 0.178 0.384 (Lps/(Lps + Lpu))high96 0.212 0.266 0.386 0.325Training 0.308 0.462 0.589 0.493 (Lps/(Lps + Lpu))coll96 0.008 0.039 0.026 0.079dp,highu,96 0.017 0.128 0.057 0.233 dn,collu,96 0.115 0.319 0.044 0.205dp,highs,96 0.342 0.475 0.152 0.359 dn,colls,96 0.607 0.489 0.323 0.469No. Obs. 5706 4410D = 0 D = 1 D = 0 D = 1Outcome Variable(b) Mean S.D. Mean S.D. Outcome Variable Mean S.D. Mean S.D.ln(Lps/Lpu)06 -0.553 1.657 0.490 1.676 (Lps/(Lps + Lpu))06 0.340 0.333 0.559 0.337ln(Lps/Lpu)06 -1.202 1.202 -0.853 1.261 (Lps/(Lps + Lpu))06 0.166 0.253 0.278 0.266ln(Ls/Lu)high06 -0.577 1.683 0.645 1.722 (Ls/(Ls + Lu))high06 0.399 0.315 0.604 0.312ln(Ls/Lu)coll06 -3.050 1.090 -2.665 1.166 (Ls/(Ls + Lu))coll06 0.037 0.066 0.081 0.096ln(Ln/Lp)06 -1.791 1.082 -1.639 1.093 (Ln/(Ln + Lp))06 0.184 0.156 0.208 0.156Notes: (a) The sample statistics for the explanatory variables that are used to estimate the decisions to import inTable B.8. (b) The sample statistics for the outcome variables that are used to estimate the skill demand equation(B.1). The superscript “high” and “coll” denote variables that are measured using highschool or college as the skillthreshold, respectively.492.3. DataThe set of explanatory variables, X, includes the lagged value of the outcome variablein 1996, denoted by using the subscript “96” in place of “06,” dummy variables for plantsthat did not hire any skilled or unskilled workers in each occupation in 1996, denoted bydjs,96 and dju,96 for j = p, n, and the relative wage ratios in 1996, denoted by ln(Ws/Wu)96.38In addition, X contains the plant’s current export status, our estimate of Hicks-neutralproductivity ϕ, the plant’s capital stock, the local skilled-unskilled wage ratio,39 a largeset of dummy variables to capture differences across foreign ownership, R&D expenditures,worker training expenditures, industries and provinces. Using production function (2.1), weestimate a model-consistent measure of Hicks-neutral productivity ϕ based on the frameworksdeveloped by Olley and Pakes (1996), Levinsohn and Petrin (2003), Ackerberg et al. (2015)and Gandhi et al. (2013) as described in Appendix B.3. For robustness, we also estimated aconventional measure of TFP from a standard Cobb-Douglas production function and usedthis in place of our Hicks-neutral productivity measure.2.3.4 InstrumentsWe expect that the decision to import for any given plant is likely to be endogenously deter-mined with its decision to hire skilled labour. The identification strategy we outline belowrelies on the presence of instruments. Our primary instrument set includes location-specifictransport costs and industry-specific measures of the fraction of imports shipped by air.We also consider industry-specific measures of imported input weight, changes in product-specific input and output tariffs, and tariff-based measures of export market access to checkour benchmark results or control for the potential endogeneity of plant export behavior. Wediscuss the construction of each instrument in turn.Since we do not observe transport costs to the port directly, we construct the measureof transport cost for each plant as follows. To incorporate geographical information, we firstdivide Indonesia into cells of one kilometer squared and assign a value of 1-10 to each cell,where “10” is the highest cost (Steepness of Slope, Sea vs. Land). Then, we use ArcGIS tofind the least accumulative-cost path between any plant and its nearest port. Finally, ourmeasure of transport cost is obtained from the least accumulative-cost after dividing it bythe sample standard deviation.For the fraction of imports shipped by air, we rely on research that suggests that differ-ences in trade responses across industries can arise from differences in the nature of delivery(e.g. air vs. water) or heaviness of the output. We extend this literature by combining38We include an explanatory variable that equals zero if either ds,96 = 0 or du,96 = 0 and equals ln(Lps/Lpu)otherwise. For notational brevity, we indicate this variable by ln(Lps/LPu )9639The wage ratio is measured through a series of Mincer regressions described in Appendix A.1. We proceedin this fashion so to isolate the local difference in wages due to education alone, rather than have differencesin the wage ratio reflect differences in demographics, experience, etc across regions.502.3. Datameasures the fraction of imports shipped by air (or of the weight of imports to Indonesia),industry-by-industry, with an Indonesian import input-output table to construct an industry-level measure of the share of imports shipped by air.40 The intuition for the transport modeinstrument (air vs. water) comes directly from Hummels and Schaur (2013) which arguesthat exporters pay a premium to ship goods by air for faster delivery. Similarly, as Cos¸arand Demir (2015) recognize, heavier imports will be more costly to ship.For each industry j, the variables airsharej and weightj measure the fraction of Indone-sian imports shipped by air and the weight of Indonesian imports,airsharej =air valuejair valuej + ocean valuej, and, weightj = ln(ocean weightjocean valuej),where air valuej denotes the value of air shipments to Indonesia in 2006 and ocean valuej(ocean weightj) denotes the value (weight) of shipments to Indonesia by ocean in the sameyear.41 We then combine this information with an import input-output table which providesus with the share of imports purchased from each industry in Indonesia.42 Letting shareijrepresent industry i’s import expenditure share on from industry j, our measures of importair share and import weight in industry i are constructed asImport airsharei =∑jshareij × airsharej , Import weighti =∑jshareij × weightj ,where∑j shareij = 1 ∀ i by construction. Not surprisingly, these two instruments are highlycorrelated. Because of this we largely focus on the variable capturing the fraction of importsshipped by air since the import weight variable adds little statistically significant variationto our first stage regressions.For the fourth and fifth instruments, we match each plant in our manufacturing survey toproduct-level (5-digit ISIC) output and input tariffs constructed byAmiti and Konings (2007)and use the change in output and input tariff rates between 1996 and 2001 as our instrument.The sixth instrument is a tariff-based measure of market access for Indonesian exporters indestination markets. For each industry and year, we calculate the average tariff faced by firmsin export markets where export shares are used as weights. We then compute the changein export market access for each Indonesian industry. Full details of the construction of all40The import input-output table is produced by BPS Indonesia.41Specifically, we compute these measures for shipments to Indonesia from the US and Europe. We thentake a simple average across both import sources. We thank Kerem Cos¸ar who provided us with his Stata dofiles that compute variables weightj and airsharej from EU and US trade data sets.42BPS Indonesia produces detailed input-output tables measuring of total purchases (all sources) at theindustry-level or total domestic purchases. We construct measures import flows and import shares by sub-tracting the information in the domestic input-output table from the comparable information in the totalinput-output table. All input-output data is measured in 1995.512.4. Resultsinstruments can be found in Appendix A.1.Naturally, we are concerned that the empirical estimates we find may be biased if theinstruments we use are not exogenous. For the transport cost variable, it is possible thatplants with a high-return from importing will choose to locate closer to ports. To addressthe potential concern for endogenous location choice, we focus on the sample of plants whichinitially did not import in 1996. In this fashion, we can consider the impact of transport costs(and tariffs) on plants who made their location decision well before they began using importedmaterials. Likewise, we use the 1996 industry affiliation when assigning the import airshareto each plant. In this fashion, we guard against bias that would arise from plants whichstrategically switch to new industries in response to changes in the trade environment.432.4 Results2.4.1 Benchmark IV FindingsTable 2.5 presents the results from estimating equation (2.5) by OLS and IV for productionand non-production workers separately. Consistent with the model presented in Section2.2, columns (1), (2), (5) and (6) use the log of the ratio of skilled production to unskilledproduction workers as its dependent variable. Unfortunately, because numerous plants havenot hired even one skilled worker, using the log skill ratio leads to a non-trivial loss of plants.To address this potential source of bias, we repeat our exercise using the fraction of skilledworkers in the plant’s workforce as the dependent variable in columns (3), (4), (7) and (8).In all cases we restrict attention to the set of plant’s which were not importing in 1996 so toisolate the impact of importing on plants who made their location decisions and determinedtheir main product before they began using imported materials.44Columns (1)-(4) present results for production workers where a skilled production workeris defined as one who has successfully completed highschool. The OLS point estimate incolumn (1) suggests that importing significantly increases the relative demand for skilledworkers within the production occupation by 48 log basis points. Similarly, column (3) indi-cates that importing increases the skilled fraction of the plant’s workforce by 5.5 percentagepoints. While these effects seem widely different at first blush, they are roughly consistentwith each other since the fraction of skilled employees hired by initial non-importers prior toimporting is typically quite small.Columns (2) and (4), which instrument import status using both the distance to a major43When using import weight or the tariff instruments we also match plants according to their 1996 industryaffiliation.44Capital, R&D, and training variables could be endogenously determined. When we drop capital, R&D,and training variables from the set of regressors in Table 2.5, the results are very similar. See Table B.6 inthe appendix.522.4. ResultsIndonesian port and fraction of imports shipped by air, suggest substantially larger effects. Infact, our findings suggest that importing increases the relative demand for skilled productionworkers by 371 log basis points and similarly increases the skilled fraction of the plant’s laborforce by 99 percentage points. Given the large magnitude of these estimates, it would be nat-ural for the reader to be concerned that our point estimates suffer from the presence of weakinstruments. However, as documented in Table B.8 of the Appendix, the first stage resultssuggest that our instruments are sufficiently strong (individually and jointly) to confidentlyestimate the causal impact of importing on the demand for skill. Furthermore, the p-valuesof Hansen’s J test support the validity of the overidentification restrictions, providing someevidence that instruments are uncorrelated with the error term.In a “standard” setting where we assume that there is no heterogeneity in β across plantsin equation (2.5), the finding that the IV estimate is much larger than the OLS estimatecould be viewed as puzzling since the OLS bias may likely be upward in this case. When thecoefficient β is random, however, finding a large IV estimate is less puzzling because the IVestimator identifies the local average treatment effect in the sense that it only captures theimpact of importing on plants that change their import status in response to variation in theinstrumental variable. Our results suggest that, on average, only those plants with very highvalues of β—interpreted as plants with a better ability to adopt skill biased technology—choose to change their import status. One possible explanation is that starting to importis very costly: when the start-up cost of importing is large, only those plants which receivesufficiently large benefits from changing their import status (represented by high values ofβ) will choose to start importing. Moreover, it is important to recall the context of thisestimate. In our sample, as reported in Table 2.4, the ratio of skilled to unskilled workersamong importers is nearly double that of the average non-importer.Columns (5)-(8) consider the same experiment for non-production workers. We againfind that the IV estimates indicate that importing consistently has a large, positive andstatistically significant impact on the demand skilled workers. It important to recognize thatour definition of ‘skill’ has changed in this experiment; we now define a skilled worker asone with a college degree. Our findings suggest an important change in the organizationof non-production activities, although non-production workers represent a relatively smallfraction of the workforce as documented is Section 2.3.45 In fact, the IV estimates imply a364 log basis point increase in demand skilled non-production workers and a 70 percentagepoint increase in the fraction of skilled non-production workers.The control variables in Table 2.5 generally report consistent and intuitive coefficients.The estimated coefficients on plant-level export status are often negative, which reflects the45Our findings are strongest using the skill thresholds documented in Table 2.5. However, we continue tofind marginally significant effects if we use alternative skill thresholds in each case. See the Appendix fordetails.532.4. ResultsTable 2.5: Skill Demand Equation Across OccupationsOccupation Production Non-ProductionThreshold Highschool CollegeDependent Variable ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)OLS IV OLS IV OLS IV OLS IV(1) (2) (3) (4) (5) (6) (7) (8)Import Status 0.479*** 3.705*** 0.055*** 0.990*** 0.249*** 3.643*** 0.023 0.698***[0.108] [1.320] [0.017] [0.280] [0.094] [1.130] [0.015] [0.238]Export Status -0.043 -0.263** 0.018 -0.053* -0.174** -0.486*** 0.018 -0.034[0.074] [0.127] [0.013] [0.028] [0.076] [0.143] [0.012] [0.023]Wagej06 -0.101 -0.161 -0.010 -0.015 -0.064 -0.195 -0.030* -0.026[0.164] [0.191] [0.023] [0.029] [0.120] [0.148] [0.016] [0.019]Capital 0.124*** 0.072** 0.024*** 0.012** -0.015 -0.080** 0.014*** 0.004[0.018] [0.029] [0.003] [0.005] [0.017] [0.031] [0.003] [0.005]Hicks-neutral, ϕ -0.259*** -0.316*** -0.007 -0.021* 0.020 -0.054 0.030*** 0.020*[0.049] [0.061] [0.008] [0.011] [0.049] [0.061] [0.008] [0.010]Foreign-Owned 0.068 -0.243 0.007 -0.105* 0.124 -0.312 0.015 -0.067[0.150] [0.250] [0.030] [0.057] [0.146] [0.264] [0.028] [0.048]R&D 0.025 -0.169 0.029* -0.014 0.072 -0.148 0.024 -0.010[0.102] [0.151] [0.017] [0.030] [0.094] [0.151] [0.016] [0.025]Training 0.212*** 0.124 0.048*** 0.026* 0.020 -0.086 0.045*** 0.025*[0.061] [0.079] [0.010] [0.015] [0.058] [0.081] [0.009] [0.013]Wagej96 -0.543*** -0.657*** -0.114*** -0.137*** 0.331** 0.165 0.030* 0.013[0.190] [0.226] [0.030] [0.040] [0.133] [0.181] [0.018] [0.023]ln(Ljs/Lju)96 0.362*** 0.335*** 0.279*** 0.216***[0.023] [0.029] [0.031] [0.043]dju 0.282 0.033 -0.025 -0.036[0.201] [0.264] [0.124] [0.149]djs -0.993*** -0.952*** -0.444*** -0.315***[0.074] [0.085] [0.076] [0.100](LjsLjs+Lju)960.439*** 0.383*** 0.191*** 0.163***[0.020] [0.031] [0.026] [0.031]Industry FE Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes YesR2 0.343 — 0.395 — 0.169 — 0.143 —Hansen J p-value — 0.153 — 0.141 — 0.336 — 0.185No. Obs 3,139 3,111 4,445 4,410 2,108 2,089 4,021 3,988Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initialnon-importers is used in all regressions. The education threshold used to determine a skilled production workeris a highschool diploma, while the threshold used for a skilled non-production worker is a college degree. Importstatus is treated as an endogenous variable in columns (2), (4), (6) and (8). It is instrumented with both thedistance to port and the share of imports shipped by air.542.4. Resultsfact that Indonesia has a comparative advantage in unskilled-labor intensive goods. Thesignificant positive capital and training coefficients indicate both capital and training arecomplementary to hiring skilled labor. On the other hand, foreign ownership is often neg-atively associated with the demand for skilled labor, suggesting that foreign ownership isa substitute for skill-intensive production processes (e.g. by offshoring the skill-intensiveportion of production abroad). The estimated coefficient on Hicks-neutral productivity ϕ isnegative, which suggests a trade-off between the adoption of skill-biased technology and theadoption of technology that is unbiased across skill differences. The coefficient on relativewages in 2006 is negative, as expected, but insignificant.46 The estimated coefficient on thelagged value of the outcome variable is positive, statistically significant, and consistentlyestimated to lie between 0 and 1. This may reflect either the persistence of unobserved char-acteristics that affect the plant’s demand for skilled labor or the presence of adjustment costsassociated with changing the plant’s skill ratio.Table 2.6 considers the plant-level demand for skill across all workers in a plant. Specif-ically, columns (1)-(4) consider the impact of importing on the relative demand for workerswith at least a highschool diploma and columns (5)-(8) similarly examine the impact ofimporting on the relative demand for workers with a college degree. We also consider speci-fications where we use the log ratio of non-production to production workers or the fractionof the workforce engaged in non-production activities as a dependent variable in columns(9)-(12). These last exercises allow us to compare whether existing, common measures ofskill-intensity, namely the fraction non-production workers in a plant, provide meaningfullydifferent results from education-based measures of skill.47The first two exercises in Table 2.6 present results which are similar to those in Table 2.5.In particular, the results from the regressions using highschool as the skill threshold closelyresemble the results for production workers, while the results from the regressions whichdefine college as the skill threshold are similar to our results which examine non-productionworkers alone. As above, in all cases we find that importing has a large, positive and highlysignificant impact on the demand for skilled labor. In comparison, the estimated coefficientson non-production intensity are positive in columns (9)-(12) but only marginally significantin one of four columns. The results are broadly consistent with our decomposition analysisand suggest that importing is mainly inducing skill-upgrading within each occupation group46In each case, the relative wage variable, the lagged dependent variable, and the lagged indicator variablesare defined consistently with the skill threshold used in each regression. For instance, in columns (1)-(4) it isthe relative differences between workers with a highschool degree and those without, while in columns (5)-(8)it measures the wage differences will college educated workers and those without a college degree. Details onthe construction of these variables can be found in the Appendix A.1.47See Bernard and Jensen (1997a), Harrison and Hanson (1999), Pavcnik (2003), and Biscourp and Kramarz(2007) for examples of papers which use the ratio of non-production to production workers as a measure ofskill.552.4. Resultswhile the skill upgrading through reallocation from production workers to non-productionworkers plays, at best, the secondary role.On the surface, our results might appear inconsistent with the result from Amiti andCameron (2012) (pages 285-286) which “shows that relative education intensity of produc-tion workers relative to nonproduction workers actually declined between 1996 and 2006 inimporting and exporting firms relative to domestically-oriented firms.” However, it is im-portant to distinguish key differences across these empirical exercises. Specifically, Amitiand Cameron study the correlation of current import status with the relative growth ofeducation-intensity across occupations, we focus on impact of starting to import on within-plant or within-plant-and-occupation skill upgrading.48Tables 2.7 and 2.8 report a number of robustness checks for our benchmark results. Wefirst estimate our specification in first differences by the IV regression. The differencedspecification inherently controls for any time-invariant unobserved heterogeneity in equation(2.5) at the plant-level. Moreover, including both industry and region dummies allows us tocondition our results on any differential trends across regions or industry. Last, in columns(2), (4), (6), (8) and (10), we drop changes in city-level relative wages as a control variableand instead directly control for the change in relative supply of skilled to unskilled labor ineach location.49Table 2.7 reports point estimates that are consistently large and positive regardless ofwhether we control for changes in relative wages or the relative supply of skilled labor.Within occupations, we continue to find that importing has a highly significant impact onthe relative demand for skill among production workers, but less so among non-productionworkers. Examining all workers together, we again find strongly significant results when weuse either skill threshold and marginally significant point estimates when we consider thefraction of non-production workers. Across all cases, we recover substantially smaller pointestimates in the first differenced specification relative to the benchmark results.50Table 2.8 presents a series of further robustness checks. Although we only report key48Table B.4 in the Appendix replicates Column 2 of Table 8 inAmiti and Cameron (2012) to the best ofour ability and, furthermore, we investigate the relationship between changes in relative education intensityand plant-level importing dynamics. Using the change in import status between 1996 and 2006 in place ofthe 1996 import status in their specification leads to a result where the change in the relative educationintensity is positively correlated with the change in import status between 1996 and 2006. One possibleinterpretation of this positive correlation between the change in relative education intensity and the changein import status is that starting to import induces more education-upgrading within production workers thanwithin non-production workers.49Regional supply of skilled and unskilled workers are simple counts of working age population with edu-cational attainment above and below the skill threshold (high-school or college). Across all columns we focuson the skilled fraction of the plant’s workforce because there are many plants that did not hire any skilledworkers in 1996, which forces us to drop more plants from our sample in the log specification.50The first differenced specification could bias the point estimates downwards if the specification with laggeddependent variable is the correct specification. The Appendix B.4 provides a detailed argument that extendsthe argument of Angrist and Pischke (2008) (pp. 184-185) in the context of the IV regression.562.4. Resultscoefficients in Table 2.8, a full set of controls are included in each regression. The top panel(regressions (1)-(10)) reconsiders our benchmark framework but uses our alternative measureof the relative supply of skilled labor in each location in place of the skill premium. Importingcontinues to have a large, positive impact on the demand for skill, while the coefficients onthe relative supply of skilled labor are positive.572.4.ResultsTable 2.6: Skill Demand Equation for All Workers in LevelsThreshold Highschool College OccupationDependent Variable ln(Ls/Lu)(LsLs+Lu)ln(Ls/Lu)(LsLs+Lu)ln(Ln/Lp)(LnLn+Lp)OLS IV OLS IV OLS IV OLS IV OLS IV OLS IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)Import Status 0.409*** 4.503*** 0.042*** 0.930*** 0.162* 3.313*** 0.016*** 0.278*** 0.020 1.411* 0.007 0.101[0.099] [1.585] [0.015] [0.258] [0.085] [1.093] [0.005] [0.073] [0.060] [0.755] [0.009] [0.107]Export Status 0.020 -0.293* 0.016 -0.052** -0.230*** -0.543*** -0.009*** -0.030*** -0.099** -0.222*** -0.015** -0.023**[0.067] [0.153] [0.012] [0.026] [0.065] [0.139] [0.003] [0.007] [0.049] [0.083] [0.007] [0.011]Wagej06 -0.094 -0.157 0.008 0.004 -0.072 -0.177 -0.007* -0.008 0.031 0.048 -0.002 -0.001[0.139] [0.171] [0.021] [0.027] [0.110] [0.144] [0.003] [0.005] [0.102] [0.105] [0.015] [0.014]Capital 0.116*** 0.050 0.021*** 0.010** 0.028* -0.042 0.005*** 0.002 0.027*** 0.004 0.005*** 0.003[0.016] [0.033] [0.003] [0.005] [0.016] [0.033] [0.001] [0.001] [0.010] [0.016] [0.001] [0.002]Hicks-neutral, ϕ -0.188*** -0.259*** -0.010 -0.024** 0.015 -0.051 0.004* 0.000 -0.118*** -0.129*** -0.007 -0.007[0.047] [0.062] [0.007] [0.010] [0.044] [0.058] [0.002] [0.003] [0.031] [0.035] [0.005] [0.005]Foreign-Owned 0.013 -0.314 0.017 -0.091* 0.144 -0.415 0.005 -0.027* -0.051 -0.249* -0.011 -0.024[0.152] [0.269] [0.027] [0.054] [0.130] [0.278] [0.009] [0.016] [0.099] [0.147] [0.014] [0.020]R&D 0.192** -0.039 0.030* -0.012 0.163** 0.021 0.022*** 0.010 0.168** 0.105 0.025** 0.022*[0.092] [0.163] [0.016] [0.028] [0.077] [0.123] [0.006] [0.009] [0.075] [0.090] [0.012] [0.013]Training 0.278*** 0.203*** 0.044*** 0.023 0.114** -0.004 0.017*** 0.009** 0.061 0.021 0.012** 0.009[0.055] [0.075] [0.009] [0.014] [0.053] [0.085] [0.002] [0.004] [0.037] [0.047] [0.005] [0.006]Wagej96 -0.619*** -0.769*** -0.114*** -0.134*** 0.427*** 0.219 0.017*** 0.011* 0.109 0.034 0.034** 0.026[0.172] [0.217] [0.027] [0.038] [0.119] [0.177] [0.004] [0.006] [0.119] [0.125] [0.016] [0.016]ln(Ls/Lu)96 0.449*** 0.409*** 0.384*** 0.311***[0.023] [0.033] [0.026] [0.044]du -0.074 -0.063 -0.313** -0.231[0.047] [0.060] [0.136] [0.187]ds 0.058 0.022 0.139** 0.104[0.061] [0.075] [0.058] [0.081](LsLs+Lu)960.484*** 0.436*** 0.282*** 0.225***[0.018] [0.029] [0.038] [0.050]ln(Ln/Lp)96 0.392*** 0.397***[0.019] [0.020]dp -0.119*** -0.131***[0.035] [0.037]dn 0.094*** 0.082**[0.032] [0.035](LnLn+Lp)96-0.119*** -0.131***[0.035] [0.037]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesR2 0.395 — 0.440 — 0.340 — 0.247 — 0.251 — 0.237 —Hansen J p-value — 0.234 — 0.202 — 0.782 — 0.284 — 0.360 — 0.874No. Obs 3,434 3,405 4,445 4,410 1,657 1,641 4,445 4,410 4,021 3,988 4,445 4,410Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in all regressions. Importstatus is treated as an endogenous variable in columns (2), (4), (6), (8), (10) and (12). It is instrumented with both the distance to port and the share of importsshipped by air.582.4.ResultsTable 2.7: Robustness Checks: The Skill Demand Equation in DifferencesOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationDep. Var. ∆(LpsLps+Lpu)∆(LnsLns+Lnu)∆(LsLs+Lu)∆(LsLs+Lu)∆(LnLn+Lp)IV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)∆ Import Status 0.779** 0.801** 0.255 0.251 0.540** 0.541** 0.175** 0.178** 0.259* 0.261*[0.318] [0.325] [0.217] [0.219] [0.250] [0.252] [0.070] [0.071] [0.157] [0.158]∆ Export Status -0.010 -0.010 0.002 0.001 0.003 0.005 -0.007 -0.007 -0.015 -0.015[0.022] [0.022] [0.017] [0.017] [0.018] [0.018] [0.005] [0.005] [0.011] [0.011]∆ Wage 0.046 -0.012 0.056** -0.007* -0.008[0.028] [0.016] [0.023] [0.004] [0.015]∆ Skill Supply 0.034** -0.007 0.039*** -0.002 0.001[0.016] [0.008] [0.013] [0.002] [0.008]∆ Capital -0.005 -0.005 0.007** 0.007** -0.005 -0.004 0.001 0.001 0.000 0.000[0.004] [0.004] [0.003] [0.003] [0.003] [0.003] [0.001] [0.001] [0.002] [0.002]∆ Hicks-neutral, ϕ -0.043*** -0.045*** 0.024** 0.024*** -0.040*** -0.041*** 0.001 0.001 -0.013** -0.013**[0.011] [0.011] [0.009] [0.009] [0.010] [0.009] [0.003] [0.003] [0.006] [0.006]∆ Foreign-Owned -0.002 -0.003 -0.022 -0.021 0.006 0.003 -0.006 -0.006 0.014 0.014[0.038] [0.038] [0.031] [0.031] [0.030] [0.030] [0.012] [0.012] [0.017] [0.017]∆ R&D -0.010 -0.009 -0.020 -0.021 -0.012 -0.011 -0.005 -0.005 -0.011 -0.012[0.027] [0.027] [0.019] [0.019] [0.021] [0.021] [0.007] [0.007] [0.014] [0.014]∆ Training -0.011 -0.011 0.012 0.012 -0.003 -0.003 0.002 0.002 0.001 0.001[0.014] [0.014] [0.011] [0.011] [0.011] [0.012] [0.003] [0.003] [0.007] [0.007]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesHansen J p-value 0.334 0.311 0.369 0.392 0.214 0.253 0.159 0.142 0.137 0.118No. Obs 3,366 3,361 3,366 3,345 3,366 3,361 3,366 3,345 3,366 3,361Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in all regressions. The educationthreshold used to determine a skilled production worker is a highschool diploma, while the threshold used for a skilled non-production worker is a college degree.Import status is treated as an endogenous variable in all regressions. It is instrumented with both the distance to port and the share of imports shipped by air.592.4. ResultsThe second panel (regressions (11)-(20)) includes our additional instruments for the de-cision to import; specifically, we augment our benchmark instrument set with the weight ofimported goods and the tariffs faced by importers on intermediate inputs. In each case weobserve a coefficient which is similar is size and magnitude to our benchmark IV findings inTables 2.5 and 2.6.The third panel (regressions (21)-(30)) includes both a measure of the plant-level intensityof importing, the fraction of total intermediates imported from abroad, along with the importstatus dummy variable. By including both import status and import intensity we investigatethe degree to which the impact of importing on plant-level skill composition is manifestedthrough changes in the intensive or extensive import margins. The import status coefficientis always positive and nearly always statistically significant. In contrast, import intensity isconsistently estimated to negative and is never significant. As such, we conclude the plant-level changes in skill demand largely occur through the extensive margin of importing.51In the fourth panel (regressions (31)-(40)) we replace our measure of Hicks-neutral pro-ductivity with a conventional measure of TFP. Specifically, we replace Hicks-neutral TFPwith an estimate of the Solow residual from a Cobb-Douglas production function which usescapital, materials, production workers and non-production workers as inputs. Again, werecover very similar point estimates relative to our benchmark results.52The fifth panel addresses the concern that exporting is likely to be an endogenous decisionin this context. For instance, given the evidence that importing and exporting are closelyrelated activities (see Kasahara and Lapham (2013)), ignoring the endogeneity of the plant’sexport decision may lead to bias in the estimated coefficient on import status. Regressions(41)-(50) estimate the skill equation while instrumenting both import and export status withtransport costs, air share, and the changes output tariffs and market access tariffs as definedin Section 2.3.53 We continue to find that importing has a large, positive impact of thedemand for skilled workers. In contrast, the impact of exporting on the demand for skill isestimated to be insignificantly different from zero in all but one column.Feenstra and Hanson (1999) and Feenstra and Hanson (1997) present a model with acontinuum of goods where the most skill-intensive goods in developing countries correspondto the least skill-intensive goods in developed countries. Trade liberalization induces the most51If we exclude import status, the import intensity is always positive and statistically significant.52Hicks-neutral productivity is estimated to take a negative coefficient in Tables 2.5-2.6, while our naivelyestimated TFP takes the opposite sign in the fourth robustness check of Table 2.8, even if it is alwaysinsignificant in this latter case. While these results may seem contradictory, they are exactly what we shouldexpect in this instance. By ignoring the skill-biased component of productivity, the conventional TFP measureconfuses both the skill-biased and Hicks-neutral components and, as a result, is likely to be positively correlatedwith the demand for skilled labour. In contrast, the Hicks-neutral productivity term we estimate disentanglesthese two components of productivity. Plants with larger values of skill-biased productivity will naturally bemore likely to have smaller measured Hicks-neutral productivity.53First stage results for the decision to export can be found in the Appendix.602.4. Resultsskill-intensive goods in developing countries to be exported to developed countries, leading toan increase in the demand for skilled labor in developing countries. The Feenstra and Hansenhypothesis does not appear to hold in our data since none of the estimated coefficients onexporting are significantly positive in regression (41)-(50).Although we document evidence that importing leads to an increase in the demand forskill, the mechanism behind this result has been largely investigated thus far. As we discussed,one plausible mechanism is that importing induces the adoption of skill-biased technology.While there is no direct data on foreign technology adoption, our data set includes a vari-able which captures whether a plant adopts a standardized production process, such as thoserecognized by the International Organization for Standardization (ISO) or the InternationalElectrotechnical Commission (IEC).54 The use of standards may allow for improved coordi-nation with foreign suppliers, facilitating the adoption of foreign skill-biased technology.In the sixth panel, we estimate the effect of standards on the demand for skilled productionworkers using our benchmark specifications of Table 2.5 and 2.6 but replacing the importdummy with a dummy for standardized production, where we focus on the sample of non-exporters since standardization is closely associated with exporting activities.55 Regressions(51)-(60) indicate that the adoption of standards significantly increases the demand for bothskilled production and non-production workers.To further explore the relationship between standardization and importing, we also con-sider a linear regression model of the decision to adopt standardized production. In particular,we regress our standardization dummy variable on import status and a full set of controlsusing the sample of non-exporting firms.56 In all columns of Table 2.9 we find that the pointestimate on importing is both large and positive. Moreover, in 8 out of 10 columns of Table2.9 the estimate is reported to be at least marginally significant. Although these resultsare hardly overwhelming, they are consistent with the hypothesis that importing inducesstandardization and, thus, skill-biased technological change.54Specifically, the survey question asks “Does this establishment use standard of production process?” withthe following list of standards: ISO (International Organisation for Standardization), IEC (International Elec-trotechnical Commission), ITU (International Telecommunication Union), CAC (Codex Alimentarius Com-mission), AFNOR (Association Francaise de Normalisation), ANSI (American National Standard Institute),BIS (Bureau of India Standard), BSI (British Standards Insitution), DIN (Deutshes Institute for Nonnungev), JISC (Japanese Industrial Standartds Commitee), SAL (Standards Australia), SNI (Standar NasionalIndonesia), ASTM (American Society for Testing and Material), ASME (American Standard of MechanicalEngineering), and NFPA (National Fire Protection Association). Unfortunately, no further information onwhich standards are used is available.55Transport costs are typically strong predictors of the use of standards.56Explicitly instrumenting for endogenous export decisions returns very similar point estimates and statis-tical significance for the import status variable.612.4. ResultsTable 2.8: Robustness ChecksOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationDep. Var. ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)ln(Ls/Lu)(LsLs+Lu)ln(Ls/Lu)(LsLs+Lu)ln(Ln/Lp)(LnLn+Lp)IV IV IV IV IV IV IV IV IV IVSkill Supply Controls(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Import Status 2.448* 0.778*** 3.783*** 0.513** 3.388** 0.660*** 3.041** 0.226*** 0.786 0.024[1.364] [0.274] [1.344] [0.231] [1.684] [0.238] [1.268] [0.070] [0.812] [0.120]Skill Supply06 0.244*** 0.050*** 0.131** 0.017** 0.185*** 0.053*** 0.029 0.004* 0.035 0.003[0.070] [0.012] [0.061] [0.008] [0.065] [0.011] [0.059] [0.002] [0.039] [0.006]Skill Supply96 -0.006 -0.008 -0.075 0.009 0.019 -0.006 0.038 0.003 0.052 0.006[0.083] [0.015] [0.073] [0.008] [0.080] [0.013] [0.068] [0.002] [0.045] [0.006]Large IV set(11) (12) (13) (14) (15) (16) (17) (18) (19) (20)Import Status 3.240*** 0.864*** 3.498*** 0.672*** 4.105*** 0.818*** 3.191*** 0.267*** 1.557** 0.116[1.247] [0.254] [1.049] [0.230] [1.496] [0.235] [1.072] [0.069] [0.746] [0.108]Import Intensity(21) (22) (23) (24) (25) (26) (27) (28) (29) (30)Import Status 7.309* 1.640** 5.498 1.420** 8.128* 1.458** 3.825 0.445** 2.945 0.148[3.974] [0.658] [3.640] [0.675] [4.632] [0.600] [2.554] [0.179] [1.936] [0.238]Import Share -8.143 -1.479 -2.563 -1.445 -7.487 -1.200 -0.715 -0.354 -3.417 -0.116[8.230] [1.365] [6.897] [1.323] [8.824] [1.233] [3.979] [0.375] [3.768] [0.475]TFP Measurement(31) (32) (33) (34) (35) (36) (37) (38) (39) (40)Import Status 3.151*** 0.947*** 3.885*** 0.724*** 3.968*** 0.885*** 3.264*** 0.275*** 1.226* 0.089[1.211] [0.270] [1.133] [0.238] [1.451] [0.248] [1.072] [0.071] [0.729] [0.105]Solow Residual 0.001 0.011 0.023 0.014 0.022 0.008 0.031 0.003 -0.032 -0.000[0.044] [0.009] [0.050] [0.008] [0.046] [0.008] [0.048] [0.002] [0.027] [0.004]Instrumenting Export Status(41) (42) (43) (44) (45) (46) (47) (48) (49) (50)Import Status 2.977** 0.909*** 4.240** 0.888*** 4.446** 0.947*** 0.933 0.191** 0.739 0.014[1.289] [0.305] [1.701] [0.330] [1.840] [0.300] [0.893] [0.075] [0.825] [0.121]Export Status -0.551 0.073 0.232 0.136 0.142 0.088 -1.419*** -0.019 -0.719 -0.093[0.558] [0.124] [0.859] [0.116] [0.620] [0.120] [0.492] [0.029] [0.470] [0.070]Standards(51) (52) (53) (54) (55) (56) (57) (58) (59) (60)Standards 5.354* 1.365** 1.931* 0.828** 5.976* 1.220** 0.502 0.222** 1.053 -0.078[3.000] [0.618] [1.148] [0.393] [3.083] [0.559] [0.816] [0.093] [1.230] [0.178]Capital-Skill Complementarity(61) (62) (63) (64) (65) (66) (67) (68) (69) (70)Import Status 5.716*** 1.476*** 3.639* 0.525 7.500*** 1.460*** 3.113** 0.273** 2.609** 0.184[2.130] [0.464] [2.060] [0.378] [2.901] [0.465] [1.521] [0.137] [1.159] [0.126]Capital-Skill Comp. 0.388 0.081 -0.021 -0.017 0.513 0.114 -0.143 0.002 0.246 0.004[0.511] [0.090] [0.567] [0.093] [0.862] [0.108] [0.423] [0.031] [0.189] [0.016]Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in allregressions. Import status is treated as an endogenous variable in all columns. Industry and region fixed effects are included in all regressions.Columns (51)-(60) focus on the sample of non-exporters since exporting and production standardization are highly correlated activities.622.4. ResultsTable 2.9: Importing and Standardized ProductionDep. Var. StandardsOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationIV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Import Status 1.491* 1.425* 1.819* 1.622* 2.268 1.395* 1.113 1.851* 1.467* 1.517*[0.877] [0.851] [1.002] [0.966] [2.112] [0.843] [1.282] [1.055] [0.850] [0.826]Control Vars. Yes Yes Yes Yes Yes Yes Yes Yes Yes YesIndustry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesHansen J p-value 0.460 0.428 0.399 0.379 0.647 0.421 0.202 0.411 0.472 0.426No. Obs 3,329 3,329 3,329 2,958 2,720 3,329 1,194 3,329 2,958 3,329Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initialnon-importers is used in all regressions. The education threshold used to determine a skilled production worker is ahighschool diploma, while the threshold used for a skilled non-production worker is a college degree. Import status istreated as an endogenous variable in all columns. It is instrumented with both the distance to port and the share ofimports shipped by air. All full set of control variables is included in each regression. Detailed results are reported inthe Appendix.Finally, we also consider whether importing leads to skill-upgrading through a capital-skill complementarity mechanism, rather than the framework posited in Section 2.2. Wefirst extend our model to allow for potential capital-skill complementarity as presented inAppendix B.5. The firm’s cost minimization problem will then imply that the relative demandfor skilled labor can be written as:LjsLju=(βWuWs)σj(Aj)σj−1(K(Lps + Lns ))(σj−1)(1−β). (2.10)It is clear that capital-skill complementarity implies adding one additional variable to ourbenchmark empirical specification, the log ratio of capital to total (production and non-production) skilled labor, ln (K/(Lps + Lns )). Unfortunately, including ln (K/(Lps + Lns )) onthe right-hand side of equation (2.5) is likely to induce endogeneity bias since skilled labordetermines both outcome and explanatory variables. As such, regressions (61)-(70) of Table2.8 document IV estimates of the impact of importing on the demand skilled labor whilealso instrumenting the endogenous capital-skill control using lagged (i.e., 1996) values ofln(K/(Ljs + Ljs))as an additional instrument. We find that the impact of importing onthe demand for skilled production workers is nearly unaffected by controlling for capital-skillcomplementarity; the point estimates on the import status variable are of a similar magnitudeand significance as the benchmark regressions in Tables 2.5 and 2.6.632.5. Marginal Treatment Effects0 0.2 0.4 0.6 0.8 1 1.2p00.10.20.30.40.50.6f(p)Density of Import Probabilities for Lsp/(Lup+Lsp)D=0D=1Figure 2.2: Support of Estimated Propensity Scores2.5 Marginal Treatment Effects2.5.1 Propensity ScoresWe estimate the import probabilities for each plant using a logit specification where we includethe interaction terms between instruments and the lagged value of the outcome variable in1996 as well as dummy variables for plants that did not hire any skilled or unskilled workersin 1996. Transport costs, air shares, and weights are included as instruments. Table B.3 inthe Appendix reports the estimate of the coefficient and marginal derivative for each variablewith bootstrapped standard errors, where we find that transport costs and air shares arealways a strong predictor of importing.Figure 2.2 plots the distribution of estimated propensity scores for importing and non-importing plants. It is evident that the common support of the propensity scores acrossimporting and non-importing plants does not span the full unit interval. For this reason,we restrict our computation of treatment effects to the region where there is significantoverlap between the propensity scores of non-importing and importing plants as reported inthe second to last row of Table 2.10; specifically, treatment effects are computed over theregion where the minimum and maximum values are given by the 1st percentile and the99th percentile values of the estimated propensity scores for which we have common support,respectively. Because there are very few non-importing plants with propensity scores beyondthe upper bound of this range, it is difficult to apply nonparametric methods and confidentlyestimate the MTE outside of this range.642.5. Marginal Treatment Effects2.5.2 Treatment EffectsFigure 2.3 plots the estimated relationships between the MTE and UD along with 90 percent(equal-tailed) bootstrap confidence bands across 5 different outcome variables using the shareof skilled workers as the dependent variable.57 As shown in Figure 2.3(a)-(d), the estimatedMTE curve for production workers is well above zero for small values of UD and is downwardsloping for all of our four education-based skill measures. These findings provide evidencethat plants self-selected into importing based on the plant-specific, unobserved component ofthe net benefit from importing—among plants that choose to import when their propensityscores, estimated in terms of observables, are low, the unobserved ability to adopt the skilledbiased technology must be high. On the other hand, in Figure 2.3(e), when we use theshare of non-production workers as the outcome variable, the estimated MTE curve is notsignificantly different from zero across all values of UD.Figure 2.4 graphs the estimated weights for computing different treatment parameterswhen we use Lps/(Lps + Lpu) as the outcome variable. While the TT heavily overweightsindividual plants with low levels of UD, the TUT overweights those with high levels of UD.By construction, the ATE equally weights different values of UD. If the MTE curve were flat,there would be no self-selection based on the unobservable gains, and the ATE would equal tothe TT and the TUT. The fact that the estimated MTE curve is downward sloping suggeststhe presence of selection bias from the unobserved, heterogeneous return to importing andinvalidates the use of a simpler propensity score matching methods for estimating the ATEin our context.Table 2.10 reports the estimates of various summary measures of the impact of importingon skill demand: the ATE, the TT, the TUT, and policy relevant treatment effects (theMPRTEs and the PRTEs). These treatment effects are computed as the weighted averagesof the MTE, where these weights integrate to one over the restricted support reported in thesecond to last row of Table 2.10. Bootstrap standard errors and the 90 percent equal-tailedbootstrap confidence interval are reported in square brackets and parentheses, respectively.Appendix B.2 discusses the details of our estimation procedure.The first two columns of Table 2.10 report the estimated treatment effects for productionworkers. The ATE, the TT, and the TUT for production workers are estimated to be positiveand statistically significant, indicating that importing increases the demand for educatedproduction workers across different groups of plants. Furthermore, the TT is estimatedto be substantially larger than the ATE which, in turn, is larger than the TUT. This is57The import decision model is estimated for each bootstrap sample so that the first stage estimation erroris taken into account. Table B.1 in the Appendix reports the estimates of the skill demand equation usingthe sample of plants for which the estimated propensity scores are on the estimated common support. Theresults for using the log of the skill ratios as the outcome variables are reported in Figure B.4 and are similarto those in Figure 2.3.652.5. Marginal Treatment Effects0 0.1 0.2 0.3 0.4 0.5 0.6UD-0.500.511.522.533.544.5MTEMTE(p) for Lsp/(Lup+Lsp) with 90 percent confidence band(a) Production, Highschool0 0.1 0.2 0.3 0.4 0.5 0.6UD-0.500.511.522.53MTEMTE(p) for Lsn/(Lun+Lsn) with 90 percent confidence band(b) Non-Production, College0 0.1 0.2 0.3 0.4 0.5 0.6UD-10123456MTEMTE(p) for Ls/(Lu+Ls) for highschool+ with 90 percent confidence band(c) All, Highschool0 0.1 0.2 0.3 0.4 0.5 0.6UD-0.100.10.20.30.40.50.60.70.80.9MTEMTE(p) for Ls/(Lu+Ls) for college+ with 90 percent confidence band(d) All, College0 0.1 0.2 0.3 0.4 0.5 0.6UD-0.4-0.200.20.40.60.811.21.4MTEMTE(p) for Ln/(Ln+Lp)with 90 percent confidence band(e) All, OccupationFigure 2.3: Estimated MTE662.5. Marginal Treatment Effects0 0.1 0.2 0.3 0.4 0.5 0.6UD00.050.10.150.20.250.3WeightWeight function for PRTE and MPRTE (outcome variable: Lsp/(Lup+Lsp))ATETTTUTPRTE: (1-,)TCMPRTE: P,=P+,MPRTE: Zk,=Zk+,Figure 2.4: Estimated Weights for ATE, TT, TUT, MPRTEs, and PRTE (De-pendent variable: Lps/(Lps + Lpu))potentially indicative of substantial unobserved heterogeneity in the effect of importing onskill demand across plants and that plants with greater returns from importing self-selectinto importing. While plants that were induced to import witnessed large increases in thedemand for skilled production workers, the counterfactual impact of importing on the demandfor skilled production workers is substantially smaller among plants that choose not to importin 2006.As Columns (3)-(8) of Table 2.10 show, we observe similar patterns for the orderingamong the TT, the ATE, and the TUT across different education-based measures of skilldemand. However, the skill measures based on the ratio of non-production to productionworkers reported in columns (9)-(10) indicate that the ATE, the TT, and the TUT are notsignificantly different from zero.Table 2.11 examines the robustness of our results using different specifications and esti-mation methods, where we focus on the share of skilled workers within production workers ornon-production workers as the outcome variable. Columns (1) and (5) report the estimatesof treatment effects when we use conventional TFP in place of our estimated Hicks-neutralproductivity.672.5.MarginalTreatmentEffectsTable 2.10: Treatment Effects of Importing on Skill DemandOccupation Production Non-Production All All AllThreshold Highschool College Highschool College Occupation(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Dependent Var. ln(LpsLpu)06(LpsLpu+Lps)06ln(LnsLnu)06(LnsLnu+Lns)06ln(LsLu)06(LsLu+Ls)06ln(LsLu)06(LsLu+Ls)06ln(LpLp)06(LpLp+Lp)06ATE 1.897 0.690 3.231 0.623 2.390 0.651 2.099 0.188 0.496 0.084[0.926] [0.175] [0.883] [0.146] [0.734] [0.199] [0.751] [0.046] [0.608] [0.102](0.70,3.82) (0.51,1.07) (2.75,5.67) (0.45,0.95) (1.72,4.08) (0.46,1.13) (1.31,3.71) (0.14,0.29) (-0.34,1.69) (-0.04,0.29)TT 5.041 2.127 4.193 1.235 6.377 2.477 3.153 0.442 1.663 0.368[2.158] [0.581] [1.521] [0.408] [2.223] [0.703] [1.255] [0.113] [1.704] [0.312](2.66,9.61) (1.59,3.46) (3.11,7.95) (0.74,2.14) (4.62,12.04) (1.96,4.18) (1.91,6.09) (0.34,0.73) (-0.31,5.36) (-0.01,0.98)TUT 1.543 0.550 3.093 0.563 2.027 0.492 1.905 0.163 0.379 0.057[0.871] [0.145] [0.824] [0.131] [0.671] [0.164] [0.694] [0.043] [0.553] [0.086](0.46,3.30) (0.39,0.85) (2.67,5.38) (0.39,0.85) (1.35,3.52) (0.32,0.88) (1.10,3.35) (0.11,0.25) (-0.40,1.42) (-0.06,0.23)MPRTE 4.379 1.859 3.904 1.104 5.692 2.091 2.932 0.393 1.471 0.313(P ∗α = P + α) [1.867] [0.494] [1.356] [0.351] [1.911] [0.588] [1.135] [0.098] [1.455] [0.267](2.34,8.32) (1.39,2.99) (2.98,7.32) (0.69,1.89) (4.18,10.48) (1.65,3.51) (1.79,5.59) (0.31,0.64) (-0.19,4.58) (-0.01,0.85)MPRTE 4.241 1.793 3.834 1.078 5.500 2.000 2.896 0.384 1.425 0.301(Zk∗α = Zk + α) [1.783] [0.477] [1.319] [0.340] [1.800] [0.568] [1.114] [0.095] [1.372] [0.257](2.27,8.07) (1.34,2.88) (2.90,7.16) (0.68,1.83) (4.11,10.01) (1.55,3.36) (1.76,5.53) (0.30,0.62) (-0.13,4.37) (-0.00,0.81)PRTE 4.518 1.915 3.949 1.138 5.819 2.185 2.951 0.405 1.507 0.323[6.640] [0.518] [1.394] [0.366] [2.236] [0.623] [1.143] [0.102] [1.494] [0.273](2.43,8.73) (1.43,3.10) (2.99,7.40) (0.70,1.95) (4.30,10.74) (1.72,3.68) (1.81,5.64) (0.31,0.66) (-0.19,4.69) (-0.00,0.87)Support(c) [0.01,0.58] [0.01,0.58] [0.00,0.57] [0.01,0.57] [0.01,0.60] [0.01,0.59] [0.01,0.54] [0.01,0.56] [0.00,0.56] [0.01,0.54]No. of Obs.(d) 2820 3997 1898 3992 3452 3985 2216 3967 3960 4000Notes: (a) The bootstrap standard errors are in square brackets. (b) The bootstrap equal-tailed 90 percent confidence intervals are in parentheses. (c) The minimumand the maximum values of support over which treatment effects are computed; various treatment effects are computed by restricting the weights to integrate to onein the restricted support, for which minimum and maximum values are determined by the 1st percentile and the 99th percentile of observations in the commonsupport, respectively. (d) The sample size for estimating the MTE curve.682.5.MarginalTreatmentEffectsTable 2.11: Robustness Check: Treatment Effects of Importing on Skill Demand for Production WorkersOccupation Production Non-ProductionThreshold Highschool CollegeDep. Var.(LpsLpu+Lps)06(LnsLnu+Lns)06(1) (2) (3) (4) (5) (6) (7) (8)Replace ϕ Use Sieve No Treatment Replace Use Sieve No Treatmentwith in place of Interaction Effects over with in place of Interaction Effects overTFP Local Poly. with Z Common Support TFP Local Poly. with Z Common SupportATE 0.690 0.645 0.819 0.539 0.623 0.502 0.547 0.911[0.173] [0.158] [0.209] [0.281] [0.146] [0.150] [0.139] [0.270](0.51,1.07) (0.48,1.00) (0.60,1.27) (0.18,1.04) (0.45,0.95) (0.28,0.77) (0.39,0.86) (0.82,1.74)TT 2.127 2.168 2.370 2.140 1.235 0.970 1.573 1.242[0.583] [0.560] [0.731] [0.592] [0.406] [0.473] [0.454] [0.415](1.59,3.46) (1.63,3.49) (1.50,3.90) (1.56,3.46) (0.76,2.14) (0.14,1.78) (1.11,2.57) (0.76,2.15)TUT 0.550 0.509 0.671 0.435 0.563 0.460 0.447 0.910[0.143] [0.135] [0.169] [0.298] [0.131] [0.139] [0.118] [0.292](0.39,0.86) (0.36,0.81) (0.50,1.04) (0.01,0.95) (0.40,0.85) (0.24,0.70) (0.30,0.69) (0.79,1.81)MPRTE 1.859 1.855 2.055 1.861 1.104 0.877 1.343 1.108[0.496] [0.466] [0.607] [0.500] [0.350] [0.389] [0.383] [0.355](1.41,2.96) (1.44,2.94) (1.34,3.32) (1.33,2.97) (0.70,1.87) (0.23,1.56) (0.96,2.19) (0.70,1.90)MPRTE2 1.793 1.776 1.996 1.795 1.078 0.858 1.296 1.080[0.478] [0.447] [0.589] [0.483] [0.339] [0.371] [0.372] [0.347](1.37,2.86) (1.37,2.81) (1.30,3.21) (1.29,2.85) (0.68,1.82) (0.25,1.52) (0.92,2.11) (0.68,1.84)PRTE 1.915 1.928 2.140 1.966 1.138 0.902 1.403 1.162[0.519] [0.492] [0.644] [1.291] [0.364] [0.413] [0.402] [0.425](1.45,3.08) (1.48,3.09) (1.39,3.49) (1.34,3.19) (0.72,1.94) (0.20,1.61) (0.99,2.30) (0.70,2.01)Support(c) [0.01,0.58] [0.01,0.58] [0.01,0.56] [0.00,0.87] [0.01,0.57] [0.01,0.57] [0.01,0.55] [0.00,0.87]No. of Obs.(d) 3997 3997 4006 3997 3992 3992 3993 3992Notes: (a) The bootstrap standard errors are in square brackets. (b) The bootstrap equal-tailed 90 percent confidence intervals are in parentheses. (c) The minimumand the maximum values of support over which treatment effects are computed; various treatment effects are computed by restricting the weights to integrate to onein the restricted support, for which minimum and maximum values are determined by the 1st percentile and the 99th percentile of observations in the commonsupport, respectively. (d) The sample size for estimating the MTE curve.692.5. Marginal Treatment EffectsIn columns (2) and (6), we estimate the partial linear model (B.1) where we use a sieveestimator based on the 4th order polynomials in P (Z) instead of the local polynomial esti-mator. Columns (3) and (7) consider a specification of the skill demand equation withoutany interactions between the instruments and the lagged outcome variable, while columns (4)and (8) estimate treatment effects over the estimated common support instead of the subsetof the common support defined by the 1st percentile and the 99th percentile of observationsthat are on the common support.58 The estimates of the ATE, the TT, and the TUT incolumns (1)-(8) of Table 2.11 are significantly positive and exhibit the patterns similar tothose reported in columns (2) and (4) of Table 2.10.2.5.3 Policy ExperimentOur IV and MTE estimates confirm that importing has a substantial impact on the demandfor skilled production workers. Nonetheless, it is less clear how much more skill-upgradingwould be induced by further changes in policy related variables. To examine this issue,we consider alternative policies that change the probability of importing but do not affectpotential outcomes or the unobservables related to import decisions, (S0, S1, V ) defined in(2.6)-(2.7), and compute the mean effect of going from a baseline policy to an alternativepolicy per plant shifted into importing. This treatment effect is called the Policy RelevantTreatment Effect (PRTE) as proposed by Heckman and Vytlacil (2005) and Heckman andVytlacil (2007b). Let P ∗(Z) and P (Z) denote the propensity scores under an alternativepolicy and a baseline policy, respectively.We consider the alternative policy of reducing the cost of shipping goods to the nearestport by 10 percent so that P ∗(Z) is set to the propensity score under the alternative transportcost of TC∗ = 0.9TC. The PRTE under this alternative policy captures the causal impactthat a marginal improvement in roads and infrastructure would have on the relative demandfor skilled workers across plants. Note that this policy change will have a heterogeneousimpact across plants. We compute the estimate of what the PRTE would be when we restrictthe support of the propensity scores to the restricted support reported in the second to the lastrow of Table 2.10.59 We also compute the marginal version of the PRTE called the MarginalPolicy Relevant Treatment Effect (MPRTE) proposed by Carneiro et al. (2010). Given asequence of alternative policies indexed by a scalar variable α such that limα→0 P ∗α(Z) =58To estimate the treatment effects reported in Table 2.11, we use the same specification for the decision toimport as the specification reported in Table B.3 except that, in columns (3) and (7), we exclude the interactionterms between the instruments and the 1996 value of the log of skill ratio from the set of explanatory variables.59As discussed in Carneiro et al. (2010), the PRTE is not identifiable without strong support conditions.To compute the estimate of what the PRTE would be on the restricted support, we replace the value of thepropensity scores with the maximum value of the support whenever the value of the propensity scores underthe alternative policy are larger than the maximum value of the restricted support so that all of the propensityscores under the alternative policy lie on the restricted support.702.6. ConclusionP (Z), the MPRTE is defined as the limit of a sequence of PRTEs as α approaches zero.We consider two policy sequences as described in Carneiro et al. (2010): (i) a policy thatincreases the probability of importing by α so that P ∗α = P + α and (ii) a policy that shiftsone of the components in Z, say Zk, so that Zkα = Zk + α.As reported in columns (1)-(8) of the lower panel of Table 2.10, the estimates of theMPRTEs and the PRTE indicate that the subset of plants that would be induced to startimporting by further policy change would substantially increase their demand for skilledworkers when we use the education-based skill measures. These estimates are not sensitive tochanges in specifications and estimation methods on the whole as shown in the lower panelof Table 2.11. In contrast, when we use the share of non-production workers as the outcomevariable, the estimates of the MPRTEs and the PRTE are not significantly different from zero,providing no evidence that further importing would affect the demand for non-productionworkers relative to production workers.2.6 ConclusionThis paper studies the impact that importing foreign materials has on the demand for edu-cated workers among Indonesian manufacturing plants. We develop a model of heterogeneousmanufacturing plants where the decision to import may be influenced by the adoption ofskill-biased technology. In our model the degree to which importing induces skill-biased tech-nological change is potentially heterogenous across plants and unobservable to the researcher.To the extent that importing affects skill-biased productivity we would expect that it willdirectly impact mix of skilled and unskilled workers hired by Indonesian manufacturers.To estimate the impact of importing on the demand for skilled workers we exploit detaileddata from the Indonesian manufacturing survey. Our data documents the education level ofevery worker in every manufacturing plant with at least 20 employees. Defining a skilledworker as one with a high school education for production workers and one with a collegeeducation for non-production workers, we find that importing greatly increases the demandfor educated workers among Indonesian importers within each of occupation categories. Wealso document evidence that the effect of importing on the demand for educated workersis heterogeneous across plants. In particular, plants that were induced to import duringour sample period were estimated to be those with generally high returns from importing.We further find that policies that improve transportation infrastructure in Indonesia wouldencourage new plants to start importing and increase the demand for educated workers amongnew importers. Notably, however, when we repeat our experiments using a conventionalmeasure of relative skill demand, defined as the ratio of non-production to production workers,we often find no significant impact of importing on the demand for skilled labor.71Chapter 3Reallocation and CompositionEffects of Trade on the Demand forSkilled Workers. A DecompositionAnalysis using IndonesiaManufacturing Data3.1 IntroductionIn the past two decades, Indonesia has experienced a large increase in the supply of educatedworkers. During 1996 to 2006, the portion of employed workers with at least high schooleducation increased from 40 percent to 58 percent, and the share of workers with more thancollege education more than doubled (9 percent to 19 percent).60 On the demand side, therole of opening to trade in this large developing economy is of great interest to researchersand policy makers. See Pavcnik (2003) as an example, and Goldberg and Pavcnik (2007)provides a review of studies that discuss the wage effect of trade in developing countries.Though the majority of the schooling increase were absorbed by the demand of service sector(69 percent), most of the remaining (20 percent) went into manufacturing industries, whichare more related with the exchange of goods in the open economy.61Empirically, there exists significant heterogeneity in the plants demand of skilled workers,even within narrowly defined industries. As a result, the change in the aggregate skill demandreflects not only the skill composition changes within plants , but also the market sharerelocations across plants. The skill demand change among continuing plants, as well asthe entry and exit of plants, contribute to the dynamics of aggregate skill demand. Usingthe Indonesia census of manufacturing plants for the period 1996 to 2006, we examine therelationship between trading decisions of plants and the aggregate skill demand at different60Data source: Indonesian labor force survey61The change in the total skill share caused by each industry equals to the change in the product of the skillshare and employment share of the industry.723.1. Introductionmargins (reallocation of workers across plants versus the skill composition changes withinplants), and we address the role of the entry and exit of plants in the aggregate skill demandchange. It turns out the impact of entry and exit of plants to the overall skill demand isnegligible, as they are similar to the surviving plants. Focusing on the surviving plants, wefind that plants that switched from domestic to import or export grew in employment shares,at the expense of the shrinking of the plants that switched from trading to not trading. Forplants that started importing, this expansion is more skill biased, but skill and unskilledworkers grew evenly for plants that started exporting. Consequently, the growth in size andincrease in the skill intensities of the plants that switched from non-importing to importingcontribute to the increase in the aggregate demand for skilled workers. Plants that stoppedimporting or exporting laid off workers, and this employment reduction towards more tounskilled workers. Thus, though it is possible to observe the skill intensity to increase forplants that quited international trade, they did not demand more skilled workers. The plantsthat continue importing or exporting grew in size, but the growth did not bias towardsworkers of any skill. Given that the plants that always trade are most skill intensive, theirgrowth (even not skill biased) contribute positively to the aggregate skill demand.Studies that base on plant level data typically use occupation categories, such as non-production or white-collar workers, as a proxy for skilled workers. See Bernard and Jensen(1997a), Harrison and Hanson (1999), Pavcnik (2003), Biscourp and Kramarz (2007) asexamples.62 Though the occupation categories are correlated with the education levels ofthe workers, they do not capture the variations in the skill compositions within groups.Our data capture a much more precise measure of skill at the plant level. Specifically, ourpanel data record the education levels of workers in every Indonesian manufacturing plantwith at least 20 employees. Moreover, we are able to observe the education distributionwithin each occupation category. To understand the performance of occupation categoriesas a proxy for workers’ skill, we decompose the overall increase in the share of educatedworkers into the increase within occupation categories and the increase due to reallocationbetween occupation categories. We find that more than 95 percent of the skill upgradinghappened within production and within non-production workers, suggesting that focusingon the relative demand for non-production workers to production workers provide limitedinsights on the pattern of skill upgrading, and on how trading affects the demand for educatedworkers.Section 2 introduces our data and discuss the measure of workers’ skill to be carriedthroughout later analysis. Section 3 analyze the impact of entry and exit of plants to theaggregate skill demand. Section 4 looks into the relationship between the trading decisionsof plants and their skill demand, by decomposition exercises and regressions analysis that62See footnote 29 in Chapter 2 for exceptions733.2. Datacontrol for relevant plant characteristics. Lastly, section 5 concludes.3.2 Data3.2.1 Data SourceSame as Chapter 2, our primary source of data is the Indonesian manufacturing survey be-tween 1995 and 2007, where we mainly use the data recorded in the census years 1996 and2006 because, in these two years, the Indonesian manufacturing survey records the distri-bution of academic achievement in two distinct occupation categories (non-production andproduction) in each plant. Specifically, in each plant we observe the number of workers withprimary, secondary and post-secondary education. We construct relative skill measures thatare directly based on the workers’ education levels for each occupation category. In addition,we augment the manufacturing survey with the Indonesian household survey to control forregional labor market conditions. Details about the data coverage, advantages, limitationsand variable construction are discussed in section 2.3 of Chapter 2.3.2.2 Measure of SkillWhile education is a natural way of measuring the skills of workers, due to data limitation,studies that exploit plant or firm level data typically use occupation categories, such as non-production or white-collar workers, as a proxy of skill. For instance, Amiti and Cameron(2012) investigate the impact of trade liberalization on the wages of production workersrelative to non-production workers. They find that falling input tariffs has caused the wageof non-production workers to fall relative to the wage of production workers within Indonesianmanufacturing firms that import their intermediate inputs.The adoption of production/non-production category as a proxy for workers’ skill is sup-ported by the the close relationship between occupation and education levels of workers.However, the extent to which this proxy is satisfactory depends on how much the variation inthe demand of skilled workers was caused by the variation in the need of workers in differentjobs. If the variation of skill demand across plants and over time is a result of differencesin skill compositions within occupations, this proxy will provide limited insights on the pat-tern of skill demand. Figure 3.1 presents the pattern of skilled worker share under differentmeasurement of skill during the period 1996-2006.63 While the share of educated workers63To compute the skill shares, we use the census of Indonesia manufacturing plants when skill is defined bynon-production occupation, and the labor force survey when skill is defined by workers’ education levels. Inthe later case, the sample is restricted to the workers employed in manufacturing sector. This sample does notexactly match the census of manufacturing plants as it contains individuals who are hired in small plants thatcontains less than 20 employers, and thus not surveyed in the manufacturing plants census. The educationlevels of workers in smaller plants tend to be smaller, as a result, the average skill share in this sample is743.2. DataFigure 3.1: Trend of Skilled Worker Share, by Different Skill MeasurementsData Source: 1996-2006 Indonesia Census of Manufacturing Plants and Labor ForceSurvey. Sample of the Labor Force Survey is restricted to employed workers in themanufacturing sector.(defined by education above high-school) increased from 32 percent to 48 percent, the shareof non-production workers remained stable over time. The overall share of college educatedworkers also increased, but by a much smaller magnitude.To precisely characterize the development of Indonesia labor market, we decompose thechange in the share of educated workers hired in the Indonesia manufacturing plants intotwo components. The first component captures the reallocation of labor between the produc-tion and non-production occupations (“between”), and the second component captures theeducation upgrading within each occupation (“within”):∆LsL= ∆LpsLp× L¯pL︸ ︷︷ ︸within prod.+ ∆LnsLn× L¯nL︸ ︷︷ ︸within non-prod.+(L¯psLp− L¯nsLn)×∆LpL︸ ︷︷ ︸between(3.1)where L represents number of workers, superscripts “p” and “n” denote production and non-production occupations, and subscripts “s” and “u” denote skilled and unskilled workers.∆ (Ls/L) = (Ls/(Ls + Lu))06−(Ls/(Ls + Lu))96 is the change in the overall share of educatedsmaller than that from the census of manufacturing plants. Nevertheless, the changes in the skill share from1996 to 2006 from the two samples are consistent.753.2. Dataworkers between 1996 and 2006, ∆(Ljs/Lj)=(Ljs/(Ljs + Lju))06−(Ljs/(Ljs + Lju))96is thechange in the share of educated workers within occupation j, ∆ (Lp/L) = (Lp/(Lp + Ln))06−(Lp/(Lp + Ln))96 is the change in the share of workers in the production occupation, and(Lj/(Lp + Ln)) =[(Lj/(Lp + Ln))96+(Lj/(Lp + Ln))06]/2 and(Ljs/(Ljs + Lju))=[(Ljs/(Ljs + Lju))96+(Ljs/(Ljs + Lju))06]/2 are the average of the corresponding share vari-ables in the two periods.Table 3.1: Decompose by occupation groupDef. Skill High School+ College+(1) (2) (3) (4)Panel A: levelsYear 1996 2006 1996 2006Ls/(Ls + Lu) 0.448 0.582 0.045 0.064Lps/(Lps + Lpu) 0.402 0.541 0.021 0.027Lns /(Lns + Lnu) 0.651 0.778 0.150 0.243Ln/(Lp + Ln) 0.183 0.174 0.183 0.174Panel B: decompose the overall changes∆(Ls/(Ls + Lu)) 0.135 0.020within prod. 0.114 0.005within non-prod. 0.023 0.017between -0.002 -0.001a. Data Source: Indonesia Manufacturing Surveyin 1996 and 2006.Panel A of Table 3.1 reports the share of educated workers in all workers, in productionand non-production occupations separately, and the share of non-production workers, in 1996and 2006. Panel B of the same table presents the decompositions of the changes in the shareof educated workers. In column (1) and (2), where we define educated workers as workerswith at least a high school diploma, the overall share of educated workers increased by 13.5percentage points from 0.448 to 0.582 between 1996 and 2006, and almost all of this increaseis explained by the skill upgrading within occupations, with 84 percent explained by “withinproduction” term. Columns (3) and (4) present the same skill share levels and repeat thedecomposition exercise, but but here skilled workers are defined as those workers with aneducational attainment of no less than college. The overall share of college educated workersincreased by 1.9 percentage points from 0.045 to 0.064 between 1996 and 2006, and the“within non-production” term in the decomposition accounts for more than 85 percent.This simple decomposition exercise suggests that little of the skill-upgrading at the plant-level can be explained by the reallocation of workers across occupations. As a result, focusingon the relative demand for non-production workers to production workers provide limited763.3. Entry and Exit of Plantsinsight to the dynamics of the overall skill demand. Moreover, the skill upgrading happenswithin production and non-production occupations at different levels. While plants needmore workers with at least high-school education to do production jobs, their demandededucation for non-production jobs is more than college. We carry both education thresholdsfor the definition of “skilled” workers throughout our analysis.3.3 Entry and Exit of PlantsEmpirically, there exists significant heterogeneity in the plants’ demand for skilled workers,even within narrowly defined industries. As a result, the change in the aggregate skill de-mand reflects not only the skill composition changes within plants , but also the market sharereallocations across plants. The skill demand change among continuing plants, as well as theentry and exit of plants, contributes to the dynamics of aggregate skill demand.64 How muchdid entry and exit contribute to the aggregate skill demand for skilled labor between 1996and 2006? To answer this question, we decompose the skill demand change to get the con-tributions of continuing plants, exiting plants and new entering plants.65 The decompositionmethod follows Melitz and Polanec (2014), which was proposed to decompose the aggregatedproductivity change with entry and exit of firms. The decomposition breaks down the changein aggregate skill demand into components for the three groups of plants: continuing plants(C), entrants (E) and exiters (X). Using subscript i to index individual plants and superscripts to index skilled workers, for two periods t ∈ {1, 2} and three types of plants G ∈ {C,E,X},we define SGt ≡ LsGt/LGt =∑i∈G Lsit∑i∈G Litto be the share of skilled workers among workers inplants G, and ΦGt ≡ LGt/Lt =∑i∈G Lit∑i Litto be the employment share of plants G among allplants. The skill share in period t , St =∑i Lsit∑i Litcan be decomposed as:S1 = ΦC1SC1 + ΦX1SX1 = SC1 + ΦX1(SX1 − SC1)S2 = ΦC2SC2 + ΦE2SE2 = SC2 + ΦE2(SE2 − SC2).From this, we have the skill share change being decomposed into the three components:∆S = (SC2 − SC1)︸ ︷︷ ︸cont. plants+ ΦX1(SC1 − SX1)︸ ︷︷ ︸exiters+ ΦE2(SE2 − SC2)︸ ︷︷ ︸entrants. (3.2)64Researchers find that reallocation of resources across heterogeneous firms is an important driver of produc-tivity changes (See Bartelsman et al. (2013)). Also,less productive firms exit the market and more productivefirms expand following trade opening, rise the aggregate productivity (e.g. Melitz (2003), Pavcnik (2002))65Notice that since our data only include plants that hire more than 20 workers, the entry and exit weobserve could be a result of plants growing/shrinking across the threshold.773.3. Entry and Exit of PlantsSince the skill share of the entrants in the first period and that of the exiters in the secondperiod are not observed, the skill shares of continuing plants in the two periods are used asbenchmarks. Entrants/exiters contribute to the overall skill demand change only if their skillintensities are different from the contemporaneous continuing plants. Consider a hypotheticalexample of an economy with same skill demand for all plants. If the skill shares of therepresentative plants increase by the same amount, adding entry and exit of identical plantswill not change the skill upgrading rate. This decomposition assigns a zero contribution forentry and exit under this scenario.66Table 3.2: Decompose Skill Share Changes, by Production Dynamicsof PlantsSkill Shr.: SGt Emp. Shr.: ΦGt Decomposition1996 2006 1996 2006Skilled: High-School+All Plants 0.430 0.567 S2 − S1 0.137Cont. Plants 0.448 0.582 0.679 0.586 SC2 − SC1 0.135Exiters 0.393 - 0.321 - ΦX1(SC1 − SX1) 0.018Entrants - 0.546 - 0.414 ΦE2(SE2 − SC2) -0.015Skilled: College+All Plants 0.040 0.062 S2 − S1 0.021Cont. Plants 0.045 0.064 0.679 0.586 SC2 − SC1 0.020Exiters 0.031 - 0.321 - ΦX1(SC1 − SX1) 0.004Entrants - 0.058 - 0.414 ΦE2(SE2 − SC2) -0.003a. Data Source: Indonesia Manufacturing Survey in 1996 and 2006.Table 3.2 presents skilled worker shares of survivors, entrants and exiters in 1996 and2006, together with the decomposition results of equation (3.2). The upper panel definesskilled workers as those with more than high-school education and the lower panel uses thecollege threshold. Between 1996 and 2006, the share of workers with at least high-schooleducation increased from 0.430 to 0.567, and that of workers with more than college degreeincreased from 0.04 to 0.062. For both education thresholds, the increase in the skill demandof continuing plants counts for more than 95 percent of the overall change (0.135 out of0.137 for workers with more than high-school education, and 0.020 out of 0.021 for workerswith more than college education). Given that exiting plants employ 32 percent of all theworkers in 1996, and new entrants employ 41 percent of all the workers in 2006, their smallcontributions to the overall changes are mainly caused by the small difference between theirskill demand and that of the contemporaneous continuing plants.67 The evidence suggests66Melitz and Polanec (2014) compare their decomposition method with those implied by Griliches and Regev(1995) and Foster et al. (2001), demonstrate that other methods can bias the contribution of continuing plantsdownwards by assigning positive contributions to entrants and exiters in an economy with homogeneous firms.67We repeat the same decomposition exercises for production and non-production workers separately, the783.4. Impact of Trade on Skill Demandthat new entrants are not necessarily more skill intensive than exiters.Based on the decomposition results that suggest little contribution of entrants and exitersto the aggregate skill share change, we focus on the continuing plants in the later analysisabout the role of international trade. Unless explicitly noted, we restrict the sample tobalanced panel, and omit the subscript C for notation simplicity.3.4 Impact of Trade on Skill DemandAbout 30 percent of the Indonesia manufacturing plants have engaged in activities of ex-porting final goods or importing intermediate goods. The education compositions of tradingplants are different from those of non-trading plants. Table 3.3 documents the percentageof workers in each educational category within production or non-production workers. Forproduction workers, importing and exporting plants are found to hire systematically moreworkers with education levels above high school. While this remains true for non-productionworkers, it is much less stark. Instead, we observe that trading plants tend to hire a substan-tially greater share of college-educated non-production workers. While this study focus ondocumenting the stylized facts about the relationship between the trade decisions and skilldemand of plants, it is in line with the studies that investigate the mechanisms that drivethese observed facts. For example, our result suggest that the plants that started to importintermediate goods from foreign countries employ more production and non-production work-ers. While the employment of both skilled and unskilled workers increased, the growth wasmore toward skilled workers, especially for production occupations. This finding is consistentwith Kasahara et al. (2016), which particularly discusses that the skill biased technology in-crease caused by importing intermediate goods drove up the demand of skilled workers withinIndonesia manufacturing plants. Another example is that we found plants export to foreigncountries employ more workers. This is consistent with the prediction that exporting plants(the ones who can afford the fixed exporting cost) have extra demand from foreign countriesincrease their production in Melitz (2003).3.4.1 DecompositionTo understand how plants with different trading dynamics contribute to the aggregate skilldemand, we decompose the skill share change by the plants’ import/export status in thetwo periods. In the balanced panel, there are four types of plants according to their statusof importing intermediate goods: never importers (NN), switchers from non-importing toimporting (NI), switchers from importing to non-importing (IN), and always importers (II).We use subscript g ∈ {NN,NI, IN, II} to represent the plants group. In each period t ∈results also suggest insignificant roles of entry and exit in the aggregate skill demand change.793.4. Impact of Trade on Skill Demand{1, 2}, the aggregate skill share (St ≡ Lst/Lt =∑i Lsit∑i Lit) equals the sum of the skill intensitiesof plants in different groups (Sgt ≡ Lsgt/Lgt =∑i∈g Lsit∑i∈g Lit), weighted by the employment sharesof the groups (Φgt ≡ Lgt/Lt =∑i∈g Lit∑i Lit), i.e. St =∑g∈{NN,NI,IN,II}ΦgtSgt. The change inthe skill demand is thenS2 − S1 =∑g∈{NN,NI,IN,II}(Φg2Sg2 − Φg1Sg1) (3.3)=∑g∈{NN,NI,IN,II}Ng · (φg2Sg2 − φg1Sg1).Where, Ng in the second line is the number of plants in group g, and φgt ≡ 1NgΦgt is theaverage labor share of plants in group g. The per-plant contribution of group g to theaggregate skill demand change is φg2Sg2 − φgSg1. Similarly, we group plants into four typesby their exporting status and compute the per-plant contribution of each group.Columns (1)-(2) and columns (5)-(6) of panel A in Table 3.4 and 3.5 present the decom-position results of equation (3.3) for all workers. When we focus on workers with more thanhigh-school education (columns (1)-(2)), the per-plant contribution to the aggregate skill de-mand is the highest among the plants that trade in the second period. Continuing tradershave highest per-plant contributions to the overall skill intensity change (last row). Whilethe per-plant contributions of switchers from not trading to trading are smaller(second lastrow), they are 2-5 times larger than the contributions of switchers from trading to not tradingand never traders (first two rows). Similar patterns hold when we look at workers with morethan college education, the magnitude in the overall change and the per-plant contributionsof each group is much smaller.This is consistent with the flat pattern of college worker sharepresented in Figure 3.1.We repeat the same decomposition exercises for production and non-production workersseparately, shown in panel B and panel C in in table 3.4 and 3.5. Again we find that theper-plant contribution to the aggregate skill demand is highest among the plants that tradein the second period. One exception is that plants stopped exporting increased the ratio ofcollege educated workers. We will show later that this increase in skill share is mainly causedby laying off the workers with lower education, rather than hiring more educated workers.Firms that stopped exporting require less non-production workers, and this adjustment waslargely implemented by laying off less education ones. The order between always tradersand switchers from not trading to trading alters when we use high-school as threshold fornon-production workers or when we use college as threshold for production workers. Amongproduction workers, many Indonesian plants do not hire any workers with college or post-graduate training. As a result, defining a skilled worker as a “college graduate” in our803.4. Impact of Trade on Skill Demandsample of production workers would eliminate a significant number of plants that are whollycomposed of workers without college education. On the other hand, using a high-schooleducation threshold would potentially obscure a key margin on which firms upgrade employeeskills in response to trade among non-production workers because the skill upgrading abovecollege level happened mostly within non-production workers, as documented in Table 3.1.For these reasons, skilled workers are defined as those with more than high school degree forproduction workers, and those with at least a college degree for non-production workers infuture discussions. We present the results of alternative definitions (college for productionworkers and high school for non-production workers) for completeness.It is clear from equation (3.3) that individual plants contribute to the aggregate skilldemand along two margins: reallocation of workers across plants with different skill demand,and skill composition change within plants. Plants exporting and importing decisions caninfluence their choices of employment and skill intensity. To analyze these two margins, wefurther decompose the per-plant contribution of group g byφg2Sg2 − φg1Sg1 = (φg2 − φg2)︸ ︷︷ ︸emp. reallocationS¯g + (Sg2 − Sg1)︸ ︷︷ ︸composition changeφ¯g for g ∈ {NN,NI, IN,NN}.(3.4)Columns (3)-(4) and (7)-(8) of panel A in table 3.4 and 3.5 present this decomposition resultsfor all workers. The employment reallocation columns indicate that plants import or exportin the second period grow in size, and workers left the plants that never trade or stop trading.The growth of skill intensity of continuing importers is larger than that of plants just startingimporting. On the other hand, plants that just start exporting grow the most in size, morethan the plants that always export. Given that the skill demand of never traders and switchersfrom trading to not trading are lower, this employment reallocation contributes positively tothe aggregate skill demand. The columns of composition changes do not show clear differenceamong plants with different trading dynamics. However, as we will show in the regressionanalysis section, once we control for relevant plants’ characteristics, plants started to importbecame more skill biased.3.4.2 Regression AnalysisThe plants import and export decisions are highly correlated, and there are other plantcharacteristics that determine both import/export status and skill demand of plants. Weuse regressions analysis to check the reallocation effect and composition effect of trade withcontrols of observed plant characteristics.813.4. Impact of Trade on Skill DemandWe run the following regressions that are consistent with the decomposition exercises:∆Yi = β0 +∑g∈Ωmβmg 1(i ∈ g) +∑g∈Ωxβxg1(i ∈ g) + β2∆Xi + i (3.5)Yi takes the value of log employment for analyzing the reallocation effect, and it takes thevalue of the log of skill share for analyzing the composition effect. 1(i ∈ g) equals one if thetrade dynamics of plant i is in group g, and zero otherwise. Ωm and Ωx are the four groupsof import and export dynamics, as described at the beginning of section 4.1. Xi are plantcharacteristics including capital, productivity measure, dummy of foreign ownership, dummyof non-zero R & D expenses, dummy of non-zero training expenses, geographic location dum-mies and 3-digit industry dummies. Choosing never importers (never exporters) as omittedgroups for Ωm (Ωx), the coefficients βmg (βxg ) tells how much plants in group g are differentfrom the plants that never importer (export) in terms of ∆Yi.Reallocation EffectTable 3.6 presents the regression results that capture the employment effect of trade, forall workers (columns (1)-(3)) and for production and non-production workers separately(columns (4)-(9)). In all columns, switchers from domestic to import or export grew inemployment shares, at the expense of the shrinking of the plants that switched back to do-mestic. The plants that shifted from import to not import shrunk mostly non-productionworkers, while those shifted from export to non-export reduced both types of workers. Thereduction of non-production workers of them is probably because those workers were largelyused to support trading activities, such as translating, contacting foreign firms, report tocustom, etc. Exporters laid off production workers after they turned to domestic is prob-ably caused by reduction in production caused by cutting the foreign demand. Continuingimporters did not significantly increase their hiring. Continuing exporters increased theiremployment, and the increase was almost all caused by production worker hiring, possiblyfor increasing production to meet growing foreign demand.823.4.ImpactofTradeonSkillDemandTable 3.3: Trade and Skill Intensities of Plants in 2006<Primary Primary Jr. High High College+ <Primary Primary Jr. High High College+Panel A: Production WorkersExporters Non-ExportersImporters 0.010 0.095 0.245 0.605 0.045 0.027 0.108 0.299 0.530 0.035(0.051) (0.151) (0.204) (0.253) (0.085) (0.109) (0.210) (0.221) (0.281) (0.065)Non-Importers 0.028 0.182 0.316 0.454 0.021 0.052 0.239 0.347 0.347 0.015(0.103) (0.242) (0.237) (0.294) (0.045) (0.141) (0.300) (0.266) (0.310) (0.051)Panel B: Non-Production WorkersExporters Non-ExportersImporters 0.003 0.032 0.104 0.532 0.329 0.006 0.029 0.153 0.561 0.251(0.020) (0.093) (0.150) (0.245) (0.256) (0.068) (0.120) (0.200) (0.238) (0.210)Non-Importers 0.012 0.052 0.134 0.571 0.232 0.010 0.054 0.203 0.571 0.161(0.085) (0.149) (0.192) (0.277) (0.243) (0.077) (0.180) (0.280) (0.321) (0.229)a. Data Source: Indonesia Manufacturing Survey in 2006.b. Sample restricted to balanced panel.833.4.ImpactofTradeonSkillDemandTable 3.4: Decompose Skill Share Changes, by Importing Dynamics of PlantsSkilled: High-school+ Skilled: College+Imp. Statusoverall contribution emp. reallocation composition overall contribution emp. reallocation composition # ofchange per 1000 plants per 1000 plants change change per 1000 plants per 1000 plants change plants(S2 − S1) (φg2Sg2 − φg1Sg1) (φg2 − φg1) (Sg2 − Sg1) (S2 − S1) (φg2Sg2 − φg1Sg1) (φg2 − φg1) (Sg2 − Sg1) Ng(1) (2) (3) (4) (5) (6) (7) (8) (9)Panel A: all workersnonimp-nonimp0.1350.005 -0.003 0.1170.0200.001 -0.003 0.020 7465imp-nonimp 0.005 -0.024 0.139 0.001 -0.024 0.022 941nonimp-imp 0.022 0.014 0.109 0.004 0.014 0.021 658imp-imp 0.052 0.023 0.147 0.006 0.023 0.018 1476Panel B: production workersnonimp-nonimp0.1390.006 -0.002 0.1190.0060.000 -0.002 0.008 7465imp-nonimp 0.007 -0.022 0.145 0.001 -0.022 0.010 941nonimp-imp 0.022 0.013 0.116 0.002 0.013 0.010 658imp-imp 0.051 0.020 0.154 0.001 0.020 0.001 1476Panel C: non-production workersnonimp-nonimp0.1270.004 -0.005 0.1310.0930.004 -0.005 0.084 7465imp-nonimp -0.003 -0.031 0.134 0.006 -0.031 0.080 941nonimp-imp 0.022 0.016 0.079 0.014 0.016 0.082 658imp-imp 0.057 0.037 0.109 0.033 0.037 0.097 1476a. Data Source: Indonesia Manufacturing Survey in 2006.b. Sample restricted to balanced panel.843.4.ImpactofTradeonSkillDemandTable 3.5: Deompose Skill Share Changes, by Exporting Dynamics of PlantsSkilled: High-school+ Skilled: College+Imp. Statusoverall contribution emp. reallocation composition overall contribution emp. reallocation composition # ofchange per 1000 plants per 1000 plants change change per 1000 plants per 1000 plants change plants(S2 − S1) (φg2Sg2 − φg1Sg1) (φg2 − φg1) (Sg2 − Sg1) (S2 − S1) (φg2Sg2 − φg1Sg1) (φg2 − φg1) (Sg2 − Sg1) Ng(1) (2) (3) (4) (5) (6) (7) (8) (9)Panel A: all workersnonexp-nonexp0.1350.006 -0.002 0.1390.0200.001 -0.002 0.023 7524exp-nonexp 0.013 -0.022 0.130 0.003 -0.022 0.023 961nonexp-exp 0.032 0.021 0.127 0.004 0.021 0.015 654exp-exp 0.041 0.015 0.130 0.005 0.015 0.016 1401Panel B: production workersnonexp-nonexp0.1390.006 -0.002 0.1450.0060.000 -0.002 0.008 7524exp-nonexp 0.015 -0.023 0.132 0.000 -0.023 0.003 961nonexp-exp 0.030 0.021 0.122 0.002 0.021 0.008 654exp-exp 0.043 0.017 0.135 0.001 0.017 0.004 1401Panel C: non-production workersnonexp-nonexp0.1270.006 0.000 0.1210.0930.004 0.000 0.090 7524exp-nonexp 0.006 -0.023 0.130 0.019 -0.023 0.136 961nonexp-exp 0.046 0.025 0.157 0.014 0.025 0.049 654exp-exp 0.034 0.006 0.125 0.023 0.006 0.092 1401a. Data Source: Indonesia Manufacturing Survey in 2006.b. Sample restricted to balanced panel.853.4. Impact of Trade on Skill DemandTo see in detail whether skilled or unskilled workers get hired/fired when plants ex-pand/shrink, we repeat the same regressions for skilled and unskilled workers separately.Table 3.7 and table 3.8 present the result, using high school and college as two different skillthresholds. Panel A of the two tables present the results for skilled worker employment andpanel B are for unskilled worker employment. We find that plants that shifted from domes-tic to importer or exporter, increased the employment of both skilled and unskilled workers(above high-school for production workers and above college for non-production workers),for both production and non-production occupations. For plants that switched importingstatus, this hiring extension is more toward skilled, especially for production workers. Thisis consistent with Kasahara et al. (2016), which conclude that importing intermediate goodscan increase skill demand by rising the skill biased technology. The plants that stoppedtrade (imp-nonimp and exp-nonexp) reduced both skilled and unskilled hiring for productionworkers. For non-production workers, import quiters laid off unskilled workers while exportquiters reduced skilled ones. A possible explanation is that importing-induced technologychange still need skilled non-production workers to maintain, but skilled non-productionworkers only support exporting activities that are not related with the production process ofgoods.Table 3.6: total employment and trade status, first differencedDep. Variable: ∆log(L)All Workers Production Workers Non-production Workers(1) (2) (3) (4) (5) (6) (7) (8) (9)imp-nonimp -0.028 -0.038 -0.017 -0.030 -0.130** -0.139***[0.032] [0.033] [0.036] [0.036] [0.051] [0.052]nonimp-imp 0.189*** 0.174*** 0.159*** 0.139*** 0.244*** 0.230***[0.044] [0.044] [0.049] [0.049] [0.067] [0.068]imp-imp 0.068** 0.045 0.069** 0.039 0.040 0.020[0.028] [0.030] [0.031] [0.033] [0.045] [0.047]exp-nonexp -0.102** -0.113** -0.098* -0.107** -0.101* -0.107*[0.047] [0.047] [0.051] [0.051] [0.060] [0.061]nonexp-exp 0.195*** 0.184*** 0.199*** 0.191*** 0.187*** 0.182***[0.045] [0.046] [0.049] [0.049] [0.065] [0.066]exp-exp 0.094*** 0.078** 0.117*** 0.104*** 0.078 0.070[0.031] [0.033] [0.035] [0.037] [0.050] [0.052]Observations 4,739 4,739 4,739 4,739 4,739 4,739 4,739 4,739 4,739R-squared 0.100 0.100 0.101 0.090 0.091 0.092 0.059 0.060 0.060a. Robust standard errors in brackets: *** p<0.01, ** p<0.05, * p<0.1b. Data Source: Indonesia Manufacturing Survey in 1996 and 2006.c. Sample restricted to balanced panel.d. Control variables include province fixed effects, 3-digit industry fixed effects; first differenced capital, productivity, foreignownership; and dummy of R&D and of worker training. Columns (1) (4) and (7) control for change in export status.Columns (2) (5) and (8) control for change in import status.863.4. Impact of Trade on Skill DemandTable 3.7: skilled and unskilled employment (high-school+) and trade status, firstdifferencedAll Workers Production Workers Non-production WorkersPanel A: Dependent Variable = log of skilled workers (∆log(Ls))(1) (2) (3) (4) (5) (6) (7) (8) (9)imp-nonimp -0.124** -0.135*** -0.094 -0.103 -0.074 -0.089[0.050] [0.051] [0.065] [0.066] [0.054] [0.054]nonimp-imp 0.189*** 0.172** 0.254*** 0.239*** 0.214*** 0.191**[0.070] [0.070] [0.090] [0.091] [0.075] [0.075]imp-imp 0.050 0.026 0.092 0.069 0.070 0.039[0.046] [0.048] [0.059] [0.061] [0.048] [0.049]exp-nonexp -0.101 -0.106 -0.094 -0.106 -0.121* -0.130**[0.066] [0.067] [0.087] [0.087] [0.063] [0.064]nonexp-exp 0.227*** 0.223*** 0.137* 0.126 0.316*** 0.307***[0.066] [0.066] [0.077] [0.078] [0.066] [0.066]exp-exp 0.085* 0.077 0.106* 0.086 0.108** 0.094*[0.047] [0.049] [0.059] [0.062] [0.049] [0.051]Observations 4,293 4,293 4,293 3,233 3,233 3,233 3,997 3,997 3,997R-squared 0.077 0.078 0.078 0.071 0.071 0.072 0.070 0.071 0.071Panel B: Dependent Variable = log of unskilled workers (∆log(Lu))(1) (2) (3) (4) (5) (6) (7) (8) (9)imp-nonimp -0.136** -0.132** -0.135** -0.136** -0.227* -0.201[0.055] [0.056] [0.056] [0.056] [0.131] [0.132]nonimp-imp 0.032 0.036 0.031 0.028 0.144 0.178[0.075] [0.075] [0.074] [0.074] [0.137] [0.141]imp-imp -0.063 -0.060 -0.083 -0.090 -0.019 0.019[0.052] [0.055] [0.054] [0.056] [0.104] [0.107]exp-nonexp -0.187*** -0.177** -0.174** -0.160** -0.125 -0.125[0.071] [0.072] [0.072] [0.073] [0.157] [0.158]nonexp-exp 0.015 0.026 0.059 0.072 -0.231 -0.230[0.073] [0.074] [0.074] [0.075] [0.158] [0.159]exp-exp -0.008 0.010 0.012 0.036 -0.142 -0.145[0.049] [0.052] [0.051] [0.053] [0.105] [0.111]Observations 4,354 4,354 4,354 4,291 4,291 4,291 1,626 1,626 1,626R-squared 0.089 0.089 0.090 0.093 0.092 0.093 0.096 0.098 0.098a. Robust standard errors in brackets: *** p<0.01, ** p<0.05, * p<0.1b. Data Source: Indonesia Manufacturing Survey in 1996 and 2006.c. Sample restricted to balanced panel.d. Control variables include province fixed effects, 3-digit industry fixed effects; first differenced capital, productivity, foreignownership; and dummy of R&D and of worker training. Columns (1) (4) and (7) control for change in export status.Columns (2) (5) and (8) control for change in import status.e. Education threshold to define skilled worker: high-school873.4. Impact of Trade on Skill DemandTable 3.8: skilled and unskilled employment (college+) and trade status, first dif-ferencedAll Workers Production Workers Non-production WorkersPanel A: Dependent Variable = log of skilled workers (∆log(Ls))(1) (2) (3) (4) (5) (6) (7) (8) (9)imp-nonimp 0.036 0.032 0.093 0.082 0.032 0.022[0.070] [0.070] [0.143] [0.143] [0.078] [0.079]nonimp-imp 0.247*** 0.240** 0.088 0.075 0.235** 0.215**[0.094] [0.094] [0.192] [0.193] [0.100] [0.101]imp-imp 0.081 0.065 0.109 0.080 0.081 0.047[0.062] [0.064] [0.129] [0.131] [0.065] [0.067]exp-nonexp -0.195** -0.214*** -0.003 -0.008 -0.167** -0.183**[0.082] [0.083] [0.169] [0.169] [0.083] [0.084]nonexp-exp 0.150* 0.137 -0.098 -0.106 0.184* 0.172*[0.085] [0.085] [0.181] [0.182] [0.094] [0.096]exp-exp 0.090 0.067 0.126 0.108 0.149** 0.131**[0.061] [0.063] [0.123] [0.125] [0.064] [0.066]Observations 2,343 2,343 2,343 752 752 752 2,077 2,077 2,077R-squared 0.082 0.081 0.083 0.116 0.118 0.118 0.085 0.086 0.087Panel B: Dependent Variable = log of unskilled workers (∆log(Lu))(1) (2) (3) (4) (5) (6) (7) (8) (9)imp-nonimp -0.050 -0.060* -0.033 -0.045 -0.108* -0.117**[0.033] [0.034] [0.037] [0.037] [0.057] [0.058]nonimp-imp 0.186*** 0.170*** 0.160*** 0.141*** 0.199** 0.188**[0.046] [0.046] [0.050] [0.050] [0.079] [0.080]imp-imp 0.062** 0.039 0.067** 0.038 0.023 0.006[0.030] [0.032] [0.033] [0.035] [0.054] [0.056]exp-nonexp -0.098** -0.107** -0.105** -0.113** -0.060 -0.063[0.048] [0.048] [0.051] [0.051] [0.073] [0.073]nonexp-exp 0.195*** 0.187*** 0.199*** 0.191*** 0.180** 0.177**[0.045] [0.046] [0.049] [0.049] [0.072] [0.072]exp-exp 0.094*** 0.080** 0.114*** 0.102*** 0.056 0.052[0.032] [0.034] [0.035] [0.037] [0.056] [0.059]Observations 4,739 4,739 4,739 4,736 4,736 4,736 4,501 4,501 4,501R-squared 0.093 0.093 0.094 0.088 0.089 0.090 0.051 0.051 0.051a. Robust standard errors in brackets: *** p<0.01, ** p<0.05, * p<0.1b. Data Source: Indonesia Manufacturing Survey in 1996 and 2006.c. Sample restricted to balanced panel.d. Control variables include province fixed effects, 3-digit industry fixed effects; first differenced capital, productivity,foreign ownership; and dummy of R&D and of worker training. Columns (1) (4) and (7) control for change in exportstatus. Columns (2) (5) and (8) control for change in import status.e. Education threshold to define skilled worker: college883.5. ConclusionComposition EffectLastly, we look at how trade decisions of plants are related to their skill intensities. Table 3.9presents the regression results that capture the composition effect of trade, for all workers(columns (1)-(3)) and for production and non-production workers separately (columns 4-9).Panel A are the results using high school as the threshold to define skilled workers andPanel B are the results with college as skill threshold. Again, we focus on the results whereproduction workers with more than high-school education and non-production workers withcollege education are defined as skilled.Columns (1)-(3) suggest that there is no significant overall composition effect of trade.However, when we look at production and non-production workers separately, we find thatplants switched from domestic to importing demand higher share of skilled workers. Plantsthat stopped exporting demanded smaller share of skilled workers, but those stopped im-porter have more share of skilled non-production workers. However, this increase in the skillshare was mainly caused by laying off less educated workers, rather than hiring more skilledworkers. Continuing importers and continuing exporters did not increase or decrease theirskill intensities. This result together with the result of continuing trader employment ex-pansion suggest that these plants increase their hiring of skilled and unskilled workers to aconstant portion.3.5 ConclusionThis paper first investigate the performance of occupation categories as a proxy of workers’skill. We found that most skill upgrading happened within production and non-productionoccupations. Less than 5 percent of the overall skill demand change were caused by changein the relative demand of production and non-production workers. Consequently, using oc-cupation categories to proxy worker skills is unsatisfactory. Education, whenever available,is a preferred measurement.In the presence of firm heterogeneity, trade opening may influence the aggregate skilldemand by relocating resources among plants with different skill intensity, so that plantdynamics such as entry and exit may contribute to skill demand changes. How much entryand exit contribute to the aggregate demand for skilled labor between 1996 and 2006? Wefound that it to be quite small. Less than 5 percent of the overall change in skill demandare from new entrants or exiters. Given that entrants and exiting plants employ 41 and32 percent of the total labor respectively, their small contributions are mainly caused bythe small difference between their skill demand and that of the contemporaneous continuingplants.Lastly, we we examine the relationship between trading decisions of plants and their hiring893.5. ConclusionTable 3.9: skilled worker share and trade status, first differencedDep. Variable: ∆log(Ls/L)Occp. All Workers Production Workers Non-production WorkersPanel B: education threshold for defining skill = high-school(1) (2) (3) (4) (5) (6) (7) (8) (9)imp-nonimp -0.010 -0.020 -0.029 -0.032 0.039 0.005[0.079] [0.079] [0.097] [0.098] [0.157] [0.158]nonimp-imp 0.186* 0.172 0.238* 0.235* -0.021 -0.080[0.109] [0.109] [0.137] [0.137] [0.188] [0.191]imp-imp 0.093 0.077 0.116 0.114 -0.022 -0.085[0.075] [0.077] [0.091] [0.094] [0.137] [0.137]exp-nonexp 0.075 0.061 0.036 0.016 -0.153 -0.146[0.096] [0.097] [0.120] [0.120] [0.183] [0.184]nonexp-exp 0.171* 0.157 0.088 0.072 0.652*** 0.657***[0.097] [0.098] [0.112] [0.113] [0.171] [0.172]exp-exp 0.056 0.032 0.037 0.004 0.225* 0.245*[0.067] [0.070] [0.082] [0.085] [0.124] [0.128]Observations 3,916 3,916 3,916 2,820 2,820 2,820 1,195 1,195 1,195R-squared 0.046 0.046 0.047 0.050 0.050 0.051 0.108 0.112 0.113Panel B: education threshold for defining skill = college(1) (2) (3) (4) (5) (6) (7) (8) (9)imp-nonimp 0.090 0.094 0.389** 0.384** 0.166 0.179*[0.071] [0.072] [0.151] [0.151] [0.101] [0.102]nonimp-imp 0.007 0.014 0.007 0.001 0.172 0.194[0.101] [0.102] [0.202] [0.202] [0.132] [0.134]imp-imp -0.014 -0.010 0.153 0.141 0.072 0.099[0.067] [0.071] [0.138] [0.142] [0.089] [0.094]exp-nonexp -0.119 -0.122 0.078 0.069 -0.238** -0.262**[0.089] [0.090] [0.186] [0.186] [0.118] [0.119]nonexp-exp -0.047 -0.049 -0.155 -0.167 -0.036 -0.055[0.085] [0.086] [0.187] [0.187] [0.122] [0.123]exp-exp -0.007 -0.008 0.073 0.042 -0.051 -0.084[0.065] [0.069] [0.129] [0.132] [0.089] [0.094]Observations 2,343 2,343 2,343 752 752 752 1,933 1,933 1,933R-squared 0.050 0.051 0.051 0.135 0.133 0.136 0.050 0.050 0.052a. Robust standard errors in brackets: *** p<0.01, ** p<0.05, * p<0.1b. Data Source: Indonesia Manufacturing Survey in 1996 and 2006.c. Sample restricted to balanced panel.d. Control variables include province fixed effects, 3-digit industry fixed effects; first differenced capital, pro-ductivity, foreign ownership; and dummy of R&D and of worker training. Columns (1) (4) and (7) controlfor change in export status. Columns (2) (5) and (8) control for change in import status.903.5. Conclusionand skill intensity decisions separately (reallocation effect and composition effect). We findthat the plants switched from domestic to import or export grew in employment shares, atthe expense of the shrinking of the plants that switched back to domestic. For plants thatstarted importing, this expansion is more skill biased, but skill and unskilled workers grewevenly for plants that started exporting. Consequently, the growth in size and increase inthe skill intensities of the plants that switched from non-importing to importing contributeto the increase in the aggregate demand of skilled workers. Plants that stopped importingor exporting laid off workers, and this employment reduction influences skilled workers more.Thus, though it is possible to observe the skill intensity to increase for plants that quitedinternational trade, they did not demand more skilled workers. The plants that continueimporting or exporting grew in size, but the growth did not bias towards workers of any skill.Given that the plants that always trade are most skill intensive, their growth (even not skillbiased) contribute positively to the aggregate skill demand.91BibliographyAcemoglu, D. and D. Autor (2011): “Skills, Tasks and Technologies: Implicationsfor Employment and Earnings,” in Handbook of Labor Economics, ed. by D. Card andO. Ashenfelter, Elsevier, vol. 4, Part B, 1043 – 1171.Ackerberg, D. A., K. Caves, and G. Frazer (2015): “Identification Properties ofRecent Production Function Estimators,” Econometrica, forthcoming.Amiti, M. and L. Cameron (2012): “Trade Liberalization and the Wage Skill Premium:Evidence from Indonesia,” Journal of International Economics, 87, 277 – 287.Amiti, M. and D. Davis (2012): “Trade, Firms, and Wages: Theory and Evidence,” Reviewof Economic Studies, 79.Amiti, M. and J. Konings (2007): “Trade Liberalization, Intermediate Inputs, and Pro-ductivity: Evidence from Indonesia,” American Economic Review, 97.Amiti, M. and S.-J. Wei (2006): “Service Offshoring and Productivity: Evidence from theUnited States,” Working Paper 11926, National Bureau of Economic Research.Angrist, J. D. and J.-S. Pischke (2008): Mostly harmless econometrics: An empiricist’scompanion, Princeton university press.Attanasio, O., G. P., and N. Pavcnik (2004): “Trade Reforms and Wage Inequality inColombia,” Journal of Development Economics, 74.Autor, D. H. and D. Dorn (2013): “The Growth of Low-Skill Service Jobs and thePolarization of the US Labor Market,” American Economic Review, 103, 1553–97.Autor, D. H., D. Dorn, and G. H. Hanson (2013): “The China Syndrome: Local LaborMarket Effects of Import Competition in the United States,” American Economic Review,103, 2121–68.Baldwin, R. and F. Robert-Nicoud (2014): “Trade-in-goods and trade-in-tasks: Anintegrating framework,” Journal of International Economics, 92, 51–62.92BibliographyBartelsman, E., J. Haltiwanger, and S. Scarpetta (2013): “Cross-country differencesin productivity: The role of allocation and selection,” The American Economic Review,103, 305–334.Bartik, T. J. (1991): Who Benefits from State and Local Economic Development Policies?,no. wbsle in Books from Upjohn Press, W.E. Upjohn Institute for Employment Research.Berman, E., J. Bound, and Z. Grilliches (1994): “Changes in the Demand for SkilledLabour within U.S. Manfuacturing: Evidence from the Annual Survey of Manufactures,”Quarterly Journal of Economics, 109.Bernard, A. B. and J. Jensen (1997a): “Exporters, skill upgrading, and the wage gap,”Journal of International Economics, 42, 3 – 31.Bernard, A. B. and J. B. Jensen (1997b): “Exporters, Skill Upgrading, and the WageGap,” Journal of International Economics, 42.Bernard, A. B., J. B. Jensen, S. J. Redding, and P. K. Schott (2007): “Firms inInternational Trade,” Journal of Economic Perspectives, 21, 105–130.Biscourp, P. and F. Kramarz (2007): “Employment, skill structure and internationaltrade: Firm-level evidence for France,” Journal of International Economics, 72, 22–51.Blinder, A. S. (2006): “Offshoring: the next industrial revolution?” FOREIGN AFFAIRS-NEW YORK-, 85, 113.——— (2009): “How Many US Jobs Might be Offshorable?” World Economics, 10, 41–78.Blinder, A. S. and A. B. Krueger (2013): “Alternative Measures of Offshorability: ASurvey Approach,” Journal of Labor Economics, 31, S97 – S128.Bloom, N., M. Draca, and J. V. Reenan (2011): “Trade Induced Technical Change:The Impact of Chinese Imports on IT and Innovation,” mimeo, Stanford University.Burstein, A., J. Cravino, and J. Vogel (2013): “Importing Skill-Biased TechnologicalChange,” American Economic Journal: Macroeconomics, 5.Burstein, A., E. Morales, and J. Vogel (2015): “Accounting for Changes in Between-Group Inequality,” NBER Working Paper 20855, National Bureau of Economic Research.Burstein, A. and J. Vogel (2012): “International Trade, Technology, and the Skill Pre-mium,” mimeo, UCLA.93BibliographyBustos, P. (2011): “The Impact of Trade on Technology and Skill Upgrading: Evidencefrom Argentina,” mimeo, University of Pompeu Fabra.Carneiro, P., J. Heckman, and E. Vytlacil (2010): “Evaluating Marginal PolicyChanges and the Average Effect of Treatment for Individuals at the Margin,” Econo-metrica, 78.——— (2011): “Estimating Marginal Returns to Education,” American Economic Review,101.Chor, D. (2010): “Unpacking Sources of Comparative Advantage: A Quantitative Ap-proach.” Journal of International Economics, 82.Cos¸ar, A. K. (2013): “Adjusting to Trade Liberalization: Reallocation and Labor MarketPolicies,” mimeo, Stockholm School of Economics.Cos¸ar, A. K. and B. Demir (2015): “Domestic Road Infrastructure and InternationalTrade: Evidence from Turkey,” mimeo, Stockholm School of Economics.Costinot, A. and J. Vogel (2010): “Matching and Inequality in the World Economy,”Journal of Political Economy, 118.Davis, D. R. and J. Harrigan (2011): “Good Jobs, Bad Jobs, and Trade Liberalization,”Journal of International Economics, 84.De Loecker, J. (2011): “Product Differentiation, Multi-Product Firms and Estimating theImpact of Trade Liberalization on Productivity,” Econometrica, 79.De Loecker, J., P. Goldberg, A. Khandelwal, and N. Pavcnik (2012): “Prices,Markups and Trade Reform,” mimeo, Princeton University.Doms, M., T. Dunne, and K. R. Troske (1997): “Workers, Wages, and Technologies,”Quarterly Journal of Economics, 112.Doraszelski, U. and J. Jaumandreu (2013): “R&D and Productivity: Estimating Pro-duction Functions When Productivity is Endogenous,” Review of Economic Studies, forth-coming.Ebenstein, A., A. Harrison, M. McMillan, and S. Phillips (2014): “Estimating theimpact of trade and offshoring on American workers using the current population surveys,”Review of Economics and Statistics, 96, 581–595.Egger, H. and U. Kreikemeier (2009): “Firm Heterogeneity and the Labor MarketEffects of Trade Liberalization,” International Economic Review, 50.94BibliographyFeenstra, R. and G. Hanson (1996): “Globalization, Outsourcing, and Wage Inequality,”American Economic Review, 86.——— (1997): “Foreign Direct Investment and Relative Wages: Evidence from Mexico’sMaquiladoras,” Journal of International Economics, 42.——— (1999): “The Impact of Outsourcing and High-Technology Capital on Wages: Esti-mates for the United States,” Quarterly Journal of Economics, 114.Felbermayr, G., J. Prat, and H. Schmerer (2011): “Globalization and labor marketoutcomes: Wage bargaining, search frictions, and firm heterogeneity,” Journal of EconomicTheory, 146.Filho, N. A. M. and M.-A. Muendler (2007): “Labor Reallocation in Response to TradeReform,” CESifo Working Paper Series 1936, CESifo Group Munich.Firpo, S., N. M. Fortin, and T. Lemieux (2011): “Occupational Tasks and Changes inthe Wage Structure,” IZA Discussion Papers 5542, Institute for the Study of Labor (IZA).Foster, L., J. C. Haltiwanger, and C. J. Krizan (2001): “Aggregate ProductivityGrowth. Lessons from Microeconomic Evidence,” in New Developments in ProductivityAnalysis, National Bureau of Economic Research, Inc, NBER Chapters, 303–372.Frazer, G. (2013): “Imports, Import Sources and Skill Utilization,” Working paper, Uni-versity of Toronto.Fr´ıas, J. A., D. S. Kaplan, and E. A. Verhoogen (2009): “Exports and Wage Premia:Evidence from Mexican Employer-Employee Data,” mimeo, Columbia University.Gandhi, A., S. Navarro, and D. Rivers (2013): “On the Identification of ProductionFunctions: How Heterogeneous is Productivity,” mimeo, University of Wisconsin.Gindling, T. and D. Robbins (2001): “Patterns and Sources of Changing Wage Inequalityin Chile and Costa Rica During Adjustment,” World Development, 29.Goldberg, P. and N. Pavcnik (2005): “Trade Protection and Wages: Evidence from theColombian Trade Reforms,” Journal of International Economics, 66.Goldberg, P. K. and N. Pavcnik (2007): “Distributional Effects of Globalization inDeveloping Countries,” Journal of Economic Literature, 45, 39–82.Gonsaga, G., N. M. Filho, and C. Terra (2006): “Trade liberalization and the evolutionof skill earnings differentials in Brazil,” Journal of International Economics, 68.95BibliographyGoos, M., A. Manning, and A. Salomons (2014): “Explaining Job Polarization:Routine-Biased Technological Change and Offshoring,” American Economic Review, 104,2509–26.Gottschalk, P., D. A. Green, and B. M. Sand (2015): “Taking Selection to Task:Bounds on Trends in Occupational Task Prices for the US, 1984-2013,” Working paper,University of British Columbia.Griliches, Z. and H. Regev (1995): “Firm productivity in Israeli industry 1979-1988,”Journal of Econometrics, 65, 175–203.Grossman, G. M. and E. Rossi-Hansberg (2008): “Trading Tasks: A Simple Theory ofOffshoring,” American Economic Review, 98, 1978–97.Halpern, L., M. Koren, and A. Szeidl (2015): “Imports and Productivity,” AmericanEconomic Review forthcoming.Hanson, G. H., R. J. Mataloni, and M. J. Slaughter (2005): “Vertical ProductionNetworks in Multinational Firms,” The Review of Economics and Statistics, 87, 664–678.Harrison, A. and G. Hanson (1999): “Who gains from trade reform? Some remainingpuzzles,” Journal of Development Economics, 59, 125–154.Heckman, J. and E. Vytlacil (1999): “Local Instrumental Variables and latent Vari-able models for Identifying and Bounding Treatment Effects,” Proceedings of the NationalAcademy of Sciences, 96.——— (2005): “Structural Equations, Treatment Effects and Econometric Policy Evalua-tion,” Econometrica, 73.Heckman, J. J. and E. J. Vytlacil (2007a): “Chapter 70 Econometric Evaluation ofSocial Programs, Part I: Causal Models, Structural Models and Econometric Policy Eval-uation,” Elsevier, vol. 6, Part B of Handbook of Econometrics, 4779 – 4874.——— (2007b): “Chapter 71 Econometric Evaluation of Social Programs, Part II: Using theMarginal Treatment Effect to Organize Alternative Econometric Estimators to EvaluateSocial Programs, and to Forecast their Effects in New Environments,” Elsevier, vol. 6, PartB of Handbook of Econometrics, 4875 – 5143.Helpman, E., O. Itskhoki, and S. Redding (2010): “Inequality and Unemployment ina Global Economy,” Econometrica, 78.96BibliographyHendayana, S., S. Asep, and H. Imansyah (2010): “Indonesia’s Issues and Challengeson Quality Improvement of Mathematics and Science Education,” Journal of InternationalCooperation in Education, 41–51.Hummels, D., R. Jrgensen, J. Munch, and C. Xiang (2014): “The Wage Effectsof Offshoring: Evidence from Danish Matched Worker-Firm Data,” American EconomicReview, 104, 1597–1629.Hummels, D. L. and G. Schaur (2013): “Time as a Trade Barrier,” American EconomicReview, 103.Imbens, G. and J. Angrist (1994): “Identification and Estimation of Local Average Treat-ment Effects,” Econometrica, 62.Jensen, J. B., L. G. Kletzer, J. Bernstein, and R. C. Feenstra (2005): “TradableServices: Understanding the Scope and Impact of Services Offshoring [with Comments andDiscussion,” Brookings Trade Forum, 75–133.Kasahara, H. and B. Lapham (2013): “Productivity and the Decision to Import andExport: Theory and Evidence,” Journal of International Economics, 89.Kasahara, H., Y. Liang, and J. Rodrigue (2015): “Does Importing IntermediatesIncrease the Demand for Skilled Workers? Plant-level Evidence from Indonesia,” Workingpaper, University of British Columbia.——— (2016): “Does Importing Intermediates Increase the Demand for Skilled Workers?Plant-level Evidence from Indonesia,” Working paper, Vancouver School of Economics,University of British Columbia; Department of Economics, Vanderbilt University.Kijima, Y. (2006): “Why did wage inequality increase? Evidence from urban India 1983-1999,” Journal of Development Economics, 81.Koren, M. and M. Csillag (2011): “Machines and machinists: Capital-Skill Complemen-tarity from an International Trade Perspective,” IEHAS Discussion Papers 1114, Instituteof Economics, Centre for Economic and Regional Studies, Hungarian Academy of Sciences.Krishna, P., J. P. Poole, and M. Z. Senses (2014): “Wage Effects of Trade Reformwith Endogenous Worker Mobility,” Journal of International Economics, 93, 239 – 252.Kugler, M. and E. Verhoogen (2009): “Plants and Imported Inputs: New Facts and anInterpretation,” American Economic Review Papers and Proceedings, 99.97BibliographyLevinsohn, J. and A. Petrin (2003): “Estimating Production Functions Using Inputs toControl for Unobservables,” Review of Economic Studies, 70.Lileeva, A. and D. Trefler (2010): “Improved Access to Foreign Markets Raises Plant-Level Production...For Some Plants,” Quarterly Journal of Economics, 125.Martins, P. S. and L. D. Opromolla (2009): “Exports, Imports and Wages: Evidencefrom Matched Firm-Worker-Product Panels,” IZA Discussion Papers 4646, Institute forthe Study of Labor (IZA).Matsuyama, K. (2007): “Beyond Icebergs: Towards A Theory of Biased Globalization,”Review of Economic Studies, 74.McCaig, B. and N. Pavcnik (2014): “Export markets and labor allocation in a low-incomecountry,” NBER Working Paper 20455.——— (2015): “Informal employment in a growing and globalizing low-income country,”American Economic Review: Papers and Proceedings, 105.Melitz, M. J. (2003): “The Impact of Trade on Intra-Industry Reallocations and AggregateIndustry Productivity,” Econometrica, 71, 1695–1725.Melitz, M. J. and S. Polanec (2014): “Dynamic Olley-Pakes Productivity Decompositionwith Entry and Exit,” Working Paper 18182, National Bureau of Economic Research.Muendler, M.-A. (2004): “Trade, Technology, and Productivity: A Study of Brazilianmanufacturers 1986-1998,” mimeo, UCSD.Olley, G. S. and A. Pakes (1996): “The Dynamics of Productivity in the Telecommuni-cations Equipment Industry,” Econometrica, 64, 1263–97.Parro, F. (2013): “Capital-Skill Complementarity and the Skill Premium in a QuantitativeModel of Trade,” American Economic Journal: Macroeconomics, 5.Pavcnik, N. (2002): “Trade Liberalization, Exit, and Productivity Improvements: Evidencefrom Chilean Plants,” The Review of Economic Studies, 69, 245–276.——— (2003): “What explains skill upgrading in less developed countries?” Journal ofDevelopment Economics, 71.Sanchez-Paramo, C. and N. Schady (2003): Off and Running? Technology, Trade, andthe Rising Demand for Skilled Workers in Latin America, The World Bank.98Topalova, P. (2007): “Trade Liberalization, Poverty and Inequality: Evidence from In-dian Districts,” in Globalization and Poverty, National Bureau of Economic Research, Inc,NBER Chapters, 291–336.Trefler, D. (2004): “The Long and Short of the Canada-U.S. Free Trade Agreement,”American Economic Review, 94.Vannoorenberghe, G. (2011): “Trade between symmetric countries, heterogeneous firms,and the skill premium,” Canadian Journal of Economics, 44.Verhoogen, E. A. (2008): “Trade, Quality Upgrading, and Wage Inequality in the MexicanManufacturing Sector,” The Quarterly Journal of Economics, 123, 489–530.Vytlacil, E. (2002): “Independence, Monotonicity, and Latent Index Models: An Equiva-lence Result,” Econometrica, 70.Wooldridge, J. (2009): “On Estimating plant-level Production Functions Using ProxyVariables to Control for Unobservables,” Economic Letters, 104.Yeaple, S. (2005): “A Simple Model of Firm Heterogeneity, International Trade, andWages,” Journal of International Economics, 65.99Appendix AAppendix for Chapter 1A.1 Solve for the Baseline Model(Proof of Proposition 1 and Proposition 2)Assume θj = θ ∀θ, we have a system of task price changes ˆPrωt(1− ρδrωt − θ) Pˆrωt +∑j[(δrωtξrjt − (1− θ)φrωjt)(∑ω′pirω′jtPˆrω′t)]=∑j(φrωjt − δrωtξrjt)Lˆrjt − (1− δrωt)Yˆ ∗rωtOmit the region and time subscript for notation simplicity, defineaω ≡ 1− ρδω − θbωj ≡ δωξj − φωj(1− θ)Lˆω ≡∑j(φωj − δωξj)Lˆjyˆ∗ω ≡ (1− δω)Yˆ ∗ω ,Then the system can be written as(A+B) Pˆ = Lˆ− yˆ∗Where A is a Ω × Ω diagonal matrix with [A]ωω = aω, B is a Ω × Ω matrix with [B]ωω′ =∑Jj=1 bωjpiω′j . Pˆ , Lˆ, and Yˆ∗ are Ω vectors with [Pˆ ]ω = Pˆω, [Lˆ]ω = Lˆω and [yˆ∗]ω = yˆ∗ω.The matrix A+B is invertable when ρ 6= 1. The solution to this linear system is thenPˆ = (A+B)−1(Lˆ− yˆ∗)The relative employment share change pˆirωjt − pˆirω′jt, according to (1.8), is same acrossj′s and equals to θ(Pˆrωt − Pˆrω′t) for any pair of tasks (ω, ω′). I first discuss the solution100A.1. Solve for the Baseline Modelwhen there are only two tasks (routine and non-routine) and then extend it to three tasks(cognitive, routine and manual).Two tasks (Routine and Non-routine). When there are only two tasks, routine (R)and non-routine (N), the solution ispˆiR − pˆiN = ρ− 1det(A+B)θ×∑j[δR(φNj − δrNjξj)− δN (φRj − δrRjξj)] Lˆj + δN (1− δR)Yˆ ∗R − δR(1− δN )Yˆ ∗NThe general equilibrium scaling factor ρ−1det(A+B) is positive :Proof 1det(A+B) =aN + J∑j=1bNjpiNjaR + J∑j=1bRjpiRj− J∑j=1bNjpiRj J∑j=1bRjpiNj=aN + J∑j=1bNjpiNjaR + J∑j=1bRjpiRj +J∑j=1bRjpiNj−aN + J∑j=1bNjpiNj J∑j=1bRjpiNj− J∑j=1bNjpiRj J∑j=1bRjpiNj= (ρ− 1)δN J∑j=1bRjpiNj− δRaN + J∑j=1bNjpiNjThus,ρ− 1det(A+B)> 0 ⇐⇒ δN J∑j=1bRjpiNj− δRaN + J∑j=1bNjpiNj > 0The two summation terms can be combined to getδN J∑j=1bRjpiNj− δR J∑j=1bNjpiNj = (θ − 1) J∑j=1(δNφRj − δRφNj)piNj101A.1. Solve for the Baseline ModelCombining all the terms to getδN J∑j=1bRjpiNj− δRaN + J∑j=1bNjpiNj= (θ − 1)J∑j=1(δNφRj − δRφNj)piNj − δR(1− ρδN − θ)= (θ − 1)1− J∑j=1φNjpiNj δR +(θ − 1) J∑j=1φRjpiNj+ ρδR δN≥ (θ − 1)1− pimaxNj J∑j=1φNj δR +(θ − 1) J∑j=1φRjpiNj+ ρδR δN> 0Q.E.DAssume this scaling factor to be same across regions, and assume the comparative ad-vantages captured by φωj − ξj to be region-invariant (a female worker is better at routinejobs wherever she works). Consider a two period change, I can add back the region and timesubscript to getpˆirR − pˆirN =J∑j=1αjrt0Lˆrj + βRrt0(1− δrRt0)Yˆ ∗R + βNrt0(1− δrNt0)Yˆ ∗Nwhere αjrt0 = δrRt0(φNjt0 − δrNt0ξjt0) − δrNt0(φRjt0 − δrRt0ξjt0) are positive or negative de-pending on the beginning of period shares {δrωt0}ω∈{R,N} and the comparative advantages ofgroup j in tasks R and N captured by φRjt0 − δrRt0ξjt0 and φNjt0 − δrNt0ξjt0 ; βRrt0 = δrNt0is positive and its magnitude increases domestic share of non-routine task; βNrt0 = −δrRt0 isnegative and its absolute value increasing with domestic share of routine task.Three tasks (Cognitive, Routine and Manual). When there are three tasks, use subscriptC,R,M to represent cognitive, routine and manual accordingly for notation simplicity. Definebkh ≡∑Jj=1 bkjpihj , the solution satisfiesRˆ − Cˆ = 1− ρdet (A+B)θ[αR−CC (LˆC − Yˆ ∗C) + αR−CR (LˆR − Yˆ ∗R) + αR−CM (LˆM − Yˆ ∗M )]102A.2. Matching ISCO68 with 1990 US Census Occupational Classificationwhere,αR−CC = δMbRM − δRaM − δRbMM > 0αR−CR = −δMbCM + δCaM + δCbMM < 0αR−CM = δRbCM − δCbRM ≷ 0αR−MC = δRbMC − δMbRC ≷ 0αR−MR = −δCbMC + δMaC + δMbCC < 0αR−MM = δCbRC − δRaC − δRbCC > 0The the general equilibrium scaling factor 1−ρdet(A+B) is positive when ρ 6= 1.68Assume this scaling factor to be constant across regions and consider a two period change,I can add back the region subscript r and time subscript t to getpˆirR− pˆirC =J∑j=1αL,R−Crjt0 Lˆrj +βC,R−Crt0(1−δC)Yˆ ∗rC +βR,R−Crt0 (1−δR)Yˆ ∗rR+βM,R−Crt0 (1−δM )Yˆ ∗rMIn this expression, αL,R−Crjt0 =∑ω=C,R,M αR−Cω,rt0(φrωjt0−δrωt0ξrjt0) can be positive or negative,depending on the impact of supply shocks on the dependent variable {αR−Crωt0 }ω∈{C,R,M} andthe comparative advantage of group j across tasks captured by earning shares of a group ineach task relative to its total earning share {φrωjt0− δrωt0ξrjt0}ω∈{C,R,M}. βR,R−Crt0 = −αR−CrRt0is positive and βC,R−Crt0 = −αR−CrCt0 is negative with the magnitude of their absolute valuesdecreasing in {δω′}{ω′ 6=ω}. The sign of βM,R−Crt0 is not clear. Same as the two task case,the magnitudes of these effects depend on foreign shares of each task. More than that, theemployment share of a third task matters. Thus, I will include controls of task shares of thethird task measures. The results for pˆirR − pˆirM are analogous.A.2 Matching ISCO68 with 1990 US Census OccupationalClassificationIndonesia adapts the International Standard Classification of Occupations 1968 version (ISCO68)while US data uses the US Census occupation classification. Harry Ganzeboom69 matchedthe 1988 version of ISCO (ISCO88) with ISCO68 and also ISCO88 with the 1900 US Censusclassification. Based on his work, I used ISCO88 to bridge ISCO68 and the 1990 US CensusClassification. The matching between ISCO88 and US Census and also between ISCO88and ISCO68 are many to many. I take average task scores when many US occupations arematched with one ISCO88 occupations and also when many ISCO88 codes are matched with68To save the space, the long proof is available upon request.69Correspondence tables available from http://www.harryganzeboom.nl/ismf/index.htm103A.3. Task Groups of Occupationsone ISCO68 code.A.3 Task Groups of OccupationsTable A.1 list the 79 consistently defined occupations grouped into cognitive, routine andmanual, ordered by the task intensity within each group.A.4 Measure the Distance to Port (Airport)To form this instrumental variable, we use the location information of individual plants. In-donesia is comprised of 33 provinces which are administratively subdivided in to 429 regenciesin our data. The dataset includes the location of the surveyed plants down to the region level.Due to the detailed administrative divisions, the variation contained in the plant location datais informative. Among all ports in Indonesia, 2 can be considered large, 16 medium and allothers remaining are either small or very small.70 The 18 large or medium sized ports arechosen to be the main destinations for our constructed measure of transportation costs71.Specifically, given these destinations, and taking the geographical features of Indonesia intoconsideration, we compute the least-cost path from the center of every regency to its nearestport by ArcGIS. The calculation divides the entire country into cells with size 1 km2 whereeach cell contains a value representing the average elevation of that area. The travel costof each cell depends on the slope from the cell to its adjacent cells and whether the celllocates on land or on sea. ArcGIS determines the optimal route for each cell by finding theleast-accumulative-cost path to its nearest major port. The transportation cost for a regencyis approximated by the the accumulative cost along the optimal route from the center cell ofthe regency. For each plant, the proxy for its transportation cost is the transportation costof the regency in which the plant is located. Details about the process of computing this costmeasure are described in the following paragraphs.Three types of data are used in ArcGIS to generate the transportation cost: raster data(R), point data (P) and table data (T). Raster data consists of a matrix of cells (pixels)organized into a grid where each cell contains a value representing information. In our data,each cell represents a 1 km2 square in the real world. Point data contains information forspecific points. Each point is composed of one coordinate pair representing its location onthe earth. Table data is used to store the attributes (e.g. names, locations, temperatures,etc.) of features.70Source of port information: World Port Source.71Measuring the distance to airport is analogous104A.4. Measure the Distance to Port (Airport)There are three main steps for computing the transportation cost. First, generate the costraster for Indonesia which defines the cost to move planimetrically through each cell accordingto geographical features. Second, given a cost raster and the main ports as destinationpoints, the “Cost Distance” tool generates the raster data in which the least accumulated costdistance for each cell to its nearest destination is calculated. Lastly, to get the measure of thetransportation cost for each regency, we extract the cost distance value for the cells locatedin the center of the regencies from the raster data obtained from second step. FigureA.1displays the process of this calculation. The ellipses in the flowchart represent data while theround-cornered squares represent tools.Step 1. The travel cost of each cell depends on the slope from the cell to its adjacentcells and whether the cell is located on land or sea. “Elevation-full” is the Indonesia elevationdata, the value of a cell in this raster data indicates the average elevation in the 1 km2. Cellsin the sea take a value of zero. The “SLOPE” tool generates the slope layer “ElevationSlope”, in which a cell value indicates the maximum rate of change between the cell and itsneighbors. A road which traverses less steep slopes is preferable. We reclassify the slopelayer, slicing the values into 10 equal intervals. A value of 10 is assigned to the most costlyslopes (steepest) and 1 is assigned to the least costly slope (flattest), values in between areranked linearly. “Reclass Slope” is the raster data after re-classification. Each cell valuebetween 1 and 10 indicates the difficulty of traveling over it. One problem with this surfaceis that traveling across the sea is considered costless since the elevation is zero (and so arethe slopes) everywhere on the sea. To solve this problem, another layer “Sea” is created.The “Sea” raster assigns value 0 for land and 1 for sea. The last step for generating the costraster overlays the rasters “Reclass Slope” and “Sea” using a common measurement scaleand weights 50 percent on each layer. Specifically, scale values of the “Reclass Slope” layerare unchanged (10 for steepest and 1 for flattest), and scale values for “Sea” layer are set tobe 1 for land (low cost) and 10 for sea (high cost), thus, the cost of travelling over cell i isCosti = 0.5×ReclassSlopei + 0.5× 10SeaiPutting all the cells on map forms the raster data “Cost Surf”.Step 2. Given the 18 main ports (“Main Ports”) as destinations, the “COST DIS-TANCE” tool calculates the accumulated distance from each cell to its nearest destinationalong the optimal path, using the “Cost Surf” data obtained in step 1 to measure the costof passing cells. The resulting raster data “Cost Dist” reports the transportation cost of allthe cells.Step 3. We extract the values of the cells located in the center of administrative regenciesfrom the transportation cost map “Cost Dist” using the tool “EXTRACT VALUES TO105A.4. Measure the Distance to Port (Airport)POINTS.”Figure A.1: Process of Measuring Transportation CostNotes: This figure displays the process of calculating the transportation cost for the regenciesin Indonesia using ArcGIS. The ellipses in the flowchart represent data and the round-corneredsquares represent tools.106A.5. First Stage ResultsA.5 First Stage ResultsTable A.2 present selected first stage results. I focus on the small instrument set sinceindividual significance of the instrumental variables is more transparent.A.6 Model with Agricultural SectorA.6.1 Model SettingsThere are measure one of workers with homogeneous preferences. At any time t, the repre-sentative worker’s optimization problem ismax{Y Nt ,Y At }α ln(Y Nt ) + (1− α) ln(Y At )s.t PAt YAt + YNt ≤ Ytwhere, Y Nt is the consumption of non-agricultural goods and YAt is the consumption ofagricultural goods, PAt is the relative price of agricultural goods, and the non-agriculturalgood is taken to be the numeraire. Yt is the total income. Then, the demand of non-agricultural and agricultural goods in this economy satisfies Y Nt = αYt and PAt YAt = (1−α)Yt.To describe the production of non-agricultural and agricultural goods, I use superscriptsto indicate goods and subscripts to indicate tasks. The non-agricultural good is producedby a set of non-agricultural tasks ΩN ≡ {1, 2, . . . ,Ω}, according to a constant elasticity ofsubstitution (CES) production functionY Nt = ∑ω∈ΩNµ1ρωYρ−1ρωtρρ−1,where ρ > 0 is the elasticity of substitution across tasks, Yωt ≥ 0 is the output of task ωand µt is the task-augmenting productivity that is assumed to be constant over time. Theagriculture good is simply produced by a single agricultural task AY At = YAtA.6.2 Equilibrium with Trade of TasksTask SupplyThe selection of workers to perform various tasks is same as the baseline model, exceptthat agriculture task option is added. Specifically, a worker i in group j working in task107A.6. Model with Agricultural Sectorω ∈ {ΩN , A} generates task outputY iωt = Tjωiωt,where, Tωj captures the group level comparative advantage and iωt is individual-specific taskproductivities. Let it = (i1t, i2t, ..., iΩt) to be drawn from a multivariate Fre´chet distribution,the fraction of workers in group j that works in task ω ispiωjt =(Pωt · Tωj)θj∑ω′∈{ΩN ,A}(Pω′t · Tω′j)θj for ω ∈ {ΩN , A}. (A.1)Again, by the Fre´chet distribution assumption, Wωjt = Wjt withWjt = γj ∑ω∈{ΩN ,A}(Pωt · Tωj)θj 1θj . (A.2)Task DemandAccording to the goods demand and profit maximization of non-agricultural and agriculturalsectors, the demand for non-agricultural task in the economy isYωt = αµωYtP−ρωt + Y∗ωt for ω ∈ ΩN , (A.3)where, Y ∗ωt is the foreign demand of task ω. The demand for agricultural task in the economyisYAt = (1− α)Yt/PAt + Y ∗At (A.4)where, Y ∗ωt is the foreign demand of task AEquilibriumIn a free-entry competitive market, the total revenue of a task equals to the total cost ofproducing it,PωtYωt =J∑j=1WjtLjtpiωjt for ω ∈ {ΩN , A}. (A.5)Lastly, the total income equals to total productionYt =∑jWjtLjt (A.6)The equilibrium of this economy is described by J(Ω + 1) + 2(Ω + 1) + J + 1 endogenous108A.6. Model with Agricultural Sectorvariables {Pωt, Yωt, piωjt, Yt,Wjt : ω ∈ {ΩN , A}, j ∈ {1, 2, ..., J}} determined by task supplyequations (A.2) and (A.1), task demand equations (A.3) and (A.4), and market clearingconditions (A.5) and (A.6).A.6.3 Solution to the EquilibriumDefine xˆ ≡ ∆x/x = d ln(x), I log differentiate the equilibrium equations. The equilibriumwith task trading can be described by the follow linear systempˆiωj = θj(Pˆω −∑ω′∈{ΩN ,A}piω′jt0Pˆω′) for ω ∈ {ΩN , A} , (A.7)Wˆj =∑ω′∈{ΩN ,A}piω′jt0Pˆω′ , (A.8)Yˆω =J∑j′=1φj′ωt0(Wˆj′ + Lˆj′ + pˆiωj′)− Pˆω for ω ∈ {ΩN , A} , (A.9)Yˆω = δωt0(Yˆ − ρPˆω) + (1− δωt0)Yˆ ∗ω for ω ∈ ΩN , (A.10)YˆA = δAt0(Yˆ − PˆA) + (1− δAt0)Yˆ ∗A , (A.11)Yˆ =J∑j′=1ξj′t0(Wˆj′ + Lˆj′), (A.12)where, φωjt0 ≡ WjLjt0piωjt0∑Jj′=1Wj′t0Lj′t0piωj′t0is the share of group j’s earnings in task ω, δωt0 ≡ 1−Y ∗ωt0Yωt0is the share for the local consumption of task ω, and ξjt0 ≡ Wjt0Ljt0∑Jj′=1Wj′t0Lj′t0is the share of groupj’s earnings in total earnings. Assume θj = θ ∀j and solve for the task prices {Pω}ω∈{ΩN ,A}to get(1− ρδωt0 − θ)Pˆω +J∑j′=1(δωt0ξj′t0 − φj′ωt0(1− θ))∑ω∈{ΩN ,A}piω′j′t0Pˆω′=J∑j′=1(φj′ωt0 − δωt0ξj′t0)Lˆj′ − (1− δωt0)Yˆ ∗ω for ω ∈ ΩNand(1− δAt0 − θ)PˆA +J∑j′=1(δAt0ξj′t0 − φj′At0(1− θ))∑ω∈{ΩN ,A}piω′j′t0Pˆω′=J∑j′=1(φj′At0 − δAt0ξj′t0)Lˆj′ − (1− δAt0)Yˆ ∗A109A.6. Model with Agricultural SectorDefineaωt0 ≡ 1− ρδωt0 − θ for ω ∈ ΩNaAt0 ≡ 1− δAt0 − θbωjt0 ≡ δωt0ξjt0 − φωjt0(1− θ) for ω ∈ {ΩN , A}lˆω ≡∑j(φωjt0 − δωt0ξjt0)Lˆj for ω ∈ {ΩN , A}yˆ∗ω ≡ (1− δωt0)Yˆ ∗ω for ω ∈ {ΩN , A}The system can be written asΛ[PˆNPˆA]=[lˆNlˆA]−[yˆNyˆA]where,Λ =[A BC D]≡[ANN +BNN BNABAN′AAA +BAA],where, ANN is an Ω × Ω diagonal matrix with [ANN ]ωω = aωt0 , BNN is an Ω × Ω matrixwith [BNN ]ωω′ =∑Jj=1 bωjt0piω′jt0 . PˆN , lˆN , and yˆ∗N are Ω vectors with [PˆN ]ω = Pˆω,[lˆN ]ω = lˆω and [yˆ∗N]ω = yˆ∗ω. BNA is an Ω-vector with [BNA]ω =∑Jj=1 bωjt0piAjt0 , BANis an Ω-vector with [BAN ]ω =∑Jj=1 bAjt0piωjt0 . AAA +BAA is a scalar that equals to aAt0 +∑Jj=1 bAjt0piAjt0 . Notice thatΛ11 is identical to coefficient matrix of task prices in the baselinemodel. This system extends the baseline task prices by adding one column and one row toinclude agricultural sector. The matrix Λ is invertible when ρ 6= 1. The solution to this linearsystem is then [PˆNPˆA]=[A BC D]−1 [lˆN − yˆNlˆA − yˆA]The relative employment share change pˆiωj − pˆiω′j , according to (A.7), is same across j′s andequals to θ(Pˆω− Pˆω′) for any pair of tasks (ω, ω′). I am interested in the relative employmentof non-agricultural tasks. Thus, the goal of solving this system of task prices is to derive{pˆiωj − pˆiω′j} for ω ∈ ΩN explicitly.Suppose there are three non-agricultural tasks: cognitive, routine and manual (ΩN ={C,R,M}), then PˆCPˆRPˆMPˆA =[A3×3 B3×1C1×3 D1×1]−1 lˆC − yˆClˆR − yˆRlˆM − yˆMlA − yA110A.6. Model with Agricultural SectorwhereA3×3 =aCt0 +∑j bCjt0piCjt0∑j bCjt0piRjt0∑j bCjt0piMjt0∑j bRjt0piCjt0 aRt0 +∑j bRjt0piRjt0∑j bRjt0piMjt0∑j bMjt0piCjt0∑j bMjt0piRjt0 aMt0 +∑j bMjt0piMjt0B3×1 =∑j bCjt0piAjt0∑j bRjt0piAjt0∑j bMjt0piAjt0 C1×3 =∑j bAjt0piCjt0∑j bAjt0piRjt0∑j bAjt0piMjt0TD1×1 = aAt0 +∑jbAjt0piAjt0Notice that the upper left block matix A is same as the baseline model. The inversion of Λcan be expressed as[A BC D]−1=[(A−BD−1C)−1 −(A−BD−1C)−1BD−1−D−1C(A−BD−1C)−1 D−1 +D−1C(A−BD−1C)−1BD−1]I am interested in the coefficients on {yˆω}ω∈ΩN and yˆA in the expression for PˆR− PˆC (i.e.1θ (pˆiR − pˆiC)). They are:βC,R−Ct0 = (Λ−1)11 − (Λ−1)21 = 1− ρdet(A−BD−1C) [δRt0(A33 −B3C3/D)− δMt0(A23 −B2C3/D)]βR,R−Ct0 = (Λ−1)21 − (Λ−1)22 = 1− ρdet(A−BD−1C) [δMt0(A13 −B1C3/D)− δCt0(A33 −B3C3/D)]βM,R−Ct0 = (Λ−1)31 − (Λ−1)32 = 1− ρdet(A−BD−1C) [δCt0(A23 −B2C3/D)− δRt0(A13 −B1C3/D)]βA,R−Ct0 = (Λ−1)41 − (Λ−1)42 = − 1D[βC,R−Ct0 B1 + βR,R−Ct0 B2 + βM,R−Ct0 B3]A.6.4 Empirical ImplicationsThe relative employment share change pˆiωj − pˆiω′j , according to (A.7), is same across j′s andequals to θ(Pˆω − Pˆω′) for any pair of tasks (ω, ω′). Suppose there are three non-agriculturaltasks: cognitive, routine and manual (ΩN = {C,R,M}), and treat each region r as a smallopen economy, the solution of the task prices providespˆirRj − pˆirCj =J∑j=1αL,R−Crjt0 Lˆrj + βC,R−Crt0(1− δrCt0)Yˆ ∗rC + βR,R−Crt0 (1− δrRt0)Yˆ ∗rR+ βM,R−Crt0 (1− δrMt0)Yˆ ∗rM + βA,R−Crt0 (1− δrAt0)Yˆ ∗rA,111A.6. Model with Agricultural Sectorwhere, βC,R−Crt0 is negative, βR,R−Crt0is positive, but the signs of other coefficients are undeter-mined.72 The magnitude of the coefficients depend on {(1− δωt0)}ω∈{ΩN ,A}. The magnitudeof each βω,RCrt0 also depends on {pirω′jt0}ω′ 6=ω.To empirically test this result, I use interaction terms to capture the influences of theseinitial-status effect. Notice that there is only one agricultural industry and only one task isused in its production. As a result, there is no regional variation in δAt0 because there is noindustry composition ”within” agricultural sector. In all regions, the share of agriculturaltask being used by foreign countries equals to the national level agricultural goods net exportover total production. Thus, in all interaction terms, (1 − δAt0) is omitted. The resultingempirical specification is thus:pˆirRj − pˆirCj =J∑j=1αL,R−Cj Lˆrj +∑ω∈ΩNβω,R−C(1− δrωt0)Yˆ ∗rω + βA,R−C Yˆ ∗rA+J∑j=1αXL,R−cj∏ω′∈ΩN(1− δrω′t0)Lˆrj+∑ω∈{ΩN ,A}βδXω,R−C∏ω′∈ΩN(1− δrω′t0)Yˆ ∗rω+∑ω∈{ΩN ,A}βpiXω,R−C∏ω′ 6=ωpirω′t0 Yˆ∗rω + urj .The expression of pˆirR − pˆirM is analogous.72The proofs of the signs are similar to the baseline model proofs, and are available upon request.112A.6. Model with Agricultural SectorTable A.1: Occupation GroupsISCO Occupation DescriptionPanel A: Non-Routine Cognitive Occupations18 Athletes, sportsmen and related workers9 Economists20 Legislative officials and government administrators14 Workers in religion12 Jurists15 Authors, journalists and related writers50 Managers (catering and lodging services)2 Architects, engineers and related technicians51 Working proprietors (catering and lodging services)1 Physical scientists and related technicians21 Managers13 Teachers70 Production supervisors and general foremen42 Sales supervisors and buyers8 Statisticians, mathematicians, systems analysts and related technicians11 Accountants5 Life scientists and related technicians6 Medical, dental, veterinary and related workers7 Medical, dental, veterinary and related workers (7)30 Clerical supervisors40 Managers (wholesale and retail trade)17 Composers and performing artists3 Architects, engineers and related technicians (3)58 Protective service workers34 Computing machine operators19 Professional, technical and related workers not elsewhere classified44 Insurance, real estate, securities and business services salesmen and auctioneers41 Working proprietors (wholesale and retail trade)57 Hairdressers, barbers, beauticians and related workers52 Housekeeping and related service supervisors43 Technical salesmen, commercial travellers and manufacturers’ agentsPanel B: Routine Occupations38 Telephone and telegraph operators91 Paper and paper board products makers37 Mail distribution clerks92 Printers and related workers75 Spinners, weavers, knitters, dyers and related workers72 Metal processers56 Launderers, dry-cleaners and pressers113A.6. Model with Agricultural SectorTable A.1: Occupation Groups (Cont.)ISCO Occupation DescriptionPanel B: Routine Occupations (Cont.)33 Bookkeepers, cashiers and related workers74 Chemical processers and related workers32 Stenographers, typists and card- and tape-punching machine operators90 Rubber and plastics product makers81 Cabinetmakers and related woodworkers87 Plumbers, welders, sheet metal and structural metal preparers and erectors88 Jewellery and precious metal workers31 Government executive officials77 Food and beverage processers16 Sculptors, painters, photographers and related creative artists78 Tobacco preparers and tobacco product makers89 Glass formers, potters and related workers39 Clerical related workers not elsewhere classified82 Stone cutters and carvers83 Blacksmiths, toolmakers and machine-tool operators73 Wood preparation workers and paper makers35 Transport and communications supervisors79 Tailors, dressmakers, sewers, upholsterers and related workers94 Production and related workers not elsewhere classified86 Broadcasting station and sound equipment operators and cinema projectionists96 Stationary engine and related equipment operators93 Painters80 Shoemakers and leather goods makers76 Tanners, fellmongers and pelt dressers53 Cooks, waiters, bartenders and related workersPanel C: Non-Routine Manual Occupations4 Aircraft and ships’ officers71 Miners, quarrymen, well drillers and related workers98 Transport equipment operators84 Machinery fitters, machine assemblers and precision instrument makers (exceptelectrical)85 Electrical fitters and related electrical and electronics workers97 Material-handling and related equipment operators, dockers and freight handlers36 Transport conductors95 Bricklayers, carpenters and other construction workers59 Service workers not elsewhere classified55 Building caretakers, charworkers, cleaners and related workers54 Maids and related housekeeping service workers not elsewhere classified114A.6. Model with Agricultural SectorTable A.2: First Stage Regressions Correspond to IV regressions in Table 1.7(1) (2) (3) (4) (5) (6)VARIABLES Yˆ ∗C Yˆ∗R Yˆ∗M Yˆ∗C Yˆ∗R Yˆ∗M̂TARIFFC -1.766*** -0.312 -0.129 -1.705*** -0.229 -0.102(0.589) (0.404) (0.313) (0.551) (0.373) (0.306)̂TARIFFR -0.00819 -0.385*** 0.0404 0.00868 -0.359** 0.0185(0.0494) (0.148) (0.0514) (0.0517) (0.140) (0.0491)̂TARIFFM 0.133 0.0373 -3.570*** 0.268 -0.185 -2.876**(0.151) (0.200) (1.142) (0.254) (0.315) (1.451)Observations 493 493 493 452 452 452F-stat Exc. IVs 20.44 6.672 4.974 10.40 7.692 8.558Robust Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1115Appendix BAppendix for Chapter 2B.1 Data DescriptionB.1.1 Manufacturing Plant DataOur plant level data comes from the Indonesian manufacturing census (Large and MediumIndustrial Statistics) in years 1994-1996 and 2004-2007. This survey data covers all manu-facturing plants in Indonesia with at least 20 employees. Key variables used in our study aredescribed below.LaborFor each plant, the survey records the education levels of all production and non-productionworkers. This dimension of the data allows us to compute the number of skilled and un-skilled workers in each occupation category. We define production workers with more thanhigh-school education or non-production workers with more than college education as skilledworkers. Using this definition we count the number of skilled and unskilled workers for eachoccupation category and each plant.Intermediate Goods and CapitalIn order to estimate plant-specific productivity, we also need the intermediate goods andcapital used for production. Intermediate goods includes imported raw materials, domesti-cally purchased raw materials and expenditures on energy. The wholesale price index formanufactured goods is used to convert nominal values into real values.We compute the real value of capital at the beginning of year t asKit = buildingit/Pbuildt +machineit/Pmacht + vehicleit × 100/P vehict + (rentit/0.1)/P rentt ,where buildingit, machineit, and vehicleit are the nominal value of buildings, machines, andvehicles at the beginning of year t; rentit is equal to the reported value of rental paymentsfor buildings and machines, where we divide the rental value by the depreciation rate (10percent) to get the appropriate capitalized value. The capital price indices are obtained116B.1. Data Descriptionfrom Badan Pusat Statistik (BPS).73 Since rent is only paid for buildings and machines, wecompute price index for rented capital asP rentt =∑i buildingit∑i(buildingit +machineit)× P buildt +∑imachineit∑i(buildingit +machineit)× Pmacht .When the capital values are not reported in 1996 or 2006, we use the reported values ofcapital in 1994, 1995 and 1997 for constructing the 1996 capital value, and similarly, thereported values of capital in 2004, 2005 and 2007 for constructing the 2006 capital valueby assuming Kit = 0.9Kit−1 + Investmentit−1 with Investmentit = Investmentbuildingsit +Investmentmachinesit +Investmentvehiclesit , where Investmentbuildingit , Investmentmachinesit , andInvestmentvehiclesit are the real values of net investment of buildings, machines, and vehiclesin year t.Some plants do not report capital values in any year between 2004 and 2007. For thoseplants, we impute the values of capital as follows. First, using the plant observations in 2005for which capital values are constructed from the data between 2004 and 2007, we run theOLS regression logKi,2005 = X′i,2005α+i,2005, where Ki,2005 is the beginning-of-period capitalin 2005; Xit−1 includes a constant, the ratio of investment to capital, the log of productionworkers, the log of non-production workers, the log of output, the log of intermediate goods,an import dummy, province dummies, industry dummies, plant age, plant age squared, adummy variable for positive investment, a dummy variable for no hiring of production work-ers, and a dummy variable for no hiring of non-production workers. Then, given the OLSestimate of α, αˆ, we compute the imputed value of capital at the beginning of year 2006 forplants with missing capital values as Kimputei,2006 = 0.9 exp(X′i,2005αˆ) + Investmenti,2005. Forthe sample of initial non-importers, we use the imputed values of capital for 11 percent ofobservations. For plants with missing capital values in 1996, we construct the imputed valueof capital at the beginning of year 1996 using 1995 data in the same way.Other Plant VariablesOther plant information contained in the data includes the percentage of foreign ownership,total expenses on research and development (R&D), and total expenses on training. Dummiesvariables for foreign ownership, R&D and training are defined as whether the above mentionedvariables are greater than zero.73Specifically, we use the price indices for construction goods, imported and domestic machines, and vehicles.The imported and domestic machines price indices are weighted according to the input-output table formanufactured goods to get one price index for machines. The building price index covers the period 1996-2006and is extended to 2007; machine and vehicle price data only covers 1998 to 2005 and is extended to the period1996-2007. The extension from 1998 to 1996 relies on the wholesale price of capital goods which is availableduring the 1992-1999 period. The GDP deflators of construction goods, machines and vehicles are used toextend the original price index to 2007.117B.1. Data DescriptionB.1.2 Regional VariablesThe plants in our data locate across 33 provinces and 397 regions (kabupaten/kota) in In-donesia. This detailed location information allows us to take use of the variations in the localwages and the transportation cost.WageWe use the household survey data (SAKERNAS—Indonesian Labour Force Survey) to esti-mate the skill premium in each region after controlling for other personal characteristics ofworkers that may affect their wages. Specifically, using the sample of employed workers inthe household survey for 1996 and 2006, we estimate the following Mincer regression:log(Wageir) = βgGenderi+βxExperiencei+βx2Experience2+βsSkilli+βsrSkill×Dr+βrDr,where Wageir is the reported wage for individual i in region r, Genderi represents individuali ’s gender, Experiencei is the years of work experience, and Dr is a regional dummy forregion r. Skilli is a skill dummy based on an education threshold of highschool or college.The estimated value of βs + βsr is then used as our measure for the log of the relativewage ratios of skilled to unskilled workers in year 1996 or 2006, denoted by ln(Ws/Wu)96 orln(Ws/Wu)06, respectively. These skill premium measures depend on whether an educationthreshold to define Skilli is highshool or college. The skill premium based on a threshold ofhighschool education are used for the regressions in columns (1)-(4) of Table 2.5 or columns(1)-(4), (9)-(12) of Table 2.6 while we use a threshold of college education for the regressionsin columns (5)-(8) of Tables 2.5-2.6 .Distance to PortThe distance to the closest port is computed following the steps described in section A.4.B.1.3 Industrial VariablesTariffsTariff data are from Amiti and Konings (2007), where they constructed the input and outputtariffs for 5-digit ISIC industries during 1996-2001 based on an input-output table that is notpublicly available. We use the plants’ 1996 industry affiliation to assign the tariff changesto individual plants. Using the initial industry affiliation prevents potential bias that wouldarise from plants which strategically switched to new industries in response to changes in thetrade environment. One potential concern with this tariff data is that it does not cover theentire period we study (1996-2006). We use the tariff data from WITS that is reported at the118B.1. Data DescriptionFigure B.1: Trend of the Average Input and Output Tariff, 1996-2006Notes: the industrial output tariffs are the effectively appliedtariff provided by WITS. Industries are classified by 4-digitISIC. The tariff of an industry is the simple average of theindustry’s tariffs charged to all trading countries.4-digit ISIC industry classification to check the tariff changes in the 2001-2006 period. FigureB.1 demonstrates that most of the reduction in Indonesian input and output tariffs occurredbefore 2001. Figure B.2 demonstrates that output tariffs have fallen across most industriesin Indonesia over the 1991-2001 period and that there is substantial variation in the initialtariff levels and the subsequent fall across 5-digit industries over the following decade. Giventhat most of the tariff reductions had occurred by 2001 and are driven by the initial tarifflevels, we choose to use the tariff rates constructed by Amiti and Konings because they areconstructed at a more disaggregated industry level, and thus provide more variation in thetariff changes across plants.Import Heaviness and AirshareThis section describes our measures of the heaviness of imported inputs (import weight)and the fraction to of imported inputs shipped by air (import airshare) as described in themain text. We first create proxy variables for transport intensity at HS6 level for Indone-sian imports using data on US and EU imports to Indonesia by mode of transportation forthe year 2006. Detailed data for U.S. exports by commodity and transport mode are pub-lished by the US census at http://www.census.gov/foreign-trade/reference/products/layouts/imdb.html#imp_detl. Similar data for EU exports is taken from the EU Interna-tional Trade Database ComExt which is published at http://epp.eurostat.ec.europa.eu/newxtweb/. The underlying data set for our EU instruments is collected in the dataset119B.1. Data DescriptionFigure B.2: Change in Tariffs, 1991-2000, Relative to 1991 Level−.50.511.5Change in Tariffs 1991−20000 .2 .4 .6 .8Tariff in 1991Notes: Tariffs fell over the sample period in all industries withthe exception of the liquors and wine industries (ISIC codes31310, 31320) and rice milling industries (ISIC codes 31161,31169).named ‘EXTRA EU Trade Since 2000 By Mode of Transport (HS6) (DS-043328).’ We thenfollow Cos¸ar and Demir (2015) to construct the heaviness and fraction of imports shipped toIndonesia at each HS6 commodity code for both the EU and US series separately.To create the measures of imported input heaviness and airshare we need to map theHS6 measures above to the import input-output matrix produced by BPS Indonesia. A keyintermediate step in this process is linking the HS6 commodity codes to ISIC 3.0 industry clas-sification in order to create industry-level import variables. To complete this task we use thecorrespondence table ‘2002 NAICS to ISIC 3.1’ as published by the U.S. Census (https://www.census.gov/cgi-bin/sssd/naics) and the correspondence table ‘isic31 to isic3’ fromthe United Nations Stats Division published online at http://unstats.un.org/unsd/cr/registry/regot.asp?Lg=1. For robustness, we repeat this concordance using the correspon-dence tables ‘2002 NAICS to ISIC 4’, ‘isic4 to isic31’ and ‘isic31 to isic3’. These producesimilar results.Last, we use the import Input-Output Table produced by BPS Indonesia (2000) to con-struct the imported input measures of heaviness and airshare. The input-output matrixprovided by BPS Indonesia allows us to determine the share of import expenditures in eachsector. Specifically, we subtract total domestic expenditures in any given sector from totalexpenditures in the same sector. For each sector we can then straightforwardly compute theshare of total expenditures on imports from each individual sector.The input-output tables also provide a concordance between ISIC 3.0 classifications and120B.2. Estimating MTE and Treatment Effects0.2.4.6Imported Input Airshare15111                                       17220                                        24119                                        26900                                        31149                                        32210                                        35293                                          37205Note:  Each bar represents a distinct (5−digit) manufacturing industry code.(a) Fraction of Import Shipments by Air0123Imported Input Weight15111                                       17220                                        24119                                        26900                                        31149                                        32210                                        35293                                          37205Note:  Each bar represents a distinct (5−digit) manufacturing industry code.(b) Weight of Import ShipmentsFigure B.3: Import Airshare and Weight Instruments Across IndustriesIndonesian IO sectors. The IO tables are comprised 175 distinct ‘sectors’ which typicallyaggregate several ISIC 3.0 classifications. To determine the sectoral heaviness or airshare, weassign equal shares to all ISIC 3.0 classifications assigned to the same sector. As described inthe main text, we then use the sectoral import expenditures shares to construct a measureimported input weight and airshare.Figure 1 documents the variation in the fraction imports shipped by air and differencesin the weight of imported inputs across industries. It is clear that there exist substantialdifferences across industries and, not surprisingly, industries which tend to import lighterinputs are also more likely to have them shipped by air, where the correlation coefficientbetween these instruments is -0.4.B.2 Estimating MTE and Treatment EffectsWe estimate the MTE and treatment parameters following a procedure similar to that ofCarneiro et al. (2011). Because the support of P (Z) for each value ofX is small, as in Carneiroet al. (2011), we assume that (X,Z) is independent of (U1, U0, UD). Then, the MTE can beidentified within the support of P (Z) as ∆MTE(x, p) = β¯(x) + E[U1 − U0|UD = p], wherethe term β¯(x) represents the average treatment effect when X = x while E[U1 − U0|UD = p]represents the component of the MTE that depends on UD. Furthermore, because X is ahigh-dimensional vector, allowing the value of β¯ to depend on all variables in X leads toimprecise estimates of β¯(X). We set β¯(X) = X˜ ′θ, where X˜ contains the lagged dependentvariable (e.g., (Lps/(Lps + Lpu))96) while it also contain dummies for plants that did not hireany skilled or unskilled workers, djs = 1(Ljs = 0) and dju = 1(Lju = 0) in 1996 when we use121B.2. Estimating MTE and Treatment Effectsthe log of the skill ratios as the dependent variable.74 Then,E[S|X = x, P (Z) = p] = x′γ + px˜′δ +K(p), ∆MTE(x, p) = x˜′δ +K ′(p), (B.1)where K(p) = E[U1 − U0|UD ≤ p]p and K ′(p) is the first derivative of K(p). We estimate γ,δ, and K(p) by a partially linear regression of S on X and P (Z) (Robinson, 1988) with localpolynomial regressions.Specifically, we estimate γ, δ, and K(p) by a partially linear regression of S on X andP (Z) (Robinson, 1988) as follows.Step 1: We estimate P (Z) using a logit specification as described in the main text. Denote theestimated value by “hat” notation so that Pˆ (Z) denotes the estimate of P (Z).Step 2: Using the subsample of observations for which the outcome variable is measurable andfor which estimated propensity scores Pˆ (Zi)’s are on the estimated common support,we estimate E[S|P (Z)], E[X|P (Z)], and E[X˜|P (Z)] by local linear regressions of S, X,and X˜ on Pˆ (Z), respectively, where we use a normal kernel and choose their bandwidthsby “leave-one-out” cross-validation.Step 3: By regressing S − Eˆ[S|P (Z)] on X − Eˆ[X|P (Z)] and P (Z)(X˜ − Eˆ[X˜|P (Z)]) withoutan intercept, we obtain the estimate of γ and θ.Step 4: We estimate K(P (Z)) and K ′(P (Z)) by using a local quadratic regression of S−X ′γˆ−Pˆ (Z)X˜ ′θˆ on Pˆ (Z), where we use cross-validation to choose the bandwidth for the localquadratic regression.To avoid numerical singularity, all continuous variables in Z, X, and X˜ are standardizedby subtracting their means and then dividing by their sample standard deviations while alldummy variables are transformed into {−1, 1}. Table B.2 reports the bandwidth choicesusing the standardized variables for Step 2 and Step 4. We set the maximum value of thebandwidth to one-half of the length of the common support of Pˆ (Z|D = 0) and Pˆ (Z|D = 1).In column (3) of Table 2.11, we use a sieve estimator to estimate the partial linear re-gression. Specifically, we estimate E[S|P (Z)], E[X|P (Z)], and E[X˜|P (Z)] in Step 2 byregressing S, X, and X˜ on the fourth order polynomials of Pˆ (Z) while we estimate K(P (Z))and K ′(P (Z)) by regressing S−X ′γˆ− Pˆ (Z)X˜ ′θˆ on the fourth order of polynomials in Pˆ (Z).Table B.1 reports the estimates of the skill demand equation (B.1) using the sample ofplants for which the estimated propensity scores are on the estimated common support when74In our preliminary investigation, when we estimated (B.1) by setting X˜ equal to all variables in X exceptfor the local wage ratios, industry dummies, and province dummies, we found that the interaction terms withother variables in X were rarely significant across different specifications.122B.2. Estimating MTE and Treatment EffectsTable B.1: Estimates of Skill Demand EquationOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationDependent Var. (Lps/(Lpu + Lps))06 (Lns /(Lnu + Lns ))06 (Ls/(Lu + Ls))06 (Ls/(Lu + Ls))06 (Ln/(Ln + Lp))06Export -0.0298 [0.0255] -0.0155 [0.0194] -0.0306 [0.0265] -0.0294 [0.0058] -0.0322 [0.0126]Capital 0.0218 [0.0060] 0.0020 [0.0045] 0.0194 [0.0058] 0.0012 [0.0012] 0.0022 [0.0029]Hicks-neutral ϕ 0.0035 [0.0124] 0.0072 [0.0104] 0.0007 [0.0119] -0.0010 [0.0028] -0.0068 [0.0060]Foreign -0.0389 [0.0437] -0.0211 [0.0404] -0.0366 [0.0398] -0.0158 [0.0120] -0.0269 [0.0192]R&D 0.0191 [0.0258] 0.0150 [0.0219] 0.0240 [0.0233] 0.0149 [0.0071] 0.0243 [0.0134]Training 0.0370 [0.0178] 0.0208 [0.0129] 0.0312 [0.0179] 0.0059 [0.0039] 0.0064 [0.0072]log(Ws/Wu)06 -0.0107 [0.0409] -0.0112 [0.0320] 0.0077 [0.0422] -0.0008 [0.0085] 0.0095 [0.0188]log(Ws/Wu)96 -0.1482 [0.0527] -0.0385 [0.0407] -0.1511 [0.0521] 0.0040 [0.0111] 0.0232 [0.0206](Ljs/(Lpu + Ljs))06 0.0449 [0.0172] 0.1067 [0.0171] -0.0328 [0.0267] 0.0207 [0.0040] 0.2414 [0.0367](Ljs/(Lpu + Ljs))06 × P (Z) -0.5867 [0.1629] -0.5345 [0.1694] -0.4451 [0.2022] 0.0266 [0.0455] 0.1619 [0.1883]No. Obs. 3997 3992 3985 3967 4000Notes: j = n, p. The bootstrap standard errors are in square brackets. Province dummies and 3-digit ISIC industrydummies are also included.we use the share of skilled workers as the dependent variable. In the first three columns ofTable B.1, the coefficient of the interaction term between the lagged dependent variable andthe propensity score is negative and significant. One possible interpretation is that plantswith high initial skill ratios may have already adopted relatively skill-biased technology and,as a result, further adoption of foreign technology induced by importing may not substantiallyincrease their demand for skilled workers. The estimates of the other explanatory variablesare similar to those of the IV regressions in Tables 2.5-2.6.As in Heckman and Vytlacil (2005), Heckman and Vytlacil (2007a), Heckman and Vytlacil(2007b) and Carneiro et al. (2010) show, various treatment effects conditional on X can beexpressed as weighted averages of the MTE as follows:ATE(x) =∫ 10∆MTE(x, p)dp, TT (x) =∫ 10∆MTE(x, p)hTT (x, p)dp,TUT (x) =∫ 10∆MTE(x, p)hTUT (x, p)dp, PRTE(x) =∫ 10∆MTE(x, p)hPRTE(x, p)dp,MPRTE(x) =∫ 10∆MTE(x, p)hPRTE(x, p)dp,(B.2)123B.2. Estimating MTE and Treatment EffectswherehTT (x, p) =1− FP (p|X = x)E(P |X = x) , hTUT (x, p) =FP (p|X = x)E(1− P |X = x) ,hPRTE(x, p) =FP ∗(p|X = x)− FP (p|X = x)E(P |X = x)− E(P ∗|X = x) ,hMPRTE(x, p) = limα→0FP ∗α(p|X = x)− FP (p|X = x)E(P |X = x)− E(P ∗α|X = x)=(∂/∂α)FP ∗α(p|X = x)|α=0∫(∂/∂α)FP ∗α(p|X = x)|α=0dp.(B.3)FP (·|X = x) and FP ∗(·|X = x) are the cumulative distributions of P and P ∗, respectively,conditional on X = x, where P ∗ is the probability of importing under an alternative policy.Treatment effects can be computed by integrating conditional treatment effects in (B.2)using the appropriate distribution of X. Because X is high dimensional, however, it is notcomputationally feasible to estimate the conditional density function of P given X. Forthis reason, exploiting the fact that fp(P |X) = fp(P |X ′θ) implies E[log(P/(1 − P ))|X] =E[log(P/(1−P ))|X ′θ], we regress log(Pˆ /(1− Pˆ )) on X and obtain a single index of X, X ′θˆ.The conditional density function of P given X ′θ, denoted by fP (p|x′θ), is estimated by theratio of the joint density of P and X ′θˆ to the marginal density of X ′θ using ‘double-kernel’local linear regression, where we choose the bandwidth by the cross-validation following thesuggestion of Fan and Yim (2004).We compute weights hTT (x′θ, p), hTUT (x′θ, p), hPRTE(x′θ, p), and hMPRTE(x′θ, p) ashTT (x, p), hTUT (x, p), hPRTE(x, p), and hMPRTE(x′θ, p) in the formula (B.3) but usingFP (p|X ′θ = x′θ) =∫ p0 fP (u|X ′θ = x′θ)du in place of FP (p|X = x). To apply (B.2) tocompute treatment effects conditioning on the single index X ′θ, we evaluate the MTE atX ′θ = x′θ instead of X = x. To do so, we estimate E[X˜ ′δ|X ′θ] by local linear regressionand define the MTE at X ′θ = x′θ as ∆ˆMTE(x′θ, p) = Eˆ[x˜′δ|X ′θ = x′θ] + Kˆ ′(p). Integrating∆ˆMTE(x′θ, p) using weights hTT (x′θ, p), hTUT (x′θ, p), hPRTE(x′θ, p) , and hMPRTE(x′θ, p)gives our estimates of the TT (x′θ), TUT (x′θ), PRTE(x′θ), and MPRTE(x′θ). To obtainthe unconditional version of treatment effects, we integrate X ′θ from TT (X ′θ), TUT (X ′θ),PRTE(X ′θ), andMPRTE(X ′θ) using the marginal distribution ofX ′θ, denoted by fX′θ(x′θ),which is estimated by local linear regression. The last three rows of Table B.2 report thebandwidth choices associated with estimating fP (p|x′θ) and fX′θ(x′θ). Figure 2.4 showsestimated weights for ATE, TT, TUT, MPRTEs, and PRTE when dependent variable isln(Lps/Lpu).Finally, because the full support condition is violated, we report estimates of ATE, TT,TUT, PRTE, and MPRTE when we restrict the weights to integrate to one in the restrictedsupport of the MTE as described in the main text. As discussed in Heckman and Vytlacil(2005) and Carneiro et al. (2010), the PRTE cannot be identified without strong support124B.3. Estimating Hicks-Neutral ProductivityTable B.2: Bandwidth Choices by Cross-validationOccupation Production Non-ProductionThreshold Highschool CollegeDependent Var. ln(LpsLpu)06(LpsLpu+Lps)06ln(LnsLnu)06(LnsLnu+Lns)06(1) (2) (3) (4)Step 2: E[S|P ] 0.03 0.03 0.21 0.05E[Export|P ] 0.03 0.05 0.07 0.05E[Capital|P ] 0.05 0.01 0.03 0.03E[ϕ|P ] 0.03 0.05 0.05 0.05E[Foreign|P ] 0.42 0.03 0.33 0.43E[R&D|P ] 0.17 0.03 0.21 0.19E[Training|P ] 0.03 0.05 0.05 0.05E[ln(Ws/Wu)06|P ] 0.42 0.03 0.42 0.05E[ln(Ws/Wu)96|P ] 0.07 0.01 0.11 0.03E[ln(Ljs/Lju)96|P ] 0.25 0.07E[dju,96|P ] 0.42 0.01E[djs,96|P ] 0.01 0.05E[(Ljs/(Ljs + Lju)96|P ] 0.11 0.01E[industry/province|P ](a) 0.03 0.03 0.09 0.03Step 4: E[S −X′γ − P (Z)X˜′θ|P ] 0.15 0.11 0.23 0.13Bandwidth for P of fP (p|x′θ) 0.01 0.01 0.01 0.01Bandwidth for X′θ of fP (p|x′θ) 0.01 0.02 0.02 0.01Bandwidth for fX′θ(x′θ) 0.02 0.03 0.04 0.02Notes: j = p, n. Columns (1)-(4) reports the cross-validation bandwidth choices that are used to estimate thetreatment effects reported in columns (1)-(4) of Table 2.10, respectively. (a) We choose the common bandwidth forindustry/province dummies by minimizing the sum of cross-validation criterion functions over industry/provincedummies.conditions. We compute the estimate of what the PRTE would be when we restrict thesupport of P and P ∗ to the restricted support for which minimum and maximum values aregiven by the 1st and the 99th percentiles of the common support. When the value of P ∗ isabove the maximum value of the support, the maximum value of P ∗ is set to the maximumvalue of the restricted support.We use 500 bootstrap replications to construct equal-tailed bootstrap confidence bandsfor ∆ˆMTE(x′θ, p) and the standard errors for treatment effects. In each bootstrap iterationwe re-estimate P (Z) so all standard errors account for the fact that P (Z) is estimated.B.3 Estimating Hicks-Neutral ProductivityOur model implies that Hicks-neutral productivity differences are potentially among the mostimportant determinants of plant-level import decisions. Unfortunately, the data do not pro-vide a convenient measure of Hicks-neutral productivity. Moreover, standard productivityestimation methods do not consider how we might separately identify skill-biased and Hicks-125B.3. Estimating Hicks-Neutral Productivity0 0.1 0.2 0.3 0.4 0.5 0.6UD-2024681012MTEMTE(p) for log(Lsp/Lup) with 90 percent confidence band(a) Production, Highschool0 0.1 0.2 0.3 0.4 0.5 0.6UD2345678910MTEMTE(p) for log(Lsn/Lun)  with 90 percent confidence band(b) Non-Production, College0 0.1 0.2 0.3 0.4 0.5 0.6UD-20246810121416MTEMTE(p) for log(Ls/Lu) for highschool+ with 90 percent confidence band(c) All, Highschool0 0.1 0.2 0.3 0.4 0.5 0.6UD-101234567MTEMTE(p) for log(Ls/Lu) for college+ with 90 percent confidence band(d) All, College0 0.1 0.2 0.3 0.4 0.5 0.6UD-3-2-101234567MTEMTE(p) for log(Ln/Lp) with 90 percent confidence band(e) All, OccupationFigure B.4: Estimated MTE126B.3. Estimating Hicks-Neutral ProductivityTable B.3: Import Decision Model using Logit for the Sample of ProductionWorkersOutcome Variable ln(Lps/Lpu) Lps/(Lps + Lpu)Coeff. S.E. Ave. Deriv. S.E. Coeff. S.E. Ave. Deriv. S.E.TC -0.4082 [0.1605] -0.0275 [0.0107] -0.5104 [0.1693] -0.0346 [0.0115]Air 0.3666 [0.1352] 0.3142 [0.1129] 0.4482 [0.1465] 0.3862 [0.1244]Wgt -0.0091 [0.1613] -0.0007 [0.0127] 0.0682 [0.1519] 0.0055 [0.0121]TC × log(LpsLpu)96 0.0412 [0.1217] 0.0014 [0.0041]TC × dpu,96 -0.0132 [0.2417] -0.0046 [0.0816]TC × dps,96 -0.1476 [0.1350] -0.0162 [0.0146]Air × log(LpsLpu)96 0.2962 [0.1380] 0.0973 [0.0444]Air × dpu,96 0.0415 [0.1216] 0.1054 [0.3048]Air × dps,96 -0.0683 [0.1779] -0.0778 [0.1994]Wgt× log(LpsLpu)96 0.141 [0.1751] 0.0049 [0.0060]Wgt× dpu,96 0.0313 [0.1447] 0.0113 [0.0515]Wgt× dps,96 0.0526 [0.1954] 0.0046 [0.0167]TC × (Lps/(Lps + Lpu))96 0.0407 [0.1586] 0.004 [0.0156]Air × (Lps/(Lps + Lpu))96 -0.2712 [0.1469] -0.2989 [0.1605]Wgt× (Lps/(Lps + Lpu))96 -0.1838 [0.1726] -0.0196 [0.0183]Export 0.373 [0.0742] 0.0525 [0.0104] 0.3857 [0.0696] 0.0545 [0.0099]Capital 0.4058 [0.0855] 0.0122 [0.0025] 0.4306 [0.0842] 0.013 [0.0026]Hicks-neutral ϕ 0.147 [0.0756] 0.0139 [0.0071] 0.1522 [0.0754] 0.0145 [0.0072]Foreign 0.1389 [0.0479] 0.0542 [0.0181] 0.1398 [0.0496] 0.0549 [0.0192]R&D 0.0765 [0.0540] 0.0172 [0.0121] 0.0818 [0.0546] 0.0185 [0.0123]Training 0.1858 [0.0739] 0.0221 [0.0087] 0.2087 [0.0793] 0.025 [0.0095]log(WsWu)06 0.0403 [0.0866] 0.0127 [0.0271] 0.041 [0.0941] 0.013 [0.0298]log(WsWu)96 0.0282 [0.0919] 0.0117 [0.0377] 0.0109 [0.0884] 0.0045 [0.0367]log(LpsLpu)96 -0.2394 [0.2104] -0.0097 [0.0084]dpu,96 0.0276 [0.2414] 0.0115 [0.0996]dps,96 0.0479 [0.2828] 0.0057 [0.0329](Lps/(Lps + Lpu))96 0.1916 [0.2243] 0.0276 [0.0321]No. Obs. 4064 4064Notes: Estimates are from the sample which uses the log of the production skill ratio as an outcome variable.Bootstrap standard errors are in square brackets. Province dummies and 3-digit ISIC industry dummies are alsoincluded. The sample excludes plants that belong to a 3-digit ISIC industry or province within which there is novariation in import status because, in such cases, the estimated coefficient of the corresponding industry or provincedummy in the logit model would be either infinity or minus infinity.127B.3. Estimating Hicks-Neutral Productivityneutral productivity.75 Accordingly, we develop an extension of the control function methodspioneered by Olley and Pakes (1996) [OP, hereafter], Levinsohn and Petrin (2003) [LP, here-after] and Ackerberg et al. (2015), among others, to estimate a Hicks-neutral productivityseries for each plant in our data.76We assume that the firm’s production function is specified asYit = eεitQit, where Qit = eα0+ωitKαkit Mαmit Lαpp,itLαnn,it (B.4)where ωit is the part of the Hicks-neutral productivity shock that is observed/anticipatedby firm i at the time which it makes input decisions while εit captures either measurementerror or an iid unanticipated shock that is not observed at the time which it makes inputdecisions. The variables Lp,it and Ln,it represent the aggregate labor inputs for productionand non-production activities, respectively, and are defined byLj,it =((AjLsj,it)σj−1σj + (Luj,it)σj−1σj) σjσj−1for j = p, n. (B.5)Here, Lsj,it and Luj,it represent the number of skilled workers and that of unskilled workers,respectively, in occupation j, where the subscript “p” indicates production workers while thesubscript “n” captures non-production workers. We assume that ωit follows a first orderMarkov process.To estimate the production function coefficients, including the elasticity of substitutionparameters, we use the implications of plant profit maximization behavior.77 The first orderconditions with respect to Luj,it and Lsj,it are given byW ut Luj,itQit= αj(Luj,it)σj−1σj(AjLsj,it)σj−1σj + (Luj,it)σj−1σjandW st Lsj,itQit= αj(AjLsj,it)σj−1σj(AjLsj,it)σj−1σj + (Luj,it)σj−1σj,(B.6)respectively, so that (Luj,itLsj,it) 1σAσj−1σjj =W stW utfor j = p, n, (B.7)where W st and Wut represent the wages in year t for skilled and unskilled workers, respectively.We assume that there is no unanticipated ex-post shock to Aj , Wst , and Wut . Substituting75Doraszelski and Jaumandreu (2014) is a key exception.76Other important contributions to this literature include Wooldridge (2009), De Loecker (2011), De Loeckeret al. (2012) and Doraszelski and Jaumandreu (2013).77Our method is broadly based on the ideas contained in Gandhi et al. (2013), but our production functionis specified using a simple Cobb-Douglas form with CES aggregators for production and non-production laborinputs so that our analysis is substantially simpler than theirs.128B.3. Estimating Hicks-Neutral Productivity(B.7) into (B.5), we getLj,it = X− σjσj−1j,it Luj,it, where Xj,it ≡W ut Luj,itW st Lsj,it +Wut Luj,it.Substituting the above equation for Lj,it into (B.4) and taking the logarithm givesyit = α0,t + αkkit + αmmit + αplup,it + βpxp,it + αnlun,it + βnxn,it + ωit + it (B.8)where βj = − σjαjσj−1 for j = p, n, and lower case letters represent the logarithm of the uppercase letters (e.g., yit ≡ ln(Yit)). Note that, if we can consistently estimate αj and βj , thenwe also have a consistent estimate of σj because −βj/αj = σjσj−1 .We recover the estimates in two stages. In the first stage, following LP, we write ωitas a function of mit, kit: ωit = ω∗t (mit, kit). Taking an expectation of (B.8) conditional on(mit, kit), and subtracting it from (B.8) givesyit − E[yit|mit, kit] = αp{lup,it − E[lup,it|mit, kit]}+ βp{xp,it − E[xp,it|mit, kit]}+αn{lun,it − E[lun,it|mit, kit]}+ βn{xn,it − E[xn,it|mit, kit]}+ it.(B.9)where E[it|mit, kit] = 0 under the assumption that it is mean zero random variable and thatit is not observed yet when a plant makes intermediate input decision.The parameters αp, βp, αn, and βp are estimated by (i) first estimating the functionsE[yit|mit, kit], E[`up,it|mit, kit], E[`un,it|mit, kit], E[xp,it|mit, kit] and E[xn,it|mit, kit] and then(ii) running a no-intercept OLS regression of (B.9) using the estimate of the conditionalexpectation terms. Note that, even though we consider the possibility of endogenous plantexit, the first stage procedure is identical to that of LP.In the second stage we identify the remaining production function parameters αk and αm.To accomplish this, we first defineφt(mit, kit) ≡ α0,t + αkkit + αmmit + ω∗t (mit, kit)andxit ≡ yit − {αplup,it + βpxp,it + αnlun,it + βnxn,it}.Further, let χit = 1 indicate plant survival in year t. We assume that a firm stays in themarket if and only if ωit ≥ ωt(kit) as in OP. Then, we may write (B.8) asxit = α0,t + αkkit + αmmit + E[ωit|ωit−1, χit = 1] + ξit + it= αkkit + αmmit + gt(ωt(kit), ωit−1) + ξit + it (B.10)129B.3. Estimating Hicks-Neutral Productivitywhere ξit = ωit − E[ωit|ωit−1, χit = 1] and gt(ωt(kit), ωit−1) ≡ α0,t + E[ωit|ωit−1, χit = 1].The survival probability conditional on ωt−1 is given byPr{χit = 1|ωit−1, kit−1,mit−1} = Pr{ωt ≥ ωt(kit)|ωit−1,mit−1, kit−1}=∫ ∞ωt(kit(mit−1,kit−1))F (dωit|ω∗t−1(mit−1, kit−1))= Pχit . (B.11)where F (·) represents the law of motion for ωit. The capital stock follows kit = (1−δ)kit−1+ιitwhere ιit is the amount of investment between t − 1 and t, δ is the depreciation rate, andwe assume that ιit is a function of (ωit−1, kit−1) = (ω∗t (mit−1, kit−1), kit−1) so that we maywrite kit as a function of mit−1 and kit−1, i.e., kit(mit−1, kit−1) in the second line of (B.11).We estimate the survival probability (B.11) using a probit with third order polynomials in(mit−1, kit−1). Given ω∗t−1(mit−1, kit−1), we may invert (B.11) with respect to ωt; therefore,we may write ωt as a function of survival probabilities, Pχit , and ω∗t−1(mit−1, kit−1) as inωt(Pχit , ω∗t−1(mit−1, kit−1)).Then, we may express gt(ωt(kit), ωit−1) in (B.10) as a (year-specific) nonlinear functionof (Pχit , ω∗t−1(mit−1, kit−1)) asgt(ωt(Pχit , ω∗t−1(mit−1, kit−1)), ω∗t−1(mit−1, kit−1))= α0,t +∫ ∞ωt(Pχit ,ω∗t−1(mit−1,kit−1))ωitF (dωit|ω∗t−1(mit−1, kit−1))∫∞ωt(Pχit ,ω∗t−1(mit−1,kit−1))F (dωit|ω∗t−1(mit−1, kit−1)).Defineqt(Pχt , α0,t−1 + ω∗t−1(mit−1, kit−1)) ≡ gt(ωt(Pχit , ω∗t−1(mit−1, kit−1)), ω∗t−1(mit−1, kit−1)),and substituting this equation into (B.10) and using α0,t−1+ω∗t−1(mit−1, kit−1) = φt−1(mit−1, kit−1)−αkkit−1 − αmmit−1, we havexit = αkkit + αmmit + qt(Pχt , hit−1) + ξit + it, (B.12)where hit = φt(mit, kit)− αkkit − αmmit. This equation corresponds to equation (12) in OP.Given the above definitions, we recover αk and αm in three distinct steps. First, letxˆit = yit − {αˆplup,it + βˆpxp,it + αˆnlun,it + βˆnxn,it}, where (αˆp, αˆn, βˆp, βˆn) are the first stageestimates of the corresponding parameters. Then we estimate φ(mit, kit) by regressing xˆiton third order polynomials in (mit, kit). Second, we estimate the survival probability byestimating the probit for survival (χit = 1) conditional on (mit−1, kit−1) using third orderpolynomials. Third, for each candidate value of (αk, αm), we compute hˆit(αk, αm) = φˆit −130B.4. First Differences, IV and Biasαkkit−αmmit and regress xˆit−{αkkit +αmmit} on third order polynomials in (Pˆχit , hˆit−1) toobtain the estimate of qt(Pχit , hit−1) as its predicted value, denoted by qˆit(αk, αm). Denotinĝ(ξit + it)(αk, αm) = xˆit − {αkkit + αmmit − qˆit(αk, αm)}, we estimate (αk, αm) using themoment conditions E[(ξit + it)mit−1] = 0 and E[(ξit + it)kit−1] = 0. Note that we donot use kit as an instrument because kit will be correlated with ξit given that we take longdifferences.We apply the above estimation procedure to the two years of data from 1996 and 2006 sothat the time subscripts t − 1 and t correspond to 1996 and 2006, respectively. The Hicks-neutral productivity, including both the unexpected shock it and the year-specific constantα0,t, is computed asϕit ≡ α0,t + ωit + it = yit − (αˆkkit + αˆmmit + αˆplup,it + βˆpxp,it + αˆnlun,it + βˆnxn,it).We find that (αk, αm, αp, αn, βp, βn) is estimated as (0.017, 0.602, 0.152, 0.110,−0.253,−0.138).Note the production function parameters are very similar to those estimated elsewhere (e.g.See Amiti and Konings (2007)). Our estimates further imply that the elasticity of substitu-tion parameters among production and non-production workers (σp, σn) are estimated to be(1.664,1.255).As an alternative measure of productivity, we also estimate the “conventional” measureof total factor productivity (TFP) under the assumption that skilled and unskilled workersare perfect substitutes with a Cobb-Douglas production function given byYit = eεitQit, where Qit = eα0+ωitKαkit Mαmit L˜αpp,itL˜αnn,it (B.13)where L˜p,it = Lsp,it + Lup,it and L˜n,it = Lsn,it + Lun,it. Repeating our estimation exercise underthis restriction we again recover the parameters (αk, αm, αp, αn) as (0.030, 0.908, 0.065, 0.074).We also use this alternative structure and estimates to construct a second measure of pro-ductivity. In the main text this second measure is denoted as “conventional” TFP.B.4 First Differences, IV and BiasThis following derivations are an extension of Section 5.4 of Angrist and Pischke (2008).Consider a setting where β is constant parameter and that the data are generated fromYit = α+ ρYit−1 + βDit + it,where E[Ditit] 6= 0 and E[it|Zit] = 0 so that we may consistently estimate β by instrumentalvariable regression. Suppose that we mistakenly estimate a first-differenced equation using131B.5. Capital-Skill ComplementarityZit as IV using the sample of initial non-importers so that so that Dit−Dit−1 = Dit for everyobservation in the sample. The first-differenced IV estimator will converge in probability toCov(Yit−Yit−1,Zit)Cov(Dit,Zit). Because Yit − Yit−1 = α+ (ρ− 1)Yit−1 + βDit + it,Cov(Yit − Yit−1, Zit)Cov(Dit, Zit)= β − (1− ρ)Cov(Yit−1, Zit)Cov(Dit, Zit).For our transport cost instrument, Z, we can empirically confirm that Cov(Yit−1,Zit)Cov(Dit,Zit) > 0 sinceCov(Yit−1, Zit) < 0 and Cov(Dit, Zit) < 0. Given that ρ is consistently estimated to liebetween 0 and 1 in Tables 2.5 and 2.6 we expect that the β estimated in the first differencedIV regressions in Table 2.7 will be biased downwards.B.5 Capital-Skill ComplementarityWe first extend our model in Section XX to include capital-skill complementarity by con-sidering the following production function: f(K,M,Ls, Lu, A, ϕ) = ϕ(Vp)αp(V n)αnMαm ,where ϕ is the firm’s Hicks-Neutral productivity shock while V j is a CES aggregator givenby V j = [(Aj(Ljs)β(Kj)1−β)1/ρj + (Lju)1/ρj ]ρj with ρj = σj/(σj − 1) for j = {n, p}. As be-fore, Aj captures skill-biased technological change as in our benchmark model. However,in this case, it augments both skilled labor, Ljs, and capital, Kj through the composite in-put (Ljs)β(Kj)1−β. Minimizing the firm’s costs, the relative demand for skilled labor can bewritten as:LjsLju=(βWuWs)σj(Aj)σj−1(KjLjs)(σj−1)(1−β), (B.14)where we again assume that skill-biased technology is potentially a function of the firm’simport decision as written in equation (2.2).There are three issues here which merit comment. First, equation (B.14) demonstratesthat if capital-skill complementarity is an important mechanism among Indonesian manufac-turers our benchmark specification may potentially suffer from omitted variable bias. Second,the relative demand for skill equation (B.14) implies that we need to partition capital intoproduction and non-production components (Kp and Kn). While the data do not provide anatural decomposition of capital across occupation, our model implies that we can decom-pose capital using the firm’s first order conditions. Specifically, the firm’s cost minimizationproblem implies that we can write the following relationships between capital and labor ofeach type:Kp =(WsWk)(1− ββ)Lps and Kn =(WsWk)(1− ββ)Lns .132B.6. Investigating Differences with Amiti and Cameron (2012)Therefore, total capital is related to total skilled labor asK = Kn +Kp =(WsWk)(1− ββ)(Lps + Lns ), (B.15)and it follows that the fraction of total capital allocated to occupation j ∈ {n, p} can bedetermined by dividing Kn or Kp by equation (B.15) asKjK=LjsLjs + Lju.Note that this result is sensitive to the assumption that the share of skilled labor and capital,β, is equal across occupations. However, the alternative assumption that β varies acrossoccupations but capital is allocated in a fashion such that each firm has the same ratio ofproduction to non-production capital (i.e., Kn = γK and Kp = (1−γ)K for some γ ∈ (0, 1))results in a nearly identical empirical structure. We do not find any significant differenceusing this alternative assumption and, as such, we omit further discussion hereafter.Finally, it is clear that capital-skill complementarity implies adding one additional vari-able to our benchmark empirical specification, the log ratio of capital to total (productionand non-production) skilled labor, ln (K/(Lps + Lns )). As noted in the main text, when in-cluding the endogenous capital-skill control variable we also use lagged (i.e., 1996) valuesof ln (K/(Lps + Lns )) as an additional instrument along with interactions of ln (K/(Lps + Lns ))with our benchmark instruments.B.6 Investigating Differences with Amiti and Cameron(2012)Column 1 of Table B.4 is our best replication of column 2 of Table 8 in Amiti and Cameron(2012). In this exercise we regress Relative education intensityf,i,2006 as defined in Amiti andCameron (2012) on import and export dummies in 1996 and include all plants in the balancedpanel. As in the Amiti and Cameron result, we estimate a negative and significant coefficienton the initial import status. In column (2), we also include the dummies for import andexport status in 2006 and find that the import status in 2006 is positive but not significant.In column (3), we investigate the relationship between the relative education intensity anda full set of import status changes: 1. import96 = 0, import06 = 0 (baseline group); 2.import96 = 0, import06 = 1; 3. import96 = 1, import06 = 0; 4. import96 = 1, import06 = 1.We observe a positive but insignificant coefficient for firms which start importing while thecoefficient for firms which quit importing is found to be negative and significant.In columns (4), (5) and (6) we repeat each experiment, but add a control variable which133B.6. Investigating Differences with Amiti and Cameron (2012)Table B.4: Investigating Differences with Amiti and Cameron (2012)Dependent Var. ∆ Relative Education Intensity (1996-2006)(1) (2) (3) (4) (5) (6) (7)export96 -0.085*** -0.061** -0.061** 0.051*** 0.028* 0.029*(0.025) (0.028) (0.028) (0.015) (0.017) (0.017)export06 -0.066** -0.066** 0.037** 0.036**(0.028) (0.028) (0.017) (0.017)import96 -0.083*** -0.088*** 0.045*** 0.003(0.025) (0.028) (0.015) (0.017)import06 0.022 0.092***(0.029) (0.017)(1-import96)×import06 0.026 0.137***(0.041) (0.025)import96×(1-import06) -0.085** 0.035*(0.034) (0.021)import96×import06 -0.067** 0.085***(0.031) (0.019)REI96 -0.925*** -0.928*** -0.929***(0.008) (0.008) (0.008)∆export 0.006(0.023)∆import 0.058**(0.024)Constant 0.403 0.403 0.403 0.735** 0.736** 0.736** 0.403(0.475) (0.475) (0.475) (0.287) (0.286) (0.286) (0.476)Observations 7,192 7,192 7,192 7,192 7,192 7,192 7,192R-squared 0.087 0.087 0.087 0.666 0.668 0.668 0.083Industry. FE Yes Yes Yes Yes Yes Yes YesSample All All All All All All AllNotes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. Column (1) replicatescolumn (1) of Table 8 in Amiti and Cameron (2012) using import status in 1996. Column (7) considers a specificationwhere we replace import and export status with the change in import and export status.captures the firm’s relative education intensity in 1996, REI96. We observe that the coeffi-cients associated with 1996 import status are now significantly positive in columns (4) and(6) while the coefficient of starting to import is substantially larger and is much more pre-cisely estimated in column (6). These results suggest that the negative coefficient on the1996 import dummy in Amiti and Cameron’s original specification may be driven by the pos-itive correlation between 1996 import status and the firm’s initial level of relative educationintensity.In column (7) we replace import and export status with the change in import and exportstatus, which is our preferred specification. In this case, we again estimate a positive andsignificant coefficient on the change in import status. One possible interpretation of thispositive correlation between the change in relative education intensity and the change inimport status in column (7) is that starting to import induces more education-upgradingwithin production workers than within non-production workers.134B.7. Additional TablesTable B.5: A Decomposition of Plant-Level Skill Growth by Import StatusPanel A: Skilled Workers, Highschool+AllInitial Non-importersswitchers non-switchers1996 2006 1996 2006 1996 2006LevelsLs/L 0.3221 0.4667 0.4115 0.5751 0.2588 0.4013Lps/Lp 0.2749 0.4234 0.3666 0.5381 0.2117 0.3569Lns /Ln 0.6846 0.7629 0.7281 0.7885 0.6518 0.7315Ln/L 0.1646 0.1687 0.1684 0.1912 0.1495 0.1543Decomposition of the overall changes∆(Ls/L) 0.1446 0.1636 0.1425within prod. 0.1248 0.1409 0.1235within non-prod. 0.0137 0.0113 0.0128between 0.0060 0.0114 0.0062Obs. 10,537 658 7,464Panel B: Skilled Workers, College+AllInitial Non-importersswitchers non-switchers1996 2006 1996 2006 1996 2006LevelsLs/L 0.0325 0.0500 0.0458 0.0727 0.0219 0.0363Lps/Lp 0.0134 0.0209 0.0221 0.0313 0.0081 0.0141Lns /Ln 0.1376 0.1964 0.1750 0.2490 0.1096 0.1618Ln/L 0.1646 0.1687 0.1684 0.1912 0.1495 0.1543Decomposition of the overall changes∆(Ls/L) 0.0175 0.0270 0.0144within prod. 0.0058 0.0067 0.0048within non-prod. 0.0106 0.0147 0.0085between 0.0011 0.0056 0.0011Obs. 10,537 658 7,464a. Source: Indonesia Manufacturing Survey in 1996 and 2006.b. Skilled workers are defined as workers with education no less than highschoolin top panel and workers with no less than college in the bottom panel. Plantswith no production workers in 1996 or 2006 are excluded (only three observa-tions). Plants with no non-production worker in either period are treated ashaving zero within-non-production changes, and the mean value of skill sharein non-production sector (Lns /Ln) is computed using the period when thenumber of non-production workers is positive. Plants with no non-productionworkers in both 1990 and 2006 simply have a zero within non-productioncomponent and zero between component.135B.7. Additional TablesTable B.6: Robustness Checks: Dropping Capital, R&D, and TrainingOccupation Production Non-ProductionThreshold Highschool CollegeDependent Variable ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)OLS IV OLS IV OLS IV OLS IV(1) (2) (3) (4) (5) (6) (7) (8)Import Status 0.597*** 3.680*** 0.077*** 0.967*** 0.247*** 3.401*** 0.041*** 0.681***[0.108] [1.258] [0.018] [0.273] [0.092] [1.055] [0.015] [0.228]Export Status 0.111 -0.181 0.054*** -0.034 -0.174** -0.546*** 0.046*** -0.022[0.071] [0.147] [0.012] [0.032] [0.074] [0.158] [0.012] [0.028]Hicks-neutral, ϕ -0.141*** -0.254*** 0.017** -0.009 0.016 -0.117* 0.048*** 0.026**[0.046] [0.070] [0.008] [0.012] [0.048] [0.070] [0.008] [0.012]Foreign-Owned 0.111 -0.207 0.016 -0.096* 0.121 -0.297 0.022 -0.060[0.153] [0.250] [0.030] [0.057] [0.145] [0.254] [0.029] [0.048]Wagej06 -0.102 -0.177 -0.008 -0.016 -0.059 -0.189 -0.035** -0.028[0.167] [0.193] [0.023] [0.029] [0.120] [0.145] [0.016] [0.019]Wagej96 -0.556*** -0.657*** -0.116*** -0.137*** 0.332** 0.203 0.024 0.011[0.190] [0.224] [0.030] [0.040] [0.133] [0.173] [0.018] [0.022]ln(Ljs/Lju)96 0.396*** 0.353*** 0.280*** 0.226***[0.023] [0.031] [0.031] [0.041]dju 0.269 0.030 -0.022 -0.023[0.211] [0.266] [0.124] [0.145]djs -1.175*** -1.049*** -0.443*** -0.264**[0.073] [0.098] [0.074] [0.104](LjsLjs+Lju)960.497*** 0.413*** 0.214*** 0.172***[0.019] [0.036] [0.026] [0.032]Constant 1.050*** 3.677*** 0.090* 0.750*** -1.137*** -0.484 -0.099* 0.001[0.338] [0.783] [0.051] [0.145] [0.277] [0.672] [0.051] [0.107]Industry FE Yes Yes Yes Yes Yes Yes Yes YesRegion FE YES YES YES YES YES YES YES YESObservations 3,139 3,111 4,445 4,410 2,108 2,089 4,021 3,988R-squared 0.324 0.114 0.376 0.168 0.126Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initialnon-importers is used in all regressions. The education threshold used to determine a skilled production worker is ahighschool diploma, while the threshold used for a skilled non-production worker is a college degree. Import status istreated as an endogenous variable in columns (2), (4), (6) and (8). It is instrumented with both the distance to portand the share of imports shipped by air.136B.7. Additional TablesB.7 Additional TablesTable B.7: Robustness Checks: Skill Threshold DefinitionsOccupation Production Non-ProductionThreshold College HighschoolDependent Variable ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)OLS IV OLS IV OLS IV OLS IV(1) (2) (3) (4) (5) (6) (7) (8)Import Status 0.262** 4.471* 0.011** 0.068** 0.213* 2.434** -0.028 0.582**[0.133] [2.588] [0.005] [0.035] [0.127] [1.019] [0.018] [0.253]Export Status -0.364*** -0.668*** -0.008*** -0.012*** 0.051 -0.123 0.019 -0.029[0.098] [0.229] [0.002] [0.004] [0.095] [0.135] [0.013] [0.026]Wagej06 -0.210 -0.757** -0.001 -0.001 -0.044 -0.069 0.079** 0.073*[0.157] [0.367] [0.002] [0.002] [0.210] [0.222] [0.036] [0.039]Capital -0.038 -0.150* 0.003*** 0.002*** 0.116*** 0.075** 0.015*** 0.006[0.027] [0.078] [0.001] [0.001] [0.021] [0.032] [0.003] [0.005]Hicks-neutral, ϕ -0.073 -0.173 0.003** 0.002 -0.161*** -0.221*** 0.010 -0.000[0.060] [0.106] [0.001] [0.001] [0.055] [0.062] [0.009] [0.011]Foreign-Owned 0.121 -0.815 0.003 -0.004 0.236 -0.120 0.006 -0.075[0.195] [0.639] [0.007] [0.008] [0.197] [0.259] [0.028] [0.049]R&D 0.232** 0.098 0.012*** 0.010** 0.253** 0.064 0.004 -0.032[0.109] [0.199] [0.004] [0.004] [0.126] [0.163] [0.018] [0.025]Training -0.168** -0.255* 0.009*** 0.007*** 0.133* 0.132 0.008 -0.009[0.084] [0.139] [0.002] [0.002] [0.077] [0.086] [0.012] [0.015]Wagej96 0.361** 0.270 0.006** 0.005 -0.310 -0.335 -0.091** -0.094**[0.171] [0.273] [0.003] [0.003] [0.240] [0.253] [0.038] [0.043]ln(Ljs/Lju)96 0.244*** 0.124 0.319*** 0.286***[0.047] [0.105] [0.036] [0.039]dju 0.239*** 0.218***[0.064] [0.069]djs -0.906*** -0.313 -0.384*** -0.392***[0.184] [0.478] [0.086] [0.089](LjsLjs+Lju)960.096*** 0.077** 0.210*** 0.211***[0.032] [0.034] [0.017] [0.018]Industry FE Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes YesR2 0.314 — 0.111 — 0.255 — 0.167 —Hansen J p-value — 0.252 — 0.052 — 0.337 — 0.366No. Obs 959 947 4,445 4,410 1,631 1,619 4,021 3,988Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initialnon-importers is used in all regressions. The education threshold used to determine a skilled production worker is acollege degree, while the threshold used for a skilled non-production worker is a highschool diploma. Import status istreated as an endogenous variable in columns (2), (4), (6) and (8). It is instrumented with both the distance to portand the share of imports shipped by air. The variable dpu is dropped from regressions (1) and (2) due to collinearity (Ittakes the same value in 99.999 percent of all observations using the college threshold as a definition of skill).137B.7. Additional TablesTable B.8: First Stage Results: Import StatusOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationOLS OLS OLS OLS OLS OLS OLS OLS OLS OLS(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Distance to Port -0.031*** -0.031*** -0.029*** -0.030*** -0.027*** -0.031*** -0.035*** -0.030*** -0.032*** -0.032***[0.008] [0.008] [0.008] [0.008] [0.009] [0.008] [0.013] [0.008] [0.008] [0.008]Import Airshare 0.399*** 0.404*** 0.388*** 0.440*** 0.401** 0.399*** 0.677*** 0.393*** 0.447*** 0.411***[0.143] [0.143] [0.144] [0.153] [0.159] [0.143] [0.232] [0.143] [0.153] [0.144]Export Status 0.084*** 0.082*** 0.084*** 0.082*** 0.085*** 0.083*** 0.103*** 0.084*** 0.084*** 0.083***[0.014] [0.014] [0.014] [0.014] [0.014] [0.014] [0.019] [0.014] [0.014] [0.014]Wagej06 0.019 0.019 0.030 0.018 0.019 0.021[0.019] [0.019] [0.022] [0.019] [0.021] [0.019]Capital 0.014*** 0.013*** 0.014*** 0.015*** 0.016*** 0.013*** 0.021*** 0.014*** 0.016*** 0.015***[0.003] [0.003] [0.003] [0.003] [0.003] [0.003] [0.004] [0.003] [0.003] [0.003]Hicks-neutral, ϕ 0.011 0.010 0.012 0.012 0.011 0.010 0.015 0.011 0.013 0.011[0.008] [0.008] [0.008] [0.008] [0.008] [0.008] [0.011] [0.008] [0.008] [0.008]Foreign-Owned 0.122*** 0.124*** 0.121*** 0.126*** 0.123*** 0.125*** 0.153*** 0.122*** 0.130*** 0.126***[0.041] [0.041] [0.041] [0.042] [0.042] [0.041] [0.049] [0.041] [0.042] [0.042]R&D 0.046** 0.045** 0.047** 0.048** 0.048** 0.046** 0.042 0.045** 0.050** 0.049**[0.022] [0.022] [0.022] [0.023] [0.023] [0.022] [0.027] [0.023] [0.023] [0.022]Training 0.028*** 0.026*** 0.030*** 0.029*** 0.025** 0.027*** 0.038** 0.029*** 0.033*** 0.031***[0.009] [0.009] [0.010] [0.010] [0.010] [0.009] [0.015] [0.009] [0.010] [0.009]Wagej96 0.036 0.038 0.002 -0.006 0.035 0.038 0.015 0.001 0.036 0.036[0.028] [0.028] [0.019] [0.021] [0.032] [0.028] [0.036] [0.019] [0.031] [0.028]ln(Ljs/Lju)96 0.008** 0.013** 0.008** 0.015**[0.004] [0.006] [0.004] [0.007]dju 0.075* -0.010 -0.002 -0.033[0.040] [0.009] [0.008] [0.028]djs -0.008 -0.025** 0.011 0.007[0.009] [0.012] [0.009] [0.018](LjsLjs+Lju)960.054*** 0.031 0.047** 0.191*[0.019] [0.024] [0.019] [0.099]ln(Ln/Lp)96 -0.005[0.004]dp 0.012[0.008]dn 0.009[0.009](LnLn+Lp)96-0.030[0.023]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesF -stat Exc. IVs. 11.75 11.83 10.73 10.38 8.04 11.58 8.23 11.16 11.35 12.62No. Obs 4,410 4,410 4,410 3,988 3,756 4,410 2,004 4,410 3,988 4,410Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used inall regressions.138B.7.AdditionalTablesTable B.9: First Stage Results: Export StatusOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationOLS OLS OLS OLS OLS OLS OLS OLS OLS OLS(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Distance to Port 0.001 0.001 0.002 -0.005 -0.007 0.002 -0.017 0.000 -0.010 -0.001[0.014] [0.014] [0.013] [0.014] [0.015] [0.014] [0.023] [0.013] [0.014] [0.014]Import Airshare -0.770*** -0.767*** -0.777*** -0.745*** -0.740*** -0.774*** -0.603* -0.740*** -0.712*** -0.735***[0.203] [0.202] [0.202] [0.216] [0.225] [0.203] [0.330] [0.205] [0.212] [0.205]∆ Output Tariff -0.005*** -0.005*** -0.005*** -0.005*** -0.006*** -0.005*** -0.009*** -0.005*** -0.005*** -0.005***[0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.002] [0.001] [0.001] [0.001]∆ Market Access -0.005 -0.005 -0.006* -0.005 -0.005 -0.005 -0.008 -0.005 -0.005 -0.004[0.003] [0.003] [0.003] [0.004] [0.004] [0.003] [0.007] [0.003] [0.004] [0.003]Control Vars. Yes Yes Yes Yes Yes Yes Yes Yes Yes YesIndustry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesF -stat Exc. IVs. 16.89 17.57 17.66 16.78 16.83 17.86 10.45 17.51 14.85 16.98No. Obs 3,498 3,498 3,498 3,208 3,048 3,498 1,612 3,498 3,208 3,498Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in all regressions.139B.7.AdditionalTablesTable B.10: First Stage Results: Import Status, Large Instrument SetOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationOLS OLS OLS OLS OLS OLS OLS OLS OLS OLS(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Distance to Port -0.031*** -0.032*** -0.031*** -0.030*** -0.027*** -0.031*** -0.036*** -0.030*** -0.033*** -0.033***[0.008] [0.008] [0.008] [0.008] [0.009] [0.008] [0.013] [0.008] [0.008] [0.008]Import Airshare 0.392*** 0.393*** 0.404*** 0.438*** 0.401** 0.388** 0.684*** 0.380** 0.454*** 0.403***[0.150] [0.150] [0.151] [0.161] [0.167] [0.150] [0.239] [0.151] [0.161] [0.151]Import Weight -0.005 -0.005 -0.005 -0.007 -0.007 -0.004 -0.010 -0.004 -0.008 -0.004[0.008] [0.008] [0.008] [0.009] [0.009] [0.008] [0.016] [0.008] [0.009] [0.008]∆ Import Tariff 0.000 0.001 0.000 0.000 0.000 0.001 0.001 0.001 0.000 0.000[0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.001]Control Vars. Yes Yes Yes Yes Yes Yes Yes Yes Yes YesIndustry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesF -stat Exc. IVs. 6.42 6.63 6.35 5.75 4.43 6.45 4.28 6.29 6.00 6.80No. Obs 4,408 4,408 4,408 3,986 3,754 4,408 2,002 4,408 3,986 4,408Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in all regressions.140B.7. Additional TablesTable B.11: Robustness Check: Skill Supply ControlOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationDependent Variable ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)ln(Ls/Lu)(LsLs+Lu)ln(Ls/Lu)(LsLs+Lu)ln(Ln/Lp)(LnLn+Lp)IV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Import Status 2.448* 0.778*** 3.783*** 0.513** 3.388** 0.660*** 3.041** 0.226*** 0.786 0.024[1.364] [0.274] [1.344] [0.231] [1.684] [0.238] [1.268] [0.070] [0.812] [0.120]Skill Supply06 0.244*** 0.050*** 0.131** 0.017** 0.185*** 0.053*** 0.029 0.004* 0.035 0.003[0.070] [0.012] [0.061] [0.008] [0.065] [0.011] [0.059] [0.002] [0.039] [0.006]Export Status -0.157 -0.035 -0.470*** -0.017 -0.202 -0.028 -0.509*** -0.025*** -0.165* -0.016[0.128] [0.027] [0.168] [0.023] [0.159] [0.024] [0.157] [0.007] [0.087] [0.012]Wagej06 -0.452 -0.028 -0.282 -0.038 -0.368 -0.004 -0.198 -0.011* 0.024 0.000[0.333] [0.034] [0.231] [0.024] [0.313] [0.031] [0.187] [0.006] [0.120] [0.015]Capital 0.095*** 0.014*** -0.071** 0.007 0.072** 0.013*** -0.033 0.002* 0.013 0.004*[0.029] [0.005] [0.036] [0.005] [0.034] [0.004] [0.037] [0.001] [0.017] [0.002]Hicks-neutral, ϕ -0.314*** -0.020** -0.058 0.022** -0.254*** -0.022** -0.047 0.001 -0.125*** -0.006[0.056] [0.010] [0.064] [0.010] [0.056] [0.009] [0.058] [0.003] [0.034] [0.005]Foreign-Owned -0.035 -0.066 -0.291 -0.039 -0.145 -0.046 -0.366 -0.019 -0.160 -0.014[0.222] [0.052] [0.295] [0.044] [0.247] [0.046] [0.296] [0.015] [0.146] [0.021]R&D -0.084 -0.001 -0.128 -0.001 0.040 0.005 0.030 0.013 0.144 0.026**[0.139] [0.027] [0.164] [0.022] [0.152] [0.024] [0.121] [0.008] [0.088] [0.013]Training 0.176** 0.034** -0.077 0.032*** 0.234*** 0.033*** 0.011 0.011*** 0.049 0.013*[0.075] [0.014] [0.087] [0.012] [0.071] [0.013] [0.089] [0.004] [0.048] [0.007]Skill Supply96 -0.006 -0.008 -0.075 0.009 0.019 -0.006 0.038 0.003 0.052 0.006[0.083] [0.015] [0.073] [0.008] [0.080] [0.013] [0.068] [0.002] [0.045] [0.006]ln(Ls/Lu)96 0.338*** 0.123*** 0.407*** 0.316***[0.027] [0.034] [0.031] [0.044]du 0.070 0.125* -0.081 -0.232[0.237] [0.070] [0.055] [0.182]ds -0.960*** 0.107 0.022 0.097[0.101] [0.136] [0.070] [0.078](LsLs+Lu)960.385*** 0.160*** 0.432*** 0.229***[0.029] [0.028] [0.025] [0.047]ln(Ln/Lp)96 0.391***[0.020]dp -0.116**[0.036]dn 0.082**[0.033](LnLn+Lp)960.363***[0.022]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesHansen J p-value 0.289 0.242 0.551 0.425 0.419 0.529 0.913 0.459 0.746 0.731No. Obs 3,109 4,408 2,087 3,986 3,403 4,408 1,639 4,408 3,986 4,408Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in allregressions. Import status is treated as an endogenous variable in all columns. It is instrumented with both the distance to port and theshare of imports shipped by air.141B.7. Additional TablesTable B.12: Robustness Check: Large Instrument SetOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationDependent Variable ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)ln(Ls/Lu)(LsLs+Lu)ln(Ls/Lu)(LsLs+Lu)ln(Ln/Lp)(LnLn+Lp)IV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Import Status 3.240*** 0.864*** 3.498*** 0.672*** 4.105*** 0.818*** 3.191*** 0.267*** 1.557** 0.116[1.247] [0.254] [1.049] [0.230] [1.496] [0.235] [1.072] [0.069] [0.746] [0.108]Export Status -0.236* -0.044* -0.454*** -0.032 -0.265* -0.044* -0.532*** -0.029*** -0.235*** -0.025**[0.122] [0.026] [0.137] [0.023] [0.146] [0.024] [0.136] [0.007] [0.083] [0.011]Wagej06 -0.149 -0.014 -0.200 -0.026 -0.149 0.004 -0.175 -0.008 0.046 -0.001[0.184] [0.027] [0.145] [0.019] [0.166] [0.026] [0.141] [0.005] [0.106] [0.014]Capital 0.080*** 0.013*** -0.067** 0.005 0.056* 0.011** -0.039 0.002 0.001 0.003[0.028] [0.005] [0.030] [0.005] [0.031] [0.004] [0.033] [0.001] [0.016] [0.002]Hicks-neutral, ϕ -0.306*** -0.019* -0.042 0.020** -0.252*** -0.022** -0.048 0.000 -0.131*** -0.007[0.059] [0.010] [0.060] [0.010] [0.059] [0.010] [0.057] [0.003] [0.036] [0.005]Foreign-Owned -0.194 -0.089* -0.251 -0.063 -0.280 -0.077 -0.393 -0.025 -0.268* -0.026[0.234] [0.052] [0.255] [0.046] [0.254] [0.049] [0.273] [0.016] [0.149] [0.020]R&D -0.140 -0.008 -0.104 -0.009 -0.014 -0.006 0.026 0.011 0.097 0.021[0.142] [0.028] [0.149] [0.024] [0.154] [0.026] [0.121] [0.009] [0.091] [0.013]Training 0.140* 0.029** -0.070 0.026** 0.212*** 0.026* 0.001 0.010*** 0.016 0.009[0.076] [0.014] [0.080] [0.013] [0.072] [0.013] [0.084] [0.004] [0.047] [0.006]Wagej96 -0.642*** -0.134*** 0.186 0.013 -0.751*** -0.131*** 0.226 0.012* 0.029 0.025[0.217] [0.038] [0.179] [0.023] [0.209] [0.035] [0.173] [0.006] [0.126] [0.016]ln(Ls/Lu)96 0.338*** 0.127*** 0.412*** 0.314***[0.028] [0.032] [0.032] [0.043]du 0.070 0.136** -0.063 -0.234[0.251] [0.068] [0.058] [0.183]ds -0.954*** 0.078 0.026 0.105[0.083] [0.132] [0.072] [0.080](LsLs+Lu)960.389*** 0.164*** 0.441*** 0.227***[0.029] [0.030] [0.027] [0.049]ln(Ln/Lp)96 0.397***[0.020]dp -0.132***[0.038]dn 0.080**[0.035](LnLn+Lp)960.368***[0.022]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesHansen J p-value 0.001 0.001 0.134 0.268 0.067 0.001 0.231 0.081 0.558 0.328No. Obs 3,109 4,408 2,087 3,986 3,403 4,408 1,639 4,408 3,986 4,408Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in allregressions. Import status is treated as an endogenous variable in all columns. It is instrumented with both the distance to port and the shareof imports shipped by air.142B.7. Additional TablesTable B.13: Robustness Check: Import IntensityOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationDependent Variable ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)ln(Ls/Lu)(LsLs+Lu)ln(Ls/Lu)(LsLs+Lu)ln(Ln/Lp)(LnLn+Lp)IV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Import Status 7.309* 1.640** 5.498 1.420** 8.128* 1.458** 3.825 0.445** 2.945 0.148[3.974] [0.658] [3.640] [0.675] [4.632] [0.600] [2.554] [0.179] [1.936] [0.238]Import Share -8.143 -1.479 -2.563 -1.445 -7.487 -1.200 -0.715 -0.354 -3.417 -0.116[8.230] [1.365] [6.897] [1.323] [8.824] [1.233] [3.979] [0.375] [3.768] [0.475]Export Status -0.288* -0.066** -0.566*** -0.053* -0.387* -0.062** -0.569*** -0.033*** -0.216** -0.021*[0.151] [0.032] [0.161] [0.030] [0.207] [0.030] [0.160] [0.009] [0.094] [0.012]Wagej06 -0.452 -0.028 -0.282 -0.038 -0.368 -0.004 -0.198 -0.011* 0.024 0.000[0.333] [0.034] [0.231] [0.024] [0.313] [0.031] [0.187] [0.006] [0.120] [0.015]Capital 0.064* 0.010* -0.089** 0.001 0.035 0.008 -0.048 0.001 -0.002 0.003[0.036] [0.006] [0.040] [0.006] [0.041] [0.005] [0.037] [0.002] [0.019] [0.002]Hicks-neutral, ϕ -0.392*** -0.032** -0.090 0.015 -0.352*** -0.035*** -0.041 -0.001 -0.146*** -0.008[0.090] [0.015] [0.085] [0.014] [0.091] [0.014] [0.081] [0.004] [0.046] [0.006]Foreign-Owned -0.451 -0.123* -0.403 -0.100 -0.582 -0.107 -0.538 -0.034* -0.300 -0.026[0.433] [0.074] [0.345] [0.065] [0.428] [0.067] [0.373] [0.020] [0.190] [0.023]R&D -0.198 -0.020 -0.198 -0.020 -0.115 -0.017 -0.005 0.009 0.110 0.023*[0.194] [0.037] [0.195] [0.033] [0.219] [0.033] [0.135] [0.011] [0.103] [0.013]Training 0.104 0.026 -0.098 0.022 0.184* 0.022 -0.025 0.008* 0.014 0.009[0.101] [0.018] [0.103] [0.017] [0.097] [0.016] [0.119] [0.005] [0.053] [0.007]Wagej96 -0.757*** -0.180*** 0.114 -0.001 -0.919*** -0.169*** 0.203 0.008 -0.055 0.023[0.275] [0.053] [0.253] [0.033] [0.283] [0.049] [0.195] [0.008] [0.160] [0.019]ln(Ls/Lu)96 0.321*** 0.116*** 0.391*** 0.298***[0.039] [0.037] [0.045] [0.061]du 0.148 0.151* -0.063 -0.169[0.340] [0.082] [0.071] [0.201]ds -0.966*** 0.044 -0.057 0.103[0.101] [0.154] [0.104] [0.087](LsLs+Lu)960.381*** 0.142*** 0.439*** 0.203***[0.037] [0.042] [0.033] [0.060]ln(Ln/Lp)96 0.398***[0.022]dp -0.141***[0.042]dn 0.079**[0.039](LnLn+Lp)960.370***[0.022]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesNo. Obs 3,024 4,287 2,038 3,876 3,308 4,287 1,604 4,287 3,876 4,287Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in allregressions. Import status is treated as an endogenous variable in all columns. It is instrumented with both the distance to port and the shareof imports shipped by air.143B.7. Additional TablesTable B.14: Robustness Check: TFP MeasurementOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationDependent Variable ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)ln(Ls/Lu)(LsLs+Lu)ln(Ls/Lu)(LsLs+Lu)ln(Ln/Lp)(LnLn+Lp)IV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Import Status 3.151*** 0.947*** 3.885*** 0.724*** 3.968*** 0.885*** 3.264*** 0.275*** 1.226* 0.089[1.211] [0.270] [1.133] [0.238] [1.451] [0.248] [1.072] [0.071] [0.729] [0.105]Export Status -0.255** -0.051* -0.495*** -0.033 -0.276* -0.051** -0.537*** -0.029*** -0.223*** -0.023**[0.121] [0.027] [0.147] [0.024] [0.143] [0.025] [0.137] [0.007] [0.082] [0.011]Wagej06 -0.108 -0.014 -0.209 -0.027 -0.119 0.005 -0.171 -0.008 0.056 -0.001[0.183] [0.028] [0.150] [0.019] [0.164] [0.026] [0.141] [0.005] [0.104] [0.014]Capital 0.055** 0.011** -0.078** 0.006 0.041 0.009* -0.044 0.002 -0.004 0.003[0.028] [0.005] [0.034] [0.005] [0.031] [0.005] [0.034] [0.001] [0.016] [0.002]Solow Residual 0.001 0.011 0.023 0.014 0.022 0.008 0.031 0.003 -0.032 -0.000[0.044] [0.009] [0.050] [0.008] [0.046] [0.008] [0.048] [0.002] [0.027] [0.004]Foreign-Owned -0.240 -0.104* -0.315 -0.068 -0.317 -0.090* -0.413 -0.027* -0.244* -0.024[0.233] [0.056] [0.273] [0.048] [0.250] [0.052] [0.275] [0.016] [0.144] [0.020]R&D -0.180 -0.016 -0.141 -0.009 -0.051 -0.014 0.011 0.010 0.097 0.021[0.142] [0.030] [0.159] [0.025] [0.152] [0.027] [0.123] [0.009] [0.088] [0.013]Training 0.114 0.025* -0.086 0.026** 0.192*** 0.022 -0.008 0.010** 0.016 0.009[0.076] [0.015] [0.084] [0.013] [0.071] [0.014] [0.085] [0.004] [0.047] [0.006]Wagej96 -0.644*** -0.137*** 0.168 0.011 -0.743*** -0.133*** 0.220 0.011* 0.045 0.026*[0.216] [0.040] [0.188] [0.023] [0.207] [0.037] [0.175] [0.006] [0.123] [0.016]ln(Ls/Lu)96 0.340*** 0.126*** 0.412*** 0.314***[0.027] [0.033] [0.031] [0.043]du 0.076 0.134* -0.053 -0.233[0.249] [0.071] [0.057] [0.186]ds -0.944*** 0.081 0.031 0.104[0.083] [0.137] [0.071] [0.081](LsLs+Lu)960.384*** 0.163*** 0.435*** 0.225***[0.031] [0.031] [0.028] [0.050]ln(Ln/Lp)96 0.395***[0.020]dp -0.123***[0.037]dn 0.088**[0.034](LnLn+Lp)960.367***[0.022]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesHansen J p-value 0.087 0.107 0.519 0.184 0.157 0.154 0.777 0.229 0.327 0.834No. Obs 3,112 4,411 2,090 3,989 3,406 4,411 1,642 4,411 3,989 4,411Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in allregressions. Import status is treated as an endogenous variable in all columns. It is instrumented with both the distance to port and the shareof imports shipped by air.144B.7. Additional TablesTable B.15: Robustness Check: Instrumenting Export StatusOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationDependent Variable ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)ln(Ls/Lu)(LsLs+Lu)ln(Ls/Lu)(LsLs+Lu)ln(Ln/Lp)(LnLn+Lp)IV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Import Status 2.977** 0.909*** 4.240** 0.888*** 4.446** 0.947*** 0.933 0.191** 0.739 0.014[1.289] [0.305] [1.701] [0.330] [1.840] [0.300] [0.893] [0.075] [0.825] [0.121]Export Status -0.551 0.073 0.232 0.136 0.142 0.088 -1.419*** -0.019 -0.719 -0.093[0.558] [0.124] [0.859] [0.116] [0.620] [0.120] [0.492] [0.029] [0.470] [0.070]Wagej06 -0.168 -0.011 -0.224 -0.023 -0.275 0.003 -0.009 -0.007 -0.227* -0.037**[0.206] [0.033] [0.172] [0.025] [0.193] [0.033] [0.125] [0.005] [0.112] [0.016]Capital 0.098** 0.009 -0.115* -0.005 0.037 0.004 0.040 0.003 0.035 0.007*[0.041] [0.008] [0.065] [0.009] [0.049] [0.008] [0.032] [0.002] [0.029] [0.004]Hicks-neutral, ϕ -0.326*** -0.024* -0.100 0.011 -0.284*** -0.028** 0.010 0.002 -0.088** -0.001[0.075] [0.014] [0.100] [0.015] [0.081] [0.014] [0.062] [0.003] [0.044] [0.007]Foreign-Owned -0.240 -0.141* -0.451 -0.115 -0.470 -0.143* 0.263 -0.016 -0.110 -0.010[0.305] [0.078] [0.470] [0.079] [0.372] [0.077] [0.262] [0.020] [0.204] [0.030]R&D 0.016 -0.021 -0.226 -0.038 -0.047 -0.034 0.321*** 0.016 0.198 0.038**[0.174] [0.040] [0.272] [0.040] [0.214] [0.040] [0.116] [0.011] [0.122] [0.018]Training 0.133 0.013 -0.200 0.002 0.117 0.006 0.200** 0.012* 0.110 0.021*[0.119] [0.025] [0.166] [0.025] [0.130] [0.025] [0.093] [0.006] [0.083] [0.012]Wagej96 -0.374 -0.123*** 0.114 0.020 -0.642*** -0.118*** 0.314** 0.016*** 0.139 0.036*[0.244] [0.046] [0.233] [0.029] [0.241] [0.045] [0.146] [0.006] [0.141] [0.019]ln(Ls/Lu)96 0.336*** 0.090** 0.397*** 0.297***[0.030] [0.042] [0.036] [0.045]du 0.193 0.106 -0.102 -0.237[0.281] [0.086] [0.066] [0.161]ds -0.903*** 0.194 0.087 0.066[0.099] [0.163] [0.093] [0.078](LsLs+Lu)960.371*** 0.087** 0.416*** 0.262***[0.036] [0.044] [0.034] [0.053]ln(Ln/Lp)96 0.378***[0.025]dp -0.112**[0.045]dn 0.034[0.040](LnLn+Lp)960.365***[0.026]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesHansen J p-value 0.151 0.343 0.569 0.704 0.510 0.609 0.225 0.084 0.073 0.279No. Obs 2,529 3,498 1,703 3,208 2,756 3,498 1,325 3,498 3,208 3,498Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in allregressions. Import status is treated as an endogenous variable in all columns. It is instrumented with both the distance to port and theshare of imports shipped by air.145B.7. Additional TablesTable B.16: Robustness Check: StandardsOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationDependent Variable ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)ln(Ls/Lu)(LsLs+Lu)ln(Ls/Lu)(LsLs+Lu)ln(Ln/Lp)(LnLn+Lp)IV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Standards 5.354* 1.365** 1.931* 0.828** 5.976* 1.220** 0.502 0.222** 1.053 -0.078[3.000] [0.618] [1.148] [0.393] [3.083] [0.559] [0.816] [0.093] [1.230] [0.178]Wagej06 0.319 0.105 -0.262 -0.028 0.481 0.118 -0.526*** -0.005 0.208 -0.006[0.446] [0.084] [0.186] [0.035] [0.439] [0.076] [0.157] [0.009] [0.173] [0.023]Capital -0.051 -0.011 -0.060 -0.007 -0.075 -0.011 0.028 -0.000 0.003 0.008[0.108] [0.018] [0.045] [0.013] [0.104] [0.016] [0.031] [0.003] [0.037] [0.006]Hicks-neutral, ϕ -0.414*** -0.050** -0.109 -0.003 -0.317*** -0.053** -0.040 -0.005 -0.139*** -0.003[0.125] [0.026] [0.085] [0.019] [0.118] [0.023] [0.070] [0.005] [0.052] [0.008]Foreign-Owned -0.337 -0.091 0.350 -0.069 -0.490 -0.067 0.137 -0.022 -0.159 -0.012[0.672] [0.132] [0.392] [0.085] [0.729] [0.115] [0.217] [0.022] [0.205] [0.030]R&D -0.752 -0.173 -0.268 -0.095 -0.502 -0.143 0.121 0.003 0.058 0.048[0.546] [0.113] [0.278] [0.076] [0.490] [0.103] [0.133] [0.020] [0.219] [0.033]Training -0.176 -0.076 -0.128 -0.042 -0.271 -0.066 0.150 -0.004 -0.039 0.020[0.313] [0.071] [0.160] [0.051] [0.368] [0.065] [0.110] [0.011] [0.151] [0.021]Wagej96 -0.426 -0.083 0.441** 0.043 -0.432 -0.082 0.802*** 0.024*** 0.151 0.046**[0.359] [0.067] [0.217] [0.036] [0.356] [0.060] [0.198] [0.008] [0.140] [0.018]ln(Ls/Lu)96 0.307*** 0.138*** 0.416*** 0.392***[0.058] [0.040] [0.054] [0.034]du 0.344 0.247*** 0.187 -0.323[0.423] [0.091] [0.164] [0.225]ds -0.751*** 0.151 0.292** 0.166[0.178] [0.137] [0.146] [0.121](LsLs+Lu)960.309*** 0.160*** 0.385*** 0.249***[0.081] [0.042] [0.070] [0.051]ln(Ln/Lp)96 0.412***[0.023]dp -0.077[0.049]dn 0.111**[0.052](LnLn+Lp)960.389***[0.027]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesHansen J p-value 0.707 0.643 0.541 0.640 0.993 0.550 0.653 0.662 0.496 0.322No. Obs 2,186 3,329 1,318 2,958 2,435 3,329 924 3,329 2,958 3,329Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in allregressions. Import status is treated as an endogenous variable in all columns. It is instrumented with both the distance to port and theshare of imports shipped by air.146B.7. Additional TablesTable B.17: Robustness Check: Capital-Skill ComplementarityOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationDependent Variable ln(Lps/Lpu)(LpsLps+Lpu)ln(Lns /Lnu)(LnsLns+Lnu)ln(Ls/Lu)(LsLs+Lu)ln(Ls/Lu)(LsLs+Lu)ln(Ln/Lp)(LnLn+Lp)IV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Import Status 5.716*** 1.476*** 3.639* 0.525 7.500*** 1.460*** 3.113** 0.273** 2.609** 0.184[2.130] [0.464] [2.060] [0.378] [2.901] [0.465] [1.521] [0.137] [1.159] [0.126]Capital-Skill Comp. 0.388 0.081 -0.021 -0.017 0.513 0.114 -0.143 0.002 0.246 0.004[0.511] [0.090] [0.567] [0.093] [0.862] [0.108] [0.423] [0.031] [0.189] [0.016]Export -0.141 -0.035 -0.498*** -0.077*** -0.166 -0.014 -0.559*** -0.043*** -0.124 -0.025*[0.315] [0.062] [0.143] [0.024] [0.504] [0.073] [0.137] [0.009] [0.130] [0.015]Wagej06 -0.176 -0.057 -0.097 -0.017 -0.216 -0.045 -0.088 -0.009 -0.056 -0.007[0.286] [0.053] [0.210] [0.037] [0.300] [0.055] [0.181] [0.011] [0.147] [0.017]Capital -0.230 -0.054 -0.045 -0.003 -0.351 -0.081 0.081 -0.004 -0.185 -0.002[0.378] [0.065] [0.454] [0.075] [0.636] [0.078] [0.342] [0.025] [0.142] [0.012]Hicks-neutral, ϕ -0.265*** -0.004 -0.038 0.014 -0.219* -0.004 -0.060 -0.000 -0.093* -0.007[0.098] [0.022] [0.090] [0.016] [0.120] [0.024] [0.089] [0.005] [0.052] [0.006]Foreign-Owned -0.527 -0.139* -0.288 -0.051 -0.607 -0.124 -0.373 -0.031 -0.376** -0.030[0.366] [0.076] [0.291] [0.057] [0.419] [0.076] [0.248] [0.020] [0.188] [0.022]R&D -0.161 -0.007 -0.047 -0.016 -0.005 0.008 -0.050 0.007 0.145 0.018[0.226] [0.050] [0.245] [0.040] [0.325] [0.053] [0.159] [0.014] [0.116] [0.014]Training 0.212 0.051 -0.082 -0.031 0.440 0.065 -0.039 0.003 0.103 0.008[0.231] [0.051] [0.224] [0.038] [0.468] [0.061] [0.160] [0.012] [0.100] [0.011]Wagej96 -0.993* -0.214** 0.226 0.042 -1.255* -0.236** 0.214 0.021 -0.260 0.002[0.515] [0.092] [0.223] [0.038] [0.746] [0.102] [0.202] [0.013] [0.212] [0.022]ln(Ls/Lu)96 0.379*** 0.134* 0.485*** 0.298***[0.078] [0.071] [0.165] [0.047]du -0.267 0.169** -0.138 -0.358[0.402] [0.075] [0.158] [0.229]ds -1.098*** 0.309* -0.209 0.176[0.329] [0.172] [0.359] [0.163](LsLs+Lu)960.418*** 0.114* 0.516*** 0.218**[0.329] [0.068] [0.137] [0.087]ln(Ln/Lp)96 0.413***[0.032]dp -0.233**[0.119]dn 0.060[0.046](LnLn+Lp)960.380***[0.029]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesHansen J p-value 0.101 0.291 0.636 0.763 0.021 0.164 0.954 0.834 0.022 0.070No. Obs 2,542 3,244 1,795 2,036 3,012 3,244 1,487 2,082 3,115 3,244Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initial non-importers is used in allregressions. Import status is treated as an endogenous variable in all columns. It is instrumented with both the distance to port and theshare of imports shipped by air.147B.7. Additional TablesTable B.18: Importing and Standardized ProductionDep. Var. StandardsOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationIV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Import Status 1.491* 1.425* 1.819* 1.622* 2.268 1.395* 1.113 1.851* 1.467* 1.517*[0.877] [0.851] [1.002] [0.966] [2.112] [0.843] [1.282] [1.055] [0.850] [0.826]Wagej06 -0.114** -0.114** -0.060 -0.039 -0.136 -0.114*** -0.059 -0.060 -0.117** -0.114**[0.045] [0.045] [0.039] [0.040] [0.098] [0.044] [0.099] [0.039] [0.052] [0.046]capital 0.014 0.014 0.007 0.013 0.005 0.014 0.012 0.011 0.011 0.014[0.010] [0.009] [0.011] [0.012] [0.025] [0.009] [0.020] [0.012] [0.011] [0.010]Hicks-neutral, ϕ 0.008 0.010 0.002 0.006 -0.000 0.010 0.033 0.005 0.006 0.009[0.020] [0.020] [0.023] [0.024] [0.034] [0.019] [0.038] [0.023] [0.022] [0.020]Foreign-Owned 0.052 0.046 0.031 0.043 0.066 0.048 0.030 0.039 0.052 0.046[0.104] [0.102] [0.118] [0.111] [0.119] [0.101] [0.105] [0.117] [0.104] [0.105]R&D 0.121** 0.122** 0.102 0.117* 0.072 0.122** 0.078 0.115* 0.117* 0.119**[0.059] [0.057] [0.065] [0.063] [0.089] [0.057] [0.061] [0.066] [0.060] [0.060]Training 0.077*** 0.078*** 0.063* 0.084** 0.076 0.079*** 0.088** 0.072** 0.080*** 0.077***[0.029] [0.028] [0.034] [0.033] [0.047] [0.028] [0.044] [0.034] [0.031] [0.030]Wagej96 -0.032 -0.029 -0.006 0.007 -0.074 -0.029 -0.049 -0.007 -0.005 -0.032[0.059] [0.058] [0.043] [0.046] [0.091] [0.057] [0.086] [0.043] [0.064] [0.059]ln(Ls/Lu)96 0.007 -0.018 0.005 -0.024[0.009] [0.012] [0.012] [0.021]du -0.132 -0.024 0.070[0.107] [0.024] [0.104]ds -0.033 -0.038 -0.003 -0.072[0.023] [0.044] [0.031] [0.047](LsLs+Lu)960.042 0.009 0.051 -0.088[0.055] [0.046] [0.049] [0.270]ln(Ln/Lp)96 0.001[0.008]dp -0.028*[0.017]dn -0.040**[0.018](LnLn+Lp)960.072[0.051]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesHansen J p-value 0.460 0.428 0.399 0.379 0.647 0.421 0.202 0.411 0.472 0.426No. Obs 3,329 3,329 3,329 2,958 2,720 3,329 1,194 3,329 2,958 3,329Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initialnon-importers is used in all regressions. The education threshold used to determine a skilled production worker is ahighschool diploma, while the threshold used for a skilled non-production worker is a college degree. Import status istreated as an endogenous variable in all columns. It is instrumented with both the distance to port and the share ofimports shipped by air.148B.7. Additional TablesTable B.19: Exporting, Initial Skill-Levels, and SBTCDep. Var. Export StatusOccupation Production Non-Production All All AllThreshold Highschool College Highschool College OccupationIV IV IV IV IV IV IV IV IV IV(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Import Status -0.579 -0.580 -0.739 -0.553 -0.532 -0.584 -0.264 -0.681 -0.342 -0.530[0.413] [0.408] [0.478] [0.432] [0.544] [0.412] [0.450] [0.462] [0.359] [0.386]Wagej06 0.035 0.035 -0.025 -0.028 0.061 0.035 -0.005 -0.026 0.038 0.035[0.031] [0.031] [0.025] [0.026] [0.045] [0.031] [0.047] [0.025] [0.032] [0.031]Capital 0.033*** 0.033*** 0.035*** 0.036*** 0.037*** 0.033*** 0.033*** 0.036*** 0.032*** 0.035***[0.007] [0.007] [0.008] [0.008] [0.010] [0.007] [0.012] [0.008] [0.007] [0.007]Hicks-neutral, ϕ 0.029** 0.029** 0.030** 0.034** 0.034** 0.029** 0.004 0.032** 0.030** 0.030**[0.012] [0.012] [0.013] [0.013] [0.014] [0.012] [0.017] [0.013] [0.012] [0.012]Foreign-Owned 0.246*** 0.246*** 0.260*** 0.237*** 0.244*** 0.248*** 0.201** 0.259*** 0.220*** 0.245***[0.086] [0.087] [0.096] [0.087] [0.092] [0.087] [0.092] [0.093] [0.077] [0.085]R&D 0.112*** 0.112*** 0.118*** 0.114*** 0.115** 0.114*** 0.084** 0.121*** 0.106*** 0.118***[0.042] [0.041] [0.046] [0.043] [0.045] [0.042] [0.042] [0.044] [0.039] [0.041]Training 0.082*** 0.082*** 0.086*** 0.085*** 0.079*** 0.084*** 0.087*** 0.090*** 0.078*** 0.087***[0.020] [0.019] [0.023] [0.021] [0.022] [0.020] [0.027] [0.022] [0.019] [0.020]Wagej96 0.064 0.064 0.016 0.025 0.077* 0.063 0.004 0.015 0.073* 0.060[0.040] [0.040] [0.028] [0.030] [0.046] [0.040] [0.053] [0.028] [0.039] [0.039]ln(Lps/Lpu)96 0.009 -0.001 0.004 -0.029**[0.006] [0.012] [0.006] [0.012]dpu 0.040 0.018 -0.012 0.033[0.065] [0.016] [0.011] [0.054]dps -0.033** -0.043* -0.012 -0.049*[0.014] [0.023] [0.014] [0.028](LpsLps+Lpu)960.067** 0.065* 0.053* 0.025[0.033] [0.035] [0.031] [0.107]ln(Ln/Lp)96 -0.013**[0.005]dp -0.017*[0.010]dn -0.025**[0.011](LnLn+Lp)96-0.074**[0.036]Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes YesHansen J p-value 0.199 0.195 0.358 0.289 0.273 0.192 0.356 0.323 0.118 0.170No. Obs 3,619 3,619 3,619 3,226 2,994 3,619 1,395 3,619 3,226 3,619Notes: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are in square brackets. The sample of initialnon-importers is used in all regressions. The education threshold used to determine a skilled production worker is ahighschool diploma, while the threshold used for a skilled non-production worker is a college degree. Import status istreated as an endogenous variable in all columns. It is instrumented with both the distance to port and the share ofimports shipped by air.149

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0308074/manifest

Comment

Related Items