(pslulfdo Essays On +hwhurjhqhrxv )lupv Dqg

Transcript

201-9 .DOHE*LUPD$EUHKD PhD Thesis (PSLULFDOEssays on +HWHURJHQHRXV )LUPV DQG,QWHUQDWLRQDO 7UDGH DEPARTMENT OF ECONOMICS AND BUSINESS AARHUS UNIVERSITY  DENMARK Empirical Essays on Heterogeneous Firms and International Trade Kaleb Girma Abreha A PhD thesis submitted to School of Business and Social Sciences, Aarhus University, in partial fulfillment of the requirements of the PhD degree in Economics and Business May 2015 Preface This thesis was written between February 2012 and 2015 during my enrollment as a PhD student at the Department of Economics and Business, Aarhus University. I am grateful to the department for giving me the opportunity to join the graduate school and the financial support, without which this thesis would not have been remotely possible. I am also grateful to the Tuborg Research Centre for Globalization and Firms for the financial support and excellent research environment. My sincere gratitude goes to my supervisor and coauthor Val´erie Smeets for her guidance from the very beginning. Her suggestions have extensively contributed to the content and style of this thesis. Further, her attention to details has inspired me to be extra careful with my findings and writings. I also thank my co-supervisor Philipp Schr¨oder, who is the best leader in my experience. I have benefited from his feedbacks and learnt quite a lot from his leadership style. I have had a privilege to work with Fr´ed´eric Warzynski in a joint project. During the course of this project, his questions and suggestions have helped me to be objective with what I am trying to accomplish. I have improved the content of this thesis by incorporating many of his insightful comments. I also truly appreciate his willingness to provide me with a helping hand anytime. Between August and December 2014, I was visiting the Department of Economics at the Pennsylvania State University. I thank Mark Roberts for hosting me during my visit. I have benefited tremendously from his lectures and his valuable comments on early drafts of my papers. I thank the assessment committee members Tor Erikson (Aarhus University), Mark Roberts (Pennsylvania State University) and Johannes Van Biesebroeck (KU Leuven) for carefully reading the thesis manuscript and providing me with invaluable comments and suggestions. I have included some of them in this version, and I look forward to incorporating the rest in my future work. Over the past three years, I came to know many wonderful people. I thank you all, especially those in the Tuborg Centre, for a number of interesting conversations and fun times we had. I would also like to thank Susan Stilling, Susanne Christensen and Ann-Marie Gabel for their timely assistance with a range of practical issues. Their help saved me a lot of time and effort. I am also highly indebted to my family and friends for their patience, assistance and more importantly encouragement. Thank you very much! Kaleb Girma Abreha, Aarhus 2015 i Contents Introduction iii Dansk resum´ e vii 1 Coping with the Crisis and Export Diversification 1 2 A Dynamic Model of Firm Activities: Evidence from Danish Manufacturing 34 3 Importing and Firm Productivity in Ethiopian Manufacturing 80 4 Imported Inputs and Firm Absorptive Capacity in Ethiopian Manufacturing 117 ii Introduction Research in international trade over the last two decades has witnessed a proliferation of studies which use microdata sets. A pioneering paper by Bernard and Jensen (1995) and several subsequent studies have found that firms which participate in international trade constitute only a small fraction. At the same time, these firms are more productive and outperform their domestic counterparts in a variety of performance measures. In the literature, these performance differentials are mainly attributed to substantial entry costs in export markets (see e.g. Roberts and Tybout 1997; Bernard and Jensen 1999). The presence of these costs indicates that exporting firms are inherently more productive prior to their export market participation. Also, there are expected productivity gains by participating in international trade, albeit mixed findings in the literature (see e.g. Bernard and Jensen 1999; Clerides et al. 1998). On the other hand, the accumulation of these empirical findings, particularly firm heterogeneity, has led to the development of new models of international trade with a prominent focus on firms instead of countries and industries (see e.g. Melitz 2003; Bernard et al. 2003). In the light of this vast and still growing literature, this thesis explores the behavior of firms in international trade in two different economies; Denmark and Ethiopia. Denmark is a small open economy with a heavy reliance on the world market as a source and an outlet for domestic economic activities. In this respect, this thesis analyzes the reactions of exporting firms which have faced a massive shock during the recent economic crisis while accounting for the product and the destination dimensions of their exporting activities. Additionally, there is a prevalence of simultaneous exporting and importing within the Danish manufacturing firms. Hence, this thesis examines the determinants of a firm’s entry decision into exporting and importing and the resulting impact of this decision on its activity scope and productivity growth. On the contrary, Ethiopia is a low-income economy, with a small industrial base and a macroeconomic environment characterized by capacity constraints and market distortions. The main questions addressed in this thesis include whether or not firms operating under different economic and policy environments behave similarly as their counterparts in the developed and developing economies. Further, the question of whether or not there are benefits for firms from participating in international trade is considered. As a typical case of the least developed countries, this examination of firm internationalization in the Ethiopian manufacturing sector is insightful given that firms in these economies have not been subjected to the same degree of scrutiny as those in the developed and developing countries. iii This thesis is related to several studies, both theoretical contributions and empirical findings, which have looked into the behavior of firms in international trade. It adds to the existing literature in three main directions. First, it assesses how exporting firms have managed to cope with a large shock associated with the recent economic crisis and the role of export diversification in that regard. Second, it analyzes the complementarity between exporting and importing in determining a firm’s trade participation decision in a unified empirical framework which accommodates multiple dimensions of firm heterogeneity. Third, inspired by the empirical fact that firms in the least developed countries heavily rely on imported inputs, it examines whether or not such a heavy reliance is translated into any productivity gain. Further, it explores the determinants of the distribution of the productivity gain, if there is any, across importing firms. This thesis contains four chapters. Each chapter is a separate self-contained essay. The first two chapters asses the characteristics of firm trade participation in the Danish economy. The last two chapters study the extent of firm internationalization as well as the significance and determinants of trade-driven knowledge spillovers in the Ethiopian economy. In line with the standard models of heterogeneous firms and international trade, the focus is primarily on firms in the manufacturing sector undertaking their core economic activities within the national boundaries of the countries under study. Chapter 1—Coping with the Crisis and Export Diversification—establishes a set of stylized facts which summarizes salient features of Danish international production over the last decade. We document a significant variation in the export participation of firms across sectors and industries, a positive correlation between the scope and the scale of firms’ exporting activities, the presence of carry-along trade, and a highly uneven distribution of sales not only across firms but also within the export basket of firms. Focusing on the recent economic crisis, we decompose aggregate export growth rates into firm, product and destination components. In line with previous findings, we show that most of the decline is attributed to the intensive margin; that is, incumbent firms reducing their export shipment. We follow a recent literature (Amador and Opromolla 2013; Gopinath and Neiman 2014) by further decomposing the intensive margin into sub-intensive (the contribution to total export change of continuing products) and sub-extensive margins (the contribution coming from product churning). We find that the sub-intensive margin played the most important role during the crisis, so that firms continued to ship the same products to the same destinations but in lower quantities. We also find a significant effect of the sub-extensive margin, particularly during the recovery, as firms started shipping new products to new destinations. More importantly, we also find that the economic conditions of markets where firms ship matter in that export diversification into fast-growing economies like China or more generally the BRICS was associated with better export performance, so that trade reorientation helped firms to cope with the crisis. Chapter 2—A Dynamic Model of Firm Activities: Evidence from Danish Manufacturing— explores the nature of complementarity between exporting and importing within firms. Considering iv firms in the manufacturing sector over the period 2000-2007, a simple description of the data uncovers a widespread occurrence of simultaneous exporting and importing within firms. The data also reveals a significant export and import activity premia which follow a ranking where two-way trading firms are the best performers followed by import-only, export-only and lastly domestic firms. It is also found that there is a high persistence in the activity scope of firms. Motivated by these empirical facts, I specify a dynamic discrete choice model of exporting and importing following a modeling approach by Aw et al. (2011). In the model, firms are defined to be heterogeneous in terms of their size (capital holding), factor payment (wage), and productivity. The model provides a framework to analyze the determinants of a firm’s decision to export and import while allowing for this decision to affect its future productivity trajectory. It also enables the analysis of how large of a role market costs play in a firm’s entry decisions into exporting and importing. The parameter estimates exhibit a marked difference in the intensity of competition firms face and their pricing strategies in the domestic and export markets where export markets are characterized by a more elastic demand and a lower markup. The greater sensitivity of demand in export markets accords with the fact that these markets host a larger number of firms and product varieties. On the cost side, the estimates show that firms with a large capital holding and paying higher wages are cost-efficient even after controlling for their productivity. By extending the Levinsohn and Petrin (2003) algorithm to account for endogenous evolution of firm productivity, I find that there is a learning-by-doing effect from exporting and importing, which is especially greater from importing. As in Das et al. (2007), the start-up and running costs of operations in the export and import markets are estimated using a Bayesian Markov Chain Monte Carlo. In line with the self-selection hypothesis, the estimated sunk and fixed costs of exporting and importing are substantial. And, these costs are greater for large firms, which are more likely to operate in several markets simultaneously each involving non-negligible costs. The learning effects in addition to the market entry costs further drive the selection of firms into exporting and importing highlighting the complementarity between these trading activities. Chapter 3—Importing and Firm Productivity in Ethiopian Manufacturing—analyzes the causal relationship between importing and firm productivity. The motivation for this study comes from the fact that the vast majority of the literature on firms in international trade has been restricted to manufacturing firms in advanced economies and a few developing countries in Asia and Latin America. Hence, African manufacturing firms have been greatly neglected. Even among a handful of existing studies, utmost focus has been on exporting (see e.g. Bigsten et al. 2004; Bigsten and Gebreeyesus 2009; Mengistae and Pattillo 2004; Van Biesebroeck 2005). In this respect, the literature on African manufacturing firms remains largely incomplete, especially in view of high import-to-GDP ratios and import shares of manufacture goods in these economies. Using a panel of firms in Ethiopian manufacturing over the period 1996-2011, it is shown that more productive firms self-select into import markets implying that importing involves irreversible v and periodic costs which only the most productive firms are able to absorb. To examine the causal effect of importing on a firm’s productivity, I follow Kasahara and Rodrigue (2008) in the specification of a structural model in which the static and dynamic effects of importing are separately estimated. The estimation results provide evidence of learning-by-importing albeit an initial, temporary productivity decline. Further, the results reveal that an intensive use of imported inputs is associated with greater productivity improvement among importing firms. Chapter 4—Imported Inputs and Firm Absorptive Capacity in Ethiopian Manufacturing—is basically an extension of chapter 3, and it emphasizes the role of absorptive capacity regarding productivity impact of imported inputs. This chapter is motivated by the fact that the productivity gains from importing are relatively small in Ethiopian manufacturing sector as compared to findings from studies in other countries. To this end, I estimate a standard production function in which import and absorptive capacity (measured by the share of skilled employees in a firm’s workforce) are included as additional variables. The estimates show that imported inputs are beneficial if a firm has the necessary skill composition to absorb the embodied knowledge in those inputs. This implies that imported inputs have no special purpose if a firm has no absorptive capacity at all. Alternatively, I adopt a threshold regression and use a sample splitting technique developed by Hansen (2000). The technique splits the entire sample into different regimes based on firms’ absorptive capacities. The estimates show that the effect of imported inputs is greater for firms with a sufficiently high absorptive capacity. At the same time, the threshold estimate indicates that most firms have absorptive capacity below the threshold requirement. Consequently, despite the widespread use of imported inputs in the sector, the benefits of enhanced access to foreign technology are confined to only few firms. These results on the importance of absorptive capacity provide firm-level support to the prevailing macroeconomic evidences which identify limited absorptive capacity as an impediment to knowledge spillovers to the least developed countries. vi Dansk resum´ e Forskningen i international handel har gennem de sidste 2 ˚ artier oplevet en opblomstring af undersøgelser, som bruger mikrodatasæt. En banebrydende artikel af Bernard and Jensen (1995) samt adskillige efterfølgende artikler har fundet, at virksomheder, som deltager i international handel, kun udgør en lille del. Samtidig er disse virksomheder mere produktive og klarer sig bedre end deres hjemlige konkurrenter p˚ a mange forskellige parametre. I litteraturen tilskrives disse forskellige præstationer hovedsagelig omkostningerne ved indtrædelse p˚ a eksportmarkeder (se fx Roberts and Tybout 1997; Bernard and Jensen 1999). Tilstedeværelsen af disse omkostninger angiver, at eksporterende virksomheder i sig selv er mere produktive, før de indtræder p˚ a eksportmarkedet. P˚ a trods af blandede resultater i litteraturen, forventes der at v˚ are produktivitetsgevinster ved at deltage i international handel (se fx Bernard and Jensen 1999; Clerides et al. 1998). P˚ a den anden side har akkumuleringen af disse empiriske resultater, især ang˚ aende virksomheders heterogenitet, ført til udvikling af nye modeller i international handel, som har markant fokus p˚ a virksomheder i stedet for lande og brancher (se fx Melitz 2003; Bernard et al. 2003). I lyset af denne omfattende og stadigt voksende litteratur undersøger denne afhandling adfærden hos virksomheder i international handel i to forskellige økonomier: Danmark og Etiopien. Danmark er en lille ˚ aben økonomi med stor afhængighed af verdensmarkedet som kilde og som afsætningsmulighed for hjemlige, økonomiske aktiviteter. I den henseende analyserer denne afhandling reaktionerne hos eksporterende virksomheder, som har oplevet et massivt chok under den seneste, økonomiske krise, samtidig med at de har skullet st˚ a til regnskab for omfanget af deres eksportaktiviteter. Desuden er der en udbredt forekomst af samtidig eksport og import i de danske fremstillingsvirksomheder. S˚ aledes undersøger denne afhandling, hvilke faktorer der er afgørende for en virksomheds beslutning om at eksportere og importere, samt denne beslutnings indflydelse p˚ a virksomhedens aktivitetsomfang og produktivitetsvækst. Etiopien er derimod en lavindkomstøkonomi med et lille industrielt grundlag og et makroøkonomisk miljø, som er kendetegnet ved kapacitetsbegrænsninger og skævvridning af markedet. De vigtigste problemstillinger i denne afhandling handler om, hvorvidt virksomheder, som arbejder under forskellige økonomiske og politiske forudsætninger, agerer p˚ a samme m˚ ade som de tilsvarende virksomheder i de udviklede økonomier og udviklingsøkonomierne eller ej. Desuden overvejes det, om det er fordelagtigt for virksomheder at deltage i international handel eller ej. Denne undersøgelse af virksomheders internationalisering i Etiopiens fremstillingssektor er velegnet som en typisk case fra de mindst udviklede lande, forudsat at virksomheder i disse økonomier ikke har været udsat for vii den samme grad af pres som virksomheder i de udviklede økonomier og udviklingsøkonomierne. Denne afhandling relaterer sig til adskillige undersøgelser, b˚ ade teoretiske input og empiriske resultater, som har kigget nærmere p˚ a virksomheders ageren i international handel. Afhandlingen bidrager til den eksisterende litteratur p˚ a hovedsagelig tre punkter. For det første vurderer den, hvordan eksporterende virksomheder har h˚ andteret chokket i forbindelse med den seneste, økonomiske krise samt betydningen af omstilling af eksporten i den forbindelse. For det andet analyserer den komplementariteten mellem eksport og import, n˚ ar man skal vurdere en virksomheds beslutning om at træde ind p˚ a markedet p˚ a et samlet empirisk grundlag, som imødekommer adskillige dimensioner af virksomhedsheterogenitet. For det tredje og inspireret af det empiriske faktum, at virksomheder i de mindst udviklede lande er dybt afhængige af importerede produkter, undersøger den, om en s˚ adan dyb afhængighed omsættes til nogen form for produktivitetsgevinst. Desuden udforsker den de bestemmende faktorer for fordelingen af produktivitetsgevinsten–hvis der er nogen—p˚ a tværs af importerende virksomheder. Afhandlingen indeholder fire kapitler. Hvert kapitel er en separat, uafhængig artikel. De første to kapitler vurderer karakteristikaene for virksomheders handelsdeltagelse i den danske økonomi. De sidste to kapitler undersøger omfanget af virksomheders internationalisering samt betydningen af og bestemmende faktorer for handelsrelateret videnoverførsel i den etiopiske økonomi. I overensstemmelse med standard-modellerne for heterogene virksomheder og international handel er fokus primært rettet mod virksomheder i fremstillingssektoren, som har deres kerneaktiviteter inden for de nationale grænser i de observerede lande. Kapitel 1—Coping with the Crisis and Export Diversification—p˚ aviser et sæt af stiliserede fakta, som opsummerer hovedtrækkene i dansk international produktion i det seneste ˚ arti. Vi dokumenterer en betydelig variation i virksomheders eksportdeltagelse p˚ a¡, tværs af sektorer og brancher, en positiv sammenhæng mellem rækkevidden og omfanget af virksomheders eksportaktiviteter, tilstedeværelsen af handel i halvfabrikata samt en højst ulige fordeling af salg ikke kun p˚ atværs af virksomheder, men ogs˚ a inden for gruppen af eksportvirksomheder. Med fokus p˚ a den seneste økonomiske krise opløser vi de samlede eksportvækstrater i virksomheds, produkt-og destinationskomponenter. I overensstemmelse med tidligere resultater viser vi, at det meste af nedgangen kan tilskrives den intensive margin; dvs. nuværende virksomheder som reducerer deres eksportleverancer. Vi følger den nyere litteratur (Amador and Opromolla 2013; Gopinath and Neiman 2014) ved yderligere at opløse den intensive margin i produktintensiv margin (bidraget til den totale eksportændring af vedvarende produkter) og produktekstensiv margin (bidraget fra den s˚ a kaldte product churning). Vi finder, at den produktintensive margin spillede den vigtigste rolle under krisen, idet virksomheder fortsatte med at sende de samme produkter til de samme destinationer, blot i mindre mængder. Vi finder ogs˚ a en signifikant effekt af den produktekstensive margin, specielt under det begyndende konjunkturopsving, idet virksomheder begyndte at sende nye produkter til nye destinationer. Og hvad er mere væsentligt, s˚ afinder vi ogs˚ a at de økonomiske betingelser p˚ a de markeder, som virksomheder sender til, har betydning, viii idet eksportspredning til hurtigt voksende økonomier som Kina eller mere generelt BRIKS var forbundet med bedre eksportresultater, s˚ aledes at handelsomlægning hjalp virksomhederne til at klare krisen. Kapitel 2—A Dynamic Model of Firm Activities: Evidence from Danish Manufacturing— undersøger beskaffenheden af komplementariteten mellem eksport og import i virksomheder. Kigger man p˚ a virksomheder i fremstillingssektoren i perioden 2000-2007, afslører en simpel beskrivelse af data en udbredt forekomst af samtidig eksport og import i virksomheder. Data afslører ogs˚ a en betydelig eksport- og importaktivitet, som følger en rangorden, hvor virksomheder med tovejshandel klarer sig bedst efterfulgt af kun-importerende, kun-eksporterende og til sidst indenlandske virksomheder. Der blev ogs˚ a fundet en stor udholdenhed i virksomheders aktivitetsomfang. Motiveret af disse empiriske kendsgerninger specificerer jeg en dynamisk discrete-choice-model for eksport og import og følger en modelleringsmetode af Aw et al. (2011). I denne model er virksomheder defineret til at være heterogene m˚ alt i størrelse (kapitalbeholdning), faktorbetaling (løn) og produktivitet. Modellen tilvejebringer et grundlag til at analysere de bestemmende faktorer for en virksomheds beslutning om at eksportere og importere og at lade denne beslutning p˚ avirke virksomhedens fremtidige produktivitetsbane. Den muliggør ogs˚ a en analyse af, hvor stor en rolle omkostningerne spiller for en virksomheds beslutning om at eksportere og importere. Parameterestimaterne viser en markant forskel i den konkurrenceintensitet, virksomhederne oplever, og deres prisstrategier p˚ a hjemme-og eksportmarkederne, hvor eksportmarkeder er karakteriseret ved en mere elastisk efterspørgsel og en lavere avance. Den større efterspørgselsfølsomhed p˚ a eksportmarkeder er i overensstemmelse med det faktum, at disse markeder indeholder et større antal virksomheder og et større produktudvalg. P˚ a omkostningssiden viser estimaterne, at virksomheder med en stor kapitalbeholdning, og som betaler højere lønninger, er rentable, selv efter der er taget højde for deres produktivitet. Ved at udvide Levinsohn and Petrin (2003) algoritme til at forklare endogen udvikling af virksomhedsproduktivitet finder jeg, at der er en learning-by-doing-effekt ved at eksportere og importere, som er specielt stor ved import. Som i Das et al. (2007) estimeres opstartsomkostninger og løbende driftsomkostninger p˚ a eksport-og importmarkederne ved at anvende en Bayesian Markov Chain Monte Carlo. I tr˚ ad med selvselektionshypotesen er de estimerede produktionsomkostninger og faste omkostninger ved at eksportere og importere væsentlige. Og disse omkostninger er større for store virksomheder, som højst sandsynligt vil operere samtidigt p˚ aflere markeder, og det indebærer ikke ubetydelige omkostninger. Sammen med etableringsomkostningerne driver læringseffekterne virksomhederne til at vælge at eksportere og importere, idet de fremhæver komplementariteten mellem disse handelsaktiviteter. Kapitel 3—Importing and Firm Productivity in Ethiopian Manufacturing—analyserer˚ arsagsforholdet mellem import og virksomhedsproduktivitet. Motivationen til at lave denne undersøgelse er det faktum, at det meste af litteraturen om virksomheder i international handel er begrænset til fremstillings-virksomheder i udviklede økonomier og nogle f˚ a udviklingslande i Asien og Latiix namerika. S˚ aledes er afrikanske fremstillingsvirksomheder blevet overset. Selv i den lille h˚ andfuld af eksisterende undersøgelser har fokus været p˚ aeksport (se fx Bigsten et al. 2004; Bigsten and Gebreeyesus 2009; Mengistae and Pattillo 2004; Van Biesebroeck 2005). Som det ses, er litteraturen om afrikanske fremstillingsvirksomheder ukomplet, især i lyset af høje importavancer i forhold til BNP og importandele af fremstillingsvarer i disse økonomier. Ved at anvende et panel af etiopiske fremstillingsvirksomheder i perioden 1996-2011 p˚ avises det, at mere produktive virksomheder selvselekterer ind p˚ a eksportmarkeder, hvilket antyder, at import indebærer uigenkaldelige og periodiske omkostninger, som kun de mest produktive virksomheder kan absorbere. For at undersøge ˚ arsagseffekten af import for en virksomheds produktivitet, følger jeg Kasahara and Rodrigue (2008) i specifikationen af en strukturel model, hvor de statiske og dynamiske effekter af import bliver estimeret hver for sig. Resultaterne tilvejebringer belæg for learning-by-importing p˚ atrods af en første, forbig˚ aende produktivitetsnedgang. Resultaterne afslører endvidere, at intensiv brug af importerede input er forbundet med større produktivitetsforbedringer blandt importerende virksomheder. Kapitel 4—Imported Inputs and Firm Absorptive Capacity in Ethiopian Manufacturing—er dybest set en forlængelse af kapitel 3, og kapitlet understreger betydningen af absorberende kapacitet med hensyn til produktivitetsvirkningen af importerede input. Dette kapitel er motiveret af, at produktivitetsfordelene ved at importere er relativt sm˚ a i den etiopiske fremstillingssektor sammenlignet med resultater fra undersøgelser i andre lande. Med det form˚ al estimerer jeg en standard produktionsfunktion, hvor import og absorberende kapacitet (m˚ alt ved andelen af faglærte medarbejdere i en virksomheds arbejdsstyrke) er medtaget som ekstra variabler. Estimaterne viser, at importerede input er fordelagtige, hvis virksomheden har den nødvendige, uddannede medarbejdersammensætning til at kunne absorbere den indeholdte viden i disse input. Dette betyder, at importerede input ikke har noget specielt form˚ al, hvis virksomheden ingen absorberende kapacitet har. Alternativt vælger jeg en tærskelregression og bruger en teknik til opsplitning af stikprøver udviklet af Hansen (2000). Teknikken opdeler hele prøven i forskellige ordninger baseret p˚ a virksomheders absorberende kapacitet. Estimaterne viser, at effekten af importerede input er større for virksomheder med en tilstrækkelig høj absorberede kapacitet. Samtidig indikerer tærskelestimatet, at de fleste virksomheder har en absorberende kapacitet under tærskelkravet. P˚ a trods af den udbredte anvendelse af importerede input i sektoren er fordelene ved forøget adgang til udenlandsk teknologi derfor begrænset til ganske f˚ avirksomheder. Disse resultater vedrørende vigtigheden af absorberende kapacitet understøtter p˚ a virksomhedsniveau de gældende makroøkonomiske holdepunkter, som identificerer begrænset absorberende kapacitet som en hindring for videnspredning til de mindst udviklede lande. x References Amador, J., & Opromolla, L. D. (2013). Product and Destination Mix in Export Markets. Review of World Economics, 149 (1), 23–53. Aw, B. Y., Roberts, M. J., & Xu, D. Y. (2011). R&D Investment, Exporting, and Productivity Dynamics. American Economic Review , 101 (4), 1312–1344. Bernard, A. B., Eaton, J., Jensen, J. B., & Kortum, S. (2003). Plants and Productivity in International Trade. American Economic Review , 93 (4), 1268–1290. Bernard, A. B., & Jensen, J. B. (1995). Exporters, Jobs, and Wages in U.S. Manufacturing: 1976-1987. Brookings Papers on Economic Activity. Microeconomics, 1995 , 67–119. Bernard, A. B., & Jensen, J. B. (1999). Exceptional Exporter Performance: Cause, Effect, or Both? Journal of International Economics, 47 (1), 1–25. Bigsten, A., Collier, P., Dercon, S., Fafchamps, M., Gauthier, B., Willem Gunning, J., . . . others (2004). Do African Manufacturing Firms Learn from Exporting? Journal of Development Studies, 40 (3), 115–141. Bigsten, A., & Gebreeyesus, M. (2009). Firm Productivity and Exports: Evidence from Ethiopian Manufacturing. Journal of Development Studies, 45 (10), 1594–1614. Clerides, S. K., Lach, S., & Tybout, J. R. (1998). Is Learning by Exporting Important? Microdynamic Evidence from Colombia, Mexico, and Morocco. Quarterly Journal of Economics, 903–947. Das, S., Roberts, M. J., & Tybout, J. R. (2007). Market Entry Costs, Producer Heterogeneity, and Export Dynamics. Econometrica, 75 (3), 837–873. Gopinath, G., & Neiman, B. (2014). Trade Adjustment and Productivity in Large Crises. American Economic Review , 104 (3), 793–831. Hansen, B. E. (2000). Sample Splitting and Threshold Estimation. Econometrica, 68 (3), 575–603. Kasahara, H., & Rodrigue, J. (2008). Does the Use of Imported Intermediates Increase Productivity? Plant-level Evidence. Journal of Development Economics, 87 (1), 106–118. Levinsohn, J., & Petrin, A. (2003). Estimating Production Functions Using Inputs to Control for Unobservables. Review of Economic Studies, 70 (2), 317–341. Melitz, M. J. (2003). The Impact of Trade on Intra-industry Reallocations and Aggregate Industry Productivity. Econometrica, 71 (6), 1695–1725. Mengistae, T., & Pattillo, C. (2004). Export Orientation and Productivity in Sub-Saharan Africa. IMF Staff Papers, 327–353. Roberts, M. J., & Tybout, J. R. (1997). The Decision to Export in Colombia: An Empirical Model of Entry with Sunk Costs. American Economic Review , 545–564. Van Biesebroeck, J. (2005). Exporting Raises Productivity in Sub-Saharan African Manufacturing Firms. Journal of International Economics, 67 (2), 373–391. xi Chapter 1 Coping with the Crisis and Export Diversification 1 Coping with the Crisis and Export Diversification∗ Kaleb Girma Abreha† Val´ erie Smeets‡ Fr´ ed´ eric Warzynski§ Abstract Using a highly disaggregated firm-product-destination level data, we document salient features of Danish international production over the last decade. These include a significant variation in the export participation of firms across industries, a positive correlation between the scope (number of products exported and markets served) and the scale of exporting activities, the existence of carry-along trade, and a considerable dominance of few firms and few core products as evidenced by highly uneven distribution of sales not only across firms but also within the export basket of firms. Further, we analyze how Danish exporters responded to the global recession and the recovery that followed. We find that firms reacted mainly by adjusting their scale of export shipments, and by extending their portfolio outside their core product. More importantly, we also find that the economic conditions of markets where firms sell matter: export diversification into fast-growing economies like China or more generally the BRICS was associated with better export performance, so that trade reorientation helped firms to cope with the crisis. JEL Codes: F14, F6, L60 Keyword: Trade collapse, intensive and extensive margins, export diversification, Denmark ∗ We are grateful to Mark Roberts for a useful dissucssion on the early draft of the paper. We thank the Tuborg Foundation and the Danish Council for Independent Research in Social Sciences (FSE) for generous financial support. The usual disclaimer applies. † Department of Economics and Business, Aarhus University, Denmark, E-mail: [email protected] ‡ Department of Economics and Business, Aarhus University, Denmark, E-mail: [email protected] § Department of Economics and Business, Aarhus University, Denmark, E-mail: [email protected] 2 1 Introduction Increased availability of micro-level datasets has over the last twenty years shifted the focus of research in international trade from countries and industries to firms and products. This new line of research extends a previous wave of empirical papers starting in the mid-nineties that challenged the existing theories of international trade, which entirely focused on comparative advantage, increasing returns to scale and consumer love for variety. As summarized by Bernard et al. (2007), these theories failed to capture important empirical regularities, most notably firm heterogeneity. Numerous recent studies have shown that firms are rather different in several dimensions even in narrowly defined industries. The pioneering paper in the field, Bernard and Jensen (1995), shows that the fraction of firms active in export markets is rather small in the US manufacturing. Additionally, these firms are systematically different from domestic firms in terms of size, productivity and input mix. That is, they are larger, more productive, more skill- and capital-intensive.1 This inspired new trade theories, starting with Bernard et al. (2003) and Melitz (2003), which revolutionized this field of research. More recently, researchers have documented additional stylized facts about exporters: domestic and international productions are dominated by a few firms (see e.g. Bernard et al. 2009; Mayer and Ottaviano 2008; Eriksson et al. 2009), and export sales are concentrated in a few products within firms (Arkolakis and Muendler 2010). These empirical findings led to the development of new models of multi-product firms whose production activity can further be categorized into the production of core and peripheral products (see e.g. Bernard et al. 2010; Bernard et al. 2011; Eckel and Neary 2010; Mayer et al. 2014). In this paper, we use a rich transaction level data from Denmark to document new stylized facts on the behavior of firms facing a large shock. Descriptive summaries of the data confirm a considerable variation in the export participation of firms across industries. Despite changes in the degree of firm and product participation in the export sector over time, the correlation between the scope (number of exported products and markets served) and the scale of exporting activities has remained positive, as shown by Arkolakis and Muendler (2010). Also, the export sector is characterized by the overall dominance and growing importance of multi-product and multidestination firms whose activities involve carry-along trade, in line with Bernard et al. (2012). Additionally, there is evidence of a highly skewed distribution of export sales between few core and several peripheral products within the export basket of firms. In line with previous studies, during the periods of large economic shock, we show that most of the decline can be attributed to the intensive margin; that is, incumbent firms reducing their level of 1 Interestingly, these characteristics are not restricted to the US only. Studies from other countries such as Muˆ uls and Pisu (2009) for Belgium, Alvarez and L´ opez (2005) and Kasahara and Lapham (2013) for Chile, Isgut (2001) for Colombia, Eriksson et al. (2009) for Denmark, Verardi and Wagner (2012) for Germany, Amiti and Davis (2012) for Indonesia, Ruane and Sutherland (2005) for Ireland, Castellani et al. (2010) for Italy, De Loecker (2007) for Slovenia, M´ an ˜ez-Castillejo et al. (2010) for Spain, Andersson et al. (2008) for Sweden, and Van Biesebroeck (2005) for Sub-Saharan African countries document similar characteristics of firms with international trade participation. 3 activity in export markets. We follow a recent literature (Amador and Opromolla 2013; Gopinath and Neiman 2014) by further decomposing the intensive margin into sub-intensive (the contribution to total export change of continuing products) and sub-extensive margins (the contribution coming from product churning). We find that the sub-intensive margin played the most important role during the 2008-2009 trade collapse, so that Danish firms continued to ship the same products to the same countries but in lower quantities. We also find a significant effect of the sub-extensive margin, particularly during the recovery, as firms started shipping new products to new destinations. The main contribution of this paper is to analyze the role of export diversification. We perform a set of empirical tests to better understand the choice of the product-destination portfolio of firms and how it can be related to the export performance during a crisis. We first analyze what determines the concentration of export sales in the core product on the various markets where firms operate. In line with recent theoretical contributions (e.g. Mayer et al. 2014), we find that distance, toughness of competition and wealth matter. Firms have a higher share of their core products on more distant and larger markets, while the opposite is true on richer markets. In the next step, we examine how firms’ export performance was affected by their presence in fast-growing markets and their reliance on a core product. We find that firms which export to China and the BRICS enjoy a significantly higher export growth, so that trade reorientation to these markets helped them to cope with the crisis.2 On the other hand, export growth was lower for firms’ core products, suggesting that a too heavy reliance on one product might have a negative impact on growth. Several papers have previously studied how firms reacted to large trade shocks. Bernard et al. (2009) study the reaction of the US firms to the Asian crisis of 1997 and show the eminent role of the intensive margin of trade. More recently, Gopinath and Neiman (2014) investigate the behavior of firms in response to the Argentine 2001-2002 crisis and document the considerable importance of the intensive margin as a trade adjustment mechanism, even if the sub-extensive margin played a non-negligible role in the import trade of Argentina. Bricongne et al. (2012) show that large French firms responded to the current economic crisis mainly by lowering their export shipment whereas small firms exit export markets or reduce the number of products exported and destinations served. Using Belgian data, Behrens et al. (2013) find that the trade collapse mainly resulted from a decline in quantities and prices of existing export and import firm-product transactions rather than from entry and exit of firms, products and trading partners. These studies establish that trade adjustment mechanisms at the microeconomic level are keys to understanding the aggregate trade collapse. Unlike these previous papers, our approach is focused on export diversification as a way to cope with a crisis. The rest of the paper is organized as follows. Section 2 describes the datasets that we use. Section 3 presents some stylized facts about Danish exporters during our sample period. Section 4 discusses our export growth decomposition exercises. Section 5 looks at the link between export 2 BRICS countries comprise Brazil, China, India, Russia, and South Africa. 4 growth and diversification. Section 6 concludes. 2 Data description The datasets used in this paper are all provided by Statistics Denmark. The focus of our analysis is the time period 2000-2010. We combine three different datasets and merge them relatively easily, as firms are identified by a common identification number. Our main source of information provides detailed records on export and import transactions by all trading firms in the Danish economy. It contains the value, weight and quantity of export and import transactions for each firm and destination/source market at 8-digit CN.3 The second dataset contains information about the product portfolio of firms and describes which products firms make domestically. It is based on a survey of all firms in the manufacturing sector with at least 10 employees and therefore covers fewer firms. Like the trade statistics, we have information on the value, weight and quantity of production at 8-digit CN for a panel of manufacturing firms. These two datasets do not contain information about overall sales revenue, employment size, material input usage or capital structure of firms. We therefore use a third dataset that contains accounting information for the universe of firms in the economy. This dataset covers more than 200,000 firms per year. Merging the other two datasets with this accounting data ensures that we only consider firms with real economic activity (see Tables A.1–A.6 in the Appendix on the evolution of the number of firms over our period of analysis). After dropping classified products and trading partners and only considering trading firms with at least one employee, we end up with 4,253,959 export transactions at 6-digit HS. The industry where firms operate is based on the Statistical Classification of Economic Activities in the European Community (NACE). One practical difficulty is that this industrial classification was revised twice during our period of analysis; in 2003 and 2007. Unlike the first revision which was relatively minor, the second one was substantial. The dataset is structured in such a way that firms are redefined according to the new classification. In our analysis, we use the latest Danish industrial classification which is comparable to NACE Rev.2. Finally, we deflate all nominal variables by the consumer price index (CPI) using 1995 as a base year. 3 Stylized facts about Danish exporters 3.1 Cross-industry export participation rate Table 1 shows substantial heterogeneity in terms of firm distribution as well as export participation rate across industries. We see that the majority of firms are engaged in industries which are 3 Combined Nomenclature (CN) is a Harmonized system (HS) of product classification with further subdivisions used in EU member countries. 5 inherently non–tradable such as agriculture, forestry and fishing, and service-sector industries such as construction, wholesale and retail trade, health and social work, and transport, storage and communication. In terms of international trade, exceedingly small fractions of these firms export, below 1%. In contrast, industries in the manufacturing sector have a considerably higher export market participation rate. A high percentage of firms engaged in the manufacturing of chemicals, furniture, rubber and plastics, machinery and equipment, electrical and optical equipment, and transport equipment are active in export markets. In these industries the average participation rate is around 50% over the period under study. On the contrary, firms engaged in production of basic and fabricated metals, food, beverage and tobacco, non-metallic mineral products, printing and publishing, and wood products have a relatively lower export participation rate. More generally, it is shown that there are significant differences in the distribution as well as in the export market participation of firms not only across agriculture, manufacturing and service sectors but also across industries within sectors, as demonstrated especially for the manufacturing sector.4 3.2 Scope and scale of firm export Table 2 reports that around 13,000 firms export more than 4,500 products to more than 200 countries. This represents around 100,000 product-destination combinations. While the number of exporting firms has slightly declined over time, the average value of export transaction has increased as shown in Table 3. Similarly, average export shipment per product and destination increased despite an increase in the overall number of products exported and destination markets served. This indicates a greater role of the intensive margin of export trade. We also notice that these three figures declined during the crisis. Further, a few superstar firms are reaching more than 100 countries, exporting more than a thousand different types of products, and therefore have thousands of product-destination exporting relations. Figure 1 depicts more clearly the dynamics of the number and scale of exporting firms and export transactions (defined at product-destination level) over the last decade. For comparison purpose, we include the import equivalents of these firms and transactions. The figure shows that importing firms outnumber their exporting counterparts for the entire time period.5 The average export value at the firm level is much larger and increasing relative to imports. In contrast, the number of export transactions is much higher, and the average value of an export transaction is typically lower compared to an import transaction. From Figure 1, it can clearly be seen that the number of trading firms as well as trade transactions declined during the economic recession in 2009. Unlike the 2009 recession, the number of trading firms as well as export and import transactions have not been affected by the less dramatic 4 Table A.7 in the Appendix depicts lower cross-industry variations in firm distribution and export market participation rate, and so is the pattern within industries over time. It also reveals a greater export market participation rate for each of the industries. This shows a systematic variation in export market participation by size in which larger firms are more likely to be exporters. 5 The opposite is true among firms in the manufacturing sector where importing is rarer than exporting. 6 trade collapse of 2003. However, firms greatly adjusted downward the volume of their export and import shipments in both recessions. Combining product and destination dimensions of exporting indicates that firms exporting more products and reaching more markets saw their relative number and contribution increase between 2000 and 2010. Table 4 shows that, among exporters, those exporting only one product to only one market destination constitute the largest group, 24.06% and 22.01% in 2000 and 2010 respectively, albeit the median firm is multi-product and multi-destination. However, the contribution of these firms to overall export is marginal (below 1%), while firms reaching more than five markets account for more than 90% of the export value (up from 87% in 2000). The unique nature of our dataset allows us to trace the sources of firm exports as own-produced products or not. Table 5 shows the nature of firm production activities in terms of traded versus non-traded products. While constructing the table, we restrict our sample to firms in the manufacturing sector because we observe the production portfolio for these firms only. We count the number of products each firm produces and whether they are traded, and take the mean (and median) for all the firms in a given year. The results reveal that greater fractions of products firms produce are traded, and these products have become more prevalent over time. In contrast, the fraction of products that are not traded at all has declined. This indicates a rising integration of the manufacturing sector to the global economy. Even though the contribution of non-traded products to the value of production rose, especially in the advent of the 2008-2009 crisis, their share remained significantly small. Coming to the trading element of firm activities, we see that firms export products that they do not actually produce. Both the mean and median number of non-produced export products are greater than their produced counterparts, highlighting the presence of carry-along trade in the Danish export trade. This is in line with findings by Bernard et al. (2012) for Belgium and Amador and Opromolla (2013) for Portugal on the existence of carry-along trade. Over time, we see that the number of products the average exporting firm exports increased, and more so in the case of non-produced export products. Despite the pervasiveness of carry-along trade, its contribution to the overall export is very small, below 5%. It is not surprising that firms mainly import products they do not produce, albeit there are times firms import goods that they produce themselves.6 3.3 Export concentration within firms Another interesting aspect of international trade is the extent of concentration of exporting activity within firms. To measure this, we first calculate the share of each product in the total export of firms. Then, we rank products in descending order and calculate the cumulative shares of each product in the export basket of each firm. As a final step, we take the mean of the cumulative shares across firms with the same product scope. 6 This can partly be explained by the level of aggregation used to define products. At an even more disaggregated level such as 10-digit HS, the incidence of importing products that firms produce themselves will be lower. 7 Table 6 portrays a highly skewed distribution of firm export sales across products within multiproduct firms. Irrespective of product scope, the first ranked product constitutes more than 50% and 45% of firm export sales in 2000 and 2010, respectively. Further, two thirds of firm export sales only come from the top three products. We also observe a decline in the unevenness of the distribution of export sales over the ten years window, i.e. firms tend to export a more diversified portfolio. This unevenness of export sales distribution is consistent with previous findings of Bernard et al. (2011) for the US, Arkolakis and Muendler (2010) for Brazil, and Amador and Opromolla (2013) for Portugal. The unevenness among Danish exporters is, however, lower compared to their Brazilian counterparts as shown by Arkolakis and Muendler (2010) for the year 2000.7 Then, we examine how differently firms rely on their core products depending on the nature of their production and export activities as well as the characteristics of the export destinations that they serve. To establish this point, we regress a destination-specific share of the core product in the export bundle of a firm on several destination market characteristics such as distance, market size and income level.8 log(Export share)i,j,t =β0 + β1 log(Distance)j,t + β2 log(GDP )j,t + β3 log(GDP P C)j,t Own−product RT A + β4 log(#P roducts)i,j,t + β5 Di,j,t + β6 Dj,t + i,j,t (1) where Export sharei,j,t is the share of the core export product of firm i in destination j in time period t, Distance is the geographical distance between Denmark and destination j which is timeinvariant and common for all firms exporting to that particular market, GDP refers to the GDP which is used as a proxy for the intensity of competition, GDP P C is per capita GDP which measures the level of income and size of demand in that market, #P roducts is the number of products a firm exports to a given destination, DOwn−product is a dichotomous variable showing whether a firm actually produces the core product it exports,9 DRT A denotes a dummy variable showing whether there is a trade agreement between Denmark and destination country j, and  is an error term. The estimation results in columns (1) and (2) from Table 7 show that firms serve larger and more distant markets primarily with their core product. Following Mayer et al. (2014), this can be explained by the fact that firms serve markets only with the products in which they are productive enough to absorb production and trade costs, and firms are more efficient in the production of their core products. Since more distant countries involve larger costs, firms are more likely to export their core products to these markets. Similarly, larger markets represent tougher competition and firms serve such markets primarily through products of their respective core competences. 7 Bernard et al. (2011) use a highly disaggregated 10–digit HS product classification for a single cross section year 2002 whereas Amador and Opromolla (2013) define a product at 4-digit HS and report the average over the time period 1996-2005. This makes comparison with these studies inappropriate. 8 Data on gravity variables such as bilateral distance, GDP, population, and regional trade agreements are obtained from the CEPII database. 9 We construct this variable for a restricted sample of exporting firms in the manufacturing sector because it requires information on the production portfolio, which is only available for these firms. 8 Relatedly, high purchasing power in destination markets provides an opportunity to export more products. Not surprisingly, the estimates show that the export share of the core product declines if a firm exports more products to a particular market destination. Further, trade-enhancing policy measures such as regional trade agreements could be pro-competitive, therefore forcing firms to focus on their core competences. These results are consistent with the findings of Mayer et al. (2014) for France and Carballo et al. (2013) for Costa Rica, Ecuador and Uruguay. Given the importance of carry-along trade in the economy, we run a similar regression while controlling for whether a firm actually produces the core product it exports. The estimation results are reported in columns (3) and (4). They show that distance, market size, income level and firm export scope have the same effects as above, except the insignificance of the trade agreement variable. We also see that whether or not firms actually produce their core export product matters when it comes to how large of a role it plays in their export portfolio. 4 Export growth decomposition We first decompose aggregate growth into intensive and extensive margins. As usual, we define the intensive margin as changes in export coming from incumbent exporters, and the extensive margin as changes in export resulting from entry of new exporters and exit of previously exporting firms, so that the growth rate of exports can be disaggregated as follows: X X X ∆Et = (Ei,t − Ei,t−1 ) + Ei,t − Ei,t−1 (2) i∈Ct i∈Nt i∈Xt where Ei,t refers to export of firm i at time period t, and Ct , Nt and Xt denote a group of incumbent, entering and exiting firms, respectively. The first term in equation (2) denotes the intensive margin, the second term shows an increase in export due to new entrants into exporting, and the last term sums the decline in export coming from firms exiting export markets. Extension of the decomposition to product and product-destination combinations is straightforward. Table 8 shows the dominant role of the intensive margin in the export trade. These features indicate that the main drivers of trade are incumbent firms, products and product-destination relationships. Additionally, our results show a positive and significant recovery of exports (7.42%), mostly driven by the intensive margin. Note that the two margins of trade do not always reinforce each other. For instance, unlike the 2008-2009 economic downturn where the entry and exit of firms, products and product-destination relationships reinforced the contraction in the scale of existing trade relations, the extensive margin helped to attenuate the negative export shock during 2002-2003. During the 2009-2010 period, the introduction of new products and product-destinations helped the recovery of exports. However, the recovery process was hindered by the exit of trading firms. This suggests that shocks have lasting effects on the economy in terms of firm exits and closures. 9 A small role of entry and exit of firms or products at the country level is consistent with previous studies which show that new exporters are small and therefore constitute a small share of exports. What is new to a particular firm is less likely to be new for the economy, and even if it is new, its share is mostly small. Given the multi-product and multi-destination nature of firm trading activities, firms’ decisions to adjust their product and market mix are important mechanisms to understand how they adapted to the crisis. Therefore, we extend our decomposition exercise and consider the different margins of trade adjustment at the firm level. This decomposition exercise here is equivalent to equation (2) except that only incumbent firms and their scale and scope decisions in terms of product and productdestination dimensions are considered. Following Gopinath and Neiman (2014), we define the subintensive margin as the growth rate due to changes in value of existing firm-product relationships whereas the sub-extensive margin refers to the growth rate due to continuing firms’ decisions to terminate existing product relationships or introduce new ones. X X X ∆Et = Ei,t − Ei,t−1 + Ei,t − Ei,t−1 (3) i∈Ct i∈Nt i∈Xt where Ei,t refers to firm-product export relationships i at time period t; Ct , Nt and Xt denote a set of maintained, newly added or dropped export products by incumbent firms, respectively. Again, extension to firm-product-destination combinations is straightforward. Table 9 shows that the sub-intensive margin still constitutes the most important export adjustment mechanism. However, unlike in the case of economy-wide churning of products and product-destination combinations, adjusting product and product-destinations mix at the firm level played a non-negligible role in overall export growth. In addition to attenuating the negative effects of the trade shocks during 2002-2003 and 2008-2009 recessions, export growth coming from newly added products, and especially newly added product-destinations, was particularly important in stimulating the recovery in the aftermath of the crises. In another exercise, we decompose export growth into components coming from changes in scale of export by continuing exporters, export starters and export quitters as in equation (2). Having established the prominent role of continuing exporters as primary driving forces, we then decompose export changes due to continued, newly added or dropped market destination for incumbent exporting firms as in Amador and Opromolla (2013). X X X ∆Et = (4) ∆Ei,t + ∆Ei,t + ∆Ei,t−1 i∈Ct i∈Nt i∈Xt where the additional decompositions are given by: " # X X X X X ∆Ei,t = ∆Ej,i,t + ∆Ej,i,t + ∆Ej,i,t i∈Ct i∈Ct j∈ADt j∈DDt " X j∈CDt ∆Ej,i,t = (5) j∈CDt # X X j∈CDt k∈APt ∆Ek,j,i,t + X k∈DPt 10 ∆Ek,j,i,t + X k∈CPt ∆Ek,j,i,t (6) Equations (4)-(6) summarize the steps involved in the decomposition in which E refers to export flow depending on whether firm i at time t is entering Nt , exiting Xt or incumbent Ct ; then exports by incumbent firms to different destinations depend on whether each destination j is newly added ADt , dropped DDt or continued CDt , and finally export changes to continued destinations depend on whether the exported product k is newly added APt , dropped DPt or continued CPt , respectively. Results of this decomposition exercise are shown in Table 10. As discussed previously, we see that continuing firms are more important than entering and exiting firms in the export trade. Among incumbent exporting firms, continued export destinations are more important than newly served ones highlighting the minimal effect of destination switching on export growth. For continued destinations, most of the growth comes from the changes in the export value of previously exported products, suggesting the existence of core and peripheral export products. This suggests that firms’ adjustment in their product-market portfolio might have helped them to mitigate the export collapse in continued market destinations during the 2002-2003 and 2008-2009 recessions. 5 Export diversification and growth In this section, we want to investigate how firms react to changes in the economic environment by adjusting their product mix and export destinations conditional on survival. Such export diversification decisions at the firm level usually affect the trajectory of firm export sales over time. To better understand these decisions, we start by plotting the export shares of destinations against destination-specific export growth as depicted in Figure 2. We see a negative relationship implying that the main export market destinations (mostly neighboring countries like Germany, Sweden and Norway which are affected by similar shocks) experienced slower export growth rates. This suggests that diversification into other markets could potentially improve export sales over time. We focus our attention on specific high-growth nontraditional markets namely China and more generally the BRICS. Interestingly, despite a stable or even declining number of exporting firms in the economy, a growing number of Danish firms are indeed exporting to these markets as shown in Figure 3. As a next step, we run a simple least squares regression of export growth in which nontraditional markets—China and the BRICS—are explicitly considered: M ulti−product Destination ∆Ei,t =β0 + β1 Di,t−1 + β2 Di,t−1 + β3 log(#Employees)i,t−1 (7) + β4 log(#Destinations)i,t−1 + i,t M ulti−product where ∆Ei,t = log(Ei,t ) − log(Ei,t−1 ) refers to firm i’s export growth at time period t, Di,t−1 Destination is a dummy if the firm is multi-product, Di,t−1 if the firm exports to particular markets of interest such as China and the BRICS, Employees and Destinations refer to employment size 11 and export market scope of the firm, respectively.10 We run the regression for three different time periods: period of trade collapse—2008-2009, period of trade collapse and recovery—2008-2010, and entire sample period—2000-2010.11 Table 11 shows that in the midst of the economic crisis, firms that were serving the Chinese market had higher export growth, after controlling for firm size, multi-product and multi-destination aspects of their export activity. The effect is not significant when we instead consider the BRICS as a whole. We also find that larger firms apparently sustained the shock better. After controlling for size, more diversified firms in terms of product and markets destinations had lower export growth. This can be seen as a mechanical effect, as firms involved in more destinations or selling more products are more likely to be exporting to markets which were also affected by the crisis. This implies that individual markets where firms are selling matter. During our extended sample periods 2008-2010 and 2000-2010, our main finding remains valid: export growth rates were larger for those firms exporting to China and the result extends to the BRICS. The estimates on multi-product, number of market destinations, and employment size are also very similar. We then run a similar regression at the firm-product level in which products are now indexed by j: M ulti−product Destination + β2 Di,j,t−1 ∆Ei,j,t =β0 + β1 Di,t−1 + β3 log(#Employees)i,t−1 (8) Core−product + β4 log(#Destinations)i,j,t−1 + β5 Di,j,t−1 + i,j,t Table 12 shows that export presence in China and the BRICS had a positive effect on growth over the whole period. It does not appear to have played an important role during the crisis, but both the China and BRICS effects are positive and significant over the 2008-2010 period. Another interesting finding is that firms’ core product experienced a negative shock both in the short and medium run. This suggests that firms that were too dependent on one product had lower export growth. Controlling for these factors, we also observe a significant and negative correlation between product export growth and export diversification in terms of markets served and number of products exported both in the short and medium run. We see these variables as important controls, and the effect can be seen as mechanical as discussed before. Similarly, larger firms are more likely to have faster product export growth over the whole period, although size did not help during the crisis and the recovery. 10 We find similar results in all of the regressions when mid-point growth rates are used. In this paper, we are mostly interested in the correlations between export growth and market presence. We do not explicatively consider the selection process in high growth markets. 11 12 6 Conclusion This paper uses a rich dataset from Denmark to examine the microeconomic foundations of international trade, and especially how firms adjust their behavior during a period of large shock. We find that there are significant cross-sector and cross-industry variations in firm export participation rates. Interestingly, broadening of the export activities is accompanied by deepening in that we find a positive correlation between the scope and the scale of firm export activity. Additionally, we document the dominance and increasing importance of a few superstar firms with broader export scope. Highly uneven distribution of export sales within firms provides evidence of the existence of core and peripheral products in the export portfolio of firms. This unevenness systematically varies with destination market characteristics such as bilateral distances, economic size and wealth. We also show that firms export products that they do not actually produce indicating the existence of carry-along trade. Despite the pervasiveness of carry-along trade, the export share of this trade is very small, albeit increasing over the last decade. In view of the current economic crisis, Danish firms reacted differently. Some firms quit exporting. Others changed the mix of products they export and the market destinations they serve, and this somehow helped them to mitigate the negative shocks of the crisis. However, the most important adjustment to cope with the crisis came from a few important firms reducing the shipment of their core products to major market destinations. Our findings show a positive and significant recovery of exports in the aftermath of the crisis, mostly through the intensive margin, further establishing the fact that trade performance is driven by a few superstar firms, their core products, and economic conditions of their main trading partners. We also analyze how firms reacted to a large shock by diversifying their product-destination portfolio. We stress the important role played by the intensive margin of trade. We show that export diversification into fast-growing economies like China or the BRICS was associated with better export performance. We also document that too much reliance on one core product was harmful to growth. Our results mostly display interesting cross-sectional correlations between export diversification and growth. In future work, we would like to better understand how some firms managed to successfully enter these high-growth markets; that is, introducing more dynamics and addressing the endogeneity of the diversification process. 13 References Alvarez, R., & L´opez, R. A. (2005). Exporting and Performance: Evidence from Chilean Plants. Canadian Journal of Economics, 38 (4), 1384–1400. Amador, J., & Opromolla, L. D. (2013). Product and Destination Mix in Export Markets. Review of World Economics, 149 (1), 23–53. Amiti, M., & Davis, D. R. (2012). Trade, Firms, and Wages: Theory and Evidence. Review of Economic Studies, 79 (1), 1–36. Andersson, M., L¨o¨of, H., & Johansson, S. (2008). Productivity and International Trade: Firm Level Evidence from a Small Open Economy. Review of World Economics, 144 (4), 774–801. Arkolakis, C., & Muendler, M. A. (2010). The Extensive Margin of Exporting Products: A Firm-level Analysis. NBER Working Paper #16641 . Behrens, K., Corcos, G., & Mion, G. (2013). Trade crisis? What Trade Crisis? Review of Economics and Statistics, 95 (2), 702–709. Bernard, A. B., Blanchard, E. J., Van Beveren, I., & Vandenbussche, H. Y. (2012). Carry-along Trade. NBER Working Paper #18246 . Bernard, A. B., Eaton, J., Jensen, J. B., & Kortum, S. (2003). Plants and Productivity in International Trade. American Economic Review , 93 (4), 1268–1290. Bernard, A. B., & Jensen, J. B. (1995). Exporters, Jobs, and Wages in U.S. Manufacturing: 1976-1987. Brookings Papers on Economic Activity. Microeconomics, 1995 , 67–119. Bernard, A. B., Jensen, J. B., Redding, S. J., & Schott, P. K. (2007). Firms in International Trade. Journal of Economic Perspectives, 21 (3), 105–130. Bernard, A. B., Jensen, J. B., Redding, S. J., & Schott, P. K. (2009). The Margins of U.S. Trade. American Economic Review , 99 (2), 487–493. Bernard, A. B., Redding, S. J., & Schott, P. K. (2010). Multiple-Product Firms and Product Switching. American Economic Review , 100 (1), 70–97. Bernard, A. B., Redding, S. J., & Schott, P. K. (2011). Multi-product Firms and Trade Liberalization. Quarterly Journal of Economics, 126 (3), 1271–1318. Bricongne, J. C., Fontagn´e, L., Gaulier, G., Taglioni, D., & Vicard, V. (2012). Firms and the Global Crisis: French Exports in the Turmoil. Journal of International Economics, 87 (1), 134–146. Carballo, J., Ottaviano, G. I., & Volpe Martincus, C. (2013). The Buyer Margins of Firms’ Exports. CEP Discussion Paper #1234 . Castellani, D., Serti, F., & Tomasi, C. (2010). Firms in International Trade: Importers and Exporters Heterogeneity in Italian Manufacturing Industry. World Economy, 33 (3), 424– 457. De Loecker, J. (2007). Do Exports Generate Higher Productivity? Evidence from Slovenia. Journal of International Economics, 73 (1), 69–98. Eckel, C., & Neary, J. P. (2010). Multi-product Firms and Flexible Manufacturing in the Global Economy. Review of Economic Studies, 77 (1), 188–217. Eriksson, T., Smeets, V., & Warzynski, F. (2009). Small Open Economy Firms in International Trade: Evidence from Danish Transactions-level Data. Nationaløkonomisk Tidsskrift/Danish Economic Journal , 147 (2), 175–194. Gopinath, G., & Neiman, B. (2014). Trade Adjustment and Productivity in Large Crises. American Economic Review , 104 (3), 793–831. Isgut, A. (2001). What’s Different About Exporters? Evidence from Colombian Manufacturing. Journal of Development Studies, 37 (5), 57–82. 14 Kasahara, H., & Lapham, B. (2013). Productivity and the Decision to Import and Export: Theory and Evidence. Journal of International Economics, 89 (2), 297–316. M´an ˜ez-Castillejo, J. A., Rochina-Barrachina, M. E., & Sanchis-Llopis, J. A. (2010). Does Firm Size Affect Self-selection and Learning-by-exporting? World Economy, 33 (3), 315–346. Mayer, T., Melitz, M. J., & Ottaviano, G. I. P. (2014). Market Size, Competition, and the Product Mix of Exporters. American Economic Review , 104 (2), 495–536. Mayer, T., & Ottaviano, G. I. (2008). The Happy Few: The Internationalisation of European Firms. Intereconomics, 43 (3), 135–148. Melitz, M. J. (2003). The Impact of Trade on Intra-industry Reallocations and Aggregate Industry Productivity. Econometrica, 71 (6), 1695–1725. Muˆ uls, M., & Pisu, M. (2009). Imports and Exports at the Level of the Firm: Evidence from Belgium. World Economy, 32 (5), 692–734. Ruane, F., & Sutherland, J. (2005). Export Performance and Destination Characteristics of Irish Manufacturing Industry. Review of World Economics, 141 (3), 442–459. Van Biesebroeck, J. (2005). Exporting Raises Productivity in Sub-Saharan African Manufacturing Firms. Journal of International economics, 67 (2), 373–391. Verardi, V., & Wagner, J. (2012). Productivity Premia for German Manufacturing Firms Exporting to the Euro-area and Beyond: First Evidence from Robust Fixed Effects Estimations. World Economy, 35 (6), 694–712. 15 16 1,115 6,997 Furniture Health & social work 546 Transport equipment 54,345 192,508 Others Total 670 1,011 Textiles & wearing apparel Wood & wood products 614 Rubber & plastic products 42,323 2,556 Paper, publishing & printing Wholesale & retail trade 663 Non-metallic mineral products 12,323 168 Mining & quarrying Transport, storage & communication 2,066 Machinery & equipment 80 1,712 Food, beverage & tobacco products Leather & leather products 1,618 Electrical & optical equipment 299 Chemicals & chemical products 23,983 3,579 Basic & fabricated metals Construction 35,840 # Firms Agriculture, forestry & fishing Industry 6.74% 1.67% 26.0% 15.9% 2.77% 32.6% 35.2% 52.4% 14.3% 24.9% 17.9% 40.9% 37.5% 0.10% 32.3% 16.4% 36.9% 0.89% 59.9% 18.0% 0.66% % Exporters 2000 225,516 70,015 552 40,485 21,774 260 734 519 1,101 537 162 1,491 60 14,298 353 1,496 844 31,260 253 3,161 35,161 # Firms 5.75% 1.77% 25.5% 17.0% 2.79% 47.3% 30.8% 59.2% 23.0% 24.0% 20.4% 51.2% 31.7% 0.10% 58.9% 19.4% 52.6% 0.64% 63.2% 20.1% 0.67% % Exporters 2008 220,915 75,052 523 38,547 21,029 254 657 493 1,028 506 156 1,438 56 14,375 359 1,447 822 28,616 249 2,939 32,369 # Firms 5.65% 1.59% 24.9% 17.3% 2.67% 47.6% 32.1% 59.0% 22.6% 25.7% 20.5% 51.3% 32.1% 0.06% 54.9% 20.1% 55.2% 0.69% 63.1% 20.7% 0.78% % Exporters 2009 Table 1: Industry-wise export participation rate of firms with at least one employee 22,951 78,076 487 38,577 21,688 249 618 483 975 485 152 1,442 57 15,078 355 1,423 836 27,341 242 2,859 31,528 # Firms 5.73% 1.63% 25.9% 17.7% 2.78% 49.0% 30.3% 59.6% 22.7% 24.3% 22.4% 51.9% 24.6% 0.11% 52.4% 21.0% 55.1% 0.66% 61.6% 22.0% 0.92% % Exporters 2010 Table 2: Summary on exporting firms, exported products and export destinations # # # # # # # # # # Exporting Firms Destinations: Economy-wide Destinations: Per firm, median Destinations: Per firm, maximum Exported products: Economy-wide Exported products: Per firm, median Exported products: Per firm, maximum Exported product-destinations: Economy-wide Exported product-destinations: Per firm, median Exported product-destinations: Per firm, maximum 2000 2008 2009 2010 12,973 217 2 129 4,435 3 472 78,150 4 2,257 12,919 225 2 133 4,579 4 1,432 106,215 5 8,274 12,490 225 2 143 4,547 4 1,277 104,175 5 7,813 12,782 227 2 151 4,525 4 1,438 108,978 5 9,381 Table 3: Summary on export value in million DKK 2000 2008 2009 2010 Export value per firm Mean Median Maximum 19.70 0.45 13,302.34 24.94 0.42 14,006.64 21.43 0.41 12,369.76 22.50 0.43 12,072.40 Export value per product Mean Median Maximum 57.62 3.01 7,268.63 70.35 3.53 13,814.81 58.87 3.02 7,932.45 63.55 3.32 9,179.54 Export value per destination Mean Median Maximum 1,177.65 31.11 49,355.19 1,431.76 26.41 53,429.89 1,189.73 24.65 45,777.98 1,266.71 30.31 47,330.82 17 Table 4: Product and destination scope of exporting firms # Products 1 2 Share of exporting firms Value share of exporting firms # Destinations # Destinations 3 4 5 5+ All 1 2 3 4 5 5+ All 0.39 0.28 0.18 0.08 0.06 0.20 1.17 0.23 0.28 0.14 0.14 0.06 0.39 1.23 0.27 0.19 0.25 0.12 0.05 0.58 1.46 1.36 0.10 0.15 0.15 0.05 1.45 3.26 0.05 0.09 0.14 0.07 0.14 0.65 1.15 0.82 1.77 1.65 1.75 1.91 83.82 91.72 3.13 2.71 2.50 2.30 2.28 87.08 100 0.39 0.12 0.17 0.05 0.05 0.48 1.25 0.22 0.33 0.20 0.13 0.07 0.49 1.43 0.11 0.41 0.08 0.18 0.09 1.04 1.91 0.13 0.12 0.09 0.13 0.16 1.22 1.85 0.43 0.03 0.09 0.23 0.06 0.75 1.59 0.35 1.21 1.04 1.04 1.25 86.82 91.97 1.63 2.22 1.67 1.97 1.72 90.79 100 Panel a. Year 2000 1 2 3 4 5 5+ All 24.06 5.63 2.61 1.15 0.58 1.66 35.67 2.77 5.85 2.93 1.61 0.90 2.30 16.36 1.13 1.73 2.13 1.38 0.82 2.56 9.74 0.49 0.59 0.79 0.91 0.56 2.36 5.71 0.31 0.42 0.49 0.42 0.51 2.10 4.25 0.88 1.39 1.68 1.77 1.78 20.76 28.26 29.64 15.61 10.62 7.24 5.15 31.74 100 Panel b. Year 2010 1 2 3 4 5 5+ All 22.01 6.17 2.92 1.46 0.92 2.29 35.78 1.78 5.02 3.00 1.85 1.24 3.66 16.55 0.47 1.18 1.62 1.47 1.08 3.74 9.56 0.26 0.44 0.55 0.78 0.66 3.38 6.07 0.12 0.19 0.29 0.38 0.34 2.18 3.49 0.44 0.78 0.98 1.13 1.31 23.90 28.55 25.07 13.79 9.36 7.07 5.55 39.16 100 18 19 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Year 1.15 1.14 1.10 1.03 1.04 1.06 1.04 1.02 1.01 0.97 0.98 Mean 1.68 1.99 0.72 0.59 0.56 0.64 2.53 2.38 5.50 2.76 2.25 1 1 1 1 1 1 1 1 1 1 1 Median Non-traded 1.72 1.78 1.87 1.92 1.91 1.92 2.01 1.93 1.85 1.89 1.91 Median Mean Median Produced 6.26 6.63 8.49 9.08 8.58 8.25 9.92 10.26 9.54 9.93 10.11 2 2 3 3 3 3 4 4 3 3 4 1.54 1.60 1.67 1.72 1.70 1.71 1.77 1.70 1.62 1.67 1.67 1 1 1 1 1 1 1 1 1 1 1 Panel a. Number of products (HS 6-digits) Mean Non-produced Export 98.32 98.01 99.28 99.41 99.44 99.36 97.47 97.62 94.50 97.24 97.75 3.19 3.14 2.71 3.66 3.38 4.28 3.86 3.76 4.83 3.83 5.20 96.81 96.86 97.29 96.34 96.12 95.72 96.14 96.24 95.17 96.17 94.80 Panel b. Share in value of production, export and import 1 1 1 1 1 1 1 1 1 1 1 Median Traded Mean. Production 10.66 11.23 14.14 14.35 14.81 14.85 15.98 16.72 16.14 14.56 15.95 2 3 5 5 5 5 6 6 5 4 4 1.05 1.14 1.24 1.28 1.28 1.29 1.38 1.33 1.27 1.26 1.27 79.32 80.25 82.71 81.91 80.53 81.52 83.01 83.30 82.55 85.23 82.99 0 0 0 0 0 0 0 0 0 0 0 Median Produced Mean Import Median 20.68 19.75 17.29 18.09 19.47 18.48 16.99 16.70 17.45 14.77 17.01 Mean Non-produced Table 5: Number of products produced and traded by firms in manufacturing sector Table 6: Within firm distribution of export sales 2000 2010 Scope Top 1 Top 2 Top 3 Top 4 Top 5 Top 1 Top 2 Top 3 Top 4 Top 5 1 2 3 4 5 6 7 8 9 10 10+ 100 78.51 71.05 67.21 65.17 63.07 61.61 62.03 58.52 58.22 42.90 – 100 91.41 86.81 83.96 81.64 80.21 79.57 77.68 76.54 58.84 – – 100 95.64 92.83 90.36 88.89 87.94 86.96 85.79 67.71 – – – 100 97.49 95.26 93.79 92.73 91.95 91.06 73.57 – – – – 100 98.24 96.89 95.85 94.98 94.50 77.76 100 78.40 70.78 66.56 63.02 61.86 60.07 61.15 58.35 58.35 39.18 – 100 91.71 86.48 83.24 81.30 79.22 78.60 76.85 75.53 54.08 – – 100 95.78 92.79 90.44 88.44 87.60 85.80 84.51 62.56 – – – 100 97.67 95.55 93.72 92.65 91.30 89.98 68.22 – – – – 100 98.47 96.90 95.81 94.78 93.51 72.27 Mean 52.42 67.45 74.80 79.13 82.07 45.98 60.57 68.14 72.82 75.95 20 Table 7: Core product export share and gravity variables Share of the core product (1) (2) (3) (4) Distance 0.013∗∗∗ (34.68) 0.016∗∗∗ (27.76) 0.004∗∗∗ (8.00) 0.016∗∗∗ (27.71) GDP 0.003∗∗∗ (16.91) 0.004∗∗∗ (17.45) 0.001∗ (2.18) 0.004∗∗∗ (17.36) GDP per capita -0.021∗∗∗ (-53.09) -0.022∗∗∗ (-53.61) -0.017∗∗∗ (-32.96) -0.022∗∗∗ (-53.85) # Exported products -0.018∗∗∗ (-76.58) -0.018∗∗∗ (-76.58) -0.022∗∗∗ (-55.61) -0.018∗∗∗ (-76.61) DOwn−product - - 0.049∗∗∗ (52.69) 0.049∗∗∗ (52.71) RTA - 0.007∗∗∗ (6.06) - -0.001 (-1.02) Year FE Yes Yes Yes Yes Adj. R2 0.325 0.325 0.294 0.294 N 880,976 471,912 t statistics in parentheses, ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 21 Table 8: Economy-wide decomposition of export growth Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 – – – – – – – – – – 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Firm Growth 3.50 3.99 -1.34 3.19 4.83 5.13 4.23 0.15 -16.90 7.42 Product Product-destination Intensive Extensive Intensive Extensive Intensive Extensive 5.85 4.39 -2.77 3.54 7.95 7.56 2.85 2.75 -13.35 7.92 -2.35 -0.40 1.43 -0.35 -3.12 -2.44 1.38 -2.60 -3.55 -0.51 3.12 4.23 -1.44 3.18 4.64 5.09 5.61 0.12 -16.67 6.57 0.38 -0.23 0.10 0.01 0.18 0.04 -1.38 0.03 -0.24 0.85 3.13 3.43 -2.78 2.53 4.35 4.06 4.67 -0.80 -15.94 6.19 0.37 0.57 1.44 0.67 0.47 1.07 -0.44 0.95 -0.96 1.23 Table 9: Firm-level decomposition of export growth Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 – – – – – – – – – – 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Firm-product Growth 3.50 3.99 -1.34 3.19 4.83 5.13 4.23 0.15 -16.90 7.42 Firm-product-destination Sub-intensive Sub-extensive Extensive Sub-intensive Sub-extensive Extensive 5.20 3.33 -3.66 3.10 7.12 6.47 3.07 4.02 -14.11 5.06 0.65 1.06 0.89 0.43 0.83 1.10 -0.21 -1.27 0.76 2.86 -2.35 -0.40 1.43 -0.35 -3.12 -2.44 1.38 -2.60 -3.55 -0.51 3.49 2.29 -4.67 1.40 6.46 5.30 2.21 1.05 -12.21 3.47 2.36 2.10 1.90 2.14 1.49 2.26 0.64 1.70 -1.14 4.46 -2.35 -0.40 1.43 -0.35 -3.12 -2.44 1.38 -2.60 -3.55 -0.51 22 Table 10: Export growth decomposition: destination and product margins Firms Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 – – – – – – – – – – 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Destinations Products Growth 3.50 3.99 -1.34 3.19 4.83 5.13 4.23 0.15 -16.90 7.42 Continuing Entering Exiting 5.85 4.39 -2.77 3.54 7.95 7.56 2.85 2.75 -13.35 7.92 3.51 4.16 4.89 2.64 2.52 2.16 3.58 1.13 1.31 1.95 5.86 4.56 3.46 2.99 5.64 4.60 2.21 3.72 4.86 2.46 Continued Added dest. dest. 5.86 4.39 -2.7 3.56 7.92 7.59 2.85 2.75 -13.35 7.92 23 0.00 0.00 0.00 0.07 0.03 0.00 0.00 0.00 0.00 0.00 Dropped dest. 0.01 0.00 0.00 0.08 0.00 0.02 0.00 0.00 0.00 0.00 Continued Added prod. prod. 5.54 4.67 -2.81 3.55 7.73 7.54 4.09 2.64 -13.12 7.07 0.40 4.76 0.11 0.03 0.21 0.06 5.69 0.15 0.02 1.26 Dropped prod. 0.08 5.04 0.07 0.02 0.02 0.01 6.93 0.04 0.25 0.40 Table 11: Export growth: firm-level 2008-2009 2008-2010 2000-2010 Dep. Variable (1) (2) (3) (4) (5) (6) DM ulti−product -0.582*** (-15.12) -0.583*** (-15.12) -0.583*** (-21.07) -0.581*** (-20.96) -0.517*** (-35.58) -0.513*** (-35.30) log(# Destinations) -0.104*** (-6.91) -0.106*** (-6.54) -0.097*** (-8.90) -0.110*** (-9.34) -0.094*** (-20.39) -0.109*** (-21.12) log(# Employees) 0.056*** (5.44) 0.058*** (5.64) 0.059*** (7.72) 0.060*** (8.00) 0.064*** (18.10) 0.066*** (18.45) DChina 0.100*** (2.32) - 0.093*** (2.97) - 0.084*** (6.74) - DBRICS - 0.069 (1.84) - 0.112*** (4.12) - 0.115*** (10.00) Industry FE Year FE Yes No Yes No Yes Yes Yes Yes Yes Yes Yes Yes Adj. R2 0.04 0.04 0.05 0.05 0.05 0.04 Obs. 10,245 20,497 t statistics in parentheses, * p < 0.05, ** p < 0.01, *** p < 0.001 24 108,016 Table 12: Export growth: firm-product-level 2008-2009 2008-2010 2000-2010 Dep. Variable (1) (2) (3) (4) (5) (6) DM ulti−product -0.192∗∗ (-3.10) -0.191∗∗∗ (-3.09) -0.249∗∗∗ (-5.86) -0.248∗∗∗ (-5.84) -0.166∗∗∗ (-11.51) -0.166∗∗∗ (-11.51) log(# Destinations) -0.228∗∗∗ (-35.27) -0.228∗∗∗ (-33.22) -0.207∗∗∗ (-45.17) -0.210∗∗∗ (-43.69) -0.201∗∗∗ (-98.19) -0.208∗∗∗ (-94.37) log(# Employees) -0.002 (-0.43) -0.001 (-0.36) -0.000 (-0.03) 0.000 (0.06) 0.013∗∗∗ (10.08) 0.013∗∗∗ (10.09) DCore−product -0.331∗∗∗ (-14.52) -0.329∗∗∗ (-14.46) -0.319∗∗∗ (19.86) -0.318∗∗∗ (-19.84) -0.277∗∗∗ (-48.60) -0.277∗∗∗ (-48.72) DChina 0.046 (1.49) - 0.073∗∗∗ (3.48) - 0.107∗∗∗ (10.13) - DBRICS - 0.013 (0.62) - 0.057∗∗∗ (3.79) - 0.104∗∗∗ (14.71) Industry FE Year FE Yes No Yes No Yes Yes Yes Yes Yes Yes Yes Yes Adj. R2 0.02 0.02 0.02 0.02 0.02 0.02 Obs. t statistics in parentheses, 76,192 ∗ 153,232 p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 25 698,860 16.4 2000 2002 2004 Year 2006 2008 2010 2000 2002 2004 Year 2006 2008 Number of exporting firms Average export value (log scale) Number of importing firms Average import value (log scale) 2010 14.8 60,000 14.9 70,000 15 80,000 15.1 90,000 15.2 100,000 15.3 110,000 12,000 14,000 16.6 16,000 18,000 16.8 20,000 17 22,000 Figure 1: Scope and scale of exporting and importing activities 2000 2002 2004 Year 2006 2008 2010 2000 2002 2004 Year 2006 2008 Number of export transactions Average export value (log scale) Number of import transactions Average import value (log scale) Note: A transaction refers to the number of products (HS−6) exported to/imported from a destination/source country. 26 2010 Figure 2: Export share of trade partners and export growth −6 −4 −4 −2 Export growth (log scale) −2 0 Export growth (log scale) 0 2 2 4 2010 4 2000 −15 −10 −5 −20 0 Export share (log scale) Actual values −15 −10 −5 Export share (log scale) Fitted values 27 0 600 3,000 800 3,500 1,000 4,000 1,200 4,500 1,400 1,600 5,000 Figure 3: Number of firms exporting to China and BRICS 2000 2002 2004 2006 2008 Year China (Left scale) BRICS (Right scale) 28 2010 Appendix: Tables Table A.1: Economy-wide evolution of active firms in production Producers Entering Exiting Net entry Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 # Firms # Firms % Sh. firms # Firms % Sh. firms # Firms % Sh. firms 192,508 193,039 194,152 186,749 222,698 231,270 236,112 239,685 224,516 220,915 222,951 – 27,283 27,452 23,201 57,264 36,728 36,879 35,717 31,550 34,282 34,951 – 14.13% 14.14% 12.42% 25.71% 15.88% 15.62% 14.90% 14.05% 15.52% 15.68% – 26,752 26,339 30,604 21,315 28,156 32,037 32,144 46,719 37,883 32,915 – 13.86% 13.57% 16.39% 9.57% 12.17% 13.57% 13.41% 20.81% 17.15% 14.76% – 531 1,113 -7,403 35,949 8,572 4,842 3,573 -15,169 -3,601 2,036 – 0.28% 0.57% -3.96% 16.14% 3.71% 2.05% 1.49% -6.76% -1.63% 0.91% Table A.2: Economy-wide evolution of exporting firms Exporters Entering Exiting Net entry Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 # Firms # Firms % Sh. firms # Firms % Sh. firms # Firms % Sh. firms 12,973 13,133 13,843 14,147 13,839 13,556 13,830 13,478 13,919 12,490 12,782 – 2,714 3,064 2,862 2,558 2,510 2,804 2,441 2,273 2,245 2,530 – 20.67% 22.13% 20.23% 18.48% 18.52% 20.27% 18.11% 17.59% 17.97% 19.79% – 2,554 2,354 2,558 2,866 2,793 2,530 2,793 2,832 2,674 2,238 – 19.45% 17.00% 18.08% 20.71% 20.60% 18.29% 20.72% 21.92% 21.41% 17.51% – 160 710 304 -308 -283 274 -352 -559 -429 292 – 1.22% 5.13% 2.15% -2.23% -2.09% 1.98% -2.61% -4.33% -3.43% 2.28% 29 Table A.3: Evolution of active firms in production in manufacturing sector Producers Entering Exiting Net entry Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 # Firms # Firms % Sh. firms # Firms % Sh. firms # Firms % Sh. firms 4,186 4,105 4,049 3,956 3,795 3,667 3,580 3,340 3,514 3,331 3,127 – 368 328 297 258 238 229 284 383 228 196 – 8.96% 8.10% 7.51% 6.80% 6.49% 6.40% 8.50% 10.90% 6.84% 6.27% – 449 384 390 419 366 316 524 209 411 400 – 10.94% 9.48% 9.86% 11.04% 9.98% 8.83% 15.69% 5.95% 12.34% 12.79% – -81 -56 -93 -161 -128 -87 -240 174 -183 -204 – -1.97% -1.38% -2.35% -4.24% -3.49% -2.43% -7.19% 4.951% -5.49% -6.52% Table A.4: Evolution of exporting firms in manufacturing sector Exporters Entering Exiting Net entry Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 # Firms # Firms % Sh. firms # Firms % Sh. firms # Firms % Sh. firms 3,014 2,986 3,042 2,989 2,847 2,737 2,738 2,498 2,569 2,494 2,404 – 363 393 316 252 243 283 244 304 224 237 – 12.16% 12.92% 10.57% 8.85% 8.88% 10.34% 9.77% 11.83% 8.98% 9.86% – 391 337 369 394 353 282 484 233 299 327 – 13.09% 11.08% 12.35% 13.84% 12.90% 10.30% 19.38% 9.07% 11.99% 13.60% – -28 56 -53 -142 -110 1 -240 71 -75 -90 – -0.94% 1.84% -1.77% -4.99% -4.02% 0.04% -9.61% 2.76% -3.01% -3.74% 30 Table A.5: Economy-wide evolution of importing firms Importers Entering Exiting Net entry Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 # Firms # Firms % Sh. Firms # Firms % Sh. firms # Firms % Sh. firms 15,205 15,153 17,333 17,996 18,962 19,346 19,766 20,481 20,090 18,330 18,886 – 2,987 4,608 4,117 4,513 4,476 4,288 4,662 4,270 3,582 4,515 – 19.71% 26.59% 22.88% 23.80% 23.14% 21.69% 22.76% 21.25% 19.54% 23.91% – 3,039 2,428 3,454 3,547 4,092 3,868 3,947 4,661 5,342 3,959 – 20.06% 14.01% 19.19% 18.71% 21.15% 19.57% 19.27% 23.20% 29.14% 20.96% – -52 2,180 663 966 384 420 715 -391 -1,760 556 – -0.34% 12.58% 3.68% 5.09% 1.98% 2.12% 3.49% -1.95% -9.60% 2.94% Table A.6: Evolution of importing firms in manufacturing sector Importers Entering Exiting Net entry Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 # Firms # Firms % Sh. firms # Firms % Sh. firms # Firms % Sh. firms 2,766 2,768 2,922 2,873 2,756 2,647 2,634 2,457 2,548 2,376 2,243 – 383 468 328 286 271 275 293 326 211 233 – 13.84% 16.02% 11.42% 10.38% 10.24% 10.44% 11.93% 12.79% 8.88% 10.39% – 381 314 377 403 380 288 470 235 383 366 – 13.76% 10.75% 13.12% 14.62% 14.36% 10.93% 19.13% 9.22% 16.12% 16.32% – 2 154 -49 -117 -109 -13 -177 91 -172 -133 – 0.07% 5.27% -1.71% -4.25% -4.12% -0.49% -7.20% 3.57% -7.24% -5.93% 31 32 534 1,059 407 2,571 *** 903 47 186 722 305 281 194 2,103 8,472 276 Electrical & optical equipment Food, beverage & tobacco products Furniture Health & social work Leather & leather products Machinery & equipment Mining & quarrying Non-metallic mineral products Paper, publishing & printing Rubber & plastic products Textiles & wearing apparel Transport equipment Transport, storage & communication Wholesale & retail trade Wood & wood products 20.2% 4.30% 47.5% 36.1% 10.7% 62.4% 79.0% 77.0% 36.7% 53.2% 51.1% 71.3% 94.1% 0.58% 68.8% 22.7% 71.7% 3.14% 87.2% 43.3% 8.51% % Exporters 2000 33,775 11,670 192 7,692 2,933 120 137 267 292 172 47 669 *** 2,221 221 914 358 4,066 125 977 697 # Firms 20.1% 5.25% 53.1% 37.4% 13.6% 78.3% 82.5% 85.0% 62.3% 51.2% 57.4% 84.3% 100% 0.27% 73.8% 27.5% 83.2% 3.10% 92.8% 48.4% 10.2% % Exporters 2008 Note: *** indicates the number is intentionally unreported to keep the identity of firms anonymous. 36,116 4,205 Construction Total 148 Chemicals & chemical products 11,499 1,129 Basic & fabricated metals Others 1,058 # Firms Agriculture, forestry & fishing Industry 32,091 11,822 167 7,105 2,699 100 113 233 262 144 41 612 *** 2,516 197 855 319 3,328 122 812 638 # Firms 19.6% 4.79% 51.5% 37.7% 13.4% 82.0% 87.6% 89.3% 63.4% 56.9% 61.0% 86.3% 83.3% 0.28% 75.6% 28.3% 85.6% 3.46% 91.8% 53.1% 10.0% % Exporters 2009 Table A.7: Industry-wise export participation rate of firms with at least ten employees 31,007 11,653 142 6,926 2,654 94 112 223 234 132 37 599 *** 2,301 175 820 312 3,079 113 770 628 # Firms 20.2% 5.19% 54.9% 38.7% 14.1% 81.9% 84.8% 90.6% 66.2% 58.3% 59.5% 87.1% 100% 0.48% 78.9% 29.6% 87.5% 3.38% 92.0% 56.5% 11.0% % Exporters 2010 33 Chapter 2 A Dynamic Model of Firm Activities: Evidence from Danish Manufacturing 34 A Dynamic Model of Firm Activities: Evidence from Danish Manufacturing∗ Kaleb Girma Abreha† Abstract In this paper, I structurally estimate a dynamic discrete choice model of exporting and importing. The model provides a framework to analyze the determinants of a firm’s decision to export and import while allowing for its current decision to affect its future productivity trajectory. Considering a panel of Danish manufacturing firms over the period 2000-2007, a simple description of the data reveals considerable firm heterogeneity, significant export and import activity premia, frequent incidence of simultaneous exporting and importing and high persistence in the scope of firm trading activities. The parameter estimates of the model show a marked difference in the demand elasticities in which export markets are characterized by more elastic demand, tougher competition and lower markup than the domestic market. The estimates also show that firms with larger capital holding and paying higher wages are costefficient even after controlling for their productivity. In line with the self-selection hypothesis, I find substantial sunk and fixed costs of exporting and importing. I also find a positive correlation between the size of these costs and the scale of firm operation. In addition, exporting and more importantly importing improve firm productivity, and therefore these learning effects further drive the selection of firms into trading activities. JEL Codes: F14, L60 Keyword: Firm heterogeneity, exporting, importing, dynamic discrete choice model, MCMC ∗ I am grateful to Val´eri´e Smeets, Fr´ed´eric Warzynski, Mark Roberts, Chad Syverson, Philipp Schr¨oder, Federico Clementi and participants at the Danish International Economics Workshop 2014, IAB-Aarhus Workshop 2014, Aarhus-Kiel Wokshop 2013, DGPE Workshop 2013 and Tuborg Research Centre Seminars for insightful comments and discussions. I acknowledge the Tuborg Foundation for financial support. The usual disclaimer applies. † Department of Economics and Business, Aarhus University, Denmark, E-mail: [email protected] 35 1 Introduction Inspired by the pioneering work of Bernard and Jensen (1995), numerous studies on firms in international trade have documented strong evidence of firm heterogeneity even in narrowly defined industries. The findings of these studies show that only a small fraction of firms export, and these firms are larger, more productive, more skill- and capital-intensive, and pay higher wages compared to non-exporting firms. Further, increased availability of microdata sets has triggered studies to explore the product and destination dimensions of firm exporting activities in an attempt to explain the highly uneven distribution of export sales not only across firms but also across products within the product portfolio of exporting firms. These explorations have led to the development of new models of international production where the export scope is endogenous and firms have core competency only in some of their products (see e.g. Eckel and Neary 2010; Arkolakis and Muendler 2010; Bernard et al. 2011; Mayer et al. 2014). At the same time, studies have started looking into possible explanations for the observed performance difference between exporting and non-exporting firms. Some studies point out the role of market entry costs where firms are required to make substantial and irreversible investments in export markets, and therefore it is only the productive firms which succeed in becoming exporters. Other studies resort to post-entry productivity improvements due to exposure to foreign competition and technology associated with exporting.1 It is also a possibility that part of the observed performance differential is incorrectly attributed to exporting. In this regard, Bernard et al. (2007) uncover that observed differences in performance between exporting and non-exporting firms are partly driven by firms that simultaneously export and import in US manufacturing. Similarly, Altomonte and B´ek´es (2009) for Hungary and Vogel and Wagner (2010) for Germany document that failure to control for import status of firms results in the overestimation of the export premia. On the other hand, the rapid growth rate of world trade in response to a small reduction in tariff rates has instigated research on trade in intermediate inputs which are used in the production of goods for export; vertical specialization.2 In this respect, Hummels et al. (2001) illustrate that this trade accounted for 21% of the exports and experienced a growth rate of 30% between 1970 and 1990 for a group of OECD and emerging economies. This central role of vertical specialization led to the development of new models of international fragmentation of production (see e.g. Antr`as 2003; Antr`as and Helpman 2004; Grossman and Rossi-Hansberg 2008; Yi 2003). Having shown the role of vertical specialization in global trade and the importance of accounting for simultaneous exporting and importing within firms, this paper structurally estimates a dynamic discrete choice model of exporting and importing in which firms are heterogeneous in terms of 1 Bernard et al. (2012), Hayakawa et al. (2012), Redding (2010), Wagner (2007) and Wagner (2012) provide a survey of notable theoretical and empirical contributions in this literature. 2 Yi (2003) argues that the elasticity of world trade to reductions in tariff rates is too high compared to the predictions from standard trade models. 36 size (capital holding), factor price payment (wages) and productivity. With this formulation, it is possible to characterize the intensity of competition (in terms of demand elasticity) and the reaction of firms (in terms of markups) to different intensities of market competition in domestic and export markets. It is also possible to test for the self–selection and learning–by–doing hypotheses, and identify the mechanisms (such as market entry costs) which possibly explain the export and import productivity premia. In the context of the existing literature, the main contribution of this paper comes from using an integrated approach that combines the role of market entry costs and learning opportunities from trade participation to shed light on the productivity premia and ranking observed in the data. Unlike most previous studies which only indicate the presence of markets costs of exporting and importing, this study first examines their presence and then proceeds to estimate their magnitude. It also examines the existence of any form of complementarity between exporting and importing activities and the resulting impact on firm productivity and trade participation over time. Considering a panel of Danish manufacturing firms over the period 2000-2007, a simple description of the data uncovers that most firms have active trade participation mainly through exporting. Also, these trading firms outperform their non-trading counterparts in a variety of performance measures. Further, there is a clear performance ranking in which two-way trading firms are ranked first followed by import-only and export-only firms, respectively, in line with recent findings in the literature (e.g. Altomonte and B´ek´es 2009; Castellani et al. 2010; Muˆ uls and Pisu 2009).3 Turning to the estimation results, the demand elasticities reveal that firms face intense competition in export markets, and they respond to this intense competition by charging a lower markup. On the cost side, the parameter estimates show that firms with larger capital holding and paying higher wages to their employees are more efficient even after controlling for their productivity. In support of the self-selection hypothesis, the estimates show that exporting and importing involve significant sunk and fixed costs which are, on average, greater for importing. These estimates also show that the size of market costs are positively correlated with the scale of firm operations in both export and import markets. Additionally, the non-negligible magnitude of these markets costs is manifested in terms of high state dependence in the scope of firm trading activities, which is consistent with a series of studies that establish the presence of sizable market entry costs (e.g. Bernard and Jensen 2004; Muˆ uls and Pisu 2009; Roberts and Tybout 1997). Further, the difference in sunk and fixed costs in export and import markets partly explains the prevalence of exporting vis-`a-vis other activities in the Danish manufacturing in that high sunk and fixed costs deter firms from starting or continuing importing. This difference also partially explains the productivity ranking of firms in which the import premia is the largest. Parameter estimates of the productivity evolution equation provide support to learning-bydoing in exporting and importing, which is especially greater in the case of importing. These 3 Vogel and Wagner (2010) find similar results except that export-only firms are more productive than importonly firms. 37 learning effects further reinforce the selection of firms into exporting and importing. This result on the complementarity between the learning effect and the selection process is consistent with findings of studies which examine the impact of trade in intermediate inputs on the export behavior of firms. For instance, Bas (2012), for Argentina, shows that input-tariff liberalization is associated with increased likelihood of entry into exporting. Similarly, Aristei et al. (2013), for Eastern European and Central Asian countries, find that exporting activity does not improve the chances of sourcing inputs from abroad. But, importing has a positive, significant effect on export sales mainly through its impact on productivity and product innovation. Relatedly, Bas and StraussKahn (2014), using French data, illustrate that increased access to imported inputs raises export sales and scope by enhancing firm productivity through accessibility of better technology and availability of cheaper production inputs. Also, Kasahara and Lapham (2013) demonstrate that import protection may have a detrimental impact on exporting capability in the case of Chilean manufacturing. This paper is related to several studies that aimed at investigating the relationship between exporting and importing within the framework of firm heterogeneity. The closest study to this paper is Kasahara and Lapham (2013) who investigate the export and import decisions of Chilean firms in a Melitz (2003) setting. Unlike their study, which allows for productivity gain only from increasing returns in varieties of intermediate inputs, this paper considers an endogenous evolution of firm productivity due to both exporting and importing. This paper also prominently focuses on the dynamic productivity effects of exporting and importing. Further, it explores additional dimensions of heterogeneity in the form of size and wage payments besides productivity. Methodologically, this study follows a recent contribution by Aw et al. (2011) in which export and R&D investment decisions are jointly examined. This paper is also related to Das et al. (2007) in its estimation of sunk and fixed costs using a Bayesian approach. The reminder of the paper is organized as follows. Section 2 describes the data source and the construction of the variables used in the empirical analysis. Section 3 summarizes a set of facts on exporting and importing activities. Section 4 develops a dynamic discrete choice model of exporting and importing. Section 5 presents the empirical strategy. Section 6 reports and discusses the estimation results. Section 7 concludes. 2 Data description The data used in this paper are provided by Statistics Denmark. The scope of the study is confined to the time period 2000-2007 and firms in the manufacturing sector, which covers industries 10-33 at the 2-digit according to NACE Rev.2 classification. I combine three datasets for the purpose at hand. The first dataset contains detailed information about the production portfolio of firms and describes which products firms make domestically. It covers all firms in the manufacturing sector with at least 10 employees. After constructing firm-level production data, the dataset comprises 38 39,515 firm-year observations. The second dataset has accounting information about more than 160,000 firms annually and contains 1,322,736 firm-year observations. The third dataset provides records of export and import transactions for the universe of firms in the economy. This dataset contains information on the value, weight and quantity of export and import transactions for each firm and destination/source market at the product level. After aggregation at the firm level, there are 120,761 and 164,106 firm-year observations in the export and import datasets, respectively. In the construction of the export and import variables, there are a few data issues. For exporting, the pervasiveness of carry-along trade may be a source of concern given the underlying assumptions in standard trade models where firms export products they actually produce. Despite its prevalence, carry-along trade as a source of export sales is negligibly small in the Danish manufacturing (see Abreha et al. 2014). In the case of importing, I do not distinguish between trade in primary, consumer, intermediate and capital goods. Here the data concern comes from the fact that the focus of this study primarily is on imports of capital and intermediate goods which are used in production. However, this data concern is not problematic as it seems given that international trade is exceedingly dominated by intermediate goods. According to the UNCTAD (2014) report, capital and intermediate goods constitute the most important flow of world trade especially for developed countries. Further, the focus on manufacturing sector and exclusion of firms primarily engaged in wholesale and retail activities mitigate the above data concerns. In order to make sure that firms with real economic activities are included, only those with a reported level of physical production in Denmark are considered. I delimit the time period to 2000-2007 mainly because of a series of revisions of the industry classification that pose a challenge on tracing firms and their core economic activities over time.4 Additionally, the years after 2007 are characterized by exceptionally large economic shocks which may unduly affect the outcomes of the empirical analysis. I also abstract from modeling entry and exit decisions in the local market. As a result, the final dataset comprises a balanced panel of 2,106 firms and 16,848 firm-year observations. 3 Basic facts in the Danish data This section summarizes salient features of firms in export and import markets. These summaries corroborate previous findings on the behavior of firms in international trade and provide a rationale for a joint consideration of exporting and importing activities. 4 Bernard et al. (2014) raise the issue of industry classification as it pertains to the Danish data and point out that the revisions in 1993 and 2007 were major whereas the 2003 one was minor. 39 3.1 Firm heterogeneity Table 1 reports the actual asset holdings of firms, their wage payment and productivity.5 The fact that there is a big difference in the capital holdings of the average and the median firm reveals the coexistence of a few large firms together with a significant number of small firms. This feature is further illustrated by the sizable standard deviations in each time period. In addition, firms vary in wage payment to their workers as evidenced by a non-negligible difference in the average and the median wages and moderate standard deviations. However, the wage differentials across firms are relatively small possibly due to labor mobility within and across industries and sectors. It is no surprise that firms also exhibit productivity heterogeneity particularly shown by large standard deviations as compared to the mean and the median values. Over the years, we see that firms have become larger in their capital holding, raised their wage payment and become more productive. Also, we notice that these dimensions of heterogeneities have all increased. Relatedly, Table 2 provides the size distribution of firms with different forms of trade participation. Size is defined by real capital holdings of a firm. Large firms are those with fixed assets above the median capital holding in a given year, and the converse holds for small firms. We see that small firms constitute a significant proportion of non-trading firms. On average, this group comprises 75% of the firms that exclusively delimit their production activity to the domestic market. This disproportionate representation of small firms is consistent with findings which document that larger firms are more likely to be trading. The size distribution also exhibits a discernible pattern among trading firms, albeit less uneven. On balance, large firms constitute around 58% of those engaged in exporting and importing over the entire sample period. We also observe a slightly higher prevalence of large firms in importing than in exporting. 3.2 Incidence of exporting and importing To examine the distribution of firms based on their scope of trading activities, I divide firms into four mutually exclusive categories—domestic, export-only, import-only and two-way. Domestic firms entirely restrict their activities to the local market. In contrast, export-only and import-only firms extend their operation besides the local market where the former serve export markets and the latter source inputs from the world market. There are also two-way trading firms which are exporting and importing besides their operation in the local market. Table 3 shows that exporting is the most common activity where around 74% of the firms export. Despite the relative rarity of importing (70%), it is shown that most of the exporting firms also import.6 We observe the high incidence of simultaneous exporting and importing in 5 Wages are defined at the firm level and refer to the total labor payments per employee in a given time period. These high rates of trade participation demonstrate an open manufacturing sector, which is typical of a small open economy. See Eriksson et al. (2009) for a comparison of trade participation of firms in Denmark vis-` a-vis the US. In another note, Bernard et al. (2007) also find that importing is less common in the US manufacturing sector. 6 40 which 63% of the firms engage in a two-way trading. Further, the sector has experienced increased integration to the global economy as evident from the declining share of domestic firms, albeit constituting the smallest group, and the rising share of exporting and importing firms. These patterns are suggestive of strong complementarity between exporting and importing activities. 3.3 Export and import activity premia To compare the performance of firms with a different scope of trade participation, I estimate the trading activity premia from the regression: yi,t = β0 + β1 Expi,t + β2 Impi,t + β3 Bothi,t + Controlsi,t + δt + τ + i,t where yi,t is in logarithmic scale and refers to value added, sales, capital, material, energy, wage, and employment size. Expi,t , Impi,t and Bothi,t are dummy variables taking a value of 1 if a firm is export-only, import-only or two-way, respectively. Controlsi,t includes year and industry effects and employment size (except for the last performance indicator). And, i,t is an iid error term. Table 4 reports percentage differences in a given performance indicator of export-only, importonly and two-way trading firms relative to the domestic ones. We see that exporting and importing are associated with significant premia. The estimates illustrate that trading firms are more productive, have greater sales per worker, employ more capital, material and energy per worker, and hire more employees than non-trading firms. The estimates reveal a systematic variation in the size of the premia among trading firms. The premia estimates are systematically larger in size for two-way firms. Further, the import premia are larger among firms which are exporting or importing but not both. From the premia estimates, there are two results worth noting. First, we see that production operations of import-only firms are not necessarily more capital-intensive compared to domestic firms. Second, the wage premia is the smallest vis-`a-vis the other premia estimates highlighting the crucial role of labor mobility to yield comparability of labor compensation across firms. To further establish the above performance ranking of firms with a different scope of trade participation, I run the Kolmogorov-Smirnov test for first order stochastic dominance of the productivity distributions.7 As pointed out by Arnold and Hussinger (2010), the use of panel data may violate the independence assumption necessary for the test. Consequently, the test is implemented for separate time periods. Table 5 shows that the two-tailed tests reject the null hypothesis that the productivity distributions of export-only, import-only and two-way firms are identical. On the other hand, the one-tailed test reveals that the productivity distribution of two-way trading firms stochastically dominates that of export-only and import-only firms implying that they are the most productive. Similarly, the productivity distribution of import-only firms is dominant which indicates that they are more productive than export-only firms. 7 See Delgado et al. (2002) for the details of the test procedure. 41 3.4 Persistence of exporting and importing Table 6 presents the transition probabilities of firms changing the scope of their trade participation over time. There is a strong persistence in the trading status of firms as shown by a high probability of firms maintaining their current activity status. These probabilities are 80.55%, 55.03%, 57.76% and 93.82% for domestic, import-only, export-only and two-way trading firms, respectively. We also see that the probability of retaining current status is considerably high for domestic firms indicating the difficulty of penetrating into export and import markets. It also indicates the limited capacity of domestic firms to extend their operation beyond the local market. In contrast, firms partially engaged in trading are more likely to add exporting (23.46%) or importing (26.66%) as an additional activity compared to firms engaged in neither of these activities; 11.37% and 10.47% to start exporting and importing, respectively. Relatedly, two-way firms are less likely to give up exporting (2.82%) and importing (3.82%) compared to export-only (15.58%) and import-only (21.51%) firms. The average of a cross-sectional distribution of trade participation shows that around 80% of the firms are globally active via exporting or importing highlighting a high firm trade participation in the sector. The above empirical regularities uncover that it is crucial to accommodate additional dimensions of heterogeneity in size and wage payment besides the commonly known productivity. In addition, exporting and importing activities are highly interrelated which underscores the usefulness of their joint examination for a better understanding of the behavior of firms in international trade. In the next section, I describe a behavioral framework of firm profit maximization and market entry decision which lays the foundation and provides intuition to the econometric model in the empirical section. 4 A theoretical framework The behavioral model of firm entry and exit into exporting and importing involves combining the demand and the supply components which define consumer preference, costs, revenues, profits and evolution of state variables. The theoretical framework considered here is similar to several models of market entry and exit in the literature.8 In this paper, I closely follow the modeling approach by Aw et al. (2011) to examine a firm’s choice problem of exporting and importing. Suppose a production technology of firm i at time period t is represented by its short-run marginal cost function: ci,t = β0 + βk ki,t + βw wi,t + βx xt − ωi,t (1) where ki,t refers to capital holding, wi,t firm-specific wage which varies over time, xt vector of variable input prices common to all firms in the market and ωi,t firm productivity, which is observed 8 Models of some of the key contributions in the literature include: Aw et al. (2011), Bernard and Jensen (2004), Clerides et al. (1998), Das et al. (2007) and Roberts and Tybout (1997). 42 only by the firm. Notationally, unless stated otherwise, lower case letters refer to variables in logarithmic scale. Under this formulation of the marginal cost function, firms have inherently different cost structures attributable to their capital holding, wages and productivity. We see that marginal cost is independent of output level which implies that shocks affecting output decision in the domestic market does not have any effect on a similar decision in export markets, and vice versa. As common in the standard trade literature, I consider consumers with CES preferences. The  P d −ξd
Itd d −ξd d and = Φdt Pi,t demand a firm faces takes a Dixit-Stigliz form given by Qi,t = P d Pi,td t t   −ξ x
x P Ix x −ξx zi,t Qxi,t = Ptx Pi,tx ezi,t = Φxt Pi,t e in the domestic and export markets, respectively. In t t both demand functions It refers to market size, Pt aggregate price index, Pi,t firm price index and ξd , ξx > 1 demand elasticities in the respective markets. Export demand is defined slightly differently in that an exporting firm faces a shock zi,t only inherent in export markets. zi,t can be considered broadly to accommodate factors affecting demand in export markets such as product attributes in addition to those included in the marginal cost function. zi,t is assumed to be firm-specific, time-variant and known to the firm but unobservable to the econometrician. The respective domestic and export revenue and profit functions are given by  ξd + lnΦdt + (1 − ξd ) (β0 + βk ki,t + βw wi,t + βx xt − ωi,t ) = (1 − ξd ) ln ξd − 1   ξx = (1 − ξx ) ln + lnΦxt + (1 − ξx ) (β0 + βk ki,t + βw wi,t + βx xt − ωi,t ) + zi,t ξx − 1 d ri,t x ri,t 

1 d d πi,t Φdt , ki,t , ωi,t Φdt , ki,t , ωi,t = Ri,t ξd 1 x x (Φxt , ki,t , ωi,t , zi,t ) = Ri,t (Φxt , ki,t , ωi,t , zi,t ) πi,t ξx (2) (3) In terms of market entry and exit, I abstract from entry and exit in the domestic market. It is implicitly assumed that firms do not export or import without any production for the local market. Under this setting, a firm makes entry and exit decisions with respect to exporting and importing after comparing market entry costs with expected discounted payoffs associated with each decision. These market entry costs are sunk and fixed in their nature. It is now established that exporters must acquire information about foreign markets, hire managers to look after overseas operations, and undertake complementary investments in processing, marketing, distribution and related activities. As in the case of exporting, importers incur costs related to information acquisition about foreign suppliers, controlling the quality of imported inputs and making import-related investment activities such as storage. The characteristics of these costs indicate a high degree of similarity on what constitutes sunk and fixed costs in the export and import markets. In this respect, al43 though exporting is a decision about final goods and importing about production inputs, it is not problematic to consider these decisions as symmetrically in the model. Given the dynamic nature of the choice problem, it is necessary to be explicit about the timing assumptions of the model. A firm observes its state variables, maximizes static profits in domestic and export markets, draws its sunk and fixed costs of exporting from a known cost distribution and decides whether or not to export. Afterwards, the firm observes the market costs of importing from its sunk and fixed cost distribution and chooses whether or not to source inputs from abroad. More specifically, at the beginning of each time period a firm observes its state vector si,t ≡
ki,t , ωi,t , ei,t−1 , mi,t−1 , Φdt , Φxt , zi,t . I assume that current investment ii,t becomes productive in the next period, and capital is deterministic and evolves according to ki,t = (1 − δ)ki,t−1 + ii,t−1 , where δ is the rate of depreciation. Productivity follows a controlled Markov process which is given by: ωi,t = E [ωi,t | ωi,t−1 , ei,t−1 , mi,t−1 ] + ζi,t (4) = α0 + α1 ωi,t−1 + α2 (ωi,t−1 )2 + α3 (ωi,t−1 )3 + α4 ei,t−1 + α5 mi,t−1 + α6 ei,t−1 mi,t−1 + ζi,t
where ζi,t is an error term assumed to possess a normal distribution N 0, σζ2 . By definition, ζi,t is uncorrelated with ωi,t−1 , ei,t−1 and mi,t−1 . In this specification, we see that firm productivity depends on its prior participation in export and import markets. The economic motivation for this is that both exporting and importing represent some form of exposure to foreign technology which can potentially improve a firm’s productivity, although exports represent outlets of final goods and imports inputs of production. Specifically, there may be gains from exporting which comes in the form of knowledge transfers from international buyers about product designs, quality standards, production techniques and management practices. It can also result from intense competition in the international market where firms can only survive by removing their existing inefficiencies. In the case of importing, similar channels of knowledge transfer from foreign suppliers and competition in input markets can be considered. Consequently, in the specification exporting and importing are treated symmetrically. Further, the specified productivity equation only captures dynamic effects of exporting and importing. All static effects–which are associated with immediate effects on output of imported inputs due to more variety and/or better quality of intermediates—are ruled out. Here, I rather focus on whether or not firms benefit from their trade participation in the form of productivity improvement even if they are not currently taking part in exporting, importing or both.9 This formulation of firm productivity is different from Halpern et al. (2009) who separately estimate the variety and quality effects of importing and Kasahara and Lapham (2013) which focus only on the variety effect of importing. These effects are conditional on imperfect substitutability between local and imported varieties and whether or not firms have used more varieties and/or better quality intermediates, say, following trade liberalization as in Amiti and Konings (2007) and Goldberg et 9 A similar specification to this paper is also used in Zhang (2015). 44 al. (2010). Another feature of such models is the absence of any productivity gain when the firm stops importing. The state vector also contains a firm’s previous period export and import participation ei,t−1 and mi,t−1 . It also includes the aggregate variables Φt ≡ (Φdt , Φxt ) which follow an exogenous first order Markov process. The aggregate variable related to the domestic market Φdt is captured by time dummy variables in the estimation of the revenue function and computation of firm productivity. On the other hand, the export market size indicator Φxt is estimated as an element of the parameter vector that summarizes the dynamic aspect of the entry and exit decisions. The export market shock follows an AR(1) process zi,t = ρzi,t−1 + νi,t ; νi,t ∼ N (0, σν2 ) (5) where ρ is the autocorrelation coefficient, and νi,t is a noise term. The value function of a firm with a state vector si,t choosing whether to export and import is given by Z  n  o xf d x xs E D V (si,t ) = πi,t + max πi,t − ei,t−1 γi,t − (1 − ei,t−1 )γi,t + Vi,t (si,t ) , Vi,t (si,t ) dFγ (6) ei,t x − The firm decides whether to export after comparing the expected return from exporting πi,t xf xs E D ei,t−1 γi,t − (1 − ei,t−1 )γi,t + Vi,t (si,t ) with the option value of not exporting Vi,t (si,t ). The sunk γ xs and fixed γ xf costs are firm-specific and time-variant, and they are drawn from the cost distribution dFγ (.). Assuming a fixed discount factor λ, the option value of exporting is E Vi,t Z (si,t ) = − n max λEt Vi,t+1 (si,t+1 |ei,t =1, mi,t =1) mi,t mf mi,t−1 γi,t − (1 − ms mi,t−1 )γi,t , λEt Vi,t+1 (7) o (si,t+1 |ei,t =1, mi,t =0) dFγ and that of non-exporting D Vi,t Z (si,t ) = − n max λEt Vi,t+1 (si,t+1 |ei,t =0, mi,t =1) mi,t mf mi,t−1 γi,t − (1 − ms , λEt Vi,t+1 mi,t−1 )γi,t (8) o (si,t+1 |ei,t =0, mi,t =0) dFγ We see that both option values involve a choice problem of whether to import which requires integrating each of the value functions over the import cost distributions. Here the expected value, which is integrated over the state vector, is given by Z Z Z  0   0   0  Et Vi,t+1 (si,t+1 |ei,t , mi,t ) = Vi,t+1 (si,t+1 ) dFω ω |ωi,t , ei,t , mi,t dFz z |zi,t dFΦ Φ |Φt Φ0 z0 ω0 where dFω (.), dFz (.) and dFΦ (.) represent transition densities of the stochastic state variables. To summarize, the parameter vector comprises demand parameters characterizing consumer behavior in the domestic and export markets ξ = (ξd , ξx ), marginal cost parameters β = (βk , βw ), 45 productivity transition parameters α = (α0 , α1 , α2 , α3 , α4 , α5 , α6 , σζ ), and parameters defining the dynamic aspect of the choice problem θ = (γ xs , γ xf , γ ms , γ mf , δx , ρ, σν ). δx approximates the export market size Φxt and captures the constant and the time dummy variables in the econometric specification of export revenue in equation (2).10 5 Estimation procedure In this section, I present the empirical strategy to estimate the parameter vector and describe the steps followed in some detail. Demand, marginal cost and productivity parameters Domestic and export market demand elasticities are estimated by exploiting a well defined relationship between total variable costs and domestic and export revenue functions. Under CES demand preferences and from the profit maximization condition, total variable costs can be expressed as the elasticity weighted sum of revenue from domestic and foreign sales as T V Ci,t = d Ri,t  1 1− ξd  + x Ri,t  1 1− ξx  + εi,t (9) where εi,t is appended as an iid error term capturing measurement errors in variable cost and revenue. I estimate equation (9) by ordinary least squares and easily recover the domestic and foreign demand elasticities. As in Aw et al. (2011), I use the domestic revenue function, which is available for all firms in the sample, to jointly estimate the marginal cost and productivity parameters. An iid error term νi,t is added to equation (2) to account for measurement errors in revenue from the domestic market. This yields the following estimating equation d ri,t  = (1 − ξd ) ln ξd ξd − 1  + lnΦdt + (1 − ξd ) (β0 + βk ki,t + βw wi,t + βx xt − ωi,t ) + νi,t (10) As −(1 − ξd )ωi,t + νi,t is unknown, I use a semi-parametric approach by Levinsohn and Petrin (2003) to back out the unobserved firm productivity ωi,t . For this purpose, I use electricity usage eli,t as a proxy variable that contains information about productivity; eli,t = ft (ki,t , ωi,t ). Electricity consumption is highly and instantly (that is, it is almost costless to adjust) responsive to shocks in capital and productivity, and therefore ft (.) can be viewed as monotonically increasing in ki,t and ωi,t . This monotonicity property allows inversion of f (.) and expression of ωi,t in terms of the observable variables ki,t and eli,t : ωi,t = ft−1 (ki,t , eli,t ) = gt (ki,t , eli,t ). Under the proxy variable eli,t , 10 Estimation of δx allows a counterfactual analysis of a policy change which affects the size of export markets as shown in subsection 6.5. 46 equation (10) becomes: d ri,t = δ0 + X δt Dt + (1 − ξd ) βw wit + gt (ki,t , eli,t ) + νi,t (11) t d ) + β0 ), Dt captures time-varying domestic market aggregate where δ0 denotes (1 − ξd )(ln( ξdξ−1 variable lnΦdt and common market-level factor prices (1 − ξd )βx xt , and gt (.) represents a non-linear relationship between capital, productivity and domestic revenue. Assuming gt (.) is a cubic function in ki,t and eli,t , I estimate equation (11) by ordinary least squares with industry fixed effects. From this estimation, the wage parameter βw , and the fitted values φˆi,t = (1 − ξd )(βk ki,t − ωi,t ) are obtained. These fitted values are used to express a firm’s productivity series as: ωi,t = ξd1−1 φˆi,t + βk ki,t . Substituting φˆi,t into the productivity transition function in equation (4) and rearranging the terms yields: 2    ˆ ˜ ˆ ˜ ˆ ˜ ˜ 2 φi,t−1 − βk ki,t−1 φi,t = βk ki,t − α ˜0 + α ˜ 1 φi,t−1 − βk ki,t−1 − α 3  ˜ 4 ei,t−1 − α ˜ 5 mi,t−1 − α ˜ 6 ei,t−1 mi,t−1 − ζ˜i,t +α ˜ 3 φˆi,t−1 − β˜k ki,t−1 − α (12) where β˜k = (1 − ξd )βk , α ˜ 0 = (1 − ξd )α0 , α ˜ 1 = α1 , α ˜ 2 = (1 − ξd )−1 α2 , α ˜ 3 = (1 − ξd )−2 α3 , α ˜ 4 = (1 − ξd )α4 , α ˜ 5 = (1 − ξd )α5 , α ˜ 6 = (1 − ξd )α6 and ζ˜it = (1 − ξd )ζi,t . Applying nonlinear least squares to equation (12) provides estimates of the cost coefficient βk and the productivity parameter vector α. Afterwards, the fitted values of a firm’s productivity are constructed as follows: ω ˆ i,t = ξˆ 1−1 φˆi,t + βˆk ki,t where the “hat” denotes an estimated value. d Market cost and export revenue parameters Over the entire time period, for firm i define the state vector sTi,0 = (si,0 , si,1 , ..., si,T ) and the x,T x,T T ) where eTi,0 = (ei,0 , ei,1 , ..., ei,T ), mTi,0 = (mi,0 , mi,1 , ..., mi,T ) and Ri,0 = data Di,0 = (eTi,0 , mTi,0 , Ri,0 x x x (Ri,0 , Ri,1 , ..., Ri,T ) refer to the history of export and import participation and export revenue, respectively. The contribution of firm i to the likelihood can be formulated as:
x,T T T T T T T T |ωi,0 , ki,0 , ΦT0 ) = P (eTi,0 , mTi,0 |ωi,0 , ki,0 , ΦT0 , zi,0 )h(zi,0 ) | sTi,0 ; θ ≡ P (eTi,0 , mTi,0 , Ri,0 L Di,0 (13) where h(.) is the marginal density function of zi,t which is constructed as in Das et al. (2007).11 Assume that the sunk and fixed costs are iid draws from independent exponential distribution 1 functions Fγ (γi,t ) = 1 − γ1 e− γ γi,t where γ refers to the mean value of the sunk and the fixed 11  x T The export demand shock zi,t is defined as zi,t = zi,t : Ri,t > 0 with a marginal probability density h(zi,0 )= N (0, Σνν ) where Σνν is an m × n square matrix whose diagonal elements are σν2 |m−n| , 1−ρ2 ρ ∀ m 6= n. 47 σν2 1−ρ2 and the off diagonal elements cost distributions. This independence assumption allows the construction of the joint likelihood function as a product of individual probabilities of exporting and importing. The likelihood function of exporting for a firm with a state vector si,t over the entire period can be expressed as  h iei,0 h  i1−ei,0 P eTi,0 | sTi,0 = Φ (η0 + η1 ki,0 + η2 ωi,0 + η3 zi,0 ) × Φ (η0 + η1 ki,0 + η2 ωi,0 + η3 zi,0 ) × T h Y iei,t xf x E D xs Eγ I(πi,t + Vi,t (si,t ) − Vi,t (si,t ) > ei,t−1 γi,t − (1 − ei,t−1 )γi,t ) (14) t=1 h i1−ei,t xf x E D xs × Eγ I(πi,t + Vi,t (si,t ) − Vi,t (si,t ) 6 ei,t−1 γi,t − (1 − ei,t−1 )γi,t ) iei,0 h i1−ei,0 h where Φ (η0 + η1 ki,0 + η2 ωi,0 + η3 zi,0 ) × Φ (η0 + η1 ki,0 + η2 ωi,0 + η3 zi,0 ) denotes the first period likelihood function after correcting for the initial conditions problem by using Heckman’s approach, and η 0 = (η0 , η1 , η2 , η3 )0 is vector of unconstrained parameters.12 In constructing the likelihood, the operator Eγ (.) takes the expectation over dFγ (.) if I(.) is true. Define ∆Et Vi,t+1 (si,t+1 |ei,t , mi,t ) = Et Vi,t+1 (si,t+1 |ei,t , mi,t =1)−Et Vi,t+1 (si,t+1 |ei,t , mi,t =0). The likelihood function for an importing firm over the sample period is given by:   h imi,0 h i1−mi,0 P mTi,0 | sTi,0 = Φ (µ0 + µ1 ki,0 + µ2 ωi,0 + µ3 zi,0 ) × Φ (µ0 + µ1 ki,0 + µ2 ωi,0 + µ3 zi,0 ) × T h imi,t Y mf ms Eγ I(λ∆Et Vi,t+1 (si,t+1 |ei,t , mi,t ) > mi,t−1 γi,t − (1 − mi,t−1 )γi,t (15) t=1 h i1−mi,t mf ms × Eγ I(λ∆Et Vi,t+1 (si,t+1 |ei,t , mi,t ) 6 mi,t−1 γi,t − (1 − mi,t−1 )γi,t h imi,0 i1−mi,0 h refers to the prob× Φ (µ0 + µ1 ki,0 + µ2 ωi,0 + µ3 zi,0 ) where Φ (µ0 + µ1 ki,0 + µ2 ωi,0 + µ3 zi,0 ) ability of importing in the first period after correcting for the initial conditions problem, and µ0 = (µ0 , µ1 , µ2 , µ3 )0 is a vector of unconstrained parameters as in the exporting case. There are no closed form solutions to these choice probabilities which lead to a practical difficulty in constructing the likelihood function. I obtain these probabilities by evaluating the value functions iteratively. See the steps followed in Appendix A. As pointed out by Das et al. (2007), the likelihood function may not necessarily be globally concave in the parameter vector, and thereby it is computationally difficult to find a parameter vector that maximizes the likelihood function. As a result, instead of trying to maximize the likelihood function, the posterior distribution of the parameter vector is estimated using a Bayesian Markov Chain Monte Carlo (MCMC) method. Implementation of this method requires specification of a prior distribution Π(θ) and a likelihood function of the data L(D|s; θ), where T T T , D2,0 , ..., Dn,0 ) and s = (sT1,0 , sT2,0 , ..., sTn,0 ) denote the observed data and state variables D = (D1,0 for n firms in the sample. Under a standard Bayesian inference approach, the posterior distribution 12 For a discussion on this issue, see Wooldridge 2002, pp. 493–495. 48 of the parameter vector is given by: Π (θ) L (D | s; θ) ∝ Π (θ) L (D | s; θ) P (θ | D, s) = R Π (θ) L (D | s; θ) (16) A single component Metropolis-Hastings algorithm is used to draw parameter values from the posterior distribution P (θ|D, s) to obtain the market cost and export revenue parameters.13 For this purpose, the parameter vector θ is divided into five blocks in which a block is updated one at a time. The first two blocks contain the fixed and the sunk cost parameters of exporting and importing. The third block consists of the export demand shock and market size parameters. And, the last two blocks comprise parameters included to deal with initial conditions problem with respect to export and import participation, respectively. The algorithm involves the following basic steps:
1. At iteration r = 0, define θ0 = γ xs,0 , γ xf,0 , γ ms,0 , γ mf,0 , δx0 , ρ0 , σν0 , η 0 , µ0 as starting values for each parameter θj in the vector θ.14 r ) 2. Draw a candidate parameter vector θ˜jr = θjr + κνjr from a proposal distribution q(θ˜jr |θjr , θ−j r where νj is a random draw from a multivariate standard normal distribution. θ−j is defined to be a parameter vector θr with the exception of the j th element, where the first j − 1 elements are updated and the rest still to be updated. I use κ to scale the random draws such that the candidate parameter vector lies within appropriate support. 3. Evaluate the acceptance probability of the candidate parameter vector at iteration r as : λrj  = min The j  r r r π (θ˜jr |θ−j )Lr (D|s;θ˜jr ,θ−j )q(θjr |θ˜jr ,θ−j ) ,1 r r r r r r r ˜ π (θ |θ )L (D|s;θ ,θ )q (θ |θ ,θr ) j th −j j −j j j −j block will be updated as follows: θjr =  θ˜r j with probability λrj θ r with probability 1 − λrj block is updated is given by: j The parameter vector at iteration r after the j th  (θ˜r , θr ) with probability λr j −j j θr = (θr , θr ) with probability 1 − λr j −j j 4. Repeat steps 2 and 3 until every block in each iteration is updated, and the maximum number of iteration r = rmax is reached. The mean and standard deviations of the posterior distribution the parameter vector conq of P P N r ¯ r ¯0 ditional on the data are obtained as follows θ¯ = N1 r=1 θr and N1 N r=1 (θ − θ)(θ − θ) where N = rmax − m after discarding the first m elements of the Markov chain as burn-ins. 13 For a discussion on this algorithm, see Gilks et al. 1996, pp. 9–16. For simplicity of computation in terms of defining a proposal distribution and drawing candidate parameter values, I assume that σν has a lognormal distribution as in Aw et al. (2011) and Das et al. (2007). 14 49 6 6.1 Estimation results Demand, marginal cost and productivity Table 7 presents estimates of the demand, marginal cost and productivity parameters. The demand parameter 1 − 1/ξ is estimated to be 0.6451 and 0.8644 for the domestic and export markets, respectively. We see that the difference in the demand parameters leads to a substantial difference in the implied values of demand elasticities in the domestic and export markets, which are 2.817 and 7.373, respectively.15 Because export markets host a larger number of firms and a wider variety of products than the local market, demand is highly sensitive and competition is more intense. Exporting firms respond to this market condition by charging a lower markup. We see that the markup rate in export markets (16%) is much lower as compared to the markup rate in the domestic market (55%).16 The size difference in demand elasticities as well as markup rates in the domestic and export markets are consistent with the predictions of theoretical models such as Melitz and Ottaviano (2008) and Mayer et al. (2014), which connect markups to the size of a market and its extent of trade integration. Estimates of the cost function show that a firm with a large capital holding is more costefficient. Also, the wage rate a firm pays has a negative and significant effect on the marginal cost after controlling for its capital size, productivity and prices of all other inputs used in production. This result implies that a firm can elicit increased effort by paying its workers higher wages. It also indicates that high wages are somehow proxies for a high quality workforce which contributes to the efficiency of firms. From the parameter estimates of the productivity equation, we observe a significant, nonlinear relationship between current and previous period levels of productivity, which highlights the persistence in the evolution of firm productivity over time. The nonlinearity of this relationship also holds when export and import are treated as continuous variables. In terms of the impact of previous period exporting and importing on a firm’s current productivity, we see that the estimates on α4 and α5 are positive and significant implying learning-by-doing associated with trade participation. The magnitudes of these coefficients reveal that the expected productivity gain from importing (1.42%) exceeds that of the gain from exporting (1.03%). Further, a firm simultaneously exporting and importing enjoys greater productivity improvement (2.45%) than others that are only partially engaged in international trade. However, the insignificance of α6 demonstrates that there are no greater gains from importing for an exporting firm, and so is the case for an importing firm from exporting. This result reveals that a firm that exports, imports or undertakes both enjoys a 15 By construction, demand elasticity in each market is common for all firms. Unlike the estimates in this paper, Aw et al. (2011) find almost identical demand elasticities in the two markets. They report demand elasticity estimates 6.38 and 6.10 and markups 18.6% and 19.6% in the domestic and export markets, respectively. 16 50 discernible productivity improvement from such activities which enables it to maintain its status or add another trading activity. It also indicates that the complementarity between exporting and importing is mainly through productivity. Similarly, the estimates of α4 and α5 , when export and import are defined as intensities among exporting and importing firms, demonstrate that a firm shipping a significant portion of its output abroad or hiring imported inputs more intensively experiences a greater productivity improvement. This delineates that both the scope and the scale of a firm’s trading activities determine the size of expected productivity gain from trade participation. 6.2 Market costs and export revenue To examine if there is any correlation between a firm’s size, its productivity and trade participation, I estimate a probit model of exporting and importing. Table 8 shows reduced form estimates of the variables that determine firm activity in export and import markets. The estimates show that a firm’s size and its productivity are positively and significantly correlated with the likelihood of its participation in exporting and importing. The significance of the coefficients on lagged period export and import dummy variables suggests the presence of substantial market costs firms must bear, and these costs decline with prior experience in international trade. Further, the complementarity between exporting and importing can be seen from the significance of the coefficients on mi,t and ei,t in the probit regressions of exporting and importing, respectively. In the same way, the bivariate probit estimates show that a firm with a large capital holding, which is more productive, and with a previous history of trade participation is more likely to be exporting and importing. We also see that there is a non-negligible correlation between the unobserved factors which determine a firm’s involvement in export and import markets; the estimated correlation coefficient ρ is 0.2813. As an extension, Table 9 reports reduced form estimates of the export revenue equation under the assumption that the previous history of trade participation does not affect export revenues conditional on a firm exporting. The results show that the revenue from export markets is positively and significantly associated with a firm’s productivity and its capital size. And, this result holds with and without controlling for firm fixed effects. We also see that firm fixed effects explain a significant portion of the variation in the level of export sales which is not attributed to a firm’s capital holding and its productivity, as shown by high ρ = 0.897. For comparison, I run the same regression for import expenditure and find similar results. In this case, the percentage of the variance of the error term attributed to firm fixed effects is 0.882. The above reduced-form estimates emphasize the role of market costs in a firm’s trade participation decisions. As a next step, these costs are estimated. Table 10 presents the market cost estimates with and without size heterogeneity. We see that firms incur substantial sunk and fixed costs when they export their output and source inputs from abroad. For small firms, exporting 51 requires about 2,550 (426) thousand DKK (USD) for entrants and around 99 (17) thousand DKK (USD) for incumbents.17 For large firms, the estimated sunk and fixed costs are around 5,910 (986) and 394 (66) thousand DKK (USD), respectively. With no size heterogeneity, the magnitude of these costs is estimated to be 3,008 (502) thousand DKK (USD) for new exporters and 159 (27) thousand DKK (USD) for incumbent exporting firms.18 Importing is not a costless undertaking either. Small firms incur a sunk cost of 2,429 (405), and a fixed cost of 130 (22) thousand DKK (USD). For large firms, these costs are around 6,152 (1,027) and 476 (79) thousand DKK (USD), respectively. Without size heterogeneity, the sunk and fixed costs are estimated to be 3,280 (548) and 155 (26) thousand DKK (USD), respectively. The implication of the sizes of these costs contradicts the findings by Smeets and Warzynski (2013) who report selection effects in exporting but not in importing for Danish manufacturing firms. We observe that the market costs of exporting and importing are positively correlated with a scale of firms’ operations. This indicates that large exporting firms ship multiple products to several destinations, each involving individual product-destination-specific sunk and fixed costs. This relationship between market costs and scale of operation is consistent with the predictions of theoretical models such as Arkolakis and Muendler (2010), and as empirically shown in Aw et al. (2011) for Taiwanese data. In contrast, Das et al. (2007), for Colombian manufacturing, find a lower sunk cost for large exporting firms compared to their small counterparts.19 Large importing firms also face a similar pattern of distribution of market costs. As normally expected, irrespective of firm size, the fixed costs of exporting and importing are substantially lower than the corresponding sunk costs. Further, the sizable difference in the estimated market costs for small and large firms highlights the importance of explicitly accounting for size heterogeneity in the empirical analyses of firm trade participation. We also see that importing generally involves greater sunk costs than exporting, which is different from the cost estimates of Kasahara and Lapham (2013) who find sunk costs of exporting to be larger for almost all industries they consider. This result partly explains the productivity ranking of firms shown in Table 5 in that firms must be more productive to absorb greater market costs associated with importing. This is also part of the reason why exporting is the most common activity in the Danish manufacturing sector. Very high sunk and fixed costs deter firms from becoming or maintaining their status as importers.20 Table 10 also reports estimates of the export revenue function which includes the intercept δx , 17 I use the World Development Indicators database as a source for data on exchange rate. Using 2005 as a reference year, the official exchange rate is 1 USD = 5.9911 DKK. 18 Due to differences in country currencies and choices of reference periods, comparison of the absolute magnitudes of the sunk and fixed cost estimates across studies is inappropriate. To this end, the discussion here focuses on relative sizes and distribution patterns of these costs. 19 Das et al. (2007) argue that this feature can be attributed to differences in nature of the products they export, market conditions of the destinations they serve, existing networks they possess and the like. 20 High sunk costs particularly for large importing firms may be due to the high prevalence of intrafirm trade which requires setting up production units in foreign affiliates (see e.g. Antr`as 2003; Antr`as and Helpman 2004; Nunn and Trefler 2008). 52 the serial correlation parameter ρ, and the standard deviation σν of the export demand shock. Importantly, we see that the estimate on ρ, which ranges between 0.644 and 0.693, is positive and significant. This implies that export demand shocks are strongly serially correlated highlighting that the effects of shocks on export participation and revenue carry over from one period to the next. These export revenue coefficients are robust to choice of burn-ins and whether firms are differentiated in terms of size. 6.3 In-sample model performance Here the performance of the model in fitting the actual data is examined. In this exercise, I take the initial period data for all firms and then simulate for the rest of the periods. In these simulations, I use the demand, marginal cost and productivity parameters estimates from a model where exporting and importing are treated as discrete variables. Additionally, the market cost and export revenue estimates with size heterogeneity are used. I repeat the simulations 1000 times and report the averages across these simulations. Table 11 presents the actual and predicted rates of trade participation. The model correctly predicts that two-way trading firms are the largest group followed by export-only and importonly firms. However, we observe that it overpredicts the number of export-only firms whereas it underpredicts the other groups especially when comparison is made for individual time periods. When considering all the time periods together, the model predictions fit the data reasonably well. Relatedly, Table 12 shows the actual and predicted probabilities of the empirical transition matrix. It reports two sets of predictions (Predicted 1 and 2), which are obtained after discarding the first 33% and 67% of the MCMC estimates as burn-ins, respectively. Despite small differences in some of the transition probabilities (especially with respect to export-only firms quitting exporting or maintaining their exporting status), the predicted values highly resemble the empirical transition probabilities in size and order. Besides, comparing the two predictions, we see that Predicted 2 fits the actual transition matrix much better. This is in line with the expectation that the Metropolis-Hastings algorithm, after constructing relatively long Markov chains, eventually reaches the stationary distribution of the parameter vector. See the trace plots and the running means in Figures C.1–C.6 in Appendix C. 6.4 Robustness The above parameter estimates may be sensitive to changes in functional forms and simplifying assumptions used in the model and the empirical strategy. This subsection checks for the sensitivity of the baseline estimates with respect to productivity transition, wage parameter and prior choices. 53 Linear Markov process Estimates of the productivity evolution function reported in Table 7 show that there is a significant, non-linear relationship between previous and current period productivity levels. However, it is very difficult to provide economic intuition to those coefficients. To this end, I redefine the productivity transition in equation (4) to have a linear controlled Markov process as ωi,t = θ0 + θ1 ωi,t−1 + θ2 ei,t−1 + θ3 mi,t−1 + θ4 ei,t−1 mi,t−1 + ζi,t . Under this formulation, the long-run effects of exporting on θ2 2 +θ4 productivity are 1−θ for a non-importing firm and θ1−θ for an importing firm. And, the long-run 1 1 θ3 3 +θ4 and θ1−θ for a non-exporting and an productivity improvements from importing becomes 1−θ 1 1 exporting firm, respectively. From columns (1)-(4) of Table 13, we see that there is a strong state dependence in the evolution of a firm’s productivity over time, as shown by a large and highly significant estimate on ωi,t−1 . From column (1), the long-run productivity effect of exporting is about 5.45% and 17.93% for a non-importing and an importing firm, respectively. The corresponding effect from importing is around 7.46% and 19.94% for a non-exporting and an exporting firm. These long-run effects demonstrate substantial differences in the expected productivity gains between two-way trading firms and the rest. Because the estimated coefficients used to infer the long-run effects are statistically insignificant, I run another regression with the interaction term omitted. The estimates are reported in column (3). The coefficients on ei,t−1 and mi,t−1 are now larger and highly significant while the coefficient on ωi,t−1 remains unchanged. The implied long-run productivity effects of exporting and importing are 11.54% and 15.30%, respectively. The same exercise is repeated in columns (2) and (4) except that the effects of differences in the intensities of export and import participation among trading firms are now considered. The results reveal that if the share of exports in the total sales of an exporter doubles, this will be associated with 3.38% and 3.54% increase in productivity for a non-importing and an importing firm, respectively. The effects of doubling of a firm’s import intensity are 4.66% and 4.82% for a non-exporting and an exporting firm. Without taking into account the possibility of varying effects of exporting and importing across firms with different trading scope, the expected productivity gains from increased intensities of exporting and importing are 2.64% and 4.13%, respectively. In general, these results establish that trading firms enjoy a sizable productivity gain in the long run. The observed differences in the productivity trajectories are substantial not only between trading and non-trading firms but also among exporting and importing firms with varying intensities of their export and import involvement. Wage parameter I now consider a scenario in which a firm’s wage is monotonically increasing in its size and productivity. This implies that wages contain information which can be used to back out unobserved firm productivity. If the wage a firm pays depends on its size and productivity, wi,t will be correlated 54 with ki,t and ωi,t in equation (10). In this case, it is not possible to identify βw in the first stage, and therefore βw is estimated along with βk in the second stage of the algorithm. The estimating equation in the first stage is now given by: d ri,t = δd + X δt Dt + ht (ki,t , wi,t , eli,t ) + νi,t (17) t where ht (.) is non-linear in ki,t , wi,t and eli,t and is approximated by a third-order polynomial function. From this regression, the fitted values φˆi,t = (1 − ξd )(βw wi,t + βk ki,t − ωi,t ) are recovered, and the productivity series ωi,t = ξd1−1 φˆi,t + βw wi,t + βk ki,t are constructed. In the second stage, the fitted values are inserted in the productivity transition equation which gives rise to the estimating equation    2 ˆ ˜ ˜ ˆ ˜ ˜ ˆ ˜ ˜ ˜ 2 φi,t−1 − βk ki,t−1 − βw wi,t−1 φi,t = βk ki,t + βw wi,t − α ˜0 + α ˜ 1 φi,t−1 − βk ki,t−1 − βw wi,t−1 − α (18) 3  ˆ ˜ ˜ ˜ ˜ 4 ei,t−1 − α ˜ 5 mi,t−1 − α ˜ 6 ei,t−1 mi,t−1 − ζi,t +α ˜ 3 φi,t−1 − βk ki,t−1 − βw wi,t−1 − α where β˜w = (1 − ξd )β˜w , and the other coefficients are defined in the same way as in equation (12). Equation (18) is estimated by running a nonlinear least squares program. The predicted values of a firm’s productivity are obtained from the equation ω ˆ i,t = ξˆ 1−1 φˆi,t + βˆw wi,t + βˆk ki,t . d Columns (5) and (6) of Table 13 show that the estimated wage parameter is negative and statistically significant, and its size is comparable under the discrete and continuous cases. However, as compared to βw reported in Table 7, βw obtained at the second stage is substantially lower in magnitude, albeit identical sign and significance level. This indicates that failure to account for the endogeneity of wage in the cost function results in upward bias in the estimated wage coefficient, which indicates a positive correlation between a firm’s wage and its productivity. However, the coefficients on ki,t and productivity equation are exceedingly similar both in size and significance to the original estimates in Table 7. Diffuse priors The market cost and export revenue parameters reported above are obtained by implementing the MCMC algorithm with fairly diffuse priors. It is a standard practice to undertake a prior sensitivity analysis, especially in view of the possibility that the variance of the prior distribution can have an effect on the posterior distribution. For this purpose, I double the standard deviations of all the prior distributions with the exception of the priors on the export revenue function parameters. Table 14 reports the mean and the standard deviation of the dynamic parameter estimates from the new priors along with estimates from Table 10. Comparison of the new to the baseline estimates shows that both sets of estimates are of the same order. Noticeable differences may be observed for parameters on fixed costs for small exporting firms and large importing firms, export 55 market size and autocorrelation. In addition to choices of priors, sampling error partly explains these differences in the estimates. In general, the static and dynamic parameter estimates of the model are shown to be robust to choice of functional forms and estimating assumptions. Also, the empirical model performs well in terms of fitting the actual data. In the next section, I use the estimated empirical model to do a counterfactual analysis. 6.5 Counterfactual experiment Table 15 shows the predicted effects of sunk and fixed cost reductions and export market enlargement on the transition probabilities of firms entering and exiting export and import markets. The parameter estimates used in these analyses are those obtained under size heterogeneity and discrete trade participation decisions of entering or exiting export and import markets. The first column shows the fitted values of probabilities of entry and exit in export and import markets before any change in the policy environment, which is constructed from Table 12 (Predicted 2). In the first scenario, I consider a hypothetical situation where sunk and fixed costs of exporting decline by 50% while everything else remains unchanged. We see that the predicted probabilities of entry into exporting and importing are greater than their baseline equivalents. Not surprisingly, the size of the increase in these probabilities is greater for exporting. These probabilities are also larger for firms with some form of prior involvement in international trade. Furthermore, we see that the predicted exit probabilities from both export and import markets are smaller as compared to the exit rates before the policy change. Similarly, I now examine the effect of a 50% decline in sunk and fixed costs of importing on entry and exit rates of trade participation. As in the previous case, import market cost reduction increases the likelihood of firms entering into exporting and importing and decreases exit rates from both activities. And, we observe that these effects are more pronounced for importing activities. Finally, I consider a policy environment where there is an enlargement of the export market by 50% while market entry costs remain unchanged. This market enlargement increases firm profitability from exporting and therefore induces firm export participation. At the same time, this reduces the rate of firm exit in both exporting and importing activities. The only exception here is that two-way trading firms are now more likely to quit importing. 7 Conclusion Using a panel of firms in the Danish manufacturing sector, I summarize interesting stylized facts on firms in international trade. The descriptive summaries reveal substantial firm heterogeneity in terms of size and wage payment besides productivity. In the data, we see that exporting is the most common trading activity, and there is a high frequency of simultaneous exporting and importing 56 within firms. In addition, there is a high persistence in the trading status of firms indicating a prominent role of start-up and running costs that international trade entails. In light of these empirical regularities, I structurally estimate a dynamic discrete choice model of exporting and importing. The model provides a framework to examine trade participation decisions while allowing for a firm’s current choice to affect its future productivity. The estimation results reveal that export markets are characterized by a more elastic demand, which indicates tougher competition essentially due to the presence of more firms and product varieties. In response to this tough competition, firms charge a lower markup in export markets as compared to their markup in the local market. Parameter estimates of the marginal cost function show that by providing employees with more capital to work with, firms can effectively postpone the diminishing returns to labor and enhance efficiency. Also, firms can further lower their costs of production by eliciting more effort from their employees by paying higher wages. It may also be the case that firms attract a high quality workforce and improve efficiency by paying higher wages. In terms of productivity growth over time, we see that trading firms experience post-entry productivity improvements, which are particularly larger for importing firms. Further, it is shown that exporting and importing involve substantial sunk and fixed costs of operation. And, these costs increase with the scale of operations, consistent with the expectation that large firms are likely to have numerous market-product exporting and importing relations which are costly to start up and maintain. The magnitude of these costs of trading leads to a sorting pattern where firms which are sufficiently productive succeed in becoming exporters and importers. The learning effects firms experience from trade participation further magnify the selection process into exporting and importing. The results from the counterfactual analyses illustrate that a policy change which influences the distribution of market costs of exporting and importing or profitability abroad has a significant effect on firm trade participation. Whether or not the policy change is directed towards exporting or importing, it will have direct and indirect effects on all firms due to the complementarity of these activities. For future research endeavor, it is interesting to explore these entry and exit decisions while explicitly accounting for firm product scope. This is useful given the highly skewed distribution of these activities across products within firms. It will also be insightful to examine the entry and exit decisions at the individual market level. 57 References Abreha, K. G., Smeets, V., & Warzynski, F. (2014). Coping with the Crisis and Export Diversification. Aarhus University. Altomonte, C., & B´ek´es, G. (2009). Trade complexity and productivity. IEHAS Discussion Papers # 914, Institute of Economics, Hungarian Academy of Sciences. Amiti, M., & Konings, J. (2007). Trade Liberalization, Intermediate Inputs, and Productivity: Evidence from Indonesia. American Economic Review , 97 (5), 1611–1638. Antr`as, P. (2003). Firms, Contracts, and Trade Structure. Quarterly Journal of Economics, 1375–1418. Antr`as, P., & Helpman, E. (2004). Global Sourcing. Journal of Political Economy, 112 (3), 552–580. Aristei, D., Castellani, D., & Franco, C. (2013). Firms’ Exporting and Importing Activities: Is There a Two-way Relationship? Review of World Economics, 149 (1), 55–84. Arkolakis, C., & Muendler, M. A. (2010). The Extensive Margin of Exporting Products: A Firm-level Analysis. NBER Working Paper #16641 . Arnold, J. M., & Hussinger, K. (2010). Exports versus FDI in German Manufacturing: Firm Performance and Participation in International Markets. Review of International Economics, 18 (4), 595–606. Aw, B. Y., Roberts, M. J., & Xu, D. Y. (2011). R&D Investment, Exporting, and Productivity Dynamics. American Economic Review , 101 (4), 1312–1344. Bas, M. (2012). Input-trade Liberalization and Firm Export Decisions: Evidence from Argentina. Journal of Development Economics, 97 (2), 481–493. Bas, M., & Strauss-Kahn, V. (2014). Does Importing More Inputs Raise Exports? Firm-level Evidence from France. Review of World Economics, 150 (2), 241–275. Bernard, A. B., & Jensen, J. B. (1995). Exporters, Jobs, and Wages in US Manufacturing: 1976-1987. Brookings Papers on Economic Activity. Microeconomics, 1995 , 67–119. Bernard, A. B., & Jensen, J. B. (2004). Why Some Firms Export. Review of Economics and Statistics, 86 (2), 561–569. Bernard, A. B., Jensen, J. B., Redding, S. J., & Schott, P. K. (2007). Firms in International Trade. Journal of Economic Perspectives, 21 (3), 105–130. Bernard, A. B., Jensen, J. B., Redding, S. J., & Schott, P. K. (2012). The Empirics of Firm Heterogeneity and International Trade. Annual Review of Economics, 4 (1), 283–313. Bernard, A. B., Redding, S. J., & Schott, P. K. (2011). Multiproduct Firms and Trade Liberalization. Quarterly Journal of Economics, 126 (3), 1271–1318. Bernard, A. B., Smeets, V., & Warzynski, F. (2014). Rethinking Deindustrialization. Economics Working Papers #2014-14, Aarhus University. Castellani, D., Serti, F., & Tomasi, C. (2010). Firms in International Trade: Importers’ and Exporters’ Heterogeneity in Italian Manufacturing Industry. World Economy, 33 (3), 424– 457. Clerides, S. K., Lach, S., & Tybout, J. R. (1998). Is Learning by Exporting Important? Microdynamic Evidence from Colombia, Mexico, and Morocco. Quarterly Journal of Economics, 903–947. Das, S., Roberts, M. J., & Tybout, J. R. (2007). Market Entry Costs, Producer Heterogeneity, and Export Dynamics. Econometrica, 75 (3), 837–873. Delgado, M. A., Farinas, J. C., & Ruano, S. (2002). Firm Productivity and Export Markets: A Non-parametric Approach. Journal of International Economics, 57 (2), 397–422. 58 Eckel, C., & Neary, J. P. (2010). Multi-product Firms and Flexible Manufacturing in the Global Economy. Review of Economic Studies, 77 (1), 188–217. Eriksson, T., Smeets, V., & Warzynski, F. (2009). Small Open Economy Firms in International Trade: Evidence from Danish Transactions-level Data. Nationaløkonomisk Tidsskrift/Danish Economic Journal , 147 (2), 175–194. Gilks, W. R., Richardson, S., & Spiegelhalter, D. (1996). Markov Chain Monte Carlo in Practice. New York, NY: Chapman & Hall/CRC. Goldberg, P. K., Khandelwal, A. K., Pavcnik, N., & Topalova, P. (2010). Imported intermediate inputs and domestic product growth: Evidence from India. Quarterly Journal of Economics, 125 (4), 1727–1767. Grossman, G. M., & Rossi-Hansberg, E. (2008). Trading Tasks: A Simple Theory of Offshoring. American Economic Review , 98 (5), 1978–1997. Halpern, L., Koren, M., & Szeidl, A. (2009). Imported Inputs and Productivity. CeFiG Working Papers # 08 . Hayakawa, K., Machikita, T., & Kimura, F. (2012). Globalization and Productivity: A Survey of Firm-level Analysis. Journal of Economic Surveys, 26 (2), 332–350. Hummels, D., Ishii, J., & Yi, K.-M. (2001). The Nature and Growth of Vertical Specialization in World Trade. Journal of International Economics, 54 (1), 75–96. Kasahara, H., & Lapham, B. (2013). Productivity and the Decision to Import and Export: Theory and Evidence. Journal of International Economics, 89 (2), 297–316. Levinsohn, J., & Petrin, A. (2003). Estimating Production Functions Using Inputs to Control for Unobservables. Review of Economic Studies, 70 (2), 317–341. Mayer, T., Melitz, M. J., & Ottaviano, G. I. P. (2014). Market Size, Competition, and the Product Mix of Exporters. American Economic Review , 104 (2), 495–536. Melitz, M. J. (2003). The Impact of Trade on Intra-industry Reallocations and Aggregate Industry Productivity. Econometrica, 71 (6), 1695–1725. Melitz, M. J., & Ottaviano, G. I. (2008). Market Size, Trade, and Productivity. Review of Economic Studies, 75 (1), 295–316. Muˆ uls, M., & Pisu, M. (2009). Imports and Exports at the Level of the Firm: Evidence from Belgium. World Economy, 32 (5), 692–734. Nunn, N., & Trefler, D. (2008). The Boundaries of the Multinational Firm: An Empirical Analysis. In E. Helpman, D. Marin, & T. Verdier (Eds.), The Organization of Firms in a Global Economy (pp. 55–83). Cambridge: Harvard University Press. Redding, S. J. (2010). Theories of Heterogeneous Firms and Trade. NBER Working Paper #16562 . Roberts, M. J., & Tybout, J. R. (1997). The Decision to Export in Colombia: An Empirical Model of Entry with Sunk Costs. American Economic Review , 545–564. Smeets, V., & Warzynski, F. (2013). Estimating Productivity with Multi-product Firms, Pricing Heterogeneity and the Role of International Trade. Journal of International Economics, 90 (2), 237–244. UNCTAD. (2014). Key Trends in International Merchandise Trade. New York and Geneva. Vogel, A., & Wagner, J. (2010). Higher Productivity in Importing German Manufacturing Firms: Self-selection, Learning from Importing, or Both? Review of World Economics, 145 (4), 641–665. Wagner, J. (2007). Exports and Productivity: A Survey of the Evidence from Firm-level Data. World Economy, 30 (1), 60–82. Wagner, J. (2012). International Trade and Firm Performance: A Survey of Empirical Studies 59 since 2006. Review of World Economics, 148 (2), 235–267. Wooldridge, J. M. (2002). Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press. Yi, K.-M. (2003). Can Vertical Specialization Explain the Growth of World Trade? Journal of Political Economy, 111 (1), 52–102. Zhang, H. (2015). Static and Dynamic Gains from Costly Importing of Intermediate Inputs: Evidence from Colombia. The University of Hong Kong Working Paper . 60 Table 1: Summary statistics on capital size, wage and productivity Capital Year 2000 2001 2002 2003 2004 2005 2006 2007 Wage Productivity Mean Median Std. Mean Median Std. Mean Median Std. 24.47 24.93 25.48 25.42 25.93 25.11 25.23 25.69 6.08 6.18 6.21 6.38 6.47 6.32 6.52 6.46 84.23 82.63 92.12 99.58 108.72 110.45 105.81 100.82 260.34 267.03 272.13 277.03 283.26 282.69 289.67 298.36 254.09 261.21 264.89 270.83 278.08 278.41 283.89 290.54 50.43 52.38 53.95 54.65 54.58 51.57 51.05 86.14 0.288 0.294 0.297 0.301 0.296 0.302 0.307 0.308 0.266 0.273 0.275 0.280 0.274 0.281 0.289 0.286 1.409 1.414 1.414 1.411 1.418 1.431 1.433 1.450 Note: Capital is in millions and wage in thousands of Danish Kroner. Productivity is estimated TFP using domestic revenue. All nominal values are deflated by the CPI index using 1995 as a base year. Table 2: Firm distribution by capital size Year 2000 2001 2002 2003 2004 2005 2006 2007 2000-2007 Domestic Export Import % Small % Large % Small % Large % Small % Large 70.86 75.00 76.89 77.30 78.18 75.38 74.12 73.44 75.15 29.14 25.00 23.11 22.70 21.82 24.62 25.88 26.56 24.85 41.89 41.24 42.42 42.57 41.95 42.35 43.09 42.89 42.30 58.11 58.76 57.58 57.43 58.05 57.65 56.91 57.11 57.70 39.96 39.58 41.71 41.81 41.63 41.70 41.96 42.70 41.38 60.04 60.42 58.29 58.19 58.37 58.30 58.04 57.30 58.62 Note: Small (large) firms are those with a capital size below (above) the median value of capital holding. 61 Table 3: Sample summary on firm trade participation Year Domestic Import-only Export-only Two-way # Firms % Firms # Firms % Firms # Firms % Firms # Firms % Firms 2000 2001 2002 2003 2004 2005 2006 2007 501 476 411 370 385 390 371 369 23.79 22.60 19.52 17.57 18.28 18.52 17.62 17.52 144 134 139 148 155 153 150 168 6.84 6.36 6.60 7.03 7.36 7.26 7.12 7.98 241 238 187 198 198 210 217 210 11.44 11.30 8.88 9.40 9.40 9.97 10.30 9.97 1,220 1,258 1,369 1,390 1,368 1,353 1,368 1,359 57.93 59.73 65.00 66.00 64.96 64.25 64.96 64.53 2000-2007 – 19.43 – 7.07 – 10.08 – 63.42 Table 4: Export and import participation premia Value added per worker Sales per worker Capital per worker Material per worker Energy per worker Wage Employment size Import-only Export-only Two-way 7.736∗∗∗ (0.012) 14.807∗∗∗ (0.014) 3.637 (0.036) 35.126∗∗∗ (0.026) 11.485∗∗∗ (0.027) 1.791∗∗∗ (0.006) 41.626∗∗∗ (0.030) 2.662∗∗ (0.010) 8.132∗∗∗ (0.012) 14.152∗∗∗ (0.032) 25.864∗∗∗ (0.023) 5.867∗∗ (0.024) 1.423∗∗∗ (0.005) 15.257∗∗∗ (0.027) 11.131∗∗∗ (0.008) 29.022∗∗∗ (0.009) 22.760∗∗∗ (0.024) 65.418∗∗∗ (0.018) 29.701∗∗∗ (0.019) 3.641∗∗∗ (0.004) 170.764∗∗∗ (0.0191) Standard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Industry and time fixed effects are included in these regressions. 62 Table 5: Kolmogorov-Smirnov test for equality of productivity distributions Export-only vs Import-only 2000 2001 2002 2003 2004 2005 2006 2007 ∗ H0 :Ω1 (x)-Ω0 (x) ≤ 0 H0 :Ω1 (x)-Ω0 (x)= 0 Year 0.2091∗∗∗ 0.1802∗∗∗ 0.1662∗∗∗ 0.1066 0.1024 0.1543∗∗∗ 0.1303 -0.1274 Two-way vs Import-only Two-way vs Export-only Export-only vs Import-only Two-way vs Import-only 0.2262∗∗∗ 0.2320∗∗∗ 0.1598∗∗ 0.2965∗∗∗ 0.2503∗∗∗ 0.2870∗∗∗ 0.2661∗∗∗ 0.2003∗∗∗ 0.3657∗∗∗ 0.3757∗∗∗ 0.2854∗∗∗ 0.3189∗∗∗ 0.3164∗∗∗ 0.3710∗∗∗ 0.3380∗∗∗ 0.2908∗∗∗ -0.2091∗∗∗ -0.1802∗∗ -0.1662∗∗ -0.1066 -0.1024 -0.1543∗∗ -0.1303 -0.1274 0.0000 -0.0016 -0.0059 -0.0043 -0.0022 0.0000 -0.0015 -0.0051 Two-way vs Export-only -0.0033 0.0000 -0.0015 -0.0029 -0.0066 -0.0015 0.0000 -0.0045 p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 Table 6: Transition probabilities of firms Status t+1 Import-only Export-only Domestic Status t Two-way Domestic Import-only Export-only Two-way 80.85 18.28 13.03 0.46 7.78 55.03 2.55 2.36 8.68 3.23 57.76 3.36 2.69 23.46 26.66 93.82 Cross-sectional Average 18.80 7.10 9.89 64.20 63 Table 7: Demand, marginal cost and productivity estimation Demand 1–1/ξ σ P/MC Cost Domestic Export 0.6451∗∗∗ (0.0672) 0.8644∗∗∗ (0.0246) 2.8179 7.3731 1.55 1.16 -0.0946∗∗∗ (0.0018) -0.5495∗∗∗ (0.0357) βk βw Productivity α0 α1 α2 α3 α4 α5 α6 Discrete (e, m) Continuous (e, m) -0.0115∗ (0.0066) 1.0268∗∗∗ (0.0162) -0.1014∗∗∗ (0.0210) -0.0887∗∗∗ (0.0074) 0.0102∗∗ (0.0042) 0.0141∗∗∗ (0.0048) 0.0065 (0.0061) 0.1293 0.0177 (0.0153) 0.8800∗∗∗ (0.0443) -0.2476∗∗∗ (0.0384) -0.1076∗∗∗ (0.0099) 0.0030∗∗∗ (0.0011) 0.0043∗∗∗ (0.0010) 0.0003 (0.0003) 0.1209 σζ Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Industry and time fixed effects are included in the first stage of the estimation algorithm. Table 8: Reduced-form estimation of export and import participation Probit Bivariate Probit P r(ei,t = 1) P r(ei,t = 1) P r(ei,t = 1) P r(mi,t = 1) P r(mi,t = 1) P r(mi,t = 1) P r(ei,t = 1) P r(mi,t = 1) eit−1 0.185∗∗∗ (0.0109) 1.297∗∗∗ (0.0625) – 0.117∗∗∗ (0.0135) 0.672∗∗∗ (0.0794) 2.451∗∗∗ (0.0353) 0.192∗∗∗ (0.0110) 1.844∗∗∗ (0.0643) – 0.120∗∗∗ (0.0136) 1.036∗∗∗ (0.0757) – mit−1 – 0.100∗∗∗ (0.0138) 0.444∗∗∗ (0.0804) 2.251∗∗∗ (0.0376) 0.606∗∗∗ (0.0394) Yes Yes 0.61 – 2.226∗∗∗ (0.038) Yes Yes 0.57 0.105∗∗∗ (0.0141) 0.943∗∗∗ (0.0772) 0.699∗∗∗ (0.0379) 2.000∗∗∗ (0.0360) Yes Yes 0.59 0.100∗∗∗ (0.0138) 0.452∗∗∗ (0.0805) 2.249∗∗∗ (0.0375) 0.607∗∗∗ (0.0392) Yes Yes – 0.106∗∗∗ (0.141) 0.936∗∗∗ (0.0771) 0.701∗∗∗ (0.377) 1.999∗∗∗ (0.0360) Yes Yes – kit ωit Year FE Industry FE Pseudo R2 Obs. Yes Yes 0.22 Yes Yes 0.60 Yes Yes 0.26 14,282 14,282 14,282 ρ = 0.2813 Standard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 64 Table 9: Reduced-form estimation of export revenue and import expenditure Export revenue Import expenditure (1) (2) (3) (4) ki,t 0.428∗∗∗ (0.019) 0.149∗∗∗ (0.195) 0.300∗∗∗ (0.023) 0.135∗∗∗ (0.024) ωi,t 3.120∗∗∗ (0.088) Yes Yes No 0.590∗∗∗ (0.091) Yes Yes Yes 3.523∗∗∗ (0.104) Yes Yes No 0.590∗∗∗ (0.113) Yes Yes Yes 0.45 0.03 0.38 0.04 – 0.897 – 0.882 Year FE Industry FE Firm FE R2 ρ Obs. 10,722 10,323 Standard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 65 Table 10: Market cost and export revenue parameters Firm size Firm size Priors Small Large All Small Large All γ xs 2471.749 (105.789) 5915.596 (37.977) 3013.676 (19.773) 2549.697 (63.635) 5909.658 (33.329) 3008.301 (19.930) γ xs ∼ N (0, 1000) γ xf 99.601 (11.441) 395.151 (8.887) 157.509 (6.714) 99.439 (13.841) 394.023 (10.853) 158.838 (7.095) γ xf ∼ N (0, 1000) γ ms 2390.672 (66.847) 6100.426 (71.824) 3262.507 (30.199) 2429.102 (68.356) 6152.378 (37.116) 3280.487 (27.549) γ ms ∼ N (0, 1000) γ mf 133.903 (10.041) 448.210 (32.328) 158.233 (8.742) 129.753 (11.602) 475.610 (9.168) 154.663 (5.098) γ mf ∼ N (0, 1000) δx 4.323 (0.196) 2.683 (0.392) 4.171 (0.124) 2.353 (0.228) δx ∼ N (0, 100) ρ 0.648 (0.045) 0.693 (0.041) 0.644 (0.047) 0.673 (0.043) ρ ∼ U (−1, 1) log σν -0.217 (0.004) -0.212 (0.004) -0.219 (0.002) -0.211 (0.004) log σν ∼ N (0, 10) Initial conditions η0 -4.778 (0.237) -3.955 (0.318) -4.733 (0.147) -3.778 (0.157) η0 ∼ N (0, 100) η1 1.953 (0.378) 2.604 (0.252) 2.281 (0.209) 2.678 (0.258) η1 ∼ N (0, 100) η2 1.397 (0.141) 1.711 (0.246) 1.437 (0.073) 1.670 (0.261) η2 ∼ N (0, 100) η3 -0.476 (0.246) -0.750 (0.235) -0.290 (0.196) -0.895 (0.146) η3 ∼ N (0, 100) µ0 -4.654 (0.266) -5.863 (0.208) -4.450 (0.141) -5.933 (0.243) µ0 ∼ N (0, 100) µ1 1.250 (0.131) 0.686 (0.173) 1.287 (0.148) 0.694 (0.182) µ1 ∼ N (0, 100) µ2 1.836 (0.317) 2.358 (0.194) 2.103 (0.156) 2.488 (0.133) µ2 ∼ N (0, 100) µ3 0.129 (0.059) 0.094 (0.044) 0.089 (0.034) 0.090 (0.039) µ3 ∼ N (0, 100) Iteration 30,000 30,000 Burn-in 33 % 67% Note: Standard errors in parentheses. The market cost estimates are in thousands of Danish Kroner (DKK). # Capital grids=10, # Productivity grids=100, # Export demand shock =100 and λ=0.95. 66 Table 11: In-sample model performance: trade participation rate Year Domestic Import-only Export-only Two-way Actual Predicted Actual Predicted Actual Predicted Actual Predicted 2000 2001 2002 2003 2004 2005 2006 2007 23.79 22.60 19.52 17.57 18.28 18.52 17.62 17.52 22.86 16.93 14.34 13.05 10.62 9.76 8.89 8.36 6.84 6.36 6.60 7.03 7.36 7.26 7.12 7.98 6.31 6.90 6.42 5.07 5.55 5.18 4.31 3.88 11.44 11.30 8.88 9.40 9.40 9.97 10.30 9.97 11.16 16.28 19.46 21.02 22.26 22.48 22.70 23.45 57.93 59.73 65.00 66.00 64.96 64.25 64.96 64.53 59.68 59.89 59.78 60.86 61.56 62.59 64.10 64.31 2000-2007 19.43 13.10 7.07 5.45 10.08 19.85 63.42 61.60 Table 12: In-sample model performance: transition matrix Domestic Import-only Status t Export-only Two-way Domestic Status t+1 Import-only Export-only Both Actual Predicted 1 Predicted 2 80.85 69.84 79.92 7.78 13.13 8.84 8.68 12.58 8.38 2.69 4.45 2.86 Actual Predicted 1 Predicted 2 18.28 7.58 14.25 55.03 71.35 67.17 3.23 0.87 0.95 23.46 20.20 17.63 Actual Predicted 1 Predicted 2 13.03 1.13 2.08 2.55 0.10 0.15 57.76 86.95 82.64 26.66 11.82 15.13 Actual Predicted 1 Predicted 2 0.46 0.13 0.20 2.36 0.39 0.42 3.36 15.41 9.84 93.82 84.07 89.54 Note: Predicted 1 and 2 are simulated using estimates in columns (1)-(2) and (4)-(5) of Table 10, respectively. The predicted probabilities are obtained by averaging over 1000 simulations. 67 Table 13: Robustness: Linear Markov process and wag parameter Discrete (e, m) Continuous (e, m) Discrete (e, m) Continuous (e, m) Discrete (e, m) Continuous (e, m) -0.0889∗∗∗ (0.0060) – -0.0716∗∗∗ (0.0066) – -0.0896∗∗∗ (0.0059) – -0.0723∗∗∗ (0.0065) – 2 ωi,t−1 0.9303∗∗∗ (0.0034) – 0.9378∗∗∗ (0.0040) – 0.9307∗∗∗ (0.0034) – 0.9381∗∗∗ (0.0039) – 3 ωi,t−1 – – – – ei,t−1 0.0038 (0.0041) 0.0052 (0.0047) 0.0087 (0.0060) 0.1272 0.0021∗ (0.0011) 0.0029∗∗∗ (0.0010) 0.0001 (0.003) 0.1198 0.0080∗∗∗ (0.0030) 0.0106∗∗∗ (0.0030) – 0.0016∗∗ (0.0007) 0.0025∗∗∗ (0.0006) – 0.1272 0.1198 -0.1581∗∗∗ (0.0603) -0.0758∗∗∗ (0.0084) 0.5882∗∗∗ (0.1061) -0.3578∗∗∗ (0.0506) -0.0894∗∗∗ (0.0074) 0.0102∗∗ (0.0042) 0.0138∗∗∗ (0.0048) 0.0059 (0.0061) 0.1289 -0.3485∗∗∗ (0.1357) -0.0690∗∗∗ (0.0103) 0.2046 (0.1913) -0.5281∗∗∗ (0.0767) -0.1075∗∗∗ (0.0100) 0.0028∗∗ (0.0011) 0.0042∗∗∗ (0.0010) 0.0002 (0.0003) 0.1206 Constant wi,t ωi,t−1 mi,t−1 ei,t−1 × mi,t−1 σζ Standard errors in parentheses, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 Table 14: Market cost and export revenue parameters: more diffuse priors Firm size Firm size Priors Small Large All γ xs 2549.697 (63.635) 5909.658 (33.329) 3008.301 (19.930) γ xf 99.439 (13.841) 394.023 (10.853) γ ms 2429.102 (68.356) γ mf 129.753 (11.602) Priors Small Large All γ xs ∼ N (0, 1000) 2105.818 (58.823) 5928.713 (32.487) 2923.326 (57.493) γ xs ∼ N (0, 2000) 158.838 (7.095) γ xf ∼ N (0, 1000) 166.478 (12.177) 327.058 (18.122) 132.502 (11.683) γ xf ∼ N (0, 2000) 6152.378 (37.116) 3280.487 (27.549) γ ms ∼ N (0, 1000) 2131.701 (52.929) 5734.498 (33.536) 3183.016 (36.280) γ ms ∼ N (0, 2000) 475.610 (9.168) 154.663 (5.098) γ mf ∼ N (0, 1000) 165.490 (6.228) 342.958 (7.320) 167.340 (14.115) γ mf ∼ N (0, 2000) δx 4.171 (0.124) 2.353 (0.228) δx ∼ N (0, 100) 4.276 (0.383) 3.889 (0.152) δx ∼ N (0, 100) ρ 0.644 (0.047) 0.673 (0.043) ρ ∼ U (−1, 1) 0.775 (0.059) 0.887 (0.049) ρ ∼ U (−1, 1) log σν -0.219 (0.002) -0.211 (0.004) log σν ∼ N (0, 10) -0.215 (0.003) -0.212 (0.002) log σν ∼ N (0, 10) Note: Standard errors in parentheses. The market cost estimates are in thousands of Danish Kroner. Iteration=30,000, Burn-in=67%, # Capital grids=10, # Productivity grids=100, # Export demand shock grids=100 and λ=0.95. 68 Table 15: Counterfactual analysis 0 Initial γ xs = 0 γ xf = 1 xs γ 2 1 xf γ 2 0 γ ms = 0 γ mf = 1 ms γ 2 1 mf γ 2 0 δx = 3 δ 2 x Prob. of entry into export Domestic Import only 11.24 18.58 27.59 36.64 15.03 17.52 25.00 29.82 Domestic Export only 11.70 15.28 19.41 20.22 28.60 29.26 19.25 11.39 Export only Both 2.23 0.62 1.49 0.29 2.38 0.73 0.73 0.30 Import only Both 15.20 10.04 12.71 7.97 3.73 3.48 9.15 15.28 Prob. of entry into import Prob. of exit from export Prob. of exit from import 69 Appendix A: Computation of value functions Below are the steps followed in the computation of the value functions: 1. Discretize the continuous state variables into finite set sc ∈ (sc1 , sc2 , ..., scM ), where sc = (k, ω, z). 2. Initialize the algorithm V 0 (s) and draw values of γ xs , γ xf , γ ms and γ mf from the parameter space. 0 c 3. Compute EV (s ) = 1 M M X c c |s ) V 0 (scm ) PMf (smf (s , where f (scm ) is transition density of sc . c |sc ) m=1 m m=1 4. Depending on previous period import status m−1 , the option values of exporting V E0 (m−1 ) and and non-exporting V D0 (m−1 ) are obtained as follows:   V E0 (m−1 ) = P λEV 0 (e = 1, m = 1) − λEV 0 (e = 1, m = 0) > m−1 γ mf + (1 − m−1 )γ ms × EV 0 (e = 1, m = 1) − m−1 γ mf − (1 − m−1 )γ ms   + P λEV 0 (e = 1, m = 1) − λEV 0 (e = 1, m = 0) ≤ m−1 γ mf + (1 − m−1 )γ ms × EV 0 (e = 1, m = 0) V D0  0 0 (m−1 ) = P λEV (e = 0, m = 1) − λEV (e = 0, m = 0) > m−1 γ mf + (1 − m−1 )γ ms  × EV 0 (e = 0, m = 1) − m−1 γ mf − (1 − m−1 )γ ms   0 0 mf ms + P λEV (e = 0, m = 1) − λEV (e = 0, m = 0) ≤ m−1 γ + (1 − m−1 )γ × EV 0 (e = 0, m = 0) 5. Based on previous period export status e−1 , the final step is obtaining V 1 (s)   V 1 (s) = π d + P π x + V E0 (m−1 ) − V D0 (m−1 ) > e−1 γ xf + (1 − e−1 )γ xs   x E0 xf xs × π + V (m−1 ) − e−1 γ − (1 − e−1 )γ   + P π x + V E0 (m−1 ) − V D0 (m−1 ) ≤ e−1 γ xf + (1 − e−1 )γ xs × V D0 (m−1 ) 6. Repeat steps 3-5 until |V r+1 − V r |< ε, which constitutes the convergence criterion. 70 Appendix B: Demand, cost and productivity: industry-level analysis The main analysis does not explicitly allow for variations across industries. In this appendix, I explore some industry disaggregation in the first stages of the estimation procedure, which estimate the parameters of the demand, marginal cost and productivity functions. Table B.1 reports the results of this exercise. And, it displays non-negligible cross-industry variations for the selected industries under study. For instance, the demand elasticities in the domestic market range from 2.53 to 8.01, and the implied values of the markups lie within the interval 14-65%. In the export market, the values of demand elasticities and markups are within the range 2.0-9.28 and 12-100%. We note that demand elasticities are not necessarily more elastic and markups lower in export markets as compared to the domestic market. In terms of productivity gain, the results show that firms in the basic and fabricated metals benefit from both exporting (1%) and importing (1.4%). Similarly, firms in the furniture industry gain from their past importing activity (10%). The results also show cross-industry differences in productivity change coming from how intensively firms undertake exporting and importing instead of switching their trading status. In this paper, when the market costs and export revenue parameters are estimated, I assume a cost distribution common for all firms. This assumption may not be too restrictive given that in the model firms face individual, time-varying sunk and fixed costs of exporting and importing. With regard to industry-level analysis, it is essential to mention some concerns for estimation within the current setting. The fact that some industries are populated by few firms makes the estimation task harder. Further, the small share of firms switching their trading status over the sample period prevents the exploitation of within firm variation for precise estimation of the parameters. Further, the use of a balanced panel in a sector with a relatively high firm turnover rate (15-20%) and longer time horizon (8 years) magnifies the problem, albeit it substantially reduces the practical burden of estimating the market costs and export revenue parameters. For future research purpose, adopting an alternative approach based on unbalanced panel data and which accounts for entry and exit decisions in production, without necessarily recovering the market costs of domestic operation, is necessary. 71 Table B.1: Demand, marginal cost and productivity by industry (13+14) Printing and publishing (17+18) Rubber and plastic (22) Basic and fabricated metals (24+25) Machinery and equipment (26-30) 163 1,304 26.38 67 536 13.43 163 1,304 60.74 127 1,016 33.86 384 3,072 52.60 449 3,592 29.62 119 952 47.06 Exporters (%) Importers (%) 80.67 80.60 92.72 93.66 71.01 53.30 91.04 89.27 59.21 53.13 87.86 84.21 77.63 69.33 1 − 1/ξ d 0.612∗∗∗ (0.290) 0.752∗∗∗ (0.031) 0.723∗∗∗ (0.020) 0.702∗∗∗ (0.029) 0.710∗∗∗ (0.080) 0.875∗∗∗ (0.047) 0.622∗∗∗ (0.042) 1 − 1/ξ x 0.892∗∗∗ (0.013) 0.689∗∗∗ (0.029) 0.750∗∗∗ (0.024) 0.654∗∗∗ (0.0288) 0.849∗∗∗ (0.018) 0.508∗∗∗ (0.052) 0.601∗∗∗ (0.061) σd σx 2.533 9.275 4.026 3.213 3.054 4.003 3.360 2.888 3.437 6.624 8.005 2.033 2.647 2.503 (P/M C)d (P/M C)x 1.65 1.12 1.33 1.45 1.49 1.33 1.42 1.53 1.41 1.18 1.14 2.00 1.61 1.67 βk -0.112∗∗∗ (0.009) -0.054∗∗ (0.006) -0.118∗∗∗ (0.006) -0.079∗∗∗ (0.007) -0.078∗∗∗ (0.003) -0.022∗∗∗ (0.001) -0.146∗∗∗ (0.010) βw -0.620∗∗∗ (0.139) -0.092∗∗∗ (0.150) -0.225∗∗∗ (0.081) -0.237∗∗∗ (0.559) -0.420∗∗∗ (0.0839) -0.137∗∗∗ (0.090) -0.591∗∗∗ (0.251) α0 26.090∗∗∗ (9.385) 279.590 (357.437) -66.245∗∗∗ (14.120) -675.947 (271.326) 17.294∗∗∗ (2.773) 46.986∗∗∗ (11.256) -534.979 (434.736) α1 -8.748∗∗∗ (4.341) 53.725∗∗∗ (2.127) -37.519∗∗∗ (7.706) -119.194∗∗∗ (1.058) -12.960∗∗∗ (2.385) -48.936∗∗∗ (12.307) -56.011∗∗∗ (1.359) α2 1.209∗∗∗ (0.321) 3.194∗∗∗ (0.310) -7.461∗∗∗ (1.405) -7.114∗∗∗ (0.121) 3.732∗∗∗ (0.684) 17.659∗∗∗ (4.484) -1.993∗∗∗ (0.083) α3 -0.050∗∗∗ (0.014) 0.061∗∗∗ (0.011) -0.481∗∗∗ (0.086) -0.140∗∗∗ (0.003) -0.331∗∗∗ (0.065) -2.079∗∗∗ (0.544) -0.023∗∗∗ (0.001) α4 0.011 (0.020) 0.035 (0.035) -0.007 (0.012) 0.003 (0.023) 0.010∗∗∗ (0.005) 0.002 (0.003) 0.030 (0.021) α5 -0.005 (0.020) 0.025 (0.032) 0.011 (0.018) 0.029 (0.024) 0.014∗∗∗ (0.007) 0.000 (0.005) 0.102∗∗∗ (0.029) α6 0.001 (0.026) 0.014 (0.043) 0.027 (0.021) -0.002 (0.027) 0.001 (0.008) 0.0010∗∗ (0.005) -0.044 (0.033) σζ 0.1449 0.0875 0.1323 0.1000 0.0822 0.0314 0.1581 α0 42.972∗∗∗ (6.391) 269.103 (374.058) -54.379∗∗∗ (16.866) -715.611∗∗∗ (276.469) 14.591∗∗∗ (3.401) 30.248∗∗∗ (9.987) 368.520 (250.986) α1 -18.616∗∗∗ (2.927) 49.863∗∗∗ (2.493) -32.387∗∗∗ (9.618) -127.164∗∗∗ (1.160) -12.182∗∗∗ (3.250) -30.954∗∗∗ (11.041) 25.941∗∗∗ (1.930) α2 2.971∗∗∗ (0.460) 2.808∗∗∗ (0.361) -6.822∗∗∗ (1.834) -7.639∗∗∗ (0.134) 3.957∗∗∗ (1.041) 11.232∗∗∗ (4.066) 0.487∗∗∗ (0.117) α3 -0.149∗∗∗ (0.025) 0.049∗∗∗ (0.013) -0.464∗∗∗ (0.117) -0.152∗∗∗ (0.004) -0.395∗∗∗ (0.111) -1.314∗∗∗ (0.499) 0.002 (0.002) α4 -0.003 (0.004) 0.006 (0.005) 0.001 (0.004) 0.002∗∗∗ (0.003) 0.003 (0.002) 0.002 (0.001) 0.004 (0.008) α5 -0.001 (0.005) 0.016∗∗∗ (0.008) 0.004 (0.005) 0.005 (0.003) 0.005∗∗∗ (0.002) 0.001∗∗ (0.001) 0.011∗ (0.006) α6 -0.001 (0.001) 0.001 (0.002) 0.000 (0.001) -0.000 (0.001) 0.001 (0.000) 0.000 (0.000) 0.001 (0.002) σζ 0.1347 0.0846 0.1379 0.1018 0.0747 0.0307 0.1578 NACE-Rev.2 # Firms # Firm-year Firms switching status (%) Food and beverages (10+11) Textiles Furniture (31) Discrete (e, m) Continuous (e, m) Standard errors in parentheses , ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 72 Appendix C: Figures Figure C.1: Trace plot: sunk export cost 2650 2600 2550 2500 2450 2400 2350 2300 2250 γ xs (Small firms) 6100 6050 6000 5950 5900 5850 5800 γ xs (Large firms) 3200 3150 3100 3050 3000 2950 γ xs (All firms) 73 Figure C.2: Trace plot: fixed export cost 160 150 140 130 120 110 100 90 80 γ xf (Small firms) 420 415 410 405 400 395 390 385 380 375 γ xf (Large firms) 180 170 160 150 140 130 120 γ xf (All firms) 74 Figure C.3: Trace plot: sunk import cost 2550 2500 2450 2400 2350 2300 2250 γ ms (Small firms) 6250 6200 6150 6100 6050 6000 5950 5900 γ ms (Large firms) 3350 3300 3250 3200 3150 3100 γ ms (All firms) 75 Figure C.4: Trace plot: fixed import cost 170 160 150 140 130 120 110 100 90 γ mf (Small firms) 500 480 460 440 420 400 380 γ mf (Large firms) 180 170 160 150 140 130 120 110 γ mf (All firms) 76 Figure C.5: Trace plot: export revenue parameters with size heterogeneity 5 4.8 4.6 4.4 4.2 4 3.8 δx 0.9 0.85 0.8 0.75 0.7 0.65 0.6 0.55 0.5 ρ −0.206 −0.208 −0.21 −0.212 −0.214 −0.216 −0.218 −0.22 −0.222 −0.224 −0.226 log σν 77 Figure C.6: Trace plot: export revenue parameters without size heterogeneity 4.5 4 3.5 3 2.5 2 1.5 δx 0.9 0.85 0.8 0.75 0.7 0.65 0.6 0.55 ρ −0.202 −0.204 −0.206 −0.208 −0.21 −0.212 −0.214 −0.216 −0.218 −0.22 −0.222 log σν 78 79 Chapter 3 Importing and Firm Productivity in Ethiopian Manufacturing 80 Importing and Firm Productivity in Ethiopian Manufacturing∗ Kaleb Girma Abreha† Abstract In this paper, I investigate the causal relationship between importing and firm productivity. Using a rich dataset from Ethiopian manufacturing over the period 1996-2011, I find that most firms source capital and intermediate goods from the world market. These firms are better performing as shown by significant, economically large import premia. I also find strong evidence of self-selection of more productive firms into importing, highlighting substantial import market entry costs. To examine the causal effect of importing on firm productivity, I use a model in which the static and dynamic effects of importing are separately estimated. The estimation results provide evidence of learning-by-importing. However, the small sizes of the productivity gains suggest the limited absorptive capacity of firms in the economy. JEL Codes: F14, L60 Keyword: Imported inputs, self-selection, learning-by-importing, Ethiopia ∗ I thank Val´erie Smeets, Fr´ed´eric Warzynki and Mark Roberts for helpful comments. I also thank the Tuborg Foundation for generous financial support. The usual disclaimer applies. † Department of Economics and Business, Aarhus University, Denmark, E-mail: [email protected] 81 1 Introduction International knowledge flows are considered fundamental components of globalization. Several studies identify foreign direct investment, trade, migration and others as important channels of international linkages and knowledge spillovers across countries. A pioneering contribution by Coe and Helpman (1995) on trade-driven international R&D spillovers documents significant R&D knowledge transfer across OECD countries. In a follow-up paper, Coe et al. (1997) show that the knowledge spillovers are not limited to developed countries in that developing countries substantially benefit from R&D investments elsewhere. Subsequent studies have investigated the importance of different channels of international linkages and knowledge spillovers mostly using aggregate data and adopting cross-country regressions. Recently, Acharya and Keller (2009) find that the contribution of international R&D spillovers to productivity normally exceeds that of domestic R&D, and the technology transfers are asymmetric across countries. They identify the asymmetry to emanate from differences in geographical distance between trading partners and the nature of goods traded. Other studies point out physical and human capital, policy and institutional quality, and relative backwardness from the technological frontier as important determinants of the pace and size of technology diffusion across countries. Acemoglu and Zilibotti (2001) emphasize that technology and skill mismatches lead to productivity differences across countries even when they have equal access to technology. Coe et al. (2009) demonstrate the importance of institutional factors on the degree of R&D spillovers. No (2009) shows that the scope and magnitude of international R&D spillovers depend on both production structure, pattern of international trade (in terms of volume and trading partners), national innovative and absorptive capacities of countries. However, restricting these investigations only to countries and industries masks varying roles of different technology transfer channels given that firms are characterized by marked heterogeneity in terms of global orientation, productivity, size, and factor intensity and payment even in narrowly defined industries. With increasing recognition of firm heterogeneity, the scope of research on globalization, mainly international trade, has expanded to include firms and products besides countries and industries. Consequently, the trend in the empirical trade literature has been characterized by a surge in using microeconomic data. Despite such a surge, prominent focus has been on the export side of international trade. It is only recently that studies started looking into importing and its relationship with other firm activities such as exporting. A common feature of these studies is that there is a positive, statistically significant and quantitatively large correlation between firm productivity and importing. However, the evidence on the causal relationship is mixed. For instance, Kasahara and Lapham (2013) for Chile, Vogel and Wagner (2010) for Germany, and Serti et al. (2010) and Castellani et al. (2010) for Italy find evidence of self-selection of more productive firms into importing whereas Forlani (2010) for Ireland and Smeets and Warzynski (2013) for Denmark find no supportive evidence. Evidence 82 on learning-by-importing is also inconclusive. Kasahara and Rodrigue (2008) for Chile, Smeets and Warzynski (2013) for Denmark, Halpern et al. (2009) for Hungary, Forlani (2010) for Ireland, Dovis and Milgram-Baleix (2009) and Augier et al. (2013) for Spain, and L¨o¨of and Andersson (2010) for Sweden find evidence of productivity gain from importing. On the contrary, Muendler (2004) for Brazil, Vogel and Wagner (2010) for Germany, Van Biesebroeck (2008) for Colombia and Zimbabwe find weak evidence. These mixed results are partly attributable to methodological choices or inherent differences in the nature of the import-productivity nexus across countries and over time. Therefore, a complete understanding of the causal relationship calls for further accumulation of empirical evidence from different countries. Partly due to data unavailability, with the exception of few studies, the new empirical trade literature is also characterized by neglect of firms in low-income countries. Mengistae and Pattillo (2004) find export productivity premia for Ethiopia, Ghana and Kenya. Bigsten et al. (2004) provide weak evidence of self-selection into the export market but strong evidence of learning-by-exporting for firms in Cameroon, Kenya, Ghana and Zimbabwe. Similarly, a study by Van Biesebroeck (2005) on Sub-Saharan manufacturing firms from Burundi, Cameroon, Cˆote d’Ivoire, Ethiopia, Ghana, Kenya, Tanzania, Zambia, and Zimbabwe shows that exporters are better performing, and there is a productivity gain from exporting. Relatedly, Bigsten and Gebreeyesus (2009) document evidence of selection of more productive firms into exporting as well as post-entry productivity improvement in the Ethiopian manufacturing. Foster-McGregor et al. (2014) analyze firm productivity differences in 19 Sub-Saharan African countries and show that those simultaneously exporting and importing are the most productive whereas their domestic counterparts are the least productive. The aforementioned studies on African manufacturing firms exclusively focus on the relationship between firm productivity and exporting.1 For instance, even though the Ethiopian manufacturing sector has been a subject of several empirical investigations, the importing behavior of firms in the sector has been ignored. This is puzzling given the dominant role of imports in the foreign trade of the country. Imports of goods and services form around 28% of GDP over the period 19962011 whereas the corresponding figure for exports is 13%. During the same period, manufacture imports and exports constitute around 70% and 9% of the total merchandise imports and exports, respectively. This provides a rationale for analyzing the import behavior of firms as an essential step towards a complete understanding of the nature of and gains from international trade at the firm and aggregate level in the context of a least developed country. In this paper, I investigate the causal link between importing and productivity using firms in Ethiopian manufacturing. I use a rich dataset over the period 1996-2011. Descriptive summaries of the data display that the majority of the firms are globally active, and importing is the most 1 An exception is a study by Foster-McGregor et al. (2014) which investigates productivity differences of firms importing besides exporting. One caveat of this study is its use of cross-sectional data, and therefore it neglects the dynamics over time. 83 common form of global activity. Besides, I find a positive, statistically significant and economically large import premia confirming previous findings which establish positive relationship between importing and productivity and other firm performance measures. In determining the direction of causality in the relationship between importing and productivity, I test for commonly known hypotheses—productive firms become importers, and importing makes firms productive. The main element of the first hypothesis is that there significant sunk and fixed costs of operation in import markets, and it is only firms which are sufficiently productive enough that succeed in souring inputs from abroad. To empirically test this, I consider a subsample of firms which were not importing in the past, and estimate the pre-entry import productivity premium between future importing and non-importing firms. The results of this exercises reveal a positive and significant premium, and therefore provide evidence of self-selection of more productive firms into importing. To examine whether or not there is productivity gain from becoming an importer, I distinguish between local and imported varieties of material inputs in the production function. Also, I treat import participation as an additional state variable which determines the evolution of a firm’s productivity as well as its exit and investment decisions. As an identification strategy, I exploit within firm changes in importing status as a source of variation and structurally estimate parameters of the production function and productivity evolution equation. This approach enables not only proper estimation of importing effects but also decomposition of the static and dynamic components. The estimation results show that there are dynamic productivity gains (1.1-1.2%) in the period after participating entering import market albeit a momentary adverse effect at the beginning. In the long-run, these effects amount to a firm productivity improvement of 3.5-4.9%. In addition, if the expenditure share of imported inputs doubles firm productivity by 2.1% immediately, 0.7% in the period after and 4-22% in the long-run. This paper is related to several contributions on firm international trade and economic growth literature. It is related to the large and growing literature on firm heterogeneity and international trade pioneered by Bernard and Jensen (1995). It is also related to trade-driven international knowledge spillovers literature pioneered by Coe and Helpman (1995) and Coe et al. (1997). Methodologically, I follow Kasahara and Rodrigue (2008) in the specification of the structural model. Unlike their study, possibilities of firm-level technology spillovers via imported inputs are studied in the context of a least developed country with a different macroeconomic environment, factor and product markets and at a different stage of economic development. By focusing on a typical least developed country, I accommodate not only the potentials of technology transfer but also the absorptive capacity of a country in determining the size and pace of trade-related international knowledge spillovers. In this respect, this paper is linked to a strand of literature in economic growth such as Acemoglu and Zilibotti (2001) who show a technology and worker skill mismatch leading to productivity differences even when countries have same access to technology. It is also related to Los and Timmer (2005) studying the pace of assimilation of spillovers of appropriate of 84 technology as an explanation for productivity differences across countries. At the microeconomic level, this study is connected to recent contributions by Yasar and Morrison Paul (2007) and Yasar and Paul (2008) investigating different channels of technology transfer, and Yasar (2013) and Augier et al. (2013) examining firm absorptive capacity and productivity effects of imported inputs. Unlike these studies, I consider the latest decade where the world experienced a surge in trade in intermediate goods.2 I also exploit the panel structure and longer time dimension of the data, which is rarely available for African manufacturing. To the best of my knowledge, related studies on African manufacturing firms are missing, and this study provides the first evidence. The rest of the paper is organized as follows. Section 2 provides background information on sectoral composition and international trade structure of the Ethiopian economy. Section 3 presents the data source and establishes a set of stylized facts. Section 4 provides evidence on selection of firms into importing. Section 5 develops a theoretical framework and an empirical strategy to examine any productivity gain from importing. It also discusses the estimation results. Section 6 concludes. 2 Overview of Ethiopian economy Ethiopia is one of the least developed countries according to the World Bank economic classification of countries. Typical of a least developed country, the economy has experienced a highly fluctuating growth pattern ranging from -3.46% in 1998 to 13.57% in 2004. However, the economy enjoyed nearly decades of rapid economic growth especially after 2003, albeit starting from a rather low level. On average, it expanded by 7.62% annually over the period 1996–2011. Below is a brief presentation of the salient features of the economy in terms of distribution of economic activities and reallocations across sectors. 2.1 Sectoral composition Ethiopian economy is highly agriculture-based. Table 1 shows that 48.02% of the value added in the economy come from agriculture. However, the sector experienced a decline in its contribution to the aggregate output from 55.35% in 1996 to 45.57% in 2011. This decline is due to a relatively slower sectoral growth rate—5.74% in agriculture, 6.60% in manufacturing and 10.30% in the service sector over the years 1996-2011. Owing to a heavy reliance on rain-fed agriculture, the growth rate in the sector has been characterized by extreme fluctuations ranging from -10.48% in 2003 to 16.96% in 1996. Regarding the manufacturing sector, Ethiopia has a very narrow industrial base. The share of the sector is very small, below 10%, for all the periods under the study. Despite its strong growth performance, the contribution of the sector to the value added in the economy 2 According to the UNCTAD (2014) report, trade in intermediate goods totaled USD 7 trillion and accounted for 40% of the world trade in 2011. 85 has declined recently. On the other hand, the service sector experienced a consistently high growth rate and saw its contribution rising over time. Table 2 summarizes the characteristics of the trade sector. We see that exports and imports constitute, on average, 12.99% and 28.01% of total output in the economy. These sectors registered rapid, yet fluctuating, growth rates; 12.13% and 12.77% respectively. These growth rates signify the increase in the share of exports and imports from 9.35% and 16.43% in 1996 to 17% and 32.14% in 2011, respectively, showing increasing openness and growing integration to the global economy. It is also important to note that the integration is dominantly through imports. We also see the widening of the country’s trade deficit over the years despite comparable export and import growth rates. In terms of traded items, manufacture exports constitute a very small portion of the overall merchandise export, 9.08%, whereas manufacture imports constitute a significantly higher proportion, 70.18%. While the share of manufacture exports remains more or less stable, the share of manufacture imports declined significantly from 84.49% in 1996 to 67.13% in 2011. 2.2 Geographic orientation Figure 1 shows the regional distribution of Ethiopian foreign trade. Trade is mainly concentrated in high-income countries in Europe and North America. The country trades 69.10% of the exports and 55.54% of the imports with these high-income economies. Given that advanced economies account for the largest share of R&D in the world, the concentration of trade with these countries makes trade a likely conduit for international knowledge spillover. Middle Eastern and Northern African, East Asian and Pacific, and Central and South Asian countries are the next important destinations for exports constituting 14.03%, 6.07%, and 3.60%, respectively. The respective figures for imports are 4.88%, 12.36%, and 8.40%. Trade with countries in Latin America and the Caribbean, and Sub-Saharan Africa is very small and expanded only incrementally over time. At the same time, we observe the declining importance of traditional markets namely high-income economies and the growing importance of trade partners in Asia especially on the import trade. To sum up, it is shown that the fundamental aspects of Ethiopian economy have not undergone major structural changes. However, there have been sizable changes in terms of sectoral output compositions and geographical orientations of international trade. Increasing overall openness of the economy along with the dominance and increasing importance of imports in the economy is particularly observed. These changes cause firms to adapt their behavior under different domestic and global economic circumstances. To this end, rigorous analyses of firm behavior in terms of market entry and exit decisions as well as the subsequent effects are required for a complete understanding of the nature, determinants and effects of international trade. 86 3 Basic facts from Ethiopian manufacturing In this section, I describe the data source and the variables defined. I also summarize the nature of firm international trade primarily focusing on importing activity. 3.1 Data description The dataset used in this paper comes from the Central Statistical Agency of Ethiopia. The agency conducts annual large and medium-scale surveys of firms engaged in manufacturing activities. Classification of economic activities as manufacturing is based on ISIC-Rev.3 classification and includes industries 15-37 at 2-digit ISIC. The survey covers all firms with at least 10 employees and which use power-driven machinery during the period 1996-2011. The dataset provides detailed information on the level of production, local and export sales, input usage, employee composition, and asset structure of firms. I define gross output as revenue generated from local and export sales. I construct the capital variable by exploiting the information on initial stock, investment, value sold and depreciated using the perpetual inventory method. Information on local and imported material inputs as well as energy is also available. Using information on employees, I differentiate between skilled workers: unpaid working proprietors; active partners and family workers, and administrative and technical employees, and unskilled workers: apprentice and production workers. Because the number of seasonal and temporary workers is infrequently reported, the measure of labor input is confined to the number of working proprietors; apprentices, and permanent employees. Finally, I deflate all nominal values using the consumer price index extracted from the World Development Indicators database. To have enough within industry variation, I regroup those industries with very few firms as other manufacturing. This group comprises firms in Tobacco, paper, basic metals, machinery and equipment, office equipment, electrical machinery, and motor vehicles industries. I exclude firms with zero or unreported value of production, capital stock, material input, energy expense, number of employees, and other inconsistencies to consider only firms with real economic activity. I also exclude firms appearing only once during the period under study. In the final dataset, there are 2,350 firms and 12,510 firm-year observations. 3.2 Stylized facts Here, I provide a simple description of the data that characterizes the international trade activities of Ethiopian manufacturing firms. Fact 1 There is a substantial variation in firm trade participation across industries. Table 3 shows large differences in firm export and import participation rates across industries. In 1996, textiles (20.83%), leather product (21.43%), wood products (7.69%), and wearing apparel (6.67%) were industries with relatively high export market participation rates. In contrast, food 87 and beverage (2.34%), non-metallic products (1.96%), and furniture (1.85%) had very low export participation rates. In the extreme case, there were no exporters in the printing and publishing, chemicals, rubber and plastic, and fabricated metals industries. In 2011, there has been a dramatic increase in the export participation rate of firms in each of the industries except the furniture industry. For instance, the participation rate increased to 20.20% in food and beverage, 50.00% in textile, 45.90% in leather products, and 21.05% in chemical industries. There are also substantial heterogeneities in firm import participation across industries. In 1996, chemicals (100%), rubber and plastic (100%), fabricated metals (89.47%), and textiles (83.33%) comprised industries with very high import participation rates. On the contrary, the participation rate was as low as 21.57% in the case of non-metallic products. Over the 15-year window, food and beverage, leather products, chemicals, rubber and plastic, and fabricated metals saw their import participation rates decline while the remaining industries experienced a rise. Comparing export and import participation rates, we observe that importing is the most common activity in which 69.74% of firms import while the corresponding figure for export is 5.86% over the period 1996-2011. These characteristics are against findings from manufacturing sectors of developed countries where the incidence of exporting is more common and importing is rarer.3 In terms of dynamics over time, we see that firms have become more globally engaged through exporting while a slightly lower fraction of firms import. Export participation rose from 4.72% in 1996 to 14.77% in 2011, and import participation declined from 67.87% in 1996 to 64.89% in 2011. In general, there are clear indications of increasing presence of firms in the world market mainly because more firms have started serving export markets. Fact 2 There are significant trade activity premia. I divide firms into four mutually exclusive groups based on their exporting and importing activities: domestic (neither exporting nor importing); export-only (serving domestic and export markets but not importing); import-only (serving domestic market, and importing), and two-way (serving domestic and export markets, and importing). To estimate the activity premium, I run the following regression equation: yi,t = β0 + β1 Expi,t + β2 Impi,t + β3 Bothi,t + Controlsi,t + δt + τ + i,t where yi,t refers to performance indicators TFP, output, capital, material, energy, employment size, and share of skilled workers. These performance measures are all in logarithmic scale. Expi,t , Impi,t , and Bothi,t are dummy variables taking a value of 1 if the firm is export-only, import-only or two-way, respectively. In the regression, I control for year δt and industry τ fixed effects as well as employment size (except for the last two indicators). All the estimates measure average 3 Findings by Bernard et al. (2007) for a large economy (the US) and Eriksson et al. (2009) for a small open economy (Denmark) are typical cases of greater export participation rate in the manufacturing sectors of developed countries. 88 percentage differences relative to domestic firms. Table 4 shows that two-way trading firms are the most productive whose production activity is characterized by intensive use of capital, material, energy and skilled workers. These firms are also the largest in size. Among firms partially engaged in trade, we see that export-only firms are more productive, capital-intensive and larger in size compared to import-only firms. The estimates indicate that export-only firms not necessarily use more energy per worker and hire more skilled workers. And, import-only firms are less capitalintensive, and they do not necessarily use more material per worker as compared to domestic firms. Fact 3 There is high persistence in firm trade status. Table 5 presents transition probabilities highlighting the dynamics of firm activities in terms of scope. There is a high state dependence of firms engaged in the domestic market (67.19%), import-only (84.30%) and two-way (76.95%). There is an exceptionally low incidence of state dependence among export-only firms; 33.75%. Firms engaged in either exporting or importing are more likely to add importing (38.75%) but less likely to add exporting (1.71%) as an additional activity compared to firms engaged in neither of the activities; 31.40% and 2.10% to start importing and exporting respectively. We here note the unexpectedly low probability of adding exporting as an additional firm activity. Furthermore, firms doing both are less likely to abandon any of these activities compared to those engaged in only one of them. These firms abandon exporting, importing or both with a probability of 16.43%, 10.62%, and 4.01%, respectively. In contrast, firms only exporting (importing) abandon exporting (importing) with a probability of 13.75% (13.76%). Finally, the average of the cross-sectional distribution of trade participation shows a high incidence of firm involvement in international trade; 71.41% of the firms are active in the world market via exporting, importing or both. To further highlight the extent of state dependence in firm activities, I run a probit regression of a firm’s current period import/export status Di,t on its previous period productivity tf pi,t−1 , capital stock ki,t−1 , size of employment li,t , prior import Mi,t−1 and export Xi,t−1 participation variables, and year δt and industry τ fixed effects. P r(Di,t = 1) = Φ (β0 + β1 tf pi,t−1 + β2 ki,t−1 + β3 li,t−1 + β4 Mi,t−1 + β5 Xi,t−1 + δt + τ + i,t ) Table 6 reports the marginal effects from the probit regressions. We see that productivity, capital, employment size and prior export market experiences are positively and significantly correlated with the likelihood of current period participation in export market. Comparable results are also found in the case of importing except now that large capital holding tend to lower the probability of importing. Statistical significance of the previous period trade status of firms shows the presence of market entry costs and complementarity between exporting and importing. The fact that the current period import status is significantly correlated with past export market participation strongly suggests that firms with export market experience are able to absorb at least part of the 89 import market entry costs, and therefore they are likely to become importers. Another plausible explanation is significant productivity gains from exporting further driving the self-selection of firms into importing.4 However, the complementary effect of importing in terms of increasing the likelihood of exporting is weak as can be seen from the insignificance of the marginal effect of previous importing in the export participation regression. From Tables (5) and (6), it seems that the high persistence of firms’ importing status makes the task of estimating the causal relationship between productivity and import market participation harder. In other words, without sufficient within firm variation in importing status, it is not possible to properly estimate such a relationship. However, Table 7 reports that 1,413 (60.13%) firms do not change their importing status; they either import or never use imported inputs throughout the sample period. In contrast, 937 (39.87%) firms change their importing status at least once during the same period. And 512 (27.78%) firms switch their import participation more than once. that there are enough variation in the data which can be used for precise estimation of the parameters of interest. Such within firm changes in importing provide the necessary variation required for identification of the effect of importing. 4 Selection into importing The empirical facts in the previous section establish that, on average, importing firms are more productive. They also indicate that there are substantial market entry costs associated with importing. One possible explanation for these empirical facts is that more productive firms self– select themselves into importing markets. To test the empirical validity of this argument, I plot the productivity distribution of firms in the periods prior to some of them start importing. Figure 2 shows the plot of the probability and cumulative densities of firm productivities on the vertical axis, and a normalized firm productivity (in logarithmic scale) on the horizontal axis. Here, the normalization is achieved by dividing the actual firm productivity by industry average to which the firm belongs with the objective of accounting for possible industry idiosyncrasies and the relative position of the firm in the industry. The first panel depicts that the density function of importing firms lies to the right of the productivity density of non-importing firms for all the time lags considered. Relatedly, in the second panel the cumulative density function of importing firms lies below that of non-importing firms. These features of the productivity distributions indicate that importing firms were more productive before becoming importers compared to firms currently not importing. This result implies that there is a selection of more productive firms into importing activity, in accordance with the presence of substantial market entry costs of importing. To use a standard approach in testing for the self-selection hypothesis, I run a regression of 4 Previous findings on African firms show substantial learning-by-exporting. In the Ethiopian case, using the same data source as in this paper, Bigsten and Gebreeyesus (2009) document a significant productivity gain of 15-26% from exporting over the period 1996-2005. 90 lagged values of productivity tf pi,t−s on current import status Mi,t , and control variables such as firm capital ki,t−s , employment size li,t−s , export market participation Xi,t−s as well as year δt and industry τ fixed effects. tf pi,t−s = β0 + β1 Mi,t + β2 ki,t−s + β3 li,t−s + β4 Xi,t−s + δt + τ + i,t ; s = 1, 2, 3 Table 8 presents estimates of percentage differences in productivity between current importers and non-importers periods prior to some of them becoming importers. I find a positive and highly significant estimate for the current period dummy implying that these firms were actually more productive even before they start importing. The standard approach focuses only on a single moment of the productivity distribution. To further establish the self-selection argument, I undertake the Kolmogorov-Smirnov test. This test uses all the information in the empirical productivity distribution. Th test proceeds as follows. Let x1 , x2 , ..., xn0 and xn0 +1 , xn0 +2 , ..., xn0 +n1 be random samples of size n0 and n1 independently drawn from the cumulative distribution functions Ω0 (x) and Ω1 (x). The distribution functions Ω0 (x) and Ω1 (x) represent the cumulative productivity densities of importing and non-importing firms, respectively. To test whether or not the two distributions are identical, I do a two-tailed test for the null hypothesis H0 : Ω1 (x) − Ω0 (x) = 0 against the alternative H1 : Ω1 (x) − Ω0 (x) 6= 0 where x ∈ R. The test statistic is given by D∗ = max(|Ω1 (x) − Ω0 (x)|). Similarly, the first order x stochastic dominance of the productivity distributions is checked by testing for the null hypothesis H0 : Ω1 (x) − Ω0 (x) = 0 against the alternative H1 : Ω1 (x) − Ω0 (x) ≤ 0 where x ∈ R. And, the test statistic becomes D∗ = max(Ω1 (x) − Ω0 (x)). In both cases Ω0 and Ω1 are replaced by the x P P i ≤x) i ≤x) and Ωn1 = I(i:x respectively.5 empirical distribution functions Ωn0 = I(i:x n0 n1 The Kolmogorov-Smirnov test results are shown in Table 9. The two-tailed test rejects the hypothesis that currently importing and non-importing firms have the same productivity distribution. Relatedly, the one–tailed test fails to reject the null hypothesis that the productivity distribution of importing firms stochastically dominates that of the non-importing counterparts. These results provide support to the argument that the current importers were more productive than their non-importing counterparts even before the former started importing. In conclusion, the summaries from the transition matrix and the estimates from the probit and least squares regressions clearly indicate the existence of substantial market entry costs which lead to the selection of more productive firms into importing. 5 Learning-by-importing In this section, I test for the learning-by-importing hypothesis by adopting a structural approach that addresses several estimation issues. In the test, I distinguish between static and dynamic 5 For a discussion on the test procedure, see Delgado et al. (2002). 91 effects of importing and estimate them separately. 5.1 Technology I closely follow Kasahara and Rodrigue (2008) and specify the production technology of firm i at time period t as: βk u β u s β s βe Yi,t = Ai,t Ki,t Li,t Li,t Ei,t "Z N (di,t ) m (j) θ−1 θ θ # θ−1 βm dj ;θ > 1 (1) 0 where Yi,t refers to output, Ai,t technology parameter, Ki,t capital, Lui,t and Lsi,t unskilled and skilled labor, Ei,t energy and m(j) a composite of domestic and foreign of intermediate goods. The β’s are elasticities of output with respect to inputs of production; θ > 1 elasticity of substitution between f f d d denote the number of and Ni,t where Ni,t any two intermediate goods, and N (di,t ) = Ni,t + di,t Ni,t domestically produced and imported intermediate goods, respectively, and di,t = 1 is an indicator function if the firm uses imported intermediates. In this specification, I treat different varieties of intermediate goods as horizontally differentiated with no quality difference. Assume that both domestic and foreign intermediate goods are produced and used symmetrically. That is m ¯ units of each intermediate good variety j are used, and the total material input used by firm i in time period t is Mi,t = N (di,t )m. ¯ After rearranging the terms, the production function is now given by: βm βm βk u β u s β s βe Yi,t = Ai,t N (di,t ) θ−1 Ki,t Li,t Li,t Ei,t Mi,t βm βk u βu s βs βe βm = e(ωi,t +i,t ) N (di,t ) θ−1 Ki,t Li,t Li,t Ei,t Mi,t (2) βm where e(ωi,t +i,t ) N (di,t ) θ−1 is a residual term of the production function. The residual term consists of a component which represents the impact of imported intermediate inputs due to variety effect βm N (di,t ) θ−1 , a firm’s total factor productivity ωi,t , and unobserved shock i,t which denotes, say, measurement error. Imposing a specific functional form on the evolution equation of ωi,t , it is possible to capture learning-by-importing; the effect on productivity of imported inputs one period after the firm imports. 5.2 Decision problem As in the Olley and Pakes (1996) model, a firm faces three decision problems at the beginning of each time period. First, it compares the sell-off value of exit with the continuation value of operation. If the firm decides to exit, it gets a sell-off value of Φ. If the firm stays in, it chooses levels of freely variable inputs labor and energy, and makes capital investment and import decisions. Assuming that current investment is productive in the next period, capital evolves according to ki,t = (1 − δ)ki,t−1 + ii,t−1 . Firm productivity is known to firms and follows, conditional upon 92 survival χi,t , a controlled first-order Markov process ωi,t = E [ωi,t | ωi,t−1 , di,t−1 , χi,t = 1] + ζi,t . The maximum expected discounted reward for a firm at a time period is given by the Bellman equation: ( n V (ωi,t , ki,t , di,t−1 ) = max Φi,t , max πt (ωi,t , ki,t , di,t ) − Ck (ki,t , ii,t ) − Cd (di,t , di,t−1 ) di,t ,ii,t Z +β V (ωi,t+1 , ki,t+1 , di,t ) dF (ωi,t+1 , ki,t+1 , di,t | ωi,t , ki,t , di,t , χi,t = 1) ) (3) o where ωi,t , ki,t and di,t refer to the state variables, π(.) indirect profit function, Ck (.) investment cost function, Cd (.) sunk or fixed costs of using imported intermediates depending on previous period import status, and β the discount factor. Solving the above dynamic programming problem of firm i yields three policy functions: exit rule χi,t = ω t > ωit (ki,t , di,t−1 ), discrete import di,t = dt (ωi,t , ki,t , di,t−1 ), and investment demand ii,t = it (ωi,t , ki,t , di,t−1 ) functions. The dependence of the policy functions on the import variable follows from treating previous period import decision as a state variable. Its inclusion as an additional state variable is to capture any dynamic productivity effect of imported intermediates. 5.3 Empirical strategy I now present the empirical procedure used to estimate parameters of the production function and transition equations of the model. Logarithmic transformation of the production function and inclusion of the discrete import variable di,t to capture the static effects of imported intermediates βm ln(N (di,t )) yields θ−1 u s yi,t = βk ki,t + βu li,t + βs li,t + βe ei,t + βm mi,t + βd di,t + ωi,t + i,t (4) From a controlled first-order Markov process of productivity and assuming a linear approximation, we have ωi,t = E [ωi,t | ωi,t−1 , di,t−1 , χi,t = 1] + ζi,t (5) = τ + ρ ωi,t−1 + γ di,t−1 + ζi,t where the innovation term ζi,t which is independent of ωi,t−1 and di,t−1 and with a known distribution. In the productivity equation, I condition on survival probability to control for endogenous selection of firms in the data.6 Equations (4) and (5) allow us to test both the static and dynamic effects of importing on firm output and productivity. That is, if βd > 0, it implies that using imported intermediates immediately improve output for a fixed quantity of inputs in production. On the other hand, γ > 0 indicates a dynamic productivity gain, and the long-run effect can be γ summarized as 1−ρ . 6 Gebreeyesus (2008) finds an annual firm turnover rate of (average of entry and exit rates) 22% over the years 1996-2003 in Ethiopian manufacturing. This strongly suggests the need to account for firm attrition in the data. 93 Given that a firm does not make input decisions independent of it productivity, and ωi,t is not observed in the data, estimation of equation (4) by ordinary least squares raises econometric issues due to the endogeneity of input choices. For this purpose, I adopt the widely used estimation algorithm developed by Levinsohn and Petrin (2003) in which material inputs are used as a proxy for unobserved firm productivity. I exploit the relationship that demand for material inputs depends on observed capital and unobserved productivity mi,t = mt (ki,t , ωi,t , di,t ). Under the monotonicity of mt (.), the unobserved productivity can be expressed in terms of the observables capital, material inputs and import as ωi,t = ωt (mi,t , ki,t , di,t ). The resulting estimating equation becomes: u s yi,t = βu li,t + βs li,t + βe ei,t + ϕt (mi,t , ki,t , di,t ) + i,t (6) where ϕt (mi,t , ki,t , di,t ) = βk ki,t + βm mi,t + βd di,t + ωt (mi,t , ki,t , di,t ). I estimate equation (6) by least squares in which ϕt (.) is approximated using a third-order polynomial function, and with industry and time fixed effects included. In so doing, βu , βs and βe are consistently estimated. This is because the source of correlation between the freely variable inputs and a firm’s productivity has now been controlled by the polynomial approximation, and these inputs are also uncorrelated with i,t by construction. Because ki,t , mi,t and di,t are collinear with the terms in the polynomial approximation, βk , βm and βd are unidentified in this stage. The identification assumptions to estimate βk , βm and βd crucially depends on timing. Since ki,t is determined at t − 1, it is uncorrelated with the innovation term in productivity, ζi,t , giving rise to an exogenous variation in ki,t used for identifying βk . Because the firm choose mi,t at the same time ωi,t is observed, mi,t is not independent of ζi,t . However, ζi,t is uncorrelated with mi,t−1 for it is decided at t − 1, and this condition is used to estimate βm . Identification of βd comes from the orthogonality between ζi,t and di,t−1 . That is, even if di,t is correlated with ωi,t , the innovation term ζi,t should have no correlation with past import status di,t−1 , which was decided at t − 1. After estimating equation (6) and recovering φi,t ≡ ϕˆt (ki,t , mi,t , ωi,t ) = βk ki,t + βm mi,t + βd di,t + ωi,t 7 , I run a probit regression of Pi,t = P r(χi,t = 1) = χt (ki,t , ki,t−1 , di,t−1 )) where χt (.) is approximated linearly in its arguments. The probit estimation yields firms’ predicted survival probabilities Pˆi,t for a given level of capital, productivity, and previous period import status. Afterwards, I substitute φi,t in (5) and obtain the following estimating equation: φi,t = τ + βk ki,t + βm mi,t + βd di,t + ρ(φi,t−1 − βk ki,t−1 − βm mi,t−1 − βd di,t−1 ) + γdi,t−1 + Ωt (φi,t−1 − βk ki,t−1 − βm mi,t−1 − βd di,t−1 , Pˆi,t ) + ζi,t (7) where Ωt (.) is included to control for firm attrition in the data. Equation (7) is estimated by a non-linear least squares technique. For the purpose of comparison, I also estimate equation (7) without correcting for endogenous selection of the firms. 7 Ackerberg et al. (2006) argue that using the moment conditions in the residuals of ζi,t instead of ζi,t + i,t yields precise and more stable estimates. This is due to the additional variance term associated with i,t . 94 To investigate whether intensive use of foreign varieties improves productivity, I invoke the symmetry assumption regarding the production and employment of intermediate goods. From the assumption, it follows that the ratio of imported to total intermediate inputs is given by: f Mi,t N (1)m−N ¯ ¯ N (1)−N (0) i,t (0)m = i,t Ni,t (1)m = i,t Ni,t (1)i,t . This ratio can be interpreted as the fraction of imported Mi,t ¯ inputs both in number and value in the total material input used in the production. Introducing this ratio into the production function and productivity equation, we obtain: yi,t = βk ki,t + βu Lui,t + βs Lsi,t + βe ei,t + βm mi,t + βd ni,t + ωi,t ωi,t = E [ωi,t | ωi,t−1 , ni,t−1 , χi,t = 1] + ζi,t (8) (9) = τ + ρ ωit−1 + γ ni,t−1 + ζi,t  where ni,t−1 = log f Mi,t−1 Mi,t−1  . And, the identification assumptions and estimation steps proceed in the same way as in the discrete case. To summarize, the parameter vector of interest θ = (θy , θω ) comprises production function parameters on capital, unskilled and skilled labor, energy and material θy = (βk , βl , βs , βe , βm , βd ), and productivity transition parameters θω = (ρ, µ). 5.4 Result This section discusses the least square (OLS), fixed effects (FE) and Levinsohn–Petrin (LP) estimation results. Columns (1)–(4) of Table 10 present parameter estimates where import is treated as a discrete variable. The OLS results in column (1) show that all estimates of the output elasticities are positive and significant. The magnitudes of these elasticities are also consistent with most findings in the productivity estimation literature. Importantly, the coefficient on discrete import variable is positive and significant implying that there is a productivity gain due to importing, approximately 6.18%. It is well known that least squares estimation of production functions is plagued by endogeneity problems. Assuming time–invariant firm effects, demeaning of the estimation equation yields a new estimating equation free of endogeneity. Under this assumption, I run the fixed effects regression and the results are shown in column (2). The FE estimates are very close to their OLS counterparts in terms of sign and statistical significance. However, there are size differences between them. As expected, FE estimates on capital, skilled labor and material inputs are lower than the OLS estimates. The only exception is the estimate on unskilled labor which becomes higher with the FE estimation suggesting a downward bias in the OLS estimate of the coefficient on unskilled labor. From the FE estimates we see that there are no immediate, significant productivity gains from using imported inputs, albeit a positive estimate.8 8 Consistent estimation of the input elasticities under OLS and FE prevents the estimation of ρ and γ. 95 Least squares and fixed effects regression impose restrictive assumptions. Consistent estimation of the parameters using OLS requires no correlation between freely variable inputs and serially uncorrelated firm productivity whereas FE assumes firm-specific, time-invariant unobserved productivity. Additionally, with these techniques it is not possible to endogenize productivity and distinguish between static and dynamic gains. To overcome these limitations and to impose a richer structure, I adopt the LP estimation algorithm. In columns (3) and (4), LP estimates of the coefficients on freely variable inputs are similar to OLS and FE estimates in their statistical significance. However, the LP estimates are smaller than their OLS counterparts but larger than FE estimates except for unskilled labor. Coefficients on capital, material inputs and import are estimated first without controlling for survival probability of firms and then after taking into account firm survival using the predicted probabilities from probit model. The estimates are shown in columns (3) and (4), respectively. They are very similar in statistical significance, direction and magnitude. They show that firms experience an immediate decline in productivity due to importing, 0.8%. This is not unexpected in the light of previous findings showing that firms might need to adjust their production structure to benefit from the availability of cheaper and probably better imported intermediates.9 It is also shown that there is a strong persistence in the evolution of productivity, ρ = 0.78 and 0.65, and there are dynamic productivity gains due to importing, 1.1-1.2%. Long-run effects of importing predict a firm productivity improvement around 3.5-4.9%. In columns (5)-(8) I present the estimation results in which import is treated depending on how intensive is the use of foreign varieties among importing firms. Both the OLS and FE estimates display similar patterns as their counterparts in the discrete cases. The only exception is the significance of the import variable under FE regression. The LP estimates show that a 100% increase in the share of imported input increases firm productivity by 2.1% immediately and 0.7% in the period after. In the long-run, the productivity gain is approximately 4-22%. Note that the long run productivity gain increases substantially when an endogenous selection of firms is addressed in the estimation. To see the time path, I simulate the productivity path (loosely defined as βd di,t + ωi,t ) of a hypothetical firm over time. In the simulation, I use the LP estimates while ignoring the unobserved shocks which firms experience in each period. Figure 3 shows the evolution of productivity of a firm that starts importing at period 1 and continues to do so afterwards. We see that the firm experiences a momentary decline in productivity. However, after some time the firm adjusts its production structure and is able to enjoy the productivity gains from importing. We observe that correction of an endogenous selection of firms gives rise to a rapidly converging path, albeit at lower level. In Figure 4, I repeat the same exercise in which import intensity is considered instead. Here, the hypothetical firm starts using imported inputs at period 1 and these inputs 9 For instance, for Danish manufacturing firms, Smeets and Warzynski (2013) find a temporary decline yet a continual improvement in firm productivity due to importing. 96 constitute 0.47% of the materials used in production, which is the average import share in the data. We observe a significant productivity improvement over time. We notice that correcting for the survival probability of firms makes a substantial difference. All in all, the results from the empirical analysis show that there are significant productivity gains from importing. When considering the import participation of firms only, we expect overall productivity improvements ranging from zero as in a FE estimation to 3.5-4.9% as in a LP estimation. Considering how intensive the employment of imported intermediate affects productivity, we see that there are both immediate and long–term benefits associated with a more intensive use of imported inputs. The main results of the above empirical analyses highlight the fact that even though there are temporary declines in productivity, the firm ultimately benefits from importing, and even more so if it intensifies the relative employment of imported varieties vis-`a-vis the domestic ones. However, in view of a significant portion of firms importing, and their production activities are characterized by intensive use of imported intermediates, the estimated productivity gains are relatively small. This strongly suggests the limited absorptive capacity—the capabilities and efforts to assimilate the knowledge embodied in the imported inputs—of firms in the economy. 6 Conclusion The vast majority of the literature on firm globalization has been restricted to advanced economies and a few developing countries in Asia and Latin America. African manufacturing firms have been greatly neglected because of lack of available accounting information and trade statistics. Even among a handful of existing studies, utmost focus has been on exporting. In this respect, the literature on African manufacturing remains largely incomplete, especially in light of high import–to–GDP ratios and import shares of manufacture in these economies. In this paper, with the objective of filling this void in the literature, I use a unique panel dataset from Ethiopian manufacturing. A simple description of the data uncovers that most firms source production inputs from the world market. This illustrates that firms heavily rely on imported inputs partly due to limited availability of domestically manufactured inputs. Additionally, I find a positive link between importing and productivity and other firm performance measures. Examination of the direction of causality in the import-productivity relationship shows that more productive firms self-select themselves into importing indicating significant sunk and fixed costs of importing. Additionally, to test the causal effect of importing on firm productivity, I use a framework in which the static and dynamic effects are estimated separately. The results provide evidence of learning-by-importing albeit an initial temporary decline. Furthermore, intensive use of imported inputs is associated with a greater productivity improvement among importing firms. However, the small size of the productivity gains demonstrates the limited absorptive capacity of firms in the economy. This feature is consistent with findings, mostly in economic growth literature, 97 which emphasize the mismatch between human capital of domestic workers and technological content of imported inputs as a hindrance to technology diffusion to the least developed countries. 98 References Acemoglu, D., & Zilibotti, F. (2001). Productivity Differences. Quarterly Journal of Economics, 116 (2), 563–606. Acharya, R. C., & Keller, W. (2009). Technology Transfer Through Imports. Canadian Journal of Economics, 42 (4), 1411–1448. Ackerberg, D., Caves, K., & Frazer, G. (2006). Structural Identification of Production Functions. MPRA Paper # 38349 . Augier, P., Cadot, O., & Dovis, M. (2013). Imports and TFP at the Firm Level: The Role of Absorptive Capacity. Canadian Journal of Economics, 46 (3), 956–981. Bernard, A. B., & Jensen, J. B. (1995). Exporters, Jobs, and Wages in US Manufacturing: 1976-1987. Brookings Papers on Economic Activity. Microeconomics, 1995 , 67–119. Bernard, A. B., Jensen, J. B., Redding, S. J., & Schott, P. K. (2007). Firms in International Trade. Journal of Economic Perspectives, 21 (3), 105–130. Bigsten, A., Collier, P., Dercon, S., Fafchamps, M., Gauthier, B., Willem Gunning, J., . . . others (2004). Do African Manufacturing Firms Learn from Exporting? Journal of Development Studies, 40 (3), 115–141. Bigsten, A., & Gebreeyesus, M. (2009). Firm Productivity and Exports: Evidence from Ethiopian Manufacturing. Journal of Development Studies, 45 (10), 1594–1614. Castellani, D., Serti, F., & Tomasi, C. (2010). Firms in International Trade: Importers’ and Exporters’ Heterogeneity in Italian Manufacturing Industry. World Economy, 33 (3), 424– 457. Coe, D. T., & Helpman, E. (1995). International R&D Spillovers. European Economic Review , 39 (5), 859–887. Coe, D. T., Helpman, E., & Hoffmaister, A. W. (1997). North-South R&D Spillovers. Economic Journal , 107 (440), 134–49. Coe, D. T., Helpman, E., & Hoffmaister, A. W. (2009). International R&D Spillovers and Institutions. European Economic Review , 53 (7), 723–741. Delgado, M. A., Farinas, J. C., & Ruano, S. (2002). Firm Productivity and Export Markets: A Non-parametric Approach. Journal of International Economics, 57 (2), 397–422. Dovis, M., & Milgram-Baleix, J. (2009). Trade, Tariffs and Total Factor Productivity: The Case of Spanish Firms. World Economy, 32 (4), 575–605. Eriksson, T., Smeets, V., & Warzynski, F. (2009). Small Open Economy Firms in International Trade: Evidence from Danish Transactions-level Data. Nationaløkonomisk tidsskrift, 147 (2), 175–194. Forlani, E. (2010). Irish Firms’ Productivity and Imported Inputs. CORE Discussion Papers # 015 . Foster-McGregor, N., Isaksson, A., & Kaulich, F. (2014). Importing, Exporting and Performance in Sub-Saharan African Manufacturing Firms. Review of World Economics, 150 (2), 309–336. Gebreeyesus, M. (2008). Firm Turnover and Productivity Differentials in Ethiopian Manufacturing. Journal of Productivity Analysis, 29 (2), 113–129. Halpern, L., Koren, M., & Szeidl, A. (2009). Imported Inputs and Productivity. CEFIG Working Papers # 08 . Kasahara, H., & Lapham, B. (2013). Productivity and the Decision to Import and Export: Theory and Evidence. Journal of International Economics, 89 (2), 297–316. Kasahara, H., & Rodrigue, J. (2008). Does the Use of Imported Intermediates Increase Productivity? Plant-level Evidence. Journal of Development Economics, 87 (1), 106–118. 99 Levinsohn, J., & Petrin, A. (2003). Estimating Production Functions Using Inputs to Control for Unobservables. Review of Economic Studies, 70 (2), 317–341. L¨o¨of, H., & Andersson, M. (2010). Imports, Productivity and Origin Markets: The Role of Knowledge-intensive Economies. World Economy, 33 (3), 458–481. Los, B., & Timmer, M. P. (2005). The ’appropriate technology’ Explanation of Productivity Growth Differentials: An Empirical Approach. Journal of Development Economics, 77 (2), 517–531. Mengistae, T., & Pattillo, C. (2004). Export Orientation and Productivity in Sub-Saharan Africa. IMF Staff Papers, 327–353. Muendler, M.-A. (2004). Trade, Technology, and Productivity: A Study of Brazilian Manufacturers, 1986-1998. CESifo Working Paper #1148 . No, J. Y. A. (2009). International Transmission of Technology and Trade: The Role of Crosscountry Heterogeneity. International Economic Journal , 23 (3), 427–446. Olley, G. S., & Pakes, A. (1996). The Dynamics of Productivity in the Telecommunications Equipment Industry. Econometrica, 64 (6), pp. 1263–1297. Serti, F., Tomasi, C., & Zanfei, A. (2010). Who Trades with Whom? Exploring the Links Between Firms’ International Activities, Skills, and Wages. Review of International Economics, 18 (5), 951–971. Smeets, V., & Warzynski, F. (2013). Estimating Productivity with Multi-product Firms, Pricing Heterogeneity and the Role of International Trade. Journal of International Economics, 90 (2), 237–244. UNCTAD. (2014). Key Trends in International Merchandise Trade. New York and Geneva. Van Biesebroeck, J. (2005). Exporting Raises Productivity in Sub-Saharan African Manufacturing Firms. Journal of International Economics, 67 (2), 373–391. Van Biesebroeck, J. (2008). The Sensitivity of Productivity Estimates: Revisiting Three Important Debates. Journal of Business & Economic Statistics, 26 (3), 311–328. Vogel, A., & Wagner, J. (2010). Higher Productivity in Importing German Manufacturing Firms: Self-selection, Learning from Importing, or Both? Review of World Economics, 145 (4), 641–665. Yasar, M. (2013). Imported Capital Input, Absorptive Capacity, and Firm Performance: Evidence from Firm-level Data. Economic Inquiry, 51 (1), 88–100. Yasar, M., & Morrison Paul, C. J. (2007). International Linkages and Productivity at the Plant Level: Foreign Direct Investment, Exports, Imports and Licensing. Journal of International Economics, 71 (2), 373–388. Yasar, M., & Paul, C. J. M. (2008). Foreign Technology Transfer and Productivity: Evidence from a Matched Sample. Journal of Business & Economic statistics, 26 (1), 105–112. 100 Table 1: Sectoral composition of Ethiopian economy GDP Agriculture† Manufacturing ‡ Service§ Year % Growth % GDP % Growth % GDP % Growth % GDP % Growth 6.59 1996 12.43 55.35 16.96 5.64 3.23 34.03 1997 3.13 58.87 2.00 7.96 2.97 27.83 4.51 1998 -3.46 53.54 -9.64 5.74 0.38 33.84 7.22 1999 5.16 49.60 3.40 6.25 8.71 37.10 8.31 2000 6.07 48.71 3.05 6.11 6.66 38.84 11.20 2001 8.30 46.56 9.62 6.35 3.81 40.32 4.37 2002 1.51 42.50 -1.88 6.33 1.81 43.51 5.89 2003 -2.16 40.93 -10.48 6.33 1.21 44.91 9.10 2004 13.57 43.17 16.94 5.97 7.27 42.68 4.72 2005 11.82 45.61 13.54 5.36 13.03 41.33 12.30 2006 10.83 46.80 10.91 5.10 10.30 40.44 12.92 2007 11.46 46.38 9.45 5.03 9.93 40.91 15.98 2008 10.79 49.40 7.50 4.49 9.26 39.44 16.68 2009 8.80 49.60 6.36 4.19 8.62 39.93 14.92 2010 12.55 45.64 5.13 4.38 9.21 43.97 17.04 2011 11.18 45.57 9.01 4.07 9.24 43.75 13.08 1996–2011 7.62 48.02 5.74 5.58 6.60 39.55 10.30 Source: World Development Indicators, World Bank † ‡ § Agriculture corresponds to ISIC–Rev.3 divisions 1–5. Manufacturing corresponds to ISIC–Rev.3 divisions 15–37. Services correspond to ISIC–Rev.3 divisions 50–99. 101 Table 2: Components of Ethiopian international trade Export† Import† % Merchandise trade‡ Year Level % GDP % Growth Level % GDP % Growth Manuf. Export Manuf. Import 1996 750.1 9.35 7.30 1,318.0 16.43 16.16 9.65 84.49 1997 992.0 11.99 32.26 1,559.5 18.84 18.33 6.76 66.61 1998 1,079.6 13.51 8.83 1,720.7 21.54 10.34 6.71 77.24 1999 1,015.6 12.09 -5.93 2,081.9 24.78 20.99 9.78 70.90 2000 1,083.6 12.16 6.69 2,158.5 24.22 3.68 13.43 65.19 2001 1,169.5 12.12 7.93 2,312.3 23.96 7.12 14.31 73.85 2002 1,249.3 12.75 6.83 2,635.4 26.90 13.97 11.38 64.03 2003 1,290.4 13.46 3.29 2,657.2 27.72 0.83 3.83 70.80 2004 1,635.8 15.02 26.77 3,475.4 31.92 30.79 4.58 71.97 2005 1,858.4 15.27 13.61 4,367.2 35.87 25.66 5.36 68.51 2006 1,893.7 14.04 1.90 4,991.0 36.99 14.28 13.75 76.41 2007 1,935.0 12.87 2.18 4,873.6 32.41 -2.35 9.01 60.18 2008 1,934.3 11.61 -0.04 5,236.0 31.43 7.44 8.65 71.53 2009 1,938.5 10.69 0.22 5,297.6 29.22 1.18 8.91 68.69 2010 2,826.3 13.85 45.80 6,882.5 33.73 29.92 10.37 65.27 2011 3,856.0 17.00 36.43 7,289.4 32.14 5.91 8.77 67.13 – 12.99 12.13 – 28.01 12.77 9.08 70.18 1996–2011 Source: World Development Indicators, World Bank. † ‡ Export and import of goods and services in million constant 2005 USD. Manufacture items comprise commodities in SITC–Rev.3 sections 5–8 excluding non–ferrous metals. 102 103 32 28 12 51 19 54 27 445 Chemicals Rubber and plastic Non-metallic products Fabricated metals Furniture Others Total manufacturing 42 Leather products Printing and publishing 15 Wearing apparel 13 24 Textiles Wood products 128 # Firms Food and beverage Industry 4.72 0.00 1.85 0.00 1.96 0.00 0.00 0.00 7.69 21.43 6.67 20.83 2.34 % Exporters 1996 67.87 92.59 57.41 89.47 21.57 100 100 71.88 69.23 80.95 66.67 83.33 64.06 % Importers 826 35 95 39 95 65 38 48 12 61 19 12 307 # Firms 14.77 14.29 0.00 2.56 3.16 3.08 21.05 2.08 16.67 45.90 21.05 50.00 20.20 % Exporters 2011 64.89 82.86 65.26 82.05 22.11 92.31 89.47 87.50 75.00 70.49 73.68 91.67 58.31 % Importers Table 3: International trade participation of firms in Ethiopian manufacturing 5.86 2.57 0.58 1.23 1.49 0.43 2.77 0.13 1.85 32.01 13.56 26.95 5.46 % Exporters 69.74 92.00 79.10 85.55 26.29 95.29 93.35 85.27 51.08 82.24 75.72 71.10 59.89 % Importers 1996-2011 Table 4: Export and import activity premia Export-only Import-only Two-way TFP 23.078∗∗∗ (0.055) 5.660∗∗∗ (0.014) 31.832∗∗∗ (0.033) Output per worker 46.020∗∗∗ (0.110) 8.211∗∗∗ (0.027) 77.311∗∗∗ (0.060) Capital per worker 43.952∗∗ (0.150) -14.790∗∗∗ (0.052) 39.440∗∗∗ (0.088) Material per worker 39.449∗∗ (0.143) 4.146 (0.031) 68.109∗∗∗ (0.068) Energy per worker 4.701 (0.164) 7.838∗∗ (0.032) 38.321∗∗∗ (0.077) Employment size 170.102∗∗∗ (0.121) 40.020∗∗∗ (0.023) 612.849∗∗∗ (0.059) Sh. of skilled worker -1.75 (0.077) 4.407∗∗∗ (0.013) 13.058∗∗∗ (0.027) Bootstrapped standard errors with 500 replications in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 Table 5: Transition probabilities of firm activities Domestic Status t Status t+1 Export-only Import-only Two-way Domestic Export-only Import-only Two-way 67.19 13.75 13.76 4.01 1.41 33.75 0.22 6.61 30.71 13.75 84.30 12.42 0.69 38.75 1.71 76.95 Cross-sectional Average 28.59 1.14 64.86 5.40 104 Table 6: Probit estimation of export and import participation: marginal effects Export (Xi,t ) Import (Mi,t ) (1) (3) (4) 0.030∗∗∗ (0.006) -0.004∗∗ (0.002) 0.023∗∗∗ (0.004) – 0.029∗∗∗ (0.003) -0.005∗∗ (0.002) 0.021∗∗∗ (0.004) 0.035∗ (0.019) 0.302∗∗∗ (0.006) Yes Yes (2) 0.008∗∗∗ (0.003) 0.003∗∗∗ (0.001) 0.009∗∗∗ (0.002) 0.116∗∗∗ (0.004) – 0.008∗ (0.003) 0.003∗∗∗ (0.001) 0.009∗∗∗ (0.002) 0.116∗∗∗ (0.004) -0.003 (0.004) Yes Yes Yes Yes 9,020 tf pi,t−1 ki,t−1 li,t−1 Xi,t−1 Mi,t−1 Year FE Industry FE Obs. 0.303∗∗∗ (0.006) Yes Yes 9,020 Bootstrapped standard errors with 500 replications in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 Table 7: Firms switching their importing status Switching frequency # Firms % Firms 0 1,413 60.13 1 425 18.09 2 282 12.00 3 96 4.09 4 68 2.89 5 26 1.11 5+ 40 1.71 - 2,350 100 Note: Switching frequency refers to the number of times a firm changes its importing status over the sample period. 105 Table 8: Self–selection into importing tf pi,t−3 tf pi,t−2 tf pi,t−1 11.981∗∗∗ (0.019) 11.166∗∗∗ (0.019) 9.442∗∗∗ (0.018) 8.728∗∗∗ (0.018) 10.072∗∗∗ (0.015) 9.564∗∗∗ (0.015) Xi,t−3 - 29.557∗∗∗ (0.038) - - - - Xi,t−2 - - - 28.501∗∗∗ (0.034) - - Xi,t−1 - - - - - 23.293∗∗∗ (0.031) Yes Yes 0.445 Yes Yes 0.450 Yes Yes 0.445 Yes Yes 0.450 Yes Yes 0.456 Yes Yes 0.459 Mi,t Year FE Industry FE Adj.R2 Obs. 5,832 7,118 9,020 Bootstrapped standard errors with 500 replications in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 106 Table 9: Kolmogorov-Smirnov test for equality of productivity distributions 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 ∗ H0 : Ω1 (x)-Ω0 (x) ≤ 0 H0 : Ω1 (x)-Ω0 (x)= 0 Year tf pi,t−3 tf pi,t−2 tf pi,t−1 tf pi,t−3 tf pi,t−2 tf pi,t−1 0.211∗∗∗ 0.298∗∗∗ 0.242∗∗∗ 0.155∗∗ 0.210∗∗ 0.158∗∗ 0.266∗∗∗ 0.224∗∗∗ 0.234∗∗∗ 0.205∗∗ 0.093 0.212∗∗∗ 0.129∗∗ 0.262∗∗∗ 0.211∗∗∗ 0.192∗∗∗ 0.207∗∗∗ 0.138∗∗∗ 0.209∗∗∗ 0.234∗∗∗ 0.222∗∗∗ 0.216∗∗∗ 0.274∗∗∗ 0.133∗∗ 0.154∗∗∗ 0.179∗∗∗ 0.141∗∗ 0.249∗∗∗ 0.272∗∗∗ 0.196∗∗∗ 0.195∗∗∗ 0.244∗∗∗ 0.170∗∗∗ 0.269∗∗∗ 0.166∗∗∗ 0.246∗∗∗ 0.226∗∗∗ 0.168∗∗∗ 0.148∗∗∗ 0.131∗∗∗ 0.122∗∗ 0.142∗∗∗ -0.025 -0.007 -0.004 0.000 -0.020 -0.030 -0.003 -0.009 -0.008 -0.009 -0.031 -0.003 -0.015 -0.022 -0.001 -0.014 0.000 -0.003 -0.014 -0.008 -0.006 -0.008 -0.009 -0.008 -0.005 -0.011 -0.023 -0.006 -0.022 -0.003 0.000 -0.022 -0.016 -0.002 -0.010 -0.005 -0.012 -0.012 -0.015 -0.006 -0.002 -0.006 p < 0.01, ∗∗ p < 0.05, ∗∗∗ p < 0.01 107 Table 10: Production function parameters Import participation OLS FE (1) (2) βu 0.136∗∗∗ (0.008) 0.142∗∗∗ (0.017) βs 0.128∗∗∗ (0.008) βe Import intensity OLS FE (5) (6) 0.115∗∗∗ (0.007) 0.112∗∗∗ (0.011) 0.128∗∗∗ (0.020) 0.115∗∗∗ (0.007) 0.089∗∗∗ (0.017) 0.103∗∗∗ (0.007) 0.128∗∗∗ (0.010) 0.071∗∗∗ (0.019) 0.103∗∗∗ (0.007) 0.100∗∗∗ (0.006) 0.064∗∗∗ (0.007) 0.091∗∗∗ (0.005) 0.074∗∗∗ (0.006) 0.042∗∗∗ (0.007) 0.091∗∗∗ (0.005) βk 0.059∗∗∗ (0.004) 0.033∗∗∗ (0.006) 0.042∗∗∗ (0.001) 0.043∗∗∗ (0.001) 0.070∗∗∗ (0.005) 0.040∗∗∗ (0.007) 0.037∗∗∗ (0.002) 0.038∗∗∗ (0.002) βm 0.673∗∗∗ (0.009) 0.567∗∗∗ (0.015) 0.668∗∗∗ (0.002) 0.669∗∗∗ (0.002) 0.709∗∗∗ (0.009) 0.571∗∗∗ (0.017) 0.684∗∗∗ (0.002) 0.685∗∗∗ (0.002) βd 0.060∗∗∗ (0.014) 0.012 (0.022) -0.008∗ (0.005) -0.008∗ (0.005) 0.093∗∗∗ (0.007) 0.103∗∗∗ (0.011) 0.021∗∗∗ (0.002) 0.021∗∗∗ (0.002) ρ - - 0.777∗∗∗ (0.008) 0.654∗∗∗ (0.046) - - 0.811∗∗∗ (0.009) 0.968∗∗∗ (0.056) γ - - 0.011∗∗∗ (0.004) 0.012∗∗∗ (0.004) - - 0.007∗∗∗ (0.001) 0.007∗∗∗ (0.001) N LP (3) (4) 8,282 LP (7) 5,190 Bootstrapped standard errors with 500 replications in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 108 (8) Figure 1: Geographical orientation of Ethiopian international trade Sub−Saharan Africa Middle East & North Africa 1995 2000 2005 2010 20 0 0 30 40 2 50 60 4 40 70 80 6 60 High−Income 1995 2005 2010 1995 Central & South Asia 2005 2010 2 10 1.5 20 1995 2000 2005 2010 0 0 0 5 .5 5 10 1 15 2000 Latin America & Caribbean 15 25 East Asia & Pacific 2000 1995 2000 Export share 2005 2010 1995 Import share Source: World Development Indicators Database, the World Bank 109 2000 2005 2010 Figure 2: Productivity distribution by firm import status One−period lag .6 .4 .4 pdf pdf 5 −10 −5 0 5 −10 −5 0 5 −10 −5 0 5 −5 0 5 −10 −5 0 5 .8 0 .2 .4 cdf .6 .8 0 .2 .4 cdf .6 .8 .6 cdf .4 .2 0 −10 1 0 1 −5 1 −10 0 .2 .2 0 0 .2 pdf .4 .6 Two−period lag .6 Three−period lag Importers Non−importers 110 Figure 3: Productivity effect of importing: import participation 0.02 No survival correction Survival correction 0.015 0.01 0.005 0 −0.005 −0.01 0 5 10 15 20 25 111 30 35 40 45 50 Figure 4: Productivity effect of importing: import intensity 0.4 No survival correction Survival correction 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 0 5 10 15 20 25 112 30 35 40 45 50 Appendix A: Data construction The Central Statistical Agency of Ethiopia conducts an annual survey of large and medium–scale manufacturing firms. The scope of the survey is delimited to public and private establishments with at least 10 employees and which use power–driven machinery. In the survey, an establishment is defined to be the whole of the premises under the same ownership or management with a specific location address. Below is a list of the definitions of the variables used to construct the final dataset. 1. Gross value of production: includes the sales value of all products of an establishment net of changes in the inventory of finished goods, fixed assets, and others during the reference period. The value denotes the market price inclusive of indirect taxes but exclusive of subsidies. 2. Fixed capital assets: refers to assets with a productive life at least one year. It also includes production of fixed assets for a firm’s own use. They are valued as the beginning period book value and new capital expenditures less the part sold, disposed and depreciated during the reference year. 3. Working proprietors, active partners and family workers: includes all unpaid persons who actively participate in the operation of the establishment. 4. Administrative and technical employees: refers to salaried directors and managers, technicians, research workers, engineers, scientists, accountants, and other office staff. 5. Production workers: consists of persons directly engaged in fabricating, processing, assembling, maintenance, repair, and other associated activities. 6. Seasonal and temporary workers: includes persons who are not regularly on the payroll of the establishment. 7. Basic wages and salaries: includes all payments made to employees during the reference year. It excludes commissions, bonuses, social security contributions, insurance, and professional and hardship allowances. 8. Materials: includes all raw and auxiliary materials which are consumed during the reference year. Local raw materials are those produced locally, and imported raw materials those produced in other countries and obtained directly or from local sources. The costs include factory gate purchase price, transport charges, taxes and other incidental costs. 9. Industrial cost: includes the costs of raw materials, fuels, electricity and other supplies consumed, costs of industrial services rendered by others, costs of goods bought and resold without any transformation. 113 Appendix B: Tables Table B.1: Geographical orientation of Ethiopian merchandise exports Year High-income† Sub-Saharan Africa Middle East & North Africa East Asia & Pacific Central & South Asia Latin America & Caribbean Others‡ 1996 80.78 0.18 10.21 1.51 0.00 0.00 6.53 1997 80.70 2.93 10.42 0.94 0.00 0.00 3.64 1998 82.80 0.86 9.51 0.94 0.00 0.00 4.78 1999 70.76 1.67 14.45 1.71 1.89 0.13 9.40 2000 80.90 0.60 14.21 1.03 3.17 0.08 0.02 2001 35.79 0.00 55.58 4.14 3.43 0.00 1.07 2002 71.64 0.57 15.91 3.11 7.56 0.00 1.21 2003 61.61 1.76 29.58 1.74 4.63 0.07 0.62 2004 75.19 2.62 8.00 3.74 5.57 0.23 4.65 2005 69.85 3.09 11.36 11.60 3.38 0.15 0.57 2006 69.81 3.75 9.25 11.51 3.67 0.05 1.96 2007 73.31 5.09 8.04 6.82 6.16 0.02 0.55 2008 74.49 5.87 7.03 6.40 5.25 0.14 0.82 2009 56.93 2.54 7.30 13.88 4.11 0.01 15.23 2010 58.16 1.40 6.82 14.99 4.51 0.03 14.09 2011 62.94 0.58 6.88 13.02 4.24 0.16 12.19 1996–2011 69.10 2.09 14.03 6.07 3.60 0.07 4.83 Source: Word Development Indicators, World Bank. † ‡ All high-income countries in each region are excluded from their respective regions. Includes trade with unspecified partners or with economies not covered by World Bank classification. 114 Table B.2: Geographical orientation of Ethiopian merchandise imports Year High-income† Sub-Saharan Africa Middle East & North Africa East Asia & Pacific Central & South Asia Latin America & Caribbean Others‡ 1996 69.36 3.31 4.25 3.59 3.76 0.70 13.62 1997 74.43 2.17 1.10 7.64 5.35 0.17 7.35 1998 75.10 1.59 1.70 5.98 5.62 0.31 7.56 1999 75.39 2.31 1.38 4.73 7.78 0.34 8.06 2000 49.93 1.75 22.70 8.45 7.21 0.22 9.73 2001 60.20 0.92 11.20 8.80 7.31 0.52 11.05 2002 58.92 1.18 5.95 10.75 8.79 0.58 13.84 2003 59.39 1.31 4.73 12.44 9.14 0.76 12.24 2004 57.29 0.86 3.93 12.82 8.95 0.60 15.56 2005 58.43 1.80 3.07 14.22 9.68 0.74 12.06 2006 54.72 2.22 3.97 15.74 9.38 1.29 12.68 2007 48.78 1.18 2.41 23.76 12.24 2.09 9.54 2008 42.61 2.35 1.76 19.01 15.46 0.69 18.12 2009 33.04 2.52 2.74 20.89 9.52 0.92 30.36 2010 35.91 3.35 3.53 16.89 6.10 0.64 33.58 2011 35.14 4.15 3.69 11.99 8.06 0.40 36.58 1996–2011 55.54 2.06 4.88 12.36 8.40 0.69 15.75 Source: Word Development Indicators, World Bank. † ‡ All high-income countries in each region are excluded from their respective regions. Includes trade with unspecified partners or with economies not covered by World Bank classification. 115 116 Chapter 4 Imported Inputs and Firm Absorptive Capacity in Ethiopian Manufacturing 117 Imported Inputs and Firm Absorptive Capacity in Ethiopian Manufacturing∗ Kaleb Girma Abreha† Abstract What is the role of firm absorptive capacity in trade-driven knowledge transfer? Using a firm-level dataset from Ethiopian manufacturing over the period 1996-2011, I find that imported inputs are associated with higher productivity only in firms with greater absorptive capacity. This result provides firm-level evidence on why international knowledge transfer to the least developed countries, characterized by limited absorptive capacity, are slower in pace and smaller in magnitude. JEL codes: F14, D22, L60 Keywords: Imported inputs, firm productivity, absorptive capacity, Ethiopia ∗ I am grateful to the Central Statistical Agency of Ethiopia for making the data accessible. I acknowledge financial support from the Tuborg Foundation. The usual disclaimer applies. † Department of Economics and Business, Aarhus University, Denmark, E-mail: [email protected] 118 1 Introduction It is now widely accepted that technological progress is the key to sustained economic growth and development. To this end, countries devote substantial resources to R&D activities. Findings in this field of research have shown that although these activities are highly concentrated in a handful of countries, there is significant international knowledge transfer which leads to substantial economic gains. A seminal paper by Coe and Helpman (1995), using data on a group of OECD countries, shows that there are significant gains associated with trade-driven technology transfer. In a subsequent paper, Coe et al. (1997) document that the knowledge spillovers are not entirely limited to developed economies in that developing countries greatly benefit from their trade partnership with countries which actively undertake R&D activities. Other studies establish that the stock of physical and human capital, and institutions determine the pace and the size of technological innovations and diffusions across countries. In their analysis of the impact of human capital on income per capita growth rate, Benhabib and Spiegel (1994) illustrate the positive role of human capital in the rate of technology diffusions in addition to promoting local technological innovations. Relatedly, Acemoglu and Zilibotti (2001) show that technological developments and domestic workers’ skill mismatches lead to productivity differences across countries even when they have equal access to technology. Los and Timmer (2005) find empirical evidence which pinpoints that technological assimilations are primarily driven by capital intensification which generates a potential for knowledge spillovers. Given that capital intensifications vary substantially across countries, they show that technological assimilations are slower than the predictions from the standard economic growth models. Coe et al. (2009) emphasize the importance of institutional factors on the heterogeneity of R&D spillovers across countries. Acharya and Keller (2009) identify differences in geographical distance between trading partners and nature of goods traded as a source of asymmetry in the extent of technology transfers.1 There exist some studies that look into the role of absorptive capacity in trade-driven knowledge spillovers. At the firm level, Augier et al. (2013) identify the importance of absorptive capacity (measured by skill intensity) in determining the degree of learning-by-importing for Spanish manufacturing firms. Similarly, Yasar (2013) considers trade in capital goods among Chinese firms and shows that imported capital inputs have a larger impact on productivity in importing firms with greater absorptive capacity. In this paper, I consider the Ethiopian manufacturing sector to examine trade-driven knowledge transfer at the firm level. Ethiopian manufacturing makes an interesting case for several reasons. First, the sector is relatively technologically backward indicating a great potential for technology 1 The role of absorptive capacity—which represents countries’ and firms’ capabilities and efforts to assimilate technologies and commonly measured by R&D expenditures, human capital and institutions—as a determinant of knowledge spillovers has been a subject of extensive research in the FDI literature (see e.g. Durham 2004; Girma 2005; Xu 2000). Also see Crespo and Fontoura (2007) for a survey of the literature on the topic. 119 transfer from other countries. Second, the country trades dominantly with technologically advanced economies making international knowledge spillover a possibility and trade an important conduit for such a spillover; 55% of Ethiopia’s imports come from these countries. Third, manufacture imports constitute 70% of the import trade of the country, and more than 60% of the manufacturing firms rely on imported inputs. This provides an opportunity to investigate whether such a heavy reliance on imported inputs is translated into productivity gains.2 Lastly, despite these potentials, the sector faces a shortage of skilled employees that potentially limits the extent to which firms benefit from imported inputs. Using the same dataset as in this paper, Abreha (2014) finds that there are productivity gains from using imported inputs in Ethiopian manufacturing. However, these gains are small as compared to similar findings in other developed and developing countries. For instance, Kasahara and Rodrigue (2008), for Chilean manufacturing firms, find a productivity improvement of 11.1% and 2.6% under the ordinary least squares and fixed effect regressions, respectively. Besides, using a control function approach, they report static (21.4%) and dynamic (2.4-4.1%) productivity gains. For Ethiopian manufacturing firms, the comparable estimate show a 6.18% and no significant productivity gain under least squares and fixed effects regression respectively. And, the control function estimates reveal an immediate, temporary decline and a 1.1-1.2% dynamic productivity gain. There are at least two explanations, besides the economic development of the source countries of the imported inputs, as to why the productivity gains are rather small. First, it may be the case that trade in capital goods rather than intermediate materials is the main channel for international technology diffusion. On the other hand, importing firms may have limited absorptive capacity, which refers to the ability and effort such as skill intensity and R&D investment to exploit the knowledge embodied in imported inputs. Testing the empirical validity of these alternative explanations is hindered by lack of information on trade in capital goods at the firm level. At the same time, there is no available data on firm R&D expenditures. Within this context, this paper examines the role of absorptive capacity (measured by the share of skilled workers in a firm) in the productivity effect of imported inputs. For this purpose, I estimate a Cobb-Douglas production function in which import status and absorptive capacity are included as additional variables. The parameter estimates show that imported inputs are associated with higher productivity, after controlling for the quantity of inputs used in production, if firms have the necessary skill composition to absorb the embodied knowledge in those inputs. This result suggest that mere access to technology is not a sufficient condition for transfer. In the light of the country’s lack of human capital and less effective ancillary institutions, this finding provides firm-level evidence on the limited nature and slow pace of technology transfer to the least developed countries. It must be stressed that this empirical exercise does not establish 2 Abreha (2014) summarizes the sectoral composition of the Ethiopian economy and the geographic orientation of its international trade. 120 a causal relationships rather it only displays cross-sectional correlations. This paper is closely related to a series of contributions on firm globalization and economic growth. It is linked to Augier et al. (2013) and Yasar (2013) who show the pivotal role of firm absorptive capacity regarding the productivity implications of import trade in capital goods and intermediates. Unlike these studies, the focus here is on a least developed country where there is a great potential for technology transfer yet limited absorptive capacity. Additionally, many empirical investigations on absorptive capacity and international knowledge diffusion are undertaken primarily at an aggregate level and predominantly confined to OECD countries and a few emerging economies. In this respect, this paper is one of the first studies to provide firm-level evidence from African manufacturing. The rest of the paper is organized as follows. Section 2 describes the data. Section 3 presents the econometric model and the estimation techniques. Section 4 discusses the results. Section 5 concludes. 2 2.1 Data description Data source The Central Statistical Agency of Ethiopia (CSA) conducts annual large- and medium-scale manufacturing surveys. The surveys cover all firms in the manufacturing sector with at least 10 employees and which use power-driven machinery. Classification of firms into manufacturing sector is based on ISIC-Rev.3, and it includes industries 15-37 at the 2-digit level. The focus is on the time period 1996-2011. The dataset provides detailed information on the level of production, local and export sales and fixed assets of firms. It also contains information on local and imported material inputs and energy expenses (electricity, fuel and charcoal). After excluding seasonal and temporary workers due to infrequent report of their number, I categorize workers into two broad subgroups: skilled (administrative and technical employees) and unskilled (apprentice and production workers). In this study, a firm’s absorptive capacity is approximated by the share of skilled employees in its workforce.3 Because the complementarity between skilled workers and materials is not obvious as it is in the case of capital goods, I attach in the appendix a list of raw material codes by industry groups compiled by CSA. Finally, all nominal values are deflated using a consumer price index, which is extracted from the World Development Indicators database of the World Bank.4 To restrict the empirical analysis only to firms with real economic activity, I exclude those with zero or unreported level of production, fixed asset, material input, energy expense and employment. 3 In the literature, absorptive capacity is defined in a variety of different ways. For instance, at the firm level absorptive capacity can refer to skill intensity, R&D expenditure and size. At the aggregate level, it may also include a stock of human capital, infrastructure and level of development of legal and financial institutions. 4 See Abreha (2014) for a detailed description of the data and construction of the variables used in the analysis. 121 I also exclude firms appearing only once over the entire sample period. The final dataset comprises 2,350 firms and 12,510 firm-year observations. 2.2 Absorptive capacity Table 1 reports summary statistics on firm absorptive capacities over the sample period. It shows that there are few outlying cases of absorptive capacities as indicated by large differences between the minimum and the maximum values. However, comparison of the mean and the median values demonstrates that the distribution of absorptive capacities is not greatly skewed. This is also shown by the size of the standard deviations relative to the mean values. At the same time, we notice that the mean exceeds the median implying that most firms have absorptive capacities below average, and on balance, these firms constitute around 55% of the firms in the manufacturing sector. Over time, we observe that there are no fundamental changes in the distribution of absorptive capacities with the exception of a small rise in the share of firms with absorptive capacity below the mean. 3 Methodology To explore any effect of imported inputs on the production efficiency of firms and the role of absorptive capacity in this respect, I specify a Cobb–Douglas production function: u s yi,t = β0 + βlu li,t + βls li,t + βk ki,t + βe ei,t + βm mi,t + βd di,t−1 + βs si,t−1 + βds di,t−1 si,t−1 + δt + τ + i,t (1) u s where yi,t refers to output, li,t unskilled labor, li,t skilled labor, ki,t capital, ei,t energy, mi,t material inputs, di,t−1 a dummy variable taking a value of 1 if a firm used imported in the previous period, si,t−1 absorptive capacity at the beginning of period t, δt and τ year and industry fixed effects, and i,t an iid error term. All the variables in equation (1) are in logarithmic scale with the exception of di,t−1 , si,t−1 and the fixed effects. Regressors of the parameters βd , βs and βds are one-year lagged to reduce the endogeneity problem coming from a likely correlation between the import dummy and the error term in the production function. If βd > 0, it shows that importing is positively associated with higher level of output after controlling for overall quantity of materials used in production. If βs > 0, it means that a firm whose production characterized by greater absorptive capacity exhibits greater productivity conditional on its employment size, capital holding, volume and origin of material inputs used. On the other hand, βds > 0 indicate that importing has a larger impact on the production activity of a firm with greater absorptive capacity. Also, the interaction term indicates that an increase in absorptive capacity has a larger effect in importing firms compared to otherwise similar non-importing firms. As an additional test on the productivity effect of importing, I explore whether or not firms have different output elasticities based on their absorptive capacities. For this purpose, I adopt a threshold regression and use a sample splitting technique developed by Hansen (2000). This 122 technique splits a sample into different regimes based on a threshold variable. In running the threshold regression, I do not exploit the panel structure of the data. This is due to unavailability of an estimator in the case of unbalanced panel data, albeit Hansen (1999) develops an estimator for a non-dynamic balanced panel. Also, reducing the data to a balanced panel is inappropriate due to high firm turnover rate in the sector.5 I define the production function for the threshold regression as: yi,t  β l + β k + β e + β m + β d +  = β 0 x +  ; s ≤ γ l1 i,t k1 i,t e1 i,t m1 i,t d1 i,t i,t i,t i,t 1 i,t = 0 β l + β k + β e + β m + β d +  = β x +  ; s > γ l2 i,t k2 i,t e2 i,t m2 i,t d2 i,t i,t i,t i,t 2 i,t (2) where β1 = (βl1 , βk1 , βe1 , βm1 , βd1 )0 and β2 = (βl2 , βk2 , βe2 , βm2 , βd2 )0 denote output elasticity vectors, γ a threshold parameter, xi,t = (li,t , ki,t , ei,t , di,t ) an input vector in which li,t now denotes unskilled labor, si,t a threshold variable based on absorptive capacity (= share of skilled workers), and i,t an error term. Define a threshold effect δn = β1 − β2 , an indicator function Ii,t (γ) = I{si,t ≤ γ} and xi,t (γ) = xi,t Ii,t (γ). Equation (2) can be written as 0 0 yi,t = β xi,t + δn xi,t (γ) + i,t (3) where β = β2 . After stacking the observations for each firm and time period in a vector, the regression equation becomes: Y = Xβ + Xγ δn +  (4) The sum of squared errors function from equation (4) is given by: Sn (β, δ, γ) = (Y − Xβ − Xγ δ)0 (Y − Xβ − Xγ δ) (5) which is minimized through concentration. That is, conditional on γ, equation (4) is linear in β ˆ ˆ and δn which allows estimation by ordinary least squares. The slope estimates β(γ) and δ(γ) are obtained from a regression of Y on Z =[XXγ ]. The sum  of squared errors function from this condiˆ ˆ tional regression becomes Sn (γ) ≡ Sn β(γ), δ(γ), γ = (Y − Z(Z 0 Z)−1 Z 0 Y )0 (Y − Z(Z 0 Z)−1 Z 0 Y ). Then, the threshold value is estimated by minimizing Sn (γ) with respect to γ: γˆ = argmin Sn (γ), γ which can implemented through a grid search over the quantiles of si,t . Once β, δn and γ are estimated, the next step is to test for the significance of the threshold effect; that is, H0 : β1 = β2 against H1 : β1 6= β2 . Given that γ is not identified under the H0 , the hypothesis test is nonstandard. In this respect, Hansen (1996) develops a heteroskedasticity robust Lagrange Multiplier test for which the correct p-values are calculated through a bootstrap procedure. 5 Gebreeyesus (2008) documents that the Ethiopian manufacturing exhibits a turnover rate of 20-22% annually. 123 Conditional upon rejecting the null hypothesis (that is, there is a threshold effect), the next step is to test whether or not γ is equal to a specific threshold value; H0 : γ = γ0 against H1 : γ 6= γ0 where γ0 is the true parameter value. Assuming that i,t follows an iid normal distribution, the test is undertaken using the likelihood ratio statistic: LRn (γ) = n Sn (γ) − Sn (ˆ γ) Sn (ˆ γ) Because this test statistic does not have a standard χ2 distribution, Hansen (2000) calculates the correct asymptotic critical values. The test rejects the null hypothesis for large values of LRn (γ). This test statistic is then used to construct a confidence interval for the threshold estimate. 4 Result Column (1) of Table 2 shows that the estimates on unskilled and skilled labor, capital, energy and material inputs are positive and significant. The size of these coefficients is within the range of most findings in the productivity literature. We see that the effect of using imported inputs depends on a firm’s absorptive capacity, and importing has a greater impact in firms with higher absorptive capacity. For example, for a firm with an absorptive capacity equal to 0.37, which is the average absorptive capacity in the data, the effect of imported inputs on output production is 8.10%. In a rather extreme case, imported inputs have no effect if a firm has no absorptive capacity at all, as shown by the statistically insignificant βd . Relatedly, βs is not statistically different from zero, and it shows that the share of skilled workers in the past does not affect output after controlling for the number of skilled employees in the production function. In column (2), I include the square of a firm’s absorptive capacity as an additional regressor. This is motivated by the fact that firms with high absorptive capacity may own better technology vis–`a–vis imported inputs, and therefore there is little or no expected technology transfer from using imported inputs.6 The insignificance of βss in this regression demonstrates that there is no evidence of limited scope for knowledge transfer because of proximity to the technology frontier in the case of Ethiopian manufacturing. In columns (3)–(6), I divide firms into quantiles based on their absorptive capacities and estimate the model. The coefficients on different types of labor, capital, energy and materials are positive, significant and similar across quantiles. We see that βd is insignificant and small for firms located in the lowest quantiles of the absorptive capacity distribution. On the contrary, βd becomes statistically significant and larger in size for firms in the upper quantiles. This variation in size and significance suggest that absorptive capacity plays a role in the productivity effect of importing. 6 At the aggregate level, Falvey et al. (2007) find that countries closest to or farthest from the technology frontier gain less from international transfer although absorptive capacity facilitates cross-country knowledge transfers. 124 Table 3 reports the threshold regression estimates.7 The threshold parameter γ is estimated to be 0.403. Figure 1 plots firms’ absorptive capacities against critical values and the test statistic from the testing H0 : γ = γ0 against H1 : γ 6= γ0 . We see that the likelihood ratio statistics lie below the 95% critical value for the interval [0.375, 0.406] forming the confidence interval for γ. Compared to the threshold estimate reported in Yasar (2013), the estimate here is larger in magnitude and has a narrower confidence interval.8 Besides, this threshold estimate exceeds the average and median values reported in Table 1 implying that most firms have limited absorptive capacity which prevents them from exhaustively reaping the benefits associated with using imported inputs. Based on the threshold value, the whole sample is split into two regimes, and then the output elasticities are estimated. We see that the coefficients on βl , βk , βe , and βm do not vary across the two regimes and they therefore can be considered as regime–independent. The parameter of interest βd is significant under both regimes but its statistical significance and magnitude is greater for firms with absorptive capacity above the threshold value. We notice that for firms in the second regime the βd estimate is almost twice as large. Imported inputs are associated with 2.9% and 5.3% increases in output production for firms located in the first and the second regimes, respectively. And, this difference is statistically significant as can be seen from the LM test which rejects the null hypothesis that there is no threshold effect. 5 Conclusion In this paper, I emphasize the role absorptive capacity plays in the transfer of technology through imported inputs. It is shown that the benefits of imported inputs depend on the skill composition of importing firms. The results reveal that a greater absorptive capacity is associated with a higher productivity and larger benefits of imported inputs. However, this correlation is very weak at the lower end of the absorptive capacity distribution. The threshold estimate shows that most firms have limited absorptive capacity to exploit, to a considerable extent, the embodied knowledge in the imported inputs. Despite the pervasiveness of importing and intensive use of imported inputs among Ethiopian manufacturing firms, the benefits of enhanced access to technology are confined to only firms with at least a minimum skill intensity requirement. This finding provides firm-level support to prevailing macroeconomic evidences that identify limited absorptive capacity of the least developed countries as an impediment to technology spillovers. The above results only illustrate cross-sectional correlations between absorptive capacity and the productivity effect of imported inputs. As commonly known in the productivity estimation 7 Estimation of the threshold regression and hypothesis testing are implemented using a Matlab code written by Bruce E. Hansen, which is available from the author’s website http://www.ssc.wisc.edu/~bhansen/progs/ progs threshold.html. 8 Yasar (2013) also reports IV estimate of the threshold parameter. This new estimate is smaller but still with a wider confidence interval. 125 literature, there is a bevy of econometric concerns coming from the endogeneity of input choices. An interesting future research area will be to deal with these issues by adopting an estimation strategy that exploits the panel structure of the data. 126 References Abreha, K. G. (2014). Importing and Firm Productivity in Ethiopian Manufacturing. FREIT Working Paper # 843 . Acemoglu, D., & Zilibotti, F. (2001). Productivity Differences. Quarterly Journal of Economics, 116 (2), 563–606. Acharya, R. C., & Keller, W. (2009). Technology Transfer Through Imports. Canadian Journal of Economics, 42 (4), 1411–1448. Augier, P., Cadot, O., & Dovis, M. (2013). Imports and TFP at the Firm Level: The Role of Absorptive Capacity. Canadian Journal of Economics, 46 (3), 956–981. Benhabib, J., & Spiegel, M. M. (1994). The Role of Human Capital in Economic Development Evidence from Aggregate Cross-country Data. Journal of Monetary economics, 34 (2), 143– 173. Coe, D. T., & Helpman, E. (1995). International R&D Spillovers. European Economic Review , 39 (5), 859–887. Coe, D. T., Helpman, E., & Hoffmaister, A. (1997). North-South R&D Spillovers. Economic Journal , 107 (440), 134–149. Coe, D. T., Helpman, E., & Hoffmaister, A. W. (2009). International R&D Spillovers and Institutions . European Economic Review , 53 (7), 723–741. Crespo, N., & Fontoura, M. P. (2007). Determinant Factors of FDI Spillovers – What Do We Really Know? World Development, 35 (3), 410–425. Durham, J. (2004). Absorptive Capacity and the Effects of Foreign Direct Investment and Equity Foreign Portfolio Investment on Economic Growth. European Economic Review , 48 (2), 285– 306. Falvey, R., Foster, N., & Greenaway, D. (2007). Relative Backwardness, Absorptive Capacity and Knowledge Spillovers. Economics Letters, 97 (3), 230–234. Gebreeyesus, M. (2008). Firm Turnover and Productivity Differentials in Ethiopian Manufacturing. Journal of Productivity Analysis, 29 (2), 113–129. Girma, S. (2005). Absorptive Capacity and Productivity Spillovers from FDI: A Threshold Regression Analysis. Oxford Bulletin of Economics and Statistics, 67 (3), 281–306. Hansen, B. E. (1996). Inference When a Nuisance Parameter is not Identified under the Null Hypothesis. Econometrica, 64 (2), 413–430. Hansen, B. E. (1999). Threshold Effects in Non-dynamic Panels: Estimation, Testing, and Inference. Journal of Econometrics, 93 (2), 345–368. Hansen, B. E. (2000). Sample Splitting and Threshold Estimation. Econometrica, 68 (3), 575–603. Kasahara, H., & Rodrigue, J. (2008). Does the Use of Imported Intermediates Increase Productivity? Plant-level Evidence. Journal of Development Economics, 87 (1), 106–118. Los, B., & Timmer, M. P. (2005). The ‘Appropriate Technology’ Explanation of Productivity Growth Differentials: An Empirical Approach. Journal of Development Economics, 77 (2), 517–531. Xu, B. (2000). Multinational Enterprises, Technology Diffusion, and Host Country Productivity Growth. Journal of Development Economics, 62 (2), 477–493. Yasar, M. (2013). Imported Capital Input, Absorptive Capacity, and Firm Performance: Evidence from Firm-level Data. Economic Inquiry, 51 (1), 88–100. 127 Table 1: Summaries on firm absorptive capacity Year 1996-2000 2001-2005 2006-2011 1996-2011 Minimum Maximum Mean Median Std. dev.: overall Std. dev.: between Std. dev.: within 0.04 0.93 0.38 0.36 0.18 - 0.02 0.93 0.38 0.36 0.19 - 0.01 0.96 0.36 0.33 0.20 - 0.00 0.94 0.37 0.34 0.19 0.16 0.13 % Firms below mean 53.80 54.80 56.33 55.06 Note: For each subsample of time periods, the reported summaries are calculated as simple averages of the individual years. For details, see Table A.1 in the Appendix. 128 Table 2: Production function parameters All βlu βls βk βe βm βd βs βds βss R2 Obs. Quantiles (1) (2) 1st Quantile 2nd Quantile 3rd Quantile 4th Quantile 0.140∗∗∗ (0.009) 0.118∗∗∗ (0.003) 0.059∗∗∗ (0.003) 0.101∗∗∗ (0.006) 0.673∗∗∗ (0.008) -0.0002 (0.032) -0.073 (0.065) 0.219∗∗∗ (0.071) – 0.140∗∗∗ (0.009) 0.118∗∗∗ (0.003) 0.059∗∗∗ (0.003) 0.101∗∗∗ (0.006) 0.673∗∗∗ (0.008) -0.0002 (0.032) -0.022 (0.140) 0.219∗∗∗ (0.071) -0.059 (0.137) 0.150∗∗∗ (0.029) 0.084∗∗∗ (0.026) 0.047∗∗∗ (0.008) 0.103∗∗∗ (0.011) 0.668∗∗∗ (0.016) 0.017 (0.031) – 0.117∗ (0.068) 0.157∗∗ (0.063) 0.064∗∗∗ (0.008) 0.092∗∗∗ (0.013) 0.641∗∗∗ (0.017) 0.061∗∗ (0.031) – 0.003 (0.066) 0.267∗∗∗ (0.069) 0.065∗∗∗ (0.007) 0.101∗∗∗ (0.011) 0.672∗∗∗ (0.017) 0.116∗∗∗ (0.028) – 0.126∗∗∗ (0.023) 0.148∗∗∗ (0.025) 0.050∗∗∗ (0.008) 0.105∗∗∗ (0.011) 0.702∗∗∗ (0.014) 0.155∗∗∗ (0.031) – – – – – – – – – 0.94 0.94 0.92 2,179 0.94 2,228 0.95 2,336 0.94 2,277 9,020 Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Industry and time fixed effects are included in these regressions. 129 Table 3: Threshold regression Absorptive capacity ≤γ Threshold Estimate γ 0.95 Confidence Interval >γ 0.403 [0.375, 0.406] βl βk βe βm βd R2 Obs. 0.227∗∗∗ (0.010) 0.050∗∗∗ (0.004) 0.075∗∗∗ (0.006) 0.700∗∗∗ (0.008) 0.029∗∗ (0.014) 0.238∗∗∗ (0.012) 0.052∗∗∗ (0.005) 0.111∗∗∗ (0.008) 0.694∗∗∗ (0.012) 0.052∗∗∗ (0.019) 0.922 7,805 0.935 4,705 LM test for no threshold effect test statistic LM test for no threshold effect p–value 316.115 0.000 Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Industry and time fixed are included in these regressions. Figure 1: Confidence interval construction for the threshold parameter 300 LRn(γ) 95% Critical 250 Likelihood Ratio 200 150 100 50 0 0.1 0.2 0.3 0.4 0.5 0.6 Threshold Variable: Absorptive Capacity 130 0.7 0.8 0.9 1 Appendix: Tables Table A.1: Summary statistics on firm absorptive capacity Year Minimum Maximum Mean Median St. dev. % Firms below mean absorptive cap. 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 0.04 0.05 0.03 0.04 0.04 0.02 0.03 0.02 0.03 0.02 0.02 0.01 0.02 0.02 0.00 0.00 0.94 0.95 0.94 0.89 0.93 0.91 0.92 0.96 0.95 0.91 0.93 0.98 0.95 0.97 0.97 0.95 0.37 0.38 0.38 0.37 0.38 0.40 0.39 0.38 0.37 0.37 0.37 0.36 0.36 0.37 0.34 0.36 0.36 0.36 0.36 0.34 0.36 0.38 0.36 0.36 0.35 0.35 0.33 0.33 0.33 0.33 0.30 0.33 0.18 0.19 0.18 0.17 0.19 0.19 0.19 0.18 0.20 0.18 0.19 0.20 0.20 0.21 0.19 0.20 53.71 53.33 53.78 53.49 55.28 54.51 54.03 55.74 56.30 53.41 56.81 56.26 55.68 56.34 56.82 56.05 1996–2011 0.02 0.94 0.37 0.35 0.19 55.10 131 MAJOR RAW MATERIALS CODES BY INDUSTRIAL GROUP DESCRIPTION OF RAW MATERIALS RAW MATERIALS U/M NAME U/M CODE CATTLE Ÿwƒ 0327 HEAD /lØ`/ 83 COFFEE (UNMILLED) 0401 KG,QN,TON 02,03,04 EDIBLE OIL 0318 LT,KG,BARREL,TON 65,02,81,04 FLOUR 0103 KG,QN,TON 02,03,04 GLUCOSE 0167 KG, TON 02,04 MAIZE (UNMILLED) 0009 KG,QN,TON 02,03,04 MEAT 0095 KG, TON 02,04 MILK (POWDER) 0323 KG, TON 02,04 MILK (RAW) 0006 LT,HL 65,66 OIL SEEDS /¾pvƒ IKA‹/ 0101 KG,QN,TON 02,03,04 ORANGE 0007 KG, TON,CAN* 02,04, (09-14) PULSES /Ø^Ø_/ 0324 KG,QN,TON 02,03,04 SUGAR 0112 KG,QN,TON 02,03,04 SUGAR CANE 0032 02,04,06 TEA LEAVES 0402 KG,TON, ‘000TON KG,QN,TON VEGETABLES 0147 KG,QN,TON 02,03,04 WHEAT (UN MILLED) 0008 KG,QN,TON 02,03,04 YEAST 1223 KG,QN,TON 02,03,04 ALLCOHOL 0136 LT,HL 65,66 BARLEY /Ø_ Ñwe/ 0335 KG,QN,TON 02,03,04 CARBON DIOXIDE 0176 KG, TON 02,04 ESSENCE 0111 KG, TON,UNIT** 02,04 (18-25) GRAPE & RAISING /¨Ã”/ 0021 KG, TON 02,04 FOOD 02,03,04 BEVERAGE DESCRIPTION OF RAW MATERIALS U/M NAME U/M CODE HOPS /Ñ@j/ RAW MATERIALS 0033 KG, TON 02,04 MALT /wpM/ 0114 KG,QN,TON 02,03,04 MOLASSES 0113 KG,QN,TON 02,03,04 SUGAR 0112 KG,QN,TON 02,03,04 0037 KG,QN,TON 02,03,04 TOBACCO TOBACCO LEAVES Can 100gm Code 09 “ 240gm “ 10 “ 340gm “ 11 “ 350 gm “ 12 “ 400gm “ 13 “ 850gm “ 14 Unit ** Coca-Cola code Fanta “ Mirinda “ Pepsi “ Sprite “ Teem “ Tonic “ Average “ 21 22 19 18 23 20 24 25 RAW MATERIALS U/M NAME U/M CODE RAW COTTON Ø_ ØØ 0501 KG,QN,TON 02,03,04 ACRYLIC (YARN) 0054 KG, TON 02,04 CHEMICAL & DYESTUFF 0331 KG, TON 02,04 COTTON (LINT) ¾}ÇSÖ ØØ 0058 KG, TON 02,04 COTTON (WASTE) COTTON (YARN) 0057 0061 KG, TON KG, TON 02,04 02,04 FABRICS 0062 MT,SQM,’000MT 28,51,29 FIBRE (ACRYLIC) 0320 KG, TON 02,04 FIBRE (POLYESTER) 0319 KG, TON 02,04 JUTE (FIBER) 0052 KG, TON 02,04 NYLON 0064 KG, TON 02,04 SISAL (LEAVES) 0326 KG, TON 02,04 WOOL (WASTE) 0056 KG, TON 02,04 CHEMICALS 0315 KG, ON,TON 02,03,04 HIDES & SKINS 0038 KG, TON 02,04 LEATHER LINING /¾ÝT Ñu` qÇ/ 0117 SOF,’000SQF 47,48 LEATHER SOLE 0115 KG, TON 02,04 LEATHER UPPER 0116 SOF,’000SQF 47,48 LEATHER GARMENT 0287 SQF 47 PLASTIC SOLE 0118 PAIRS, DOZZEN 42,45 PVC FOR SOLE 0102 KG, TON 02,04 SHEEP & GOAT SKINS 0039 PCS, ‘000PCS, DOZ 31,32,45 DESCRIPTION OF RAW MATERIALS TEXTILES LEATHER & FOOTWEAR DESCRIPTION OF RAW MATERIALS RAW MATERIALS U/M NAME U/M CODE CHIPWOOD 0158 PCS 31 FORMICA 0328 PCS 31 LOG /Ó”É & ”Úƒ/ 0322 Cub.m 53 PLYWOOD /¢UüMd„& óò=ƒ/ 0140 PCS 31 PLUNK /ר”„ Ø_ n 0317 KG, TON 02,04 CRAVEL /ÖÖ`/ 0269 Cub.m 53 GYPSUM /Ëf/ 0276 KG, QN,TON 02,O3,04 LIME STONE /¾•^ É”ÒÃ/ MARBLE U’u[É PUMICE 0275 0278 0272 KG, TON KG, TON,M3 Cub.m 02,04 02,04,53 53 SAND 0270 Cub.m 53 SILICA SAND/ SAND STONE 0325 KG, TON 02,04 SODA ASH 0170 KG, TON 02,04 STONE FOR GRAVEL 0119 Cub.m 53 ALUMNUM 0330 KG, TON 02,04 CHEMICAL FOR METAL 0217 KG, TON 02,04 CROWN TIN PLATE 0286 SHEET,000SHEET GAL VANIZED COILS 0289 KG, TON 02,04 IRON BARS ²”Ó w[ƒ/ ôa w[ƒ 0290 KG, TON 02,04 IRON (BILLETS)Éu