A partnership with academia

Building knowledge for trade and development

Vi Digital Library - Text Preview

Turkish Enterprise-level Response to Foreign Trade Liberalization

Working paper by UNCTAD, 2013

Download original document (English)

Trade in textiles and apparel is of special interest among international trade transactions. Removal of the final Agreement on Textiles and Clothing (ATC) quotas in 2005 brought about a division of textile- and apparel-exporting countries into groups of winners and losers. Turkey appeared as a successful country from the former category. Based firm-level data results suggest that while Turkish enterprises were more successful than most in adapting to the post-quota market in textiles and apparel, their performance paled relative to the performance of enterprises in areas not covered by the ATC. Producers that specialized in textiles and apparel during the ATC quotas removal period had ceteris paribus lower sales revenue and employment growth and a lower profit rate on average than those selling other products. The latter category of producers was also significantly more likely to fail during this period.

U n i t e d n at i o n s C o n f e r e n C e o n t r a d e a n d d e v e l o p m e n t


TURKISH ENTERPRISE-LEVEL RESPONSE
TO FOREIGN TRADE LIBERALIZATION:


The Removal of Agreements on Textiles and Clothing Quotas


POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES
STUDY SERIES No. 59













POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


STUDY SERIES No. 59






TURKISH ENTERPRISE-LEVEL RESPONSE TO FOREIGN


TRADE LIBERALIZATION:


The Removal of Agreements on Textiles and Clothing Quotas




by


Patrick Conway
University of North Carolina





Marco Fugazza
UNCTAD





M. Kerem Yuksel
Bilkent University











UNITED NATIONS
New York and Geneva, 2013


U N I T E D N AT I O N S C O N F E R E N C E O N T R A D E A N D D E V E L O P M E N T





ii


Note


The purpose of this series of studies is to analyse policy issues and to stimulate discussions in
the area of international trade and development. The series includes studies by UNCTAD staff and by
distinguished researchers from other organizations and academia. This paper represents the personal
views of the authors only, and not the views of the UNCTAD secretariat or its member States.


This publication has not been formally edited.

The designations employed and the presentation of the material do not imply the expression of
any opinion on the part of the United Nations concerning the legal status of any country, territory, city
or area, or of authorities or concerning the delimitation of its frontiers or boundaries.

Material in this publication may be freely quoted or reprinted, but acknowledgement is
requested, together with a copy of the publication containing the quotation or reprint to be sent to the
UNCTAD secretariat at the following address:



Marco Fugazza


Trade Analysis Branch
Division on International Trade in Goods and Services, and Commodities


United Nations Conference on Trade and Development
Palais des Nations, CH-1211 Geneva 10, Switzerland


Tel: +41 22 917 5772; Fax: +41 22 917 0044
E-mail: marco.fugazza@unctad.org






Series Editor:
Victor Ognivtsev
Officer-in-Charge


Trade Analysis Branch
DITC/UNCTAD













UNCTAD/ITCD/TAB/60









UNITED NATIONS PUBLICATION


ISSN 1607-8291









© Copyright United Nations 2013
All rights reserved





iii


Abstract




Trade in textiles and apparel is of special interest among international trade transactions.


Removal of the final Agreement on Textiles and Clothing (ATC) quotas in 2005 brought about a division


of textile- and apparel-exporting countries into groups of winners and losers. Turkey appeared as a


successful country from the former category. Based firm-level data our empirical results suggest that


while Turkish enterprises were more successful than most in adapting to the post-quota market in


textiles and apparel, their performance paled relative to the performance of enterprises in areas not


covered by the ATC. Producers that specialized in textiles and apparel during the ATC quotas removal


period had ceteris paribus lower sales revenue and employment growth and a lower profit rate on


average than those selling other products. The latter category of producers was also significantly more


likely to fail during this period.






Keywords: Agreement on Textiles and Clothing, Quotas, Turkey, Firms

JEL Classification: F13, F14, L67

















iv






Acknowledgements





The authors wish to thank the Turkish Statistical Institute (TUIK), and especially


Oguzhan Turkoglu, for the hospitality and strong support shown to us while conducting this


research. The conclusions drawn here are entirely those of the authors, and should not


necessarily be associated with TUIK or its employees.




Any mistakes and errors in this paper are the authors' own.






v


Contents




EXECUTIVE SUMMARY ......................................................................................................................... vii




1 INTRODUCTION .......................................................................................................................... 1




2 RELATED LITERATURE ............................................................................................................. 2




3 DATA ........................................................................................................................................... 4




3.1 The Turkish Enterprise Survey .......................................................................................... 4


3.2 The Foreign Trade database ............................................................................................. 5




4 FACTS ON THE IMPACT OF QUOTAS REMOVAL ................................................................... 7




5 THE EMPIRICAL STRATEGY ................................................................................................... 12




5.1 Trimmed data ................................................................................................................... 12


5.2 Difference-in-difference estimation ................................................................................. 13


5.3 Conditional probability of enterprise exit ......................................................................... 13




6 RESULTS .................................................................................................................................. 14




6.1 Difference-in-difference estimation ................................................................................ 14


6.2 Probit estimation ............................................................................................................. 16




7 CONCLUSIONS AND EXTENSIONS ....................................................................................... 20






REFERENCES ......................................................................................................................................... 21




APPENDIX A: Turkish firm-level surveys ........................................................................................... 23


APPENDIX B: Estimating enterprise-level productivity ..................................................................... 34





vi


List of tables




Table 1. Textiles and apparel enterprises as a share of all manufacturing enterprises ...................... 5


Table 2. Trade in manufactures ........................................................................................................... 6


Table 3. Exporters and/or importers in manufacture sectors .............................................................. 6


Table 4. Decomposing foreign trade into quota liberalization categories ........................................... 7


Table 5. Mean revenue and employment for Turkish enterprises


active throughout the period 2003–2008 .............................................................................. 8


Table 6. Transition in enterprise characteristics from 2003/2004 to 2005/2006 ................................. 9


Table 7. Transition in enterprise characteristics from 2003/2004 to 2007/2008 ............................... 10


Table 8. Transition in enterprise characteristics for ATC producers ................................................. 11


Table 9. Correlation of switching behaviour by enterprises .............................................................. 12


Table 10. Moments of panel data, with and excluding extreme values .............................................. 12


Table 11. Results for difference-in-difference estimation for Turkey .................................................. 15


Table 12. Performance and survival .................................................................................................... 17


Table 13. Duration analysis for Turkish enterprises ............................................................................ 18


Table 14. Duration analysis for Turkish enterprises (restricted) .......................................................... 19




Appendixes




Table A.1. Enterprise Survey: Totals from respondents ....................................................................... 23


Table A.2. NACE categories statistics .................................................................................................. 24


Table A.3. Enterprise Survey: Totals for manufacturing ....................................................................... 25


Table A.4. NACE categories statistics: Manufacturing ......................................................................... 26


Table A.5. Firms’ sample composition: Manufacturing ........................................................................ 27


Table A.6. Textiles and apparel enterprises as a share of all manufacturing enterprises .................... 27


Table A.7. Firms’ sample composition: Textiles and apparel ............................................................... 28


Table A.8. Foreign trade in manufactured goods ................................................................................. 29


Table A.9. Exports in manufactured goods by sector ......................................................................... 29


Table A.10. Trade in textiles and apparel ............................................................................................... 30


Table A.11. Exporters and importers in the FT database ....................................................................... 31


Table A.12. Sampling weights assigned to enterprises .......................................................................... 32


Table A.13. Sectoral composition in the Enterprise Survey ................................................................... 33


Table B.1. Production function estimation for NACE 17 firms ............................................................. 36


Table B.2. Value added function estimation for NACE 17 firms ........................................................... 37


Table B.3. Distribution of fixed and random effects for NACE 17 firms ............................................... 38


Table B.4. Production function estimation for NACE 18 firms ............................................................. 39


Table B.5. Value added function estimation for NACE 18 firms ........................................................... 40


Table B.6. Distribution of fixed and random effects for NACE 18 firms ............................................... 40


Table B.7. Production function estimation for other manufacturing ..................................................... 41


Table B.8. Value added function estimation for other manufactures ................................................... 42


Table B.9. Distribution of fixed and random effects for other manufactures in Turkey ........................ 42




Figure A1. Venn diagram: ES and FT-Export databases in 2008 ......................................................... 31






vii


EXECUTIVE SUMMARY




Liberalizing international trade creates both opportunities and threats for productive


enterprises. The opportunities stem from the opening of new markets for its products; the threats stem


from the entry of competitors to contest the liberalized market. The elimination of textiles and apparel


quotas in Canada, the United States of America and the European Union in the period 1995–2005


created opportunities and threats for textiles and apparel exporters worldwide. The system of bilateral


quotas covered by the ATC had the explicit goal of providing protection to import-competing


producers of these products in importing countries. It had the unintended goal of providing niches


within the markets of these importing countries for exports from countries not subject to binding


bilateral quotas. Those with binding quotas were constrained in the quantity that they could export to


Canada, the United States and the European Union. Those without quotas, or with non-binding quotas,


were able to expand exports beyond what would have been possible with quota-less trade. The


removal of quotas was thus not only a benefit to the consumers of ATC-importing countries, but also


the occasion of a massive dislocation among exporting countries.




The removal of ATC quotas represents a quasi-natural experiment and allows for a consistent


and robust testing of the most recent theoretical predictions. Based on firm-level Turkish data,1 this


paper provides evidence on how a trade policy shock has differentiated effects on firms within


industries. The paper also provides evidence on how the aggregate sectoral trade response is a


combination of within-firm responses and between-firm shifts in the composition of output.




Turkey as an economy responded well to this opportunity, as measured by positive growth in


export value to the countries eliminating quota. We decompose this success in this paper by observing


the changes in behaviour at the enterprise level.




• When we divide the enterprises into those producing textiles and apparel subject to quota in


2004 (i.e., ATC goods) and others, we find that those enterprises producing ATC goods grew


more rapidly in terms of real sales revenue and employment – and that they also had


significantly higher profit rates on average.




• When we divide the enterprises into exporting versus non-exporting, we find that the non-


exporting enterprises actually grew faster in both real sales and employment during the period


associated with quota elimination.




In sum, the textiles and apparel industries were the sources of faster economic growth.


However, the enterprises involved in exporting grew more slowly than those involved in exclusively


domestic sales.




We are also able to draw conclusions on enterprise-level performance of exporters. For


exporters, we found that:




• Those specializing in textiles and apparel during this period had lower sales revenue and


employment growth and a lower profit rate on average than those selling other products.




• Those exporting to the United States and the European Union had relatively higher sales


revenue and employment growth, and a higher profit rate, on average than those exporting


only to other markets.




When we consider the motivations for enterprises to cease operations, we consider a number


of alternative explanations. We find that larger enterprises were more likely to survive from one year to




1 We rely upon two databases maintained by the Turkish Statistical Institute (TUIK): the Enterprise Survey (ES) and the
Foreign Trade (FT) databases. The period retained is 2003–2008 inclusive.





viii


the next, whether size was measured by the value of sales, of employment or of the number of plants.


Independently of these factors, enterprises that had been exporters were significantly more likely to


survive from one year to the next during this period. However, this latter effect was on average


counteracted by the effect of being a producer of ATC goods: ATC producers were significantly more


likely to fail during this period.




We conclude from this that Turkey’s success in the export of textiles and apparel must be


taken in context. While Turkish enterprises were more successful than most in adapting to the post-


quota market in textiles and apparel, their performance paled relative to the performance of enterprises


in non-ATC areas. Exports to markets other than Canada, the United States and the European Union


for ATC products were reduced, while production of non-ATC products both for export and domestic


use grew rapidly. Enterprises then “voted with their feet” – significantly more enterprises in ATC


industries closed during this period than those in other industries.











Turkish Enterprise-Level Response to Foreign Trade Liberalization 1


1. INTRODUCTION




Trade in textiles and apparel is of special interest among international trade transactions. The


Agreement on Textiles and Clothing (ATC) signed in 1995 under the auspices of the World Trade


Organization was the culmination of a decades-long protective trade-policy initiative in the clothing-


and textile-importing countries of the European Union and Canada and the United States. It maintained


the pattern of bilateral quotas imposed by these importing countries on the most successful


developing-country exporters. It also introduced a fixed timetable for the removal of these quotas.


Removal was to occur in four stages, with over half of quota-restricted trade, as measured by value, to


be liberalized on 1 January 2005. Textiles and apparel products were separated into four groups, with


imports in group 1 liberalized for those importers at the beginning of 1995, imports of group 2


liberalized at the beginning of 1998, imports of group 3 liberalized at the beginning of 2002 and imports


of group 4 liberalized at the beginning of 2005.2




The system of bilateral quotas covered by the ATC had the explicit goal of providing protection


to import-competing producers of these products in the importing countries. It had the unintended


goal of providing niches within the markets of these importing countries for exports from countries not


subject to binding bilateral quotas. Those with binding quotas were constrained in the quantity that


they could export to Canada, the United States and the European Union. Those without quotas, or with


non-binding quotas, were able to expand exports beyond what would have been possible with quota-


less trade. The removal of quotas was thus not only a benefit to the consumers of ATC-importing


countries, but also the occasion of a massive dislocation among exporting countries.




Conway and Fugazza (2011) show that the removal of the final ATC quotas in 2005 brought


about a division of textile- and apparel-exporting countries into groups of winners and losers. Large


exporters previously constrained by binding quota expanded their value of exports and market share in


the ATC importing countries. Smaller exporters in nearly all cases experienced reduced export value


and market share as they faced direct competition with larger exporters. Among the larger exporters


without binding quotas, there was a division between successful and less-successful ones – some


countries were able to maintain or even expand market share, while others could not.




Turkey is a successful country from the latter category – a large exporter, especially to the


European Union that was not constrained by quotas in that market prior to 2005. Removal of the ATC


quota system put Turkish enterprises in more direct competition with the low-cost producers of East


and South Asia. Despite this increased competition, the textile and apparel sectors in Turkey


increased the value of their exports to the United States and European Union in the years 2005–2008.




The removal of ATC quotas represents a quasi-natural experiment and allows for a consistent


and robust testing of the most recent theoretical predictions. Based on firm-level Turkish data,3 this


paper provides evidence on how a trade policy shock has differentiated effects on firms within


industries as predicted by trade models à la Melitz (2003) with imperfect competition and


heterogeneous firms. The paper also provides evidence on how the aggregate sectoral trade response


is a combination of within-firm responses and between-firm shifts in the composition of output.




The major contribution of this paper is the empirical assessment of firms’ responses to trade


reform. The paper adds to the quota literature microlevel evidence from the supply side. While there is


convincing evidence from the demand side on how quotas affect product quality at the product level,


evidence on how firms react to quotas is still very scarce. In other words, there is still need to




2 Canada, the United States and the European Union identified independently the specific goods to go into each group,
and thus the order of liberalization for each good varied among the importers. In this section we use the European Union
quota groupings, as the European Union was the dominant importer of Turkish textiles and apparel products.


3 We rely upon two databases maintained by the Turkish Statistical Institute (TUIK): the Enterprise Survey (ES) and the
Foreign Trade (FT) databases. The period retained is 2003–2008 inclusive.





2 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


document supply-side responses, especially in terms of reallocation of resources both among and


within firms. The paper also represents a unique exercise in providing evidence of firms’ reaction in a


country which was quota unrestricted and thus benefited from preferential access to European and


North American markets. It goes beyond a simple theoretical curiosity as it could help consolidate


policymaking.




Turkish data suggest that the textiles and apparel industries were the sources of faster


economic growth. However, the enterprises involved in exporting grew more slowly than those involved


in exclusively domestic sales. In order to qualify more precisely enterprise-level performance of


exporters we adopt a twofold empirical strategy.




We first use a difference-in-difference-in-difference analytical technique. We limit the analysis


to manufacturing enterprises, and then divide the enterprises into three categories: textiles, apparel


and other. We also subdivide each category into exporting and non-exporting enterprises based upon


their status in 2003–2004, and among exporting enterprises in textiles and apparel we distinguish those


exporting products in 2003 and 2004 for which quotas were removed in 2005 from those with products


for which quotas were removed earlier (or never existed). We further distinguish among exporters


between those that exported to ATC countries and those that exported only to other countries. Thus


two major effects can be identified: the effect of quota removal in 2005 on economic performance by


industrial category and the effect by initial export market. We also identify the marginal contributions


for various other categories of enterprises that help to refine our analysis. We then test the average


performance of exporting enterprises for which quotas were removed against exporting enterprises in


the same category with products for which quotas were not binding in 2003 and 2004. We consider


four measures of performance: growth in real sales revenue, growth in employment, evolution of profit


rate and growth in total factor productivity – all relative to the 2003–2004 base period.




We then investigate the entry and exit of enterprises in each of these categories. One potential


dimension for adjustment is in the exit of less-productive enterprises and the expansion of the


remaining more-productive enterprises. To do so we use a probit model with random effects


controlling for factors likely to influence the survival of firms’ export status.




Our results suggest that producers that specialized in textiles and apparel during the ATC


quotas removal period had ceteris paribus lower sales revenue and employment growth and a lower


profit rate on average than those selling other products. We also find that ATC producers were


significantly more likely to fail during this period.




The rest of the paper is organized as follows. The next section contains a succinct review of


recent works dealing with the impact of the removal of ATC quotas. Section 3 briefly describes the raw


data and section 4 presents major features characterizing Turkish firms with a focus on those operating


in the textiles and apparel sectors. Section 5 presents the empirical strategy adopted. Empirical results


are summarized in section 6. The last section concludes.






2. RELATED LITERATURE




This paper relates primarily to the recent empirical quota literature centred on the ending of the


ATC in 2005. Various theoretical predictions have been tested using either disaggregated trade data or


firm-level data. Most of these predictions were established in the 1960s and 1970s and derive from the


‘non-equivalence’ of import quotas and tariffs conjecture put forward in Bhagwati (1965). Like tariffs,


binding quotas raise the price of constrained goods relative to unconstrained goods. However, quotas


give rise to additional distortions.4 A possible response to these additional distortions is the upgrading


of quality (Falvey, 1979). All these predictions however concern importing countries and are demand




4 See Anderson (1988) for a comprehensive analysis.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 3


oriented. Recent developments in trade theory following the seminal work of Melitz (2003) have allowed


a more precise understanding of possible supply-side reactions to trade reform episodes in exporting


countries, especially in terms of reallocation of resources both among and within firms.




Harrigan and Barrows (2009) examined the difference in price and quality for United States


imports in a difference-in-difference framework for the top 20 exporters to the United States: there was


a time difference, from 2004 to 2005, and a categorical difference in quota-constrained versus


unconstrained imports. The authors first measured the average adjustment in price and quality for each


country in the sample; they found a substantial downward average adjustment in price for quota-


constrained imports and a much smaller downward adjustment in quality. There were no such


downward adjustments for unconstrained imports. The authors then tested across countries to


determine whether the adjustments in price and quality from 2004 to 2005 were on average


significantly different between constrained and unconstrained categories. The downward price


adjustments were statistically significant for all exporters at the 95 per cent level of confidence, for


China alone and for non-China exporters. The downward quality adjustments were significant for China


alone and for all exporters at the 90 per cent level of confidence. This work was done at a quite


detailed level of disaggregation, and signalled the expected impact of quota removal on both price and


quality. It treated the observation of a binding quota as an exogenous event, however, and this could


introduce bias.




Edwars and Sundaram (2012), using data for Indian manufacturing firms, find that the ATC


quota removal was associated with an increase in market share of quota-constrained products relative


to unconstrained products. The magnitude of these effects suggests that quota removal was


associated with an increase in sales of about 20 per cent. Estimates are robust to controlling for


unobservable firm-specific shocks affecting outcomes. The authors’ results are consistent with quality


downgrading by textile firms in response to the quota removal. The ATC quota removal appeared to be


associated with reallocation of market share towards low-price products. In addition the price decrease


associated with quota removal was larger for low-price products. Moreover, their evidence suggests


that the estimated effects operated through an extensive margin in the first place, through product-


switching by firms.




Brambilla, Khandelwal and Schott (2010) focus their attention on exporters of textiles and


apparel to the United States. They work as well with 10-digit HS data on imports from these countries


into the United States, and they also categorize the imports as being quota constrained versus


unconstrained using the quota classifications of the United States. They analyse carefully the impact of


the quota, and then contrast that with behaviour after quota removal: they are careful to distinguish


among the four stages of sequential quota elimination under the ATC, and to connect the changes in


quantity and price with the appropriate stage of quota removal. They find both an increase in quantity


and a reduction in price for Chinese goods that is significantly different from that observed in other


quota-constrained exporters. They do not calculate quality as in Harrigan and Barrows (2009), and thus


cannot draw conclusions on the impacts of price versus quality. They also treat the quota-constrained


period as an exogenous event.




Using Chinese customs data on exports collected at the firm level, Bernhofen, Upward and


Wang (2011) found an average price drop of about 30 per cent due to the removal of the ATC quotas.


Of this overall drop in price, more than half was found to be caused by firm turnover or changes in the


composition of firms in the export market, indicating that quotas probably had an effect more on firm


entry than on product composition within existing firms in the export market. Their analysis points to a


predominant quality downgrading effect if compared with the competition effect in the fall of prices.


This conclusion was reached because differentiated products exhibited significantly greater price


reduction compared to homogeneous products.




The same database was used by Khandelwal, Schott and Wei (2011). The authors were


interested specifically in the productivity gains to China due to the ATC quota phase-down. In


particular, they decomposed productivity gains into gains from removal of the trade barrier and gains





4 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


from the removal of export licensing under the ATC quotas. Their results show that quota removal


coincided with substantial reallocation of export activity from incumbents to entrants, and that this


reallocation was inconsistent with an ex ante assignment of quotas by the Government of China on the


basis of firm productivity. As a result, the standard productivity growth expected from the removal of


the quota was magnified by the concomitant elimination of inefficient institutions and practices related


to the allocation of quota licences. In their counterfactual analysis, productivity growth from quota


removal appears to be 27 per cent higher than it would be if quotas had been allocated according to


firm efficiency. They also found that evidence for quality downgrading was not very strong. In


particular, estimates of such effect were not robust to the inclusion of country-product pair controls.




This study also relates to the literature using microdata to look at the impact of trade


liberalization on firm behaviour.5






3. DATA




Data used in our investigation are drawn from two sources. The first is the Enterprise Survey


conducted by the Turkish Statistical Institute (with Turkish acronym TUIK), soliciting survey responses


from roughly 80,000 enterprises each year.6 The second is the Foreign Trade database, reporting (from


customs sources) the roughly 500,000 annual export and import transactions by enterprises. These


two can be matched by unique enterprise number and by year so that we have not only the production


choices of the enterprise, but also the export and import transactions. We consider the 2003–2008


time period. We briefly examine the two databases in turn below. Details are provided in appendix A.






3.1. THE TURKISH ENTERPRISE SURVEY


The Turkish Statistical Institute conducts a large-scale survey annually. This survey is


conducted at the level of the enterprise, with multi-plant enterprises aggregating information for all


plants.7 The survey includes questions on the enterprise’s characteristics, its uses of inputs, the value


of its sales, and whether it is involved in export activity. In recent years it has tabulated responses from


more than 80,000 firms. The number of enterprises surveyed has stayed roughly constant through the


years, with the sharp drop in 2005 indicating a simple reduction in the number of enterprises surveyed


rather than a fall in total Turkish economic activity.8




Light manufactures represent the near totality of sales revenues in the surveys, while other


categories (agriculture, some services) make minor contributions to sales but more important


contributions to employment.




There are 46 two-digit NACE categories for which at least some enterprises report economic


activity.9




5 See for instance Pavcnik (2002), Goldberg, Khandelwal, Pavcnik and Topalova (2010) and Tybout (2000 and 2003).


6 The Enterprise Survey does not distinguish individual plants, but aggregates to the enterprise level. An earlier survey by
TUIK solicited information at the plant level, but from 2003 the responses are by enterprise. We limit ourselves to this
aggregation in this report.


7 Prior to 2003, the Survey was conducted at the plant level, and it is thus difficult to compare pre-2003 responses to
those of 2003 and later.


8 The appendix describes a reduction in 2005 in the number of enterprises invited automatically to participate in the
survey. This change could have caused the reduction in total respondents.


9 The two-digit NACE categories used by TUIK correspond in most regards to the categories of the International
Standard Industrial Classification (ISIC). In particular, the categories from 10 to 14 represent mining, the categories from
15 to 37 represent various manufacturing activities, and the higher categories represent retail, wholesale and service
(including public service) activities. The categories 17 and 18 are associated with textiles and apparel manufacture,
respectively. NACE and ISIC differ in their third and higher digits; the appendix includes a short description.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 5


In 2003, manufacturing enterprises represented about one-third of the total number of


enterprises surveyed. The reported sales of manufacturing enterprises were nearly 99 per cent of the


total sales reported by all enterprises. Manufacturing enterprises employed over half of the workers


reported employed by the surveyed firms. From 2003 to 2008, these percentages evolved: the share


of manufacturing enterprises among all surveyed rose to over 40 per cent while the share of total sales


by these enterprises dropped slightly, to nearly 98 per cent. The share of manufacturing employees in


total employees covered by the survey dropped to just over 43 per cent.




Further breaking down manufacturing enterprises surveyed by three-digit NACE code, we


obtain that the top categories in terms of sales in 2003 were apparel, textiles, motor vehicles, iron and


steel and chemical products.10 However, there were significant differences across sectors.




Apparel (182) was the top-ranked manufacturing sector in 2003 by sales revenue. It obtained


this ranking in large part because of the large number of enterprises operating in that sector. Given


that 12.6 per cent of all manufacturing enterprises reported to be apparel producers, it could be


surprising that only 7.3 per cent of all manufacturing sales revenue came from the sector. Table 1


reports the evolution of these shares over time. Although the share in total respondents remained


relatively constant, we observe that the incidence of textiles and apparel enterprises had been


declining in total sales and employment. Motor vehicles (341), iron and steel (271) and petroleum


products (232) were the next three ranked sectors in manufacturing in 2003 (and the top three in 2008),


but they reached these rankings with a miniscule share of the enterprises surveyed. The three sectors


together represented only 0.7 per cent of the enterprises and 5.0 per cent of the employees in 2003,


but they reported 18.1 per cent of the sales revenue. This was due to their characteristics of much


larger-than-average size and relative capital intensity – they were less labour-using than average. The


category of food products (158) was similar to apparel in having a relatively large share (7.0 per cent) of


the manufacturing enterprises surveyed and a relatively smaller share of sales revenues (4.7 per cent).


It was not as labour-using as apparel.






Table 1


Textiles and apparel enterprises as a share of all manufacturing enterprises




Year
Number of
enterprises


Total
sales revenue


Number of
employees


2003 0.25 0.21 0.36


2004 0.25 0.18 0.35


2005 0.28 0.17 0.34


2006 0.26 0.16 0.32


2007 0.26 0.16 0.31


2008 0.25 0.13 0.29


Source: TUIK Enterprise Survey database.




3.2. THE FOREIGN TRADE DATABASE


The Enterprise Survey includes two questions on foreign trade: enterprises are asked to


provide the Turkish lira value of all exports and (separately) all imports of the enterprise during the year.


While these are available, more detailed data about enterprise foreign trade activity is available through


the foreign trade statistics collected by the Customs Department. This is a separate database.


However, it includes the unique enterprise identification code that allows merging of the two


databases.




10 The top ten three-digit NACE categories are 182, apparel; 341, motor vehicles; 271, basic iron and steel; 232, refined
petroleum products; 158, other food products; 171, spinning and weaving of textiles ; 172, manufacture of other textiles;
244, pharmaceuticals ; 241, basic chemicals; 297, domestic appliances.





6 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


The Foreign Trade database identifies specific transactions in goods, either export or import,


undertaken by Turkish enterprises. Table 2 provides summary statistics for international trade in


manufactured goods divided into two groups: textiles and apparel, and other manufactures. The value


of exports and the value of imports have increased for both groups during 2003–2008. Figures clearly


reflect the decline of the textile and apparel sectors in manufactures sales observed in the Enterprise


Survey database. Textile and apparel exports increased by 50 per cent, while exports in other


manufacture sectors increased by almost 300 per cent.






Table 2


Trade in manufactures


(Billions of United States dollars)




Exports Imports


Textiles/apparel Other Total Textiles/apparel Other Total


2003 5.1 17.8 22.8 5.5 30.8 36.3


2004 6.1 26.5 32.7 6.3 47.3 53.5


2005 6.1 32.1 38.1 6.3 55.7 62.0


2006 6.9 39.5 46.4 6.3 67.8 74.1


2007 7.9 49.7 57.6 7.7 79.6 87.4


2008 7.7 63.6 71.3 7.1 99.1 106.2


Source: TUIK Foreign Trade database.






There is in general a great variety in export performance by group of goods (NACE two-digit),


but we can identify some salient trends. In 2003, the top three product groups for exports were motor


vehicles (34), textiles (17) and apparel (18). By 2008, motor vehicles remained the top category but iron


and steel (27), appliances (29), food products (15) and petroleum products (23) overtook textile and


apparel in terms of export value. Turkey was in fact nearly in balance with trade in textiles and apparel;


exports and imports rose through 2007 but then both declined in 2008.




Table 3 shows that for manufactures more than 45 per cent of firms present in the FT


database were “importer only” in 2003. This share remained somewhat constant during 2003–2008.


The remaining firms were either “exporter only” or “both” in equal shares. A similar pattern is observed


for firms in textiles and apparel.






Table 3


Exporters and/or importers in manufacture sectors




2003 2004 2005 2006 2007 2008


Exporter only 18 682 20 451 21 661 22 485 25 203 24 486


Importer only 27 255 29 982 32 210 33 899 39 356 37 669


Both 17 009 19 043 20 554 21 749 23 190 23 755


Total 62 946 69 476 74 425 78 133 87 749 85 910


Source: TUIK Foreign Trade database. Authors’ calculations.












Turkish Enterprise-Level Response to Foreign Trade Liberalization 7


4. FACTS ON THE IMPACT OF QUOTAS REMOVAL




Table 4 divides both exports and imports for Turkey as reported in the FT database into five


categories and observes the evolution of each of these through the sample period 2003–2008.


Category 0 comprises all trade other than in textiles and apparel; as expected, it represents almost all


of Turkey’s imports and a majority of its exports. Groups 1, 2 and 3 had already been liberalized by the


beginning of this time period; they represented a small share of exports and a very small share of


imports, and their importance to Turkish trade was in decline throughout this period. Group 4 identifies


those textiles and apparel products for which the European Union liberalized its imports in 2005. While


the total value of exports rose between 2003 and 2007, indicating that Turkey successfully weathered


the increased competition in these products, the share of these goods in total exports declined


markedly. While products in this group represented nearly 25 per cent of Turkish exports in 2003, that


share had fallen to 15 per cent in 2007 and 11 per cent in 2008.






Table 4


Decomposing foreign trade into quota liberalization categories



European
Union quota
grouping


Exports European
Union quota
grouping


Imports


year
US$


(millions)
percentage year US$ (millions) percentage


0 2003 33 120.2 70.1 0 2003 64 366.2 92.8


0 2004 46 733.9 74.0 0 2004 91 320.3 93.6


0 2005 56 495.1 76.9 0 2005 110 168.5 94.3


0 2006 67 840.2 79.3 0 2006 132 305.3 94.8


0 2007 87 876.2 81.9 0 2007 161 172.6 94.8


0 2008 113 714.6 86.1 0 2008 193 178.4 95.7




1 2003 46.6 0.1 1 2003 64.5 0.1


1 2004 58.7 0.1 1 2004 89.6 0.1


1 2005 69.1 0.1 1 2005 104.3 0.1


1 2006 75.2 0.1 1 2006 135.4 0.1


1 2007 81.2 0.1 1 2007 142.4 0.1


1 2008 89.7 0.1 1 2008 167.3 0.1




2 2003 1 304.8 2.8 2 2003 1 052.6 1.5


2 2004 1 452.9 2.3 2 2004 1 236.4 1.3


2 2005 1 375.8 1.9 2 2005 1 396.8 1.2


2 2006 1 514.9 1.8 2 2006 1 567.0 1.1


2 2007 1 602.6 1.5 2 2007 1 734.7 1.0


2 2008 1 769.1 1.3 2 2008 1 895.2 0.9




3 2003 1 403.8 3.0 3 2003 493.3 0.7


3 2004 1 525.7 2.4 3 2004 570.9 0.6


3 2005 1 410.4 1.9 3 2005 626.8 0.5


3 2006 1 389.4 1.6 3 2006 710.6 0.5


3 2007 1 454.2 1.4 3 2007 513.4 0.3


3 2008 1 462.4 1.1 3 2008 487.4 0.2




4 2003 11 377.5 24.1 4 2003 3 363.0 4.9


4 2004 13 396.0 21.2 4 2004 4 322.6 4.4


4 2005 14 126.0 19.2 4 2005 4 477.8 3.8


4 2006 14 715.0 17.2 4 2006 4 857.9 3.5


4 2007 16 257.5 15.2 4 2007 6 499.6 3.8


4 2008 14 991.4 11.4 4 2008 6 235.2 3.1


Source: TUIK Foreign Trade database. Authors’ calculations.






8 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


Table 5 reports by European Union quota grouping the mean sales revenue and employment


of Turkish enterprises operating in the textile and apparel sectors.11 There are six groups identified.


The first group (group N) includes enterprises with no exports during the period. The second group


(group 0) includes enterprises that report positive exports, but no exports in the HS categories


associated with the textile and apparel quotas. Groups 1 through 4 are defined, as above, by the


timing of quota liberalization for the product. Turkey’s producers were not constrained by quotas, but


their competitive position was weakened when quotas on other major exporters were removed. There


are two entries for each group/year combination. The two entries in group N are the mean sales


revenues in Turkish lira and the number of enterprises in the group. The two entries for the other


groups include the ratio of mean sales revenue in that group to mean sales revenue in group N and the


number of enterprises in the group.






Table 5


Mean revenue and employment for Turkish enterprises active throughout the period 2003–2008




Mean revenue


US$ (millions) As a multiple of group N


group N group 0 group 1 group 2 group 3 group 4


2003 5 702 367 4.96 0.47 2.57 1.58 2.73


3 496 3 481 3 125 84 886


2004 6 732 964 5.30 0.38 2.75 1.75 2.42


3 644 4 015 4 151 89 994


2005 6 969 197 5.42 0.77 3.05 1.74 2.29


3 621 4 294 5 145 78 1013


2006 8 696 511 5.37 1.00 2.22 1.35 2.18


3 613 4 287 5 140 66 1044


2007 8 974 675 5.78 1.49 2.33 1.43 2.37


3 671 4 314 3 128 41 998


2008 9 134 049 6.73 1.31 2.79 1.93 2.32


3 782 4 275 3 146 46 903




Mean employment


US$ (millions) As a multiple of group N


group N group 0 group 1 group 2 group 3 group 4


2003 65.9 2.35 0.76 2.37 1.73 2.86


3 496 3481 3 125 84 886


2004 72.7 2.22 0.55 2.24 2.17 2.52


3 644 4015 4 151 89 994


2005 74.6 2.21 0.90 2.67 2.21 2.57


3 621 4294 5 145 78 1013


2006 77.4 2.27 1.56 1.97 1.74 2.42


3 613 4287 5 140 66 1044


2007 75.8 2.47 2.54 2.04 1.99 2.67


3 671 4314 3 128 41 998


2008 72.6 2.68 2.84 2.28 2.45 2.73


3 782 4275 3 146 46 903




Source: TUIK Foreign Trade Database. Authors’ calculations.


Notes: Figures in italics are the number of firms in the group; group N: Enterprise with no exports; group 0:
Enterprises with exports not covered by ATC agreement; group 1: First group of liberalization (1995); group
2: Second group (1998); group 3: Third group (2002); group 4: Fourth group (2005).






11 This table is limited to enterprises observed in each year of the Enterprise Survey. A table with all reporting enterprises
has similar characteristics.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 9


Comparing mean sales revenue by group in 2003, we see that mean revenue in group 0


(including non-textile, non-apparel exporters) is nearly 5 times larger than mean revenue among non-


exporters over a similar number of enterprises (nearly 3,500 in each case). Among the textile/apparel


groups, group 4 and group 2 are the largest in mean and roughly half as large as in group 0. Group 4


is also the largest in number of enterprises (with 886), and group 2 is a distant second (with 125).


Groups 3 and (especially) 1 include a much smaller number of enterprises.




Revenues for non-exporters (group N) grew rapidly in nominal terms from 2003 to 2008. Those


for exporters of non-ATC goods (group 0) and of goods liberalized by 1995 (group1) grew even more


rapidly. Exporters of goods liberalized by 1998 (group 2) and by 2005 (group 3) experienced a slowing


down of their sales in 2006 with a rapid recovery afterwards. Sales of exporters of ATC-goods (group 4)


follow a similar trajectory. However, the inflection point for sales occurred already in 2005 and the


recovery has been shaky.




When mean employment is considered, we find that textiles and apparel enterprises were the


largest group. Employment in group N enterprises was fairly stable throughout the period, with a mean


of nearly 66 employees per firm in 2003 and of 72.6 employees per firm in 2008. Group 0 enterprises


had over twice as many employees on average, and that ratio grew during the sample period. Group 4


enterprises were even larger, with nearly three times as many employees as group N enterprises. This


ratio declined slightly through 2006 before rising through 2008.




An additional possible outcome of the removal of quotas was the change in the status of firms.


In table 6 we examine the subsample of firms for which observations are available in both 2003/2004


and 2005/2006. In panel 1, those exporting in 2003/2004 are represented in the first row. They are


divided into those enterprises that continued to export in 2005/2006 and those that ceased exporting


(but continued to produce). Those 2,654 that stopped exporting represent just over 4 per cent of the


enterprises. The second row is that of enterprises that did not export in 2003/2004; among these,


3,729 (or 5.8 per cent of the total) chose to export in 2005/2006.






Table 6


Transition in enterprise characteristics from 2003/2004 to 2005/2006




Exporting in 2005/2006 Not exporting in 2005/2006 Row total


Exporting in 2003/2004 32 770 (50.9 %) 2 654 (4.1 %) 35 424 (55.0 %)


Not Exporting in
2003/2004


3 729 (5.8 %) 25 181 (39.2 %) 28 910 (45.0 %)


Column total 36 499 (56.7 %) 27 835 (43.3 %) 64 334 (100 %)




ATC in 2003/2004 Not ATC in 2005/2006 Row total


ATC in 2003/2004 29 334 (45.6 %) 4 026 (6.3 %) 33 360 (51.9 %)


Not ATC in 2003/2004 4 277 (6.6 %) 26 697 (41.5 %) 30 974 (48.1 %)


Column Total 33 611 (52.2 %) 30 713 (47.8 %) 64 334 (100 %)





Export to European
Union/United States in
2005/2006


Not export to European
Union/ United States in
2005/2006


Row total


Export to European
Union/United States in
2003/2004


31 462 (48.9 %) 3 235 (5.0 %) 34 697 (53.9 %)


Not Export to European
Union/ United States in
2003/2004


2 649 (4.1 %) 26 988 (42.0 %) 29 637 (46.1 %)


Column Total 34 111 (53.0 %) 30 223 (47.0 %) 64 334 (100 %)


Source: TUIK Foreign Trade Database. Authors’ calculations.





10 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


In panel 2, enterprises are divided between those in the textiles and apparel sector producing


goods subject to quota in the European Union or the United States in 2003/2004 (ATC products) and


all others. There were 4,026 enterprises that stopped making ATC products (but remained in business),


representing 6.3 per cent of the total. By contrast, there were 4,277 enterprises (6.6 per cent) that


began making ATC products in 2005/2006 after not having done so in 2003/2004.




In panel 3, enterprises are divided between whether they exported to the United States and


the European Union or not. Five per cent of these chose to stop exporting to these countries in


2005/2006 after having done so in 2003/2004, and 4 per cent chose to begin exporting to these


countries after not having done so in 2003/2004.




Table 7 reports a similar analysis for enterprises operating in both 2003/2004 and 2007/2008.


We also observe that a large majority of firms did not change its status, even though a longer time


period is considered. Among those firms that switched, compared to the table 6 sample, more were


likely to export, but fewer were likely to export to the European Union/United States and less likely to


produce ATC products. Logic suggests that these patterns would be intensified if we examine only


enterprises with production of ATC goods.






Table 7


Transition in enterprise characteristics from 2003/2004 to 2007/2008




Exporting in 2007/2008 Not exporting in 2007/2008 Row total


Exporting in 2003/2004 25 439 (46.8 %) 3 433 (6.3 %) 28 872 (53.1 %)


Not exporting in
2003/2004


4 171 (7.7 %) 21 284 (39.2 %) 25 455 (46.9 %)


Column total 29 610 (54.5 %) 24 717 (45.5 %) 54 327 (100 %)




ATC in 2007/2008 Not ATC in 2007/2008 Row total


ATC in 2003/2004 23 039 (42.4 %) 4 063 (7.5 %) 27 102 (49.9 %)


Not ATC in 2003/2004 3 830 (7.0 %) 23 395 (43.1 %) 27 225 (50.1 %)


Column total 26 869 (49.4 %) 27 458 (50.6 %) 54 327 (100 %)





Export to European
Union/United States in
2007/2008


Not Export to European
Union/United States in
2007/2008


Row total


Export to European
Union/United States in
2003/2004


25 086 (46.2 %) 3 145 (5.8 %) 28 231 (52.0 %)


Not Export to European
Union/United States in
2003/2004


2 352 (4.3 %) 23 744 (43.7 %) 26 096 (48.0 %)


Column Total 27 438 (50.5 %) 26 889 (49.5 %) 54 327 (100 %)


Source: Authors’ calculations.




Table 8 reports the transition of enterprises among exporting states relative to 2003/2004, but


calculates the transitions only for those enterprises producing ATC goods in 2003/2004. When the


exporting decision is considered for the 2005/2006 horizon, a smaller percentage of these enterprises


chose to cease exporting after having exported in 2003/2004. At the same time, a larger proportion of


enterprises that did not export in 2003/2004 chose to begin exporting in 2005/2006. This pattern was


maintained for the 2007/2008 horizon, while the percentage of enterprises changing behaviour rose.


While exporting firms increased in percentage overall, the shares of enterprises exporting to the


European Union and the United States decreased. Thus there is no clear pattern among switchers


between those that stopped exporting to the European Union and the United States and those that


began exporting to those destinations.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 11


Table 8


Transition in enterprise characteristics for ATC producers




Exporting in 2005/2006 Not exporting in 2005/2006 Row total


Exporting in 2003/2004 19 501 (39.8 %) 1 925 (3.9 %) 21 426 (43.7 %)


Not exporting in
2003/2004


3 661 (7.5 %) 23 880 (48.8 %) 27 541 (56.3 %)


Column total 23 162 (47.3 %) 25 805 (52.7 %) 48 967 (100 %)




Exporting in 2007/2008 Not exporting in 2007/2008 Row total


Exporting in 2003/2004 15 622 (36.9 %) 2 569 (6.1 %) 18 191 (43.0 %)


Not exporting in
2003/2004


4 074 (9.6 %) 20 085 (47.4 %) 24 159 (57.0 %)


Column total 19 696 (46.5 %) 22 654 (53.5 %) 42 350 (100 %)





Export to European
Union/United States in
2005/2006


Not export to European
Union/United States in
2005/2006


Row total


Export to European
Union/United States in
2003/2004


19 721 (40.3 %) 2 087 (4.3 %) 21 808 (44.6 %)


Not export to European
Union/United States in
2003/2004


1 987 (4.1 %) 25 172 (51.4 %) 27 159 (55.4 %)


Column total 21 708 (44.4 %) 27 259 (55.6 %) 48 967 (100 %)





Export to European
Union/United States in
2007/2008


Not export to European
Union/United States in
2007/2008


Row total


Export to European
Union/United States in
2003/2004


16 619 (39.2 %) 2 034 (4.8 %) 18 653 (44.0 %)


Not export to European
Union/United States in
2003/2004


1 785 (4.2 %) 21 912 (51.7 %) 23 697 (56.0 %)


Column total 27 438 (43.4 %) 26 889 (49.5 %) 42 350 (100 %)


Source: Authors’ calculations.






Were these the same enterprises that changed their characteristics? For example, did an


enterprise that stopped exporting also stop producing ATC goods? Table 9 reports the correlation of


enterprise switching behaviour for both time horizons, and the correlations indicate quite common


behaviour. First, there is a negative correlation between X status and B status – those enterprises that


chose to stop exporting were more likely to be those that began selling ATC goods – although that


correlation is not perfect. Those enterprises with changing X status – i.e., that chose to stop exporting


to the European Union and the United States – exhibit a weaker correlation of the same sign.








12 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


Table 9


Correlation of switching behaviour by enterprises




DX i0506 DB i0506 DZ i0506
DXi0506 1.0
DBi0506 -0.23 1.0
DZi0506 0.28 -0.08 1.0




DX i0708 DB i0708 DZ i0708
DXi0708 1.0
DBi0708 -0.26 1.0
DZi0708 0.28 -0.08 1.0


Source: Authors’ calculations.


Notes: DXi0506: Indicator variable defined as Xi0506 – Xi0304. The Xi0506 variable is a binary variable equal to 1 if
the enterprise is an exporter in the years 2005/2006 and 0 otherwise. DBi0506: Indicator variable defined as
Bi0506 – Bi0304. The Bi0506 variable is a binary variable equal to 1 if the enterprise produced ATC goods in the
years 2005/2006 and 0 otherwise. DZi0506: Indicator variable defined as Zi0506 – Zi0304. The Zi0506 variable is a
binary variable equal to 1 if the enterprise exported to the United States and/or European Union in the years
2005/2006 and 0 otherwise.








5. THE EMPIRICAL STRATEGY




The previous section describes some major changes in features characterizing Turkish firms


coincident with the removal of quotas. In this section we further investigate the impact of the latter


policy reform by using four performance indicators as dependent variables in a difference-in-difference


estimation. We further estimate the impact of quota removal on the survival of firms by implementing a


standard probit estimation with random effects.




5.1. TRIMMED DATA


As is typical for firm-level databases, there are numerous extreme values associated with


performance measures; these do not reflect substantive differences, but rather seem to be either data-


entry mistakes or growth rates predicated on very small initial values. We exclude these extreme values


by trimming the top and bottom 1 per cent of observations from the sample we consider. The impact of


this trimming process is illustrated in table 10.12






Table 10


Moments of panel data, with and excluding extreme values




G(sales) g(employment) G(profits) g(productivity)


Full Trim Full Trim Full Trim Full Trim


Minimum -1 -0.935 -0.999 -0.864 -6966.7 -0.456 -2484.4 -26.8


Maximum 1 171 398 20.297 1 818 16.28 19 925.9 0.522 792.4 24.1


N 17 998 17 638 37 947 37 187 19 254 18 868 8 188 8 024


Source: Authors’ calculations.




12 We trimmed the top and bottom 2 per cent of the distribution in a robustness check, and the results we derived were
little different from those reported here.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 13


5.2. DIFFERENCE-IN-DIFFERENCE ESTIMATION


We use a difference-in-difference-in-difference analytical technique. We limit the analysis to


manufacturing enterprises, and then divide the enterprises into three categories: textiles, apparel and


other. We also subdivide each category into exporting (Xit=1) and non-exporting (Xit=0) enterprises


based upon their status in 2003–2004. The constant (the coefficient α) represents the average


performance of firms producing for the domestic market any product which was not an ATC product.


Among enterprises in textiles and apparel we distinguish between those that produced products in


2003 and 2004 for which quotas were removed in 2005 (Bit = 1) from those with products for which


quotas were removed earlier or never existed (Bit = 0). We also distinguish among exporters between


those that exported to ATC countries (ATCit = 1) and those that exported only to other countries (ATCit


= 0).


Yit = α + β Xit + γ Bit + δ ATCit + ε (Bit* Xit)+ η (ATCit* Bit)+ uit , uit~N(0,1) (1)




The coefficient ε identifies the average performance of exporting enterprises for which quotas


were removed against exporting enterprises in the same category with products for which quotas were


not binding in 2003 and 2004. The coefficient η identifies the effect by initial export market. It tests the


average performance of firms exporting ATC products to the European Union and the United States


relative to other markets. We consider four measures of performance (Yit): growth in real sales revenue,


growth in employment, evolution of profit rate and growth in total factor productivity – all relative to the


2003–2004 base period. All variables are either found directly in the raw data or are computed based


on these data. Enterprises reported their sales revenue in current Turkish lira. These are deflated to a


common 2003 Turkish lira value through use of producer price indices (PPI) matched with the


enterprise’s major product at the four-digit NACE level. If the four-digit PPI is not available, we use the


least aggregated index available: three-digit in most cases or rarely the two-digit indices. Enterprises


reported the average number of employees on an annual basis, and this is the measure used for


employment. As to the profit rate we use the profit rate on book value of capital reported by firms. Total


factor productivity needs to be estimated. This is done by using residuals from two different estimation


approaches. First, we take the residuals from a production function estimated at the two-digit NACE


level using four inputs (capital, labour, energy and materials); second, we take the residuals from a


value-added function in capital and labour estimated at the two-digit NACE level. Details of this


estimation and of the estimated technological coefficients are reported in appendix B.




5.3. CONDITIONAL PROBABILITY OF ENTERPRISE EXIT


The removal of quotas will lead to increased competition for Turkish textiles and apparel


enterprises in their export to the markets of the European Union and the United States. We are


interested in ascertaining whether this increased competition caused the exit of weaker enterprises.




Our empirical model is based on a standard approach in the empirical exit literature.13 We view


firms as making a decision to continue producing at the start of each year. Then if the firms are still in


operation, they decide on specific characteristics such as employment and market destinations


including foreign ones. Specifically, a firm i decides to produce a given product in year t+1 by


comparing the expected discounted sum of profits from operating, ESit+1, with the value it would earn


by exiting. Expected future profits are calculated from knowledge of the profit function Πit+1, the


observed state variables for year t (sales, employment, etc.) and knowledge of the transition process


for the time-varying state variables. If ESit+1>θ, the firm continues producing and we observe the


discrete variable y=0. If expected profits are less than the value of exit, the firm terminates production


for the market and we observe y=1.




Hence, our empirical model expresses the discrete exit variable in year t+1 as a function of


state variables in year t. State variables include past sales, past employment, whether the enterprise




13 See Caves (1998) for an early review of the literature.





14 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


has multiple plants, whether the enterprise exports, and whether it was involved in exporting to the


European Union and the United States. We then differentiate between those enterprises producing


ATC goods and those that were not.




Following the existing literature we estimate a probit model, that is




ESit+1= ZitΦ+uit, u it~ N(0,1) (2)


If ESit+1>θ, y it+1=0


If ESit+1<=θ, y it+1 = 1




Zit is a matrix of state variables at time t which includes sales in the previous year (in logs), the


level of employment in the previous years (in logs), whether the firm was a multi-plant enterprise in the


previous year, whether the firm was exporting the year before, whether the firm was a producer of ATC


products liberalized in 2005 (group 4) and the level of total factor productivity. We consider those


manufacturing enterprises observed without interruption from 2003, and we separate them into those


observed just for 2003 and 2004 (i.e., exiting after 2004), those observed for the three years from 2003


to 2005 (exiting after 2005) and similarly for 2006, 2007 and 2008.






6. RESULTS




Our empirical strategy is twofold. First we assess the impact of ATC removal on four


performance indicators. Results are presented and discussed in the next subsection. We also assess


the impact of the removal of ATC quotas on firms’ turnover. The latter results are reported and


commented in subsection 4.2.




6.1. DIFFERENCE-IN-DIFFERENCE ESTIMATION


Results of the difference-in-difference model as defined in equation (1) are reported in table


11. We consider each specification twice: first with only the explanatory variables Xit, Bit and ATCit, and


then again with each of these variables and the interactive terms Xit*Bit and X it *ATCit.




The top panel reports the regression of growth in the performance measure from its average in


2003–2004 to its average in 2005–2006 for each firm; the bottom panel reports a similar regression


from average value for 2003–2004 to average value in 2007–2008 for each firm. The independent


regressors are the binary variables defined in subsection 5.1: Xit (with coefficient β), Bit (with coefficient


γ), ATCit (with coefficient δ), the interactive variable Xit*Bit (with coefficient ε) and the interactive variable


X it*ATCit (with coefficient η). The (β) coefficient measures improvement in the performance measure


over the period in question comparing non-exporters to exporters, (γ) comparing firms producing


textiles and apparel for which quotas were removed to all other firms, (δ) comparing firms selling to the


United States and the European Union to all other firms, (ε) among exporting firms, comparing those


making textiles and apparel for which quotas were removed to all other exporters and (η) among all


exporters, comparing those exporting to the United States and the European Union to all others.




The first column of panel 1 in table 11 illustrates the stylized facts of these regressions. The


intercept α indicates that non-exporting firms making goods other than textiles and apparel (under


quota restrictions until 2005) had significantly faster growth in terms of sales than did exporting firms or


firms making textiles and apparel for the domestic market (under quota restrictions until 2005). The


coefficient β indicates that exporting firms had significantly slower sales growth than non-exporting


firms over the 2005/2006 horizon. The coefficient γ indicates that when firms making textiles and


apparel subject to a European Union quota in 2003/2004 are compared to all other firms, these textiles


and apparel firms had significantly faster sales growth for the 2005/2006 horizon. The coefficient δ







Turkish Enterprise-Level Response to Foreign Trade Liberalization 15


compares the firms exporting to the United States and the European Union to all other firms. The sales


growth of these firms was significantly greater than that of all other firms.14






Table 11


Results for difference-in-difference estimation for Turkey



Panel 1 g(sales0506) g(sales0506) g(emp0506) g(emp0506) g(profit0506)) g(profit0506) g(prod0506) g(prod0506)


α 0.86 0.76 0.66 0.63 -0.014 -0.019 0.304 -0.60


(0.10) (0.14) (0.04) (0.04) (0.005) (0.006) (0.046) (0.77)


β -0.48 -0.18 -0.37 -0.16 -0.009 0.009 -0.115 1.38


(0.03) (0.18) (0.02) (0.08) (0.002) (0.01) (0.131) (0.96)


γ 0.23 0.32 0.25 0.28 0.012 0.017 0.706 1.63


(0.10) (0.14) (0.04) (0.04) (0.005) (0.006) (0.461) (0.78)


δ 0.18 -0.09 0.05 -0.05 0.003 0.009 0.041 1.10


(0.05) (0.40) (0.03) (0.20) (0.003) (0.03) (0.194) (1.10)


ε -0.30 -0.21 -0.019 -1.53


(0.18) (0.08) (0.010) (0.97)


η 0.28 0.10 -0.006 -1.06


(0.41) (0.21) (0.03) (1.12)


N 17 638 17 638 37 187 37 187 18 868 18 868 8 024 8 024


R2 0.02 0.02 0.01 0.01 0.002 0.002 0.0004 0.001


Root MSE 1.869 1.878 1.706 1.705 0.117 0.117 5.11 5.11




Panel 2 g(sales0708) g(sales0708) g(emp0708) g(emp0708) g(profit0708) g(profit0708) g(prod0708) g(prod0708)


α 1.673 1.61 0.990 0.97 -0.010 -0.015 1.067 -0.42


(0.158) (0.22) (0.056) (0.06) (0.006) (0.007) (0.572) (0.94)


β -0.884 -0.58 -0.494 -0.30 -0.014 0.004 -0.130 2.47


(0.052) (0.30) (0.025) (0.12) (0.002) (0.01) (0.164) (1.17)


γ 0.171 0.24 0.242 0.27 0.005 0.019 0.656 2.18


(0.159) (0.22) (0.058) (0.06) (0.003) (0.007) (0.566) (0.94)


δ 0.280 -0.72 0.078 -0.12 -0.009 0.476 -1.78


(0.093) (0.29) (0.051) (0.26) (0.017) (0.268) (1.51)


ε -0.30 -0.20 -0.018 -2.65


(0.30) (0.12) (0.011) (1.18)


η 1.02 0.20 0.015 2.28


(0.30) (0.27) (0.018) (1.53)


N 15 814 15 814 32 969 32 969 16 617 16 617 6 848 6 848


R2 0.02 0.02 0.01 0.01 0.003 0.003 0.002 0.002


Root MSE 3.276 3.276 2.342 2.342 0.125 0.125 5.899 5.897


Source: Authors’ calculations.


Notes: Panel 1 dependent variables: Growth of performance indicators from 2003/2004 to 2005/2006. Top
and bottom 1 per cent of distribution trimmed in estimation. Panel 2 dependent variables: Growth from
2003/2004 to 2007/2008. Top and bottom 1 per cent of distribution trimmed in estimation.






We observe similar direct effects when considering employment growth (column 3, panel 1).


When profit rate is considered (column 5, panel 1), we observe that non-exporting firms making goods


other than textiles and apparel had significantly lower profit rates (α). Exporting firms on average also


had significantly lower profit rates (β), but exporting firms selling ATC goods had significantly higher


profit rates over this period (γ). Exporting to the European Union had a positive, though not significant,




14 This should be considered in conjunction with β. Given that exporters overall had significantly lower sales growth, δ tells
us that enterprises exporting to the markets of the United States and the European Union did significantly better than
other exporters, but still not well. The point estimate for exporters to the United States and the European Union is δ + β.





16 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


edge over other exporters (δ). For firm productivity, none of these direct effects is significantly different


from zero.




When we consider the complete specification of equation (1) new features appear. The


intercept α and coefficient γ are largely unchanged by this extension of the estimation equation; both


remain significantly different from zero. However, the coefficients β and δ become insignificantly


different from zero. The sign of β remains negative, while the sign of δ switches from positive to


negative. The coefficient ε provides a measure of the difference-in-difference effect: contingent on


being an exporter, did a firm producing textiles and apparel subject to quota gain more or less during


this period? The answer is “less”, as theory would predict when exporting firms faced increased


competition in the export market. The coefficient η provides a measure of another difference-in-


difference effect: contingent on being an exporter to the United States or the European Union, did


firms selling textiles and apparel products subject to quota in 2003/2004 tend to have better or worse


performance? The answer is “better”, though the results are not statistically significant. The point


estimates suggest that sales of exporters to the United States or the European Union grew at the rate


of non-exporters – the effect specific to these exporters offsets nearly completely the exporter effect ε.




We note in comparing the first and second columns of panel one that including the difference-


in-difference terms leads to a decomposition of the direct-effect coefficients reported in the first


column. The coefficient β in column one, for example, is equal to the sum of the coefficients β + ε from


column two. The coefficient δ in column one is equal to the sum of the coefficients δ + η in column


two.




When the complete specification for employment growth is considered (column 4, panel 1), we


observe that the coefficient β is divided into a component attributable to non-ATC goods (-0.16) and


the component attributable to ATC goods (-0.21). Those firms exporting into the market protected by


quotas will face relatively more competition and will reduce employment by relatively more than a


comparator firm in a market not liberalized in this way. The complete specification of the profit rate


(column 6, panel 1) has similar significant coefficients for α and γ to the direct-effect specification


(column 5, panel 1). The complete specification reveals, though, that the negative return to exporters in


the direct-effect specification is attributed exclusively to ATC products – when the two are separated,


the coefficient for non-ATC goods (β) becomes positive while the coefficient for ATC goods (ε) is


negative.




The results of panel 1 refer to performance over the short run – for the years 2005 and 2006


after quota elimination at the beginning of 2005. In panel 2 we examine the performance from the


2003/2004 average through the years 2007 and 2008. Comparison of column 1 in panel 2 to column 1


in panel 1 illustrates that the two periods have very similar characteristics. The coefficients in panel 2


are larger than in panel 1, as expected when we examine cumulative growth over a longer period. They


take the same sign and (for the most part) significance: exporting enterprises had significantly slower


sales growth, while enterprises producing ATC goods had more rapid (but statistically insignificant)


sales growth. Enterprises exporting to the United States and the European Union grew significantly


more rapidly. When the complete specification is considered (column 2, panel 2), the patterns of panel


1 are duplicated. It becomes clear that sales growth over this longer horizon for exporters was


significantly better for ATC products (η) than for non-ATC products (δ). The results in panel 2 for


employment growth and profit rate are qualitatively similar to those of panel 1.






6.2. PROBIT ESTIMATION


In table 12 we report the mean and median values of the four performance variables by exit


year. (We report both the mean/median and the number of enterprises observed for each


year/performance variable combination.) If we examine the median values of these, we observe that


median total factor productivity (TFP) was rising for all surviving enterprises in 2005–2007, with a fall in


2008. A breakdown by NACE category illustrates that for NACE 17 (textiles) there was a rise in median







Turkish Enterprise-Level Response to Foreign Trade Liberalization 17


TFP through 2006 and a reduction thereafter. For NACE 18 (apparel) the median TFP rose through


2007, but saw a downturn thereafter. For NACE 19 (other products) the TFP rose for each year. The


downturn in TFP is thus attributable to the downturn in textiles and apparel production. (The numbers


for the other variables – sales, employment, profit rate – were not consistent across years, and suggest


that we should recalculate what we have here.)






Table 12


Performance and survival




Last Mean Median


year Productivity Sales Employment Profit Productivity Sales Employment Profit


2004 6.5 7.46 17.06 0.1 5.3 7.69 6 0.14


1 109 3 263 3 263 3 263 1 109 3 263 3 263 3 263


2005 6.58 9.17 65.6 0.173 5.44 9.4 26 0.175


243 542 542 542 243 542 542 542


2006 7.96 9.54 55.04 0.139 5.84 9.41 19 0.14


568 974 974 974 568 974 974 974


2007 7.26 6.948 74.49 54.13 5.78 1.893 36 3.4


1 540 2 633 2 633 1 615 1 540 2 633 2 633 1 615


2008 6.92 2.53 132.32 48.88 5.66 4.736 48 3.33


41 910 55 819 55 819 42 993 41 910 55 819 55 819 42 993


Source: Authors’ calculations.


Note: The year column refers to the last year of activity of firms considered in computations.






In table 13, we report the result of probit regressions qualifying the survival of enterprises from


the previous year to the year given at the top of the column. Top panel results are derived from probit


estimation, while bottom panel results are derived from probit estimation with random effects. The data


sample includes all manufacturing enterprises in continuous operation from 2003 to the year prior to


the year indicated in each column of the table: for example, in the column indicating “2005” we


consider all enterprises operating in both 2003 and 2004. We define the dependent variable as a binary


variable, with value 1 indicating continuing operation in that year and value 0 indicating that the


enterprise was not observed in that year or in subsequent years. The explanatory variables used


include the logarithm of sales revenue in the previous year, the logarithm of employment in the


previous year, a binary variable indicating that the enterprise had more than one plant in the previous


year and a binary variable indicating that the enterprise was an exporter in the previous year. We also


include the measure of total factor productivity derived for each enterprise and a binary variable


indicating whether the enterprise was primarily a producer of the textile and apparel products


liberalized in 2005 (group 4 producers).15




The random-effect probit estimation controls for systematic differences across industries


(defined in a disaggregated fashion with over 200 groups) in the probability of survival. If we focus


upon those regressions (bottom half of the table), we observe that the scale of the enterprise’s


operation (as measured by sales revenue) is a significant and positive indicator of survival. The


employment of the enterprise, by contrast, is an irregular additional indicator – positively and


significantly associated in 2005 and 2007, but negatively and insignificantly associated with survival in


2006 and 2008. Multi-plant enterprises were significantly more likely to survive in 2006 and 2008, but


not in the other years. The total factor productivity of the enterprise was negatively and significantly




15 In table 14, we report the results of probit regressions similar to those of table 13 but excluding the productivity and
group 4 regressors. As is evident, the remaining coefficients are similar across the two sets of regressions, but the number
of observations in table 14 is larger (since it is no longer necessary to have adjacent observations to estimate the
productivity term).





18 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


associated with survival of the enterprise in each year – a surprising finding. Exporting enterprises were


ceteris paribus significantly more likely to survive. Finally, enterprises producing textiles and apparel


goods liberalized in 2005 were less likely to survive, other things equal, in each of the years 2005


through 2008. These effects are significantly different from zero in 2005 and 2008.16






Table 13


Duration analysis for Turkish enterprises



Dependent variable:
Still producing in


2005 2006 2007 2008


Probit


Constant -2.242 0.131 -1.054 0.147 -1.573 0.140 -2.280 0.309


ln(sales in previous year) 0.163 0.018 0.185 0.020 0.167 0.019 0.207 0.021


Ln(employment in previous year) 0.340 0.026 -0.018 0.029 0.144 0.028 -0.009 0.030


Multi-plant enterprise
in previous year


-0.073 0.033 0.129 0.037 0.055 0.035 0.550 0.141


Exporter in previous year 0.241 0.036 0.224 0.039 0.173 0.037 0.197 0.040


Group 4 producer -0.225 0.047 -0.127 0.049 -0.170 0.045 -0.172 0.049


Productivity -0.016 0.007 -0.029 0.008 -0.029 0.007 -0.048 0.007


N 8 738 7 998 8 068 7 178


R2 0.16 0.04 0.07 0.06


log likelihood
-4


206.83
-3 488.58 -3 950.2 -3 427.47


Probit with random effects


Constant -2.293 0.141 -1.106 0.151 -1.594 0.150 -2.339 0.315


ln(sales in previous year) 0.161 0.020 0.182 0.021 0.164 0.021 0.212 0.022


Ln(employment in previous year) 0.352 0.028 -0.012 0.030 0.156 0.030 -0.008 0.031


Multi-plant enterprise
in previous year


-0.043 0.035 0.132 0.038 0.064 0.036 0.551 0.142


Exporter in previous year 0.226 0.040 0.203 0.041 0.138 0.040 0.187 0.042


Group 4 producer -0.181 0.054 -0.075 0.055 -0.094 0.054 -0.150 0.054


Productivity -0.013 0.008 -0.026 0.008 -0.026 0.008 -0.046 0.008


N 8 738 7 998 8 068 7 178


log likelihood
-4


190.81
-3 484.93 -3 939.53 -3 426.09


Number groups 224 220 222 219


Source: Authors’ calculations.


Notes: The data sample in each case is of all enterprises in operation the year prior to the year listed and all
preceding sample years. The first panel reports the results of a probit analysis. The second panel reports the
results of random-effect (RE) probit analysis. The random effects are defined by productive activity group.






We anticipated that once we control for the size, productivity and exporter status of the


enterprise, we find that enterprises producing ATC goods will be more likely to exit in the years


following the elimination of quotas. As competition is increased in the export markets for the ATC


goods, the weaker performers will be driven from the market. The results of Table 13 reflect this; so


also may the results of Table 12 in indicating that productivity has increased over time. This could have


been the consequence of the elimination of low-productivity enterprises through market exit.






16 This effect persists despite the use of random effects conditioned on the productive activity of the enterprise. We
expect that this random-effect term will capture some of the liberalization effect, given that the goods liberalized were
concentrated in a few activity categories. When the probit is run without random effects, as in the top panel of table 13,
the negative liberalization effects for group 4 producers are negative and significant in all years.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 19


Table 14


Duration analysis for Turkish enterprises (restricted)



Dependent variable:
Still producing in


2005 2006 2007 2008


Probit


Constant -2.299 0.056 -1.81 0.059 -1.466 0.044 -2.914 0.132


ln(sales in previous year) 0.031 0.009 0.054 0.009 -0.001 0.007 0.092 0.008


Ln(employment in previous year) 0.872 0.015 0.604 0.015 0.654 0.012 0.553 0.012


Multi-plant enterprise
in previous year


-0.228 0.025 0.173 0.029 0.023 0.025 0.512 0.062


Exporter in previous year 0.228 0.025 0.025 0.025 0.228 0.022 0.322 0.025


N 30 709 25 484 34 130 33 348


R2 0.54 0.37 0.41 0.46


log likelihood
-9


819.369
-8 853.17 -13 391.53 -11 883.9


Probit with random effects


Constant -2.35 0.073 -1.656 0.073 -1.343 0.058 -2.765 0.139


ln(sales in previous year) 0.018 0.01 0.017 0.011 -0.013 0.008 0.067 0.009


Ln(employment in previous year) 0.952 0.017 0.707 0.017 0.692 0.013 0.628 0.014


Multi-plant enterprise
in previous year


-0.172 0.026 0.193 0.03 0.048 0.026 0.563 0.063


Exporter in previous year 0.22 0.026 0.247 0.073 0.213 0.023 0.304 0.026


N 30 709 25 484 34 130 33 348


log likelihood -9 380.29 -8 595.02 -13 066.73 -11 474.83


Number groups 234 233 233 233


Source: Authors’ calculations.


Notes: The data sample in each case is of all enterprises in operation the year prior to the year listed and all
preceding sample years. The first panel reports the results of a probit analysis. The second panel reports the
results of random-effect (RE) probit analysis. The random effects are defined by productive activity group.










20 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


7. CONCLUSIONS AND EXTENSIONS




Liberalizing international trade creates both opportunities and threats for productive


enterprises. The opportunities stem from the opening of new markets for its products; the threats stem


from the entry of competitors to contest the liberalized market. The elimination of textiles and apparel


quotas in the Canada, the United States of America and the European Union in the period 1995–2005


created opportunities and threats for textiles and apparel exporters worldwide.




Turkey as an economy responded well to this opportunity, as measured by positive growth in


export value to the countries eliminating quota. We decompose this success in this paper by observing


the changes in behaviour at the enterprise level.




When we divide the enterprises into those producing textiles and apparel subject to quota in


2004 (i.e., ATC goods) and others, we find that those enterprises producing ATC goods grew more


rapidly in terms of real sales revenue and employment – and they also had significantly higher profit


rates on average. When we divide the enterprises into exporting versus non-exporting ones, we find


that the non-exporting enterprises actually grew faster in both real sales and employment during the


period associated with quota elimination. In sum, the textiles and apparel industries were sources of


faster economic growth. However, the enterprises involved in exporting grew more slowly than those


involved in exclusively domestic sales.




We are also able to draw conclusions on the enterprise-level performance of exporters. For


exporters, we found that those that specialized in textiles and apparel during this period had lower


sales revenue and employment growth and a lower profit rate on average than those selling other


products. Those that exported to the United States and the European Union had relatively higher sales


revenue and employment growth, and a higher profit rate, on average than those exporting only to


other markets.




When we consider the motivations for enterprises to cease operations, we consider a number


of alternative explanations. We find that the larger enterprises are more likely to survive from one year


to the next, whether size is measured by the value of sales, of employment, or of the number of plants.


Independently of these factors, enterprises that had been exporters were significantly more likely to


survive from one year to the next during this period. However, this latter effect is on average


counteracted by the effect of being a producer of ATC goods: ATC producers were significantly more


likely to fail during this period.




We conclude from this that Turkey’s success in the export of textiles and apparel must be


taken in context. While Turkish enterprises were more successful than most in adapting to the post-


quota market in textiles and apparel, their performance paled relative to the performance of enterprises


in non-ATC areas. Exports to markets other than Canada, the United States and the European Union


for ATC products were reduced, while production of non-ATC products, both for export and domestic


use, grew rapidly. Enterprises then “voted with their feet” – significantly more enterprises in ATC


industries closed during this period than in other industries.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 21


REFERENCES



Anderson JE (1988). The Relative Inefficiency of Quotas. Cambridge, MA: MIT Press.

Bhagwati J (1965). On the Equivalence of Tariffs and Quotas. In Baldwin RE et al., eds. Trade, Growth


and the Balance of Payments: Essays in Honor of Gottfried Haberler. Chicago: Rand
McNally.



Bernhofen D, Upward R and Wang Z (2011). Quota as a Sorting Machine: Quantity Restrictions and


Price Adjustments of Chinese Textile Exports to the United States, mimeo, Nottingham
University.



Caves R (1998). Industrial Organization and New Findings on the Turnover and Mobility of Firms.


Journal of Economic Literature. 36(4):1947–1982.

Conway P and Fugazza M (2010). Taking in One Another’s Clothes? The Impact of Removal of ATC


Quotas on International Trade in Textiles and Apparel. UNCTAD Policy Issues in International
Trade Series No. 45.



Barrows G and Harrigan J (2009). Testing the Theory of Trade Policy: Evidence from the Abrupt End of


the Multifiber Arrangement. The Review of Economics and Statistics. 91(2):282–294.

Brambilla I, Khandelwal AK and Schott PK (2010). China's Experience under the Multi-Fiber


Arrangement (MFA) and the Agreement on Textiles and Clothing (ATC). In: Feenstra RC and
Wei S-J, eds. China's Growing Role in World Trade. National Bureau of Economic Research.



Edwars L and Sundaram A (2012). Trade Liberalization and Reallocation of Production an Analysis of


Indian Manufacturing. Unpublished manuscript. University of Cape Town.

Falvey RE (1979). The Composition of Trade within Import-restricted Product Categories. Journal of


Political Economy. 87(5):1105–1114.

Goldberg PK, Khandelwal A, Pavcnik N and Topalova P (2010). Multi-Product Firms and Product


Turnover in the Developing World: Evidence from India. Review of Economics and Statistics.
92(4):1042–1049.



Khandelwal A, Schott PK and Wei S-J (2011). Trade Liberalization and Embedded Institutional Reform:


Evidence from Chinese Exporters. Mimeo. Columbia Business School, United States of
America.



Melitz MJ (2003). The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry


Productivity. Econometrica. 71(6):1695–1725.

Pavcnik N (2002). Trade Liberalization, Exit, and Productivity Improvement: Evidence from Chilean


Plants. Review of Economic Studies. 69(1):245–276.

Tybout J (2000). Manufacturing firms in developing countries: How well do they do and why? Journal of


Economic Literature. 38:11–44.

Tybout J (2003). Plant and Firm-Level Evidence on the "New" Trade Theories. In Choi EK and Harrigan


J, eds. Handbook of International Trade. Blackwell: Malden, MA. 388-415. 28










Turkish Enterprise-Level Response to Foreign Trade Liberalization 23


APPENDIX A: TURKISH FIRM-LEVEL SURVEYS


The Turkish Statistical Institute has two classes of enterprises. The larger enterprises (i.e.,


those with more than 20 employees or those in selected economic sectors) are surveyed every year.


The smaller enterprises are surveyed sporadically; when surveyed, they represent themselves and


those of similar size that have been excluded for the year. For these smaller enterprises, their


responses are scaled up in estimating the gross domestic product to reflect this role of representation.


TUIK defines a weight for each enterprise between 1 (for the large firms) and a very large positive


number (for the firms representing a great number of similarly small non-surveyed enterprises).






A1. The Enterprise Survey


The Turkish Statistical Institute (with Turkish acronym TUIK) conducts a large-scale survey


annually. This survey is conducted at the level of the enterprise, with multi-plant enterprises


aggregating information for all plants.17 The survey includes questions on the enterprise’s


characteristics, its uses of inputs, the value of its sales, and whether it is involved in export activity. In


recent years it has tabulated responses from more than 80 000 firms.




Table A.1 reports the aggregates for number of enterprises, sales by enterprises and


employees of enterprises. The number of enterprises surveyed has stayed roughly constant through


the years, with the sharp drop in 2005 indicating a simple reduction in the number of enterprises


surveyed rather than a fall in total Turkish economic activity.18 The total nominal value of sales by


surveyed enterprises rose rapidly during this period. Some of this rise was due to inflation, but a large


part of this was due to the economic growth that Turkey experienced.19




Table A.1


Enterprise Survey: Totals from respondents




Total enterprises
Total sales


(in billions of TL)
Total employees


(in millions)


2003 80 213 171.17 2.56


2004 78 399 238.61 3.12


2005 63 211 269.66 3.87


2006 84 925 334.70 4.28


2007 83 844 365.98 4.50


2008 82 496 418.47 4.58


Source: Enterprise Survey database.


Note: TL, Turkish lira.




Sectoral breakdown. When enterprises are grouped by the category of their most important


product or service, it is clear that light manufactures represent the near totality of sales revenues in the


surveys, while other categories (agriculture, some services) make minor contributions to sales but more


important contributions to employment. In table A.2, enterprises are grouped by the two-digit NACE


category of their final product. We report information from 2003 and 2008 for purposes of comparison,


and list the categories by share of sales revenue in 2003.




17 Prior to 2003, the Survey was conducted at the plant level, and it is thus difficult to compare pre-2003 responses to
those of 2003 and later.


18 There was a reduction in 2005 in the number of enterprises invited automatically to participate in the survey. This
change could have caused the reduction in total respondents.


19 Annual inflation in the consumer price index as reported by the IMF was (in terms of per cent) 25.3 (2003), 8.6 (2004),
8.2 (2005), 9.6 (2006), 8.8 (2007) and 10.4 (2008) for these years.





24 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


There are 46 two-digit categories for which at least some enterprises report economic


activity.20 The categories range from 10 (mining) to 92 (other service activities). We report only the top


20 categories when ranked by sales in 2003. The entries in the table are the per cent of the total for


the year contributed by that category. Concentration ratios for the top 5, 10 and 20 categories are


reported at the bottom of the table.




Table A.2


NACE categories statistics


(Share in total)




NACE category Sales revenues Employees Enterprises


2003 2008 2003 2008 2003 2008


15 15.0 13.8 6.4 4.9 4.8 4.0


17 12.6 7.9 10.7 6.7 4.6 4.5


24 9.9 6.0 2.5 1.6 1.2 1.2


27 8.9 14.6 2.4 2.1 0.9 1.5


34 8.6 8.8 2.6 2.6 0.9 1.1


18 7.7 5.2 8.1 6.1 5.6 5.5


23 5.4 6.8 0.2 0.1 0.1 0.1


26 5.2 5.6 3.3 3.3 2.3 2.6


29 5.2 5.9 3.3 3.3 2.9 3.5


25 3.6 4.4 1.9 2.1 2.1 2.2


28 2.7 4.0 2.3 2.6 3.7 4.6


32 2.4 1.3 0.6 0.4 0.1 0.2


36 2.2 2.6 2.0 2.0 3.4 3.3


31 2.2 3.5 1.3 1.5 0.9 1.0


21 1.8 1.7 0.9 0.7 0.6 0.7


16 1.4 0.8 0.9 0.4 0.0 0.0


22 1.1 1.1 0.7 0.7 1.3 1.4


19 1.0 0.7 0.9 0.7 1.1 0.9


51 0.9 1.6 5.9 5.8 10.1 9.7


20 0.7 1.0 0.5 0.4 1.6 1.3


Top five categories 55.0 51.1 24.6 17.9 12.4 12.3


Top ten categories 82.2 79.1 41.4 32.7 25.4 26.2


Source: Enterprise Survey database.






Beginning with the top five categories in the table (all manufactured products, with 15: food


products, 17: textiles, 24: chemicals, 27: basic metals, 34: motor vehicles), we observe that these five


categories alone represent (in 2003) 55 per cent of total sales revenue as reported in the sample, but


only 25 per cent of employment and 12 per cent of the firms surveyed. When the top 20 categories are


considered, 48 per cent of the enterprises surveyed report nearly 99 per cent of the sales revenue and


57 per cent of the employment. By 2008, 49 per cent of the firms report themselves to be in these 20


categories; these report 97 per cent of the sales revenue but only 48 per cent of the employment.




Perusal of the individual entries for these top categories leads to two additional conclusions


about those responding to the survey:




20 The two-digit NACE categories used by TUIK correspond in most regards to the categories of the International
Standard Industrial Classification (ISIC). In particular, the categories from 10 to 14 represent mining, the categories from
15 to 37 represent various manufacturing activities, and the higher categories represent retail, wholesale and service
(including public service) activities. The categories 17 and 18 are associated with textiles and apparel manufacture,
respectively. NACE and ISIC differ in their third and higher digits.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 25


• The upper 10 categories in the table have the dual characteristic of representing a higher


percentage of both sales and employees than they do of total enterprises: these are sectors in


which a new enterprise generates greater-than-average sales and employment growth. This is


not true of all categories listed. For example, category 51 includes all enterprises reporting


their main activity to be purchase and sale of waste and scrap. These are about 10 per cent of


all enterprises surveyed in 2003 and 2008, and they employ nearly 6 per cent of the workers


employed in the surveyed enterprises. They contribute only about 1 per cent of the total sales


revenue in the survey.




• The manufacturing categories are those between 15 and 37, inclusive. As is evident from the


table, the top 20 categories in terms of sales revenue are all manufacturing except for category


51 mentioned above.




In much of the analysis to follow, we will focus upon manufacturing enterprises. In table A.3


we report the summary statistics for just those enterprises, and as a percentage of the totals for the


entire Enterprise Survey.




Table A.3


Enterprise Survey: Totals for manufacturing



Enterprises Sales Employees


Number Share
in total


Billions
(TL)


Share
in total


Millions


Share
in total



2003 27 138 33.8 169.17 98.8 1.34 52.3


2004 30 705 39.2 235.33 98.6 1.59 50.9


2005 25 483 40.3 264.99 98.3 1.79 46.2


2006 34 122 40.2 327.98 98.0 1.91 44.6


2007 33 345 39.8 357.15 97.6 1.97 43.7


2008 33 777 40.9 409.82 97.9 1.99 43.5


Source: Enterprise Survey database.


Note: TL, Turkish lira.






In 2003, for example, we observe that manufacturing enterprises represented about one-third


of the total number of enterprises surveyed. The reported sales of manufacturing enterprises were


nearly 99 per cent of the total sales reported by all enterprises. Manufacturing enterprises employed


over half of the workers reported employed by the firms surveyed. From 2003 to 2008, these


percentages evolved: the share of manufacturing enterprises among all surveyed rose to over 40 per


cent while the share of total sales by these enterprises dropped slightly to nearly 98 per cent. The


share of manufacturing employees in total employees covered by the survey dropped to just over 43


per cent.




We can also break down the manufacturing enterprises surveyed into groups by their final


product. After doing so by three-digit NACE code and ranking the resulting 100 groups by sales


revenue in 2003, we report characteristics of the top 20 manufacturing categories in table A.4. The top


categories are not surprising given what we discovered in table 2: apparel, textiles, motor vehicles, iron


and steel and chemical products dominate the list.21 The breakdown reported in table A2.4, however,


indicates large differences among sectors.






21 The top ten three-digit NACE categories are 182, apparel; 341, motor vehicles; 271, basic iron and steel; 232, refined
petroleum products; 158, other food products; 171, spinning and weaving of textiles; 172, manufacture of other textiles;
244, pharmaceuticals ; 241, basic chemicals; 297, domestic appliances.






26 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


• Apparel (182) was the top-ranked manufacturing sector in 2003 by sales revenue. It is so in


large part because of the large number of enterprises operating in that sector. Given that 12.6


per cent of all manufacturing enterprises reported to be apparel producers, it is a bit surprising


that only 7.3 per cent of all manufacturing sales revenue came from the sector.




• Motor vehicles (341), iron and steel (271) and petroleum products (232) are the next three


ranked sectors in manufacturing in 2003 (and the top three in 2008), but they are so with a


miniscule share of the enterprises surveyed. The three sectors together represent only 0.7 per


cent of the enterprises and 5.0 per cent of the employees in 2003, but they report 18.1 per


cent of the sales revenue. This is due to their characteristics of much larger than average size


and relative capital intensity – they are less labour-using than average. (Apparel, by contrast,


has 15 per cent of the employees but only 7 per cent of the sales revenue.)




• The category food products (158) is similar to apparel in having a relatively large share (7.0 per


cent) of the manufacturing enterprises surveyed and a relatively smaller share of sales


revenues (4.7 per cent). It is not as labour-using as apparel.




• Within manufacturing, the distribution of sales revenue is not as concentrated as was


observed in table 1.2 for all surveyed enterprises: the top 5 categories have 30 per cent of the


total, the top 10 have 46 per cent and the top 20 have 66 per cent (as compared to 99 per cent


above). The distribution of employees is roughly equally concentrated while the distribution of


enterprises is more concentrated at the top 5 level but less concentrated at the top 20 level.


(There are roughly twice as many categories in the manufacturing breakdown as in the


breakdown of table 1.2.)




Table A.4


NACE categories statistics: Manufacturing




NACE category Sales revenues Employees Enterprises


2003 2008 2003 2008 2003 2008


182 7.3 5.0 14.7 13.3 12.6 12.7


341 6.4 6.2 2.2 2.3 0.1 0.1


271 6.2 10.5 2.4 2.2 0.4 0.6


232 5.5 7.0 0.4 0.3 0.2 0.2


158 4.7 3.8 5.4 4.8 7.0 4.4


171 3.9 2.1 5.1 3.8 1.9 1.8


172 3.5 1.7 5.4 3.0 2.3 1.8


244 3.2 1.5 1.8 1.3 0.4 0.4


241 2.8 1.9 1.1 0.7 0.8 0.7


297 2.8 2.6 2.0 2.0 1.2 1.0


252 2.6 3.4 2.8 3.6 4.5 4.5


153 2.4 2.2 2.1 1.9 1.2 1.3


343 2.2 2.5 2.6 3.3 1.7 2.1


154 2.1 1.5 0.7 0.5 0.6 0.5


323 2.0 1.0 0.7 0.5 0.2 0.2


151 1.8 1.5 1.5 1.4 0.7 0.7


174 1.7 1.3 3.1 2.5 1.8 1.8


156 1.6 1.5 0.7 0.6 1.6 1.1


245 1.6 1.0 0.7 0.6 0.7 0.6


265 1.5 1.6 0.7 0.8 0.3 0.3


Top five categories 30.0 32.4 25.1 22.9 20.3 18.0


Top ten categories 46.3 42.2 40.5 33.7 27.0 23.5


Top twenty categories 65.7 59.8 55.9 49.3 40.3 36.5


Source: Enterprise Survey database.








Turkish Enterprise-Level Response to Foreign Trade Liberalization 27


Characteristics of transition of enterprise inclusion across years. Not all enterprises are


surveyed in every year. For larger enterprises, the survey is an annual event, while for smaller


enterprises the survey may be an event conducted once every two years or once every ten years.


There is a unique identifier for each enterprise, and thus these survey responses can be linked across


time. Using this identifier, we can identify those enterprises that are entering, continuing, exiting and


re-entering the panel.




Table A.5


Firms’ sample composition: Manufacturing


(Number of firms)




2003 2004 2005 2006 2007 2008 Total


Exiting 13 100 15 725 6 103 13 442 12 724


Continuing 14 037 14 984 19 381 20 687 20 625


Entering 27 137 16 672 9 654 12 702 11 043 11 029


Re-entering 846 2 046 1 619 2 127




Total 27 137 30 709 25 484 34 129 33 349 33 781 184 589




Observed only in that year 10 969 10 085 4 089 8 939 9 076 11 029


Firms always present 8 022


Source: Enterprise Survey database.






The panel is created by merging the survey observations for each year into a dataset with a


time dimension. An entering firm is one that is observed for the first time in that year. An exiting firm is


one that is observed in the database in that year, but not in the following year. A continuing firm is one


observed both the previous year and the current year. A re-entering firm is one that was not observed


in the previous year, but was observed at some earlier time period in the database. Table A.5 reports


the count of manufacturing enterprises in each of these categories in each of these years for the entire


survey. The final two rows of the table report the number of enterprises observed only in that one year,


and the number of enterprises observed in every year from 2003 to 2008 inclusive. (The difference


between the totals of table A.3 and table A.5 is due to observations in which the same enterprise


number is assigned to two observations in the same year. In that case, both are excluded in table A.5


but both are included in table A.3.)




Table A.6


Textiles and apparel enterprises as a share of all manufacturing enterprises




Year Number of enterprises Total sales revenue Number of employees


2003 0.25 0.21 0.36


2004 0.25 0.18 0.35


2005 0.28 0.17 0.34


2006 0.26 0.16 0.32


2007 0.26 0.16 0.31


2008 0.25 0.13 0.29


Source: Enterprise Survey database.





28 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


Textiles and apparel within the Turkish manufacturing sector. Textiles and apparel


enterprises are well represented among the enterprises responding to the Enterprise Survey. Table A.6


reports the share of textiles and apparel enterprises among all manufacturing enterprises. As is evident


from the first column, textiles and apparel enterprises represent about one-quarter of the total number


of respondents. This share rose slightly in 2005, but then fell back to the pre-liberalization percentage


by 2008. By contrast, textiles and apparel enterprises had been declining as a group in terms of both


share of total sales revenue and share of total employees for all survey respondents. Table A.7 reports


the pattern of transition for textiles and apparel enterprises. We observe that the number of enterprises


rose in each year, but we also observe that this was due to increased continuation of enterprises.


Entrance of new enterprises declined slightly during the years 2005–2008, in contrast to the behaviour


of manufacturing enterprises as a whole.






Table A.7


Firms’ sample composition: Textiles and apparel


(Number of firms)




2003 2004 2005 2006 2007 2008


Exiting 2 735 3 053 1 324 2 841 2 903


Continuing 3 988 4 608 5 757 5 931 5 861


Entering 6 724 3 673 2 297 2 164 2 445 2 127


Re-entering 176 291 388 330




Total 6 724 7 661 7 081 8 212 8 764 8 318




Observed only in that year 2 302 1 926 803 1 836 1 947 2 127


Firms available in every year 2 329


Source: Enterprise Survey database.






A2. The Foreign Trade database


The Enterprise Survey includes two questions on foreign trade. Enterprises are asked to


provide the Turkish lira value of all exports and (separately) all imports of the enterprise during the year.


While these are available, more detailed data about enterprise foreign trade activity is available through


the foreign trade statistics collected by the Customs Department. This is a separate database (we will


refer to it as the TUIK foreign trade (FT) database), but it includes the unique enterprise identification


code that allows merging of the two files.




The Foreign Trade database identifies specific transactions in goods, either export or import,


undertaken by Turkish enterprises. Table A.8 provides summary statistics for international trade in


manufactured goods. There were a large number of transactions – 426,954 in 2003, rising to 619,579 in


2008. For a given enterprise, there can be many transactions in each year – each destination country


(for exports) or source country (for imports) has a separate transaction, as does every different


classification of traded good. The traded good is classified both by 4-digit ISIC code and by the 12-


digit HS code associated with trade-balance accounting. In table A.8, the ISIC code is used to identify


manufactured goods.22






22 Note that table A.8 does not necessarily describe trade by manufacturing enterprises. While manufacturing enterprises
could be exporters of manufactured goods, so also could trading companies that serve as intermediaries for the purposes
of international trade. Importers of manufactured goods could be manufacturing enterprises, but they could also be
enterprises in agriculture, resource extraction or services industries.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 29


Table A.8


Foreign trade in manufactured goods


(Millions of United States dollars)



Year Transactions Export value Import value


2003 426 954 22 849.6 36 320.5


2004 497 007 32 654.0 53 516.0


2005 526 282 38 131.7 61 968.4


2006 582 534 46 355.7 74 142.4


2007 600 065 57 614.0 87 354.6


2008 619 579 71 302.4 106 178.4


Source: Foreign Trade database.




In table A.9, we limit our consideration to exported goods, but we provide a detailed


breakdown by two-digit NACE code of the exported products. There is a great variety in export


performance by group of good, but we can identify a few salient trends.






Table A.9


Exports in manufactured goods by sector


(Millions of United States dollars)




2003 2004 2005 2006 2007 2008


15 1 972.1 2 899.3 3 969.1 3 744.5 4 088.5 5 184.7


16 301.9 458.6 572.1 620.8 580.4 716.1


17 2 946.7 3 602.5 3 559.5 3 858.8 4 332.7 4 255.0


18 2 118.9 2 543.3 2 514.0 2 998.6 3 540.7 3 477.2


19 147.7 186.0 186.1 209.5 247.5 236.5


20 74.6 114.4 160.2 186.3 283.0 347.9


21 181.9 234.5 298.4 306.7 463.5 618.9


22 63.4 72.6 85.4 82.8 120.9 138.1


23 757.3 948.5 1 916.3 3 127.4 3 340.4 4 506.8


24 1 276.2 2 004.3 1 481.5 1 842.1 2 014.9 2 568.1


25 953.5 1 261.6 1 610.9 1 962.8 2 542.7 3 164.2


26 643.2 845.1 958.4 1 007.4 1 230.4 1 721.1


27 1 476.6 3 070.4 3 819.6 5 245.6 6 814.9 11 216.0


28 590.3 900.0 1 120.3 1 358.2 1 879.2 2 467.4


29 1 931.6 2 186.3 2 639.5 3 277.3 4 412.2 5 382.0


30 6.0 8.8 4.9 2.8 6.6 6.0


31 759.3 1 089.0 1 281.7 1 860.7 2 616.1 3 294.8


32 428.2 485.8 547.5 515.7 445.4 417.5


33 42.7 52.4 68.1 85.2 121.3 132.8


34 5 135.1 8 292.4 9 539.8 11 846.3 15 753.8 17 983.8


35 271.3 413.5 601.3 1 004.4 1 419.4 1 977.8


36 771.1 981.4 1 194.4 1 180.4 1 342.4 1 486.5


37 0.1 3.6 2.7 31.2 17.0 3.0



Total 22.8 32.7 38.1 46.4 57.6 71.3


Source: Foreign Trade database.





30 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


• In 2003, the top three product groups for exports were motor vehicles (34), textiles (17) and


apparel (18). By 2008, motor vehicles remained the top category but iron and steel (27),


appliances (29), food products (15) and petroleum products (23) had passed textile and


apparel in terms of export value.




• Textiles and apparel exports declined in value in 2008, but this was more than made up by


growth in other sectors. Exports grew by 24 per cent in terms of United States dollars from


2007 to 2008.




We gain another perspective on trade in textiles and apparel by breaking down the value of


exports and imports in each year into the part associated with textiles and apparel and the part


associated with all other trade. Table A.10 illustrates that Turkey was in fact nearly in balance with


trade in textiles and apparel; both exports and imports rose through 2007 but then declined in 2008,


and they were quite similar in magnitude. This is not true for the “other” category – both exports and


imports grew more rapidly during this period, but imports outstripped exports. In the “other” category,


manufactures exports were only 64 per cent of manufactures imports in 2008.






Table A.10


Trade in textiles and apparel


(Billions of United States dollars)




Exports Imports


Textiles/apparel Other Total Textiles/apparel Other Total


2003 5.1 17.8 22.8 5.5 30.8 36.3


2004 6.1 26.5 32.7 6.3 47.3 53.5


2005 6.1 32.1 38.1 6.3 55.7 62.0


2006 6.9 39.5 46.4 6.3 67.8 74.1


2007 7.9 49.7 57.6 7.7 79.6 87.4


2008 7.7 63.6 71.3 7.1 99.1 106.2


Source: Foreign Trade database.




The FT database is assembled from individual international transactions, and as such is a mix


of import and export transactions. When the database is concentrated to examine the individual


enterprises involved in international trade, the results are as given in table A.11.




There are more enterprises registering import transactions than export transactions. If we


examine the top panel of table A.11, it is evident that “Importer” only is larger as a category than


“Exporter” only. When the enterprises reporting “Both” are included, 61,425 of the 85,911 enterprises


included in the FT database in 2008 were importers, while 48,241 were exporters. Those enterprises


involved in both importing and exporting make up about 27 per cent of the total in all these years.




In the second panel of table A.11, the transactions are limited to those involving manufactured


goods. As is evident in comparing the two panels, this does not exclude many enterprises – nearly all


enterprises reporting international transactions in this period were trading in manufactured goods.




A3. Merging the FT and ES databases


When the FT and ES databases are combined, they include information on a large number of


enterprises: 102,253 in 2003, rising to 116,705 in 2008. Table A.12 provides statistics on this sample


of enterprises, and demonstrates the large degree of non-overlap in the two samples. Figure A.1


illustrates this non-overlap in a Venn diagram drawn from the 2008 observations in the top panel of


table A.12.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 31


Table A.11


Exporters and importers in the FT database




2003 2004 2005 2006 2007 2008



All products 63 968 69 483 74 428 78 134 87 751 85 911


Exporter 18 685 20 452 21 661 22 485 25 203 24 486


Importer 28 265 29 986 32 213 33 900 39 358 37 670


Both 17 018 19 045 20 554 21 749 23 190 23 755




Manufactured goods 62 946 69 476 74 425 78 133 87 749 85 910


Exporter 18 682 20 451 21 661 22 485 25 203 24 486


Importer 27 255 29 982 32 210 33 899 39 356 37 669


Both 17 009 19 043 20 554 21 749 23 190 23 755


Source: Foreign Trade database.




The 82,622 enterprises from the ES are divided into two groups, with the majority (68,464) not


exhibiting any export behaviour and about 17 per cent (14198) also registering as exporters in the FT


database. (Ignore the numbers in parentheses in figure A.1 for now.) The 48,241 enterprises reporting


exports in the FT database in 2008 include a majority (34,043) not also included in the ES, while about


30 per cent of those in the FT database (14,198) also report their activity in the ES. The substantial


degree of non-overlap is not surprising. There will be many enterprises in the ES that produce only for


the Turkish market, and as such will have no export goods to report to Customs. These will be the


enterprises among the 68,464 in figure A.1. There will also be many trading companies that are


involved in exports but do not produce the goods themselves. These will often be small firms in terms


of number of employees, and as such will not be sampled regularly in the ES. These firms contribute


to the 34043 in figure A.1.




Figure A.1


Venn diagram: ES and FT-Export databases in 2008




Enterprise Survey (ES)








Foreign Trade (FT-Export)


















Source: Enterprise Survey database and Foreign Trade database.




In the second panel of table A.12, the ES enterprises are limited to those reporting


manufacturing activity. The results for the Venn diagram are reported in parentheses in figure A.1. We


observe three large shifts in considering just this subgroup:




• There is a large reduction in the group only found in the ES (from 68,464 to 24,696). Many of


the enterprises in this category are retail and service firms, and as such are less likely to be


involved in export activity.


68,464


(24,696) 14,198


(10,905)


34,043


(37,396)





32 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


• The total number of enterprises in FT-Export does not change, but there is an increase in the


number of enterprises only found in FT (from 34,043 to 37,396). This newly excluded group of


enterprises are those trading companies that do not themselves produce manufactured


goods.




• There is a smaller, but still substantial, group of enterprises (10,905) that both report to ES and


have transactions in FT-Export. This will be the group for which we will perform our analysis.




As noted in the text, the Turkish Statistical Institute follows a two-step procedure for selecting


participants in the Enterprise Survey each year. In the first step, it selects all enterprises with


employment of 20 or more people and all enterprises in key sectors of activity (as defined by the four-


digit NACE code); this is the full-enumeration group, and members are automatically invited to


participate.23 In the second step, it assigns smaller enterprises to sampling groups by sector of activity


and selects enterprises to participate at random from those groups.




Each enterprise in the survey is assigned a weight – the weight indicates how many similar but


unselected enterprises the selected enterprise represents. Enterprises in the full-enumeration group are


assigned a weight of one – they represent only themselves. Enterprises from the sampling groups are


assigned a weight greater than one, indicating how many other enterprises in their group they


represent.24 The choice of enterprises in the sampling group occurs interactively and is designed to


ensure that a pre-determined minimum of enterprises participates from each economic sector.




In our dynamic analysis, it is important to have multiple observations for a single enterprise.


This selection process makes it likely that relatively few of the smaller enterprises will be selected


multiple times. In table A.12 below, we summarize the weights attached to two groups identified at the


bottom of table A.5 – the 54,187 enterprises observed only once in the six-year period 2003–2008, and


the 8,022 enterprises observed in all six years.




Table A.12


Sampling weights assigned to enterprises (Shares)




Weight
Firms


observed only once
Firms


always observed


1 0.289 0.992


5 0.075 0.005


10 0.075 0.002


20 0.109 0.000


30 0.074 0.000


40 0.225 0.000


50 0.072 0.000


100 0.042 0.000


500 0.038 0.000


Over 500 0.001 0.000


Source: Enterprise Survey database.


Note: The sampling weight indicates approximately the number of enterprises that this enterprise represents
for purposes of country-level aggregation. In survey design, the sampling weight is the inverse of the
probability of being sampled for an enterprise in this class.




23 These are the selection criteria governing surveys beginning in 2005. Prior to that, enterprises were in the full-
enumeration group if they had 20 or more employees or if they had more than one “local unit” – plants, shops or offices.


24 The inverse of the weight corresponds to the probability that this enterprise will be chosen at random from its group.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 33


The first row of table A.12 reports the share of the total enterprises with weight exactly 1 in the


two groups. Consistent with our description of enterprise selection, we observe that over 99 per cent of


the enterprises observed in all six years have unit weight – these will be the large enterprises described


above. There are a small number of other firms observed every year, but these generally have weights


of less than 10. Surprisingly, though, 30 per cent of the enterprises observed only once also have unit


weight. There could be two causes of this. First, it could be that the firm only met the minimum


requirements for the unit-weight group in one of the years – in the other years it was in a random-


selection group and was not selected. Second, it could be that the enterprise was selected to


participate but did not complete and return the survey. If so, it faces a fine of 1,268 Turkish lira as


defined by the Turkish Law on Statistics for non-response enterprises.25




The remaining rows of table A.12 report the percentage of the enterprises in each group with


weights that fall into the ranges indicated in the first column. For example, 7.5 per cent of the


enterprises observed only once fall into the range [1, 5], while 7.5 per cent also have weight in the


range [5, 10]. While very few of the enterprises observed in all six years fall into these ranges, a


majority of the enterprises observed only once do.




Once those with unit weight are excluded, the largest share of enterprises in this group has


weights between 30 and 40. These are then small enterprises, with a 2 to 3 per cent probability of


being selected in any year. Some, albeit very few, of these enterprises have weights greater than 500.




Table A.13


Sectoral composition in the Enterprise Survey




NACE
section


Population
Of which


exhaustively
covered


Of which
sampled


Sample size In %
Total number


of
questionnaires


A B C = A - B D E F = B + D


C 2 773 2 773 0 0 - 2773


D 348 162 20 792 327 370 17 843 5.5 38 635


E 712 712 0 0 - 712


F 69 487 3 871 65 616 1 954 3.0 5 825


G 1 134 292 12 398 1 121 894 9 062 0.8 21 460


H 244 917 2 307 242 610 650 0.3 2 957


I 365 659 2 787 362 872 1 524 0.4 4 311


K 145 801 2 479 143 322 3 109 2.2 5 588


Subtotal 2 311 803 48 119 2 263 684 34 142 1.5 82 261


M 10 135 1 719 8 416 363 4.3 2 082


N 43 396 867 42 529 820 1.9 1 687


O 120 326 1 177 119 149 2 252 1.9 3 429


Total 2 485 660 51 882 2 433 778 37 577 1.5 89 459




Source: TUIK (2004), Structural Business Statistics, p. 11.


Notes:


C, D and E Manufacturing, mining and quarrying and other industry


F Construction


G, H and I Wholesale and retail trade, transportation and storage, accommodation and food service activities


K Financial and insurance activities


M and N Professional, scientific, technical, administration and support service activities


O Public administration, human health and social work activities




25 This fine is stated explicitly on the first page of the ES questionnaire.





34 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


Table A.13 reprints a table from TUIK (2004), Structural Business Statistics describing the


division between large (full-enumeration) enterprises and smaller (sampling) enterprises. Those in the


full-enumeration group represent only 2 per cent of the 2.48 million enterprises, and another 1.5 per


cent of the total is sampled each year at random.




NACE is a derived classification of ISIC: categories at all levels of NACE are defined either to


be identical to, or form subsets of, single ISIC categories. The first level and the second level of ISIC


Rev. 4 (sections and divisions) are identical to sections and divisions of NACE Rev. 2. The third and


fourth levels (groups and classes) of ISIC Rev. 4 are subdivided in NACE Rev. 2 according to European


requirements. However, groups and classes of NACE Rev. 2 can always be aggregated into the groups


and classes of ISIC Rev. 4 from which they were derived. The aim of the further breakdowns in NACE


Rev. 2, as compared with ISIC Rev. 4, is to obtain a classification more suited to the structures of


European economies.






APPENDIX B: ESTIMATING ENTERPRISE-LEVEL PRODUCTIVITY


We begin from a four-factor description of manufacturing technology. Capital, labour, energy


and raw materials are used as factors in production, and there is as well an enterprise-specific


measure of total factor productivity. If we define variables as follows for enterprise i in time t:




Qit = quantity produced


Pit = final good price


Lit = number of workers


Wit = average wage paid to workers


Eit = quantity of energy used


PEit = price of energy


Mit = quantity of raw materials and intermediate inputs


PMit = price of raw materials and intermediate inputs


Kit = value of capital stock


Ait = total factor productivity




We will denote the logarithm of a variable by its lower-case letter, for example, qi = ln(Qi).


Using a Cobb-Douglas specification, we can then represent the technology as:




qit = ait + αK kit + αL lit + αE eit + αM mit (B.1)




where the restriction αK + αL + αE + αM = 1 can be imposed or tested econometrically. The


technological coefficients are taken as invariant through time except for ait (discussed below).




We will also consider a variant of this technology – one in which energy and materials enter in


fixed proportions, while value added (Vit) is a well-defined function of capital and labour.




Vit = PitQit – PEit Eit – PMit Mit (B.2)




vit = bit + βK kit + βL lit (B.3)




with bit the measure for total factor productivity in this specification.








Turkish Enterprise-Level Response to Foreign Trade Liberalization 35


Estimation strategy


The Enterprise Survey does not include all of these variables in the form employed here, and


so transformations are performed to obtain conformable variables.




• Enterprise-level prices (Pit) and quantities (Qit) are not observed. Producer-price indices are


matched with enterprises at the four-digit NACE level (and denoted Pjt); these are used to


deflate enterprise-level sales revenues to obtain a proxy for quantity. This quantity measure


will be equal to Qit(Pit/Pjt), and as such will differ systematically from quantity if enterprise-level


price differs systematically from four-digit NACE producer price.




• Average number of employees is used as a measure for Lit.




• Energy is measured in two forms: electricity purchases and gas purchases. These are


deflated by the producer price index for the four-digit NACE index for energy to obtain a


measure of Eit. Electricity is used in the estimations that follow.




• Raw materials and intermediate inputs are grouped together in the survey and are measured in


value terms. We use the four-digit NACE index for materials to deflate this value and obtain a


measure of Mit.




• Capital is reported in the survey in book-value terms. We use the producer price index for the


four-digit NACE category of machinery purchases to deflate the capital value to real terms.




• The variable ait is unobserved, but we conjecture that it is made up of two components:


enterprise-specific total factor productivity ωit and random error εit. The random error is


assumed to be normally distributed with errors perhaps clustered by enterprise. The total


factor productivity term ωit is assumed to be a state variable evolving according to a first-order


Markov process:




ωit = Et-1(ωit | ωit-1) + ξit (B.4)




and with ξit orthogonal to kit and ωit. As past authors have pointed out (see Olley and Pakes


(1996), Levinsohn and Petrin (2003), Ackerburg, Caves and Fraxer (2010)), this leads to the potential for


bias in the estimation of production function. We will address this in one manner below.




We estimate this production function for Turkish manufacturing enterprises in three parts – first


for NACE 17 (textiles) enterprises, second for NACE 18 (apparel) enterprises, and third for all other


manufacturing enterprises. In this first round, we use all manufacturing enterprises for which there are


non-zero sales, non-zero capital, non-zero electricity cost and non-zero raw material cost. This leads to


unbalanced panels of data of sizes specified below. We estimate five variants of the production-


function specification of (1) and (2):




• Least squares with White standard errors (robust)


• Least squares with standard errors clustered by enterprise (cluster).


• Fixed-effect (by enterprise) least squares.


• Random-effect (by enterprise) least squares.


• Least squares using the Levinsohn-Petrin (2003) correction for bias of estimation


results in total factor productivity.




We have also included year-specific effects in these regressions. These effects could be


interpreted as a time trend in total factor productivity. They are significant, and indicate rising


productivity over time, but inclusion does not change the comparison of production-function


coefficients. They will be introduced at a later stage.





36 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


Estimation Results: NACE 17 (textiles)


Estimation results for textiles enterprises are reported in table B.1. The estimated coefficients


for αK, αL, αE, and αM are reported, as is an intercept coefficient α0 common to all enterprises. Below


each coefficient is the associated standard error. All estimated coefficients are significantly different


from zero at the 95 per cent level of confidence. For the fixed-effect and random-effect estimates, we


report the variation observed in the systematic term (σu) and the unsystematic term (σe). Rho indicates


the importance of the systematic variation relative to the unsystematic variation. The F statistic


reported for the fixed-effect estimation tests whether the estimated fixed effects are jointly significant;


the null of non-significance is rejected at the 95 per cent confidence level.




Table B.1


Production function estimation for NACE 17 firms




Robust Cluster (idf)
Fixed
effects



Random
effects


Levinsohn-Petrin


α0 2.520 2.520 4.820 2.990


0.057 0.071 0.080 0.036


αK 0.027 0.027 0.008 0.015 0.013


0.004 0.004 0.003 0.002 0.004


αL 0.350 0.350 0.347 0.393 0.328


0.013 0.016 0.015 0.009 0.015


αE 0.073 0.073 0.079 0.116 0.063


0.009 0.010 0.009 0.007 0.009


αM 0.593 0.593 0.355 0.503 0.328


0.012 0.015 0.009 0.006 0.089


σ u 0.690 0.380


σ e 0.263 0.263


rho 0.873 0.675


F(u=0) 5.110




R2 0.94 0.94 0.93 0.93


R2 within 0.55 0.55


R2 between 0.94 0.95


Nobs 6 027 6 027 6 027 6 027 6 025


Ngroups 2 439 2 439 2 439


Source: Authors’ calculations.




The coefficient estimates for the “robust” and “cluster” estimations are identical by design.


The difference between the two comes from calculating standard errors – either using White residuals


(robust) or calculating the standard errors clustered by enterprise (cluster). The cluster standard errors


are in all cases larger than the robust errors.




The final three regressions (fixed-effect, random-effect and Levinsohn-Petrin) have in common


that they calculate explicitly the enterprise-specific term for total factor productivity. (The first two


regressions assume that total factor productivity is randomly distributed around the intercept.)


Levinsohn-Petrin in addition controls for the simultaneity bias potentially due to unobserved differences


in total factor productivity causing systematic adjustments in the use of variable factors such as labour.










Turkish Enterprise-Level Response to Foreign Trade Liberalization 37


While the regression results differ in small details, there are strong similarities across results:




• The coefficient associated with labour is relatively constant at about 0.35.


• The coefficient associated with energy is small but significant at 0.06 – 0.12.


• The coefficient associated with capital is quite small throughout – around 0.03 before


accounting for enterprise-specific total factor productivity, and closer to 0.01 once those


enterprise-specific corrections are made.




The significant difference in coefficient on raw materials when comparing fixed-effect and


Levinsohn-Petrin estimators to the robust, cluster and (to a lesser extent) random-effect estimators


provides evidence of the simultaneity bias associated with unobserved total factor productivity


influencing input choices. The coefficient on materials is much greater when corrections are not made


for enterprise-specific productivity differences. Levinsohn-Petrin is designed to correct for that, and to


move that “between” difference in productivity to the productivity term and away from spurious


correlation with materials. The fixed-effect estimator in this instance proves to have very similar


properties.




The role of materials as a proxy for the unobserved total factor productivity raises the question


of whether a value added specification will be a more satisfactory representation of the technology. In


table B.2 we report the results of estimation of the value added function (3) using the same five


techniques. We observe the same pattern in coefficients for capital; while they are roughly three times


higher in the value added function, they take a similar dip in value when estimated in the Levinsohn-


Petrin and fixed-effect regressions. We observe as well that the coefficient on labour, as the sole


remaining variable input, now takes on some of the characteristics of the materials coefficient in table


B.1. For the robust, cluster and random-effect estimations, this coefficient is insignificantly different


from 1, but once the correction is made for unobserved productivity in Levinsohn-Petrin the coefficient


falls to a significantly smaller 0.738. The fixed-effect estimation makes a similar, even more striking,


downward adjustment in that coefficient.




Table B.2


Value added function estimation for NACE 17 firms




Robust Cluster (idf)
Fixed


effects
Random
effects


Levinsohn-Petrin


β0 4.307 4.307 6.415 4.602


0.044 0.052 0.121 0.049


βK 0.108 0.108 0.014 0.054 0.030


0.007 0.008 0.006 0.006 0.008


βL 1.028 1.028 0.631 1.023 0.738


0.014 0.015 0.031 0.014 0.021


σ u 1.200 0.779


σ e 0.626 0.626


rho 0.786 0.607


F(u=0) 4.050


R2 0.72 0.72 0.71 0.72


R2 within 0.11 0.11


R2 between 0.76 0.77


Nobs 5 760 5 760 5 760 5 760 5 760


Ngroups 2 367 2 367 2 367


Source: Authors’ calculations.






38 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


These regressions define the technical coefficients reported in tables B.1 and B.2. They also


define implicitly an estimate of the enterprise-specific total factor productivity. In table B.3 we report


aggregate statistics about those ωit in each year. Four statistics are reported – median, mean,


skewness and kurtosis.26 Notice from these statistics the striking jump in value in 2005. Controlling for


other factors (as we did in running the regression), total factor productivity in 2005 jumped strongly.


Interestingly, it did not retain that increase in later years.




Table B.3


Distribution of fixed and random effects for NACE 17 firms




Based on the production function (B.1)


Fixed effects Random effects


Median Mean Skewness Kurtosis Median Mean Skewness Kurtosis


2003 -0.009 -0.026 -0.22 3.82 -0.022 0.008 0.49 5.8


2004 -0.015 -0.042 -0.34 3.51 -0.025 -0.01 0.38 4.62


2005 0.06 0.06 -0.19 3.6 0.009 0.032 0.438 4.2


2006 0.001 -0.06 -0.48 3.24 -0.012 0.009 0.3 4.34


2007 0.024 -0.01 -0.44 3.47 0.014 0.036 0.65 5.32


2008 0.088 0.082 -0.52 5 0.047 0.077 0.68 6.24


All 0.025 0 -0.41 3.75 0.007 0.026 0.474 5.05




Based on the value added function (B.3)


Fixed effects Random effects


Median Mean Skewness Kurtosis Median Mean Skewness Kurtosis


2003 0.03 -0.035 -0.693 4.616 0.064 0.044 -0.41 4.88


2004 0.026 -0.054 -1 7.94 0.024 0.023 -0.7 8.27


2005 0.114 0.106 -0.425 4.49 0.078 0.111 0.054 3.97


2006 0.006 -0.114 -0.531 3.47 0.042 0.049 0.069 3.82


2007 0.026 -0.024 -0.565 4.33 0.0967 0.1103 -0.033 4.57


2008 0.153 0.128 -0.66 5.27 0.172 0.187 0.06 4.3


All 0.059 0 -0.68 5.16 0.084 0.091 -0.172 5.14


Source: Authors' calculations.




Estimation results: NACE 18 (apparel)


We follow the same steps in analysing enterprise behaviour in the apparel sector, and we find


the same pattern of coefficient estimates and of productivity measures. These can be found in tables


B.4 through B.6.




In table B.4, the technological parameters α0, αK, αL, αE and αM have similar magnitudes in


these regressions. The αM are quite large when no correction is made for enterprise-specific effects,


but they are reduced in magnitude (and impact transferred on average to the enterprise-specific


effects) when the enterprise-specific corrections are introduced. The αK are once again quite small,


and with the Levinsohn-Petrin correction they disappear altogether. The labour coefficients here again


are near one-third.




26 Skewness measures the non-symmetry of the distribution. A negative value indicates a “fatter tail” at the lower end of
the distribution, while a positive value indicates a fatter tail at the upper end. Kurtosis measures the peakedness of the
distribution. The distribution is more concentrated around its mean than a normal distribution if this statistic is greater
than 3, and is less concentrated than a normal distribution if the value is less than 3.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 39


Table B.4


Production function estimation for NACE 18 firms




Robust Cluster (idf)
Fixed


effects
Random
effects


Levinsohn-Petrin


α0 2.940 2.940 4.840 3.300


0.035 0.042 0.053 0.025


αK 0.034 0.034 0.010 0.027 0.000


0.003 0.004 0.003 0.002 0.002


αL 0.340 0.340 0.358 0.370 0.280


0.009 0.011 0.010 0.007 0.010


αE 0.150 0.150 0.090 0.166 0.133


0.008 0.009 0.007 0.006 0.008


αM 0.528 0.528 0.355 0.467 0.331


0.008 0.009 0.006 0.004 0.115




σ u 0.690 0.423


σ e 0.310 0.310


rho 0.831 0.650


F(u=0) 4.700




R2 0.92 0.92 0.91 0.92


R2 within 0.57 0.56


R2 between 0.92 0.93


Nobs 12 125 12 125 12 125 12 125 12 125


Ngroups 5 245 5 245 5 245


Source: Authors’ calculations.






In table B.5, the value added formulation is used. Once again, both capital and labour shares


are magnified relative to those found in table B.4. Once again, the share of capital is reduced once the


enterprise-specific effects are allowed for. Once again (as in table B.2), the labour coefficients take on


the positive correlation with unobserved variability in total factor productivity. In the robust and cluster


regressions, the βL coefficient is much larger than that observed once the enterprise-specific


corrections are made. The Levinsohn-Petrin coefficient is an unbiased estimate, and we see once


again that the fixed-effect estimate is insignificantly different from that.




Table B.6 reports the average fixed and random effects by year. There is some evidence of an


upward bump in 2005, but this effect is less pronounced than in textiles.




Estimation results for the rest of the manufacturing sector. Our estimates for the rest of


the manufacturing sector are provided in tables B.7 through B.9. The patterns of technology


coefficients in tables B.7 and B.8 are all familiar from our analysis of textiles and apparel, although


these other industries appear to be less labour-using (that is, smaller αL than in the textiles and apparel


sectors). The Levinsohn-Petrin correction once again removes any effect of capital in the production


function.






40 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


Table B.5


Value added function estimation for NACE 18 firms




Robust Cluster (idf)
Fixed


effects
Random
effects


Levinsohn-Petrin


β0 4.620 4.620 6.600 4.980


0.032 0.037 0.074 0.033


βK 0.156 0.156 0.020 0.090 0.022


0.006 0.008 0.006 0.005 0.006


βL 0.918 0.918 0.600 0.900 0.577


0.011 0.012 0.018 0.009 0.014


σ u 1.190 0.810


σ e 0.620 0.620


rho 0.790 0.630


F(u=0) 4.720


R2 0.68 0.68 0.67 0.68


R2 within 0.14 0.14


R2 between 0.72 0.73


Nobs 11 718 11 718 11 718 11 718 11 718


Ngroups 5 125 5 125 5 125


Source: Authors’ calculations.






Table B.6


Distribution of fixed and random effects for NACE 18 firms




Based on the production function (B.1)


Fixed effects Random effects


Median Mean Skewness Kurtosis Median Mean Skewness Kurtosis


2003 -0.042 -0.065 -0.66 5.21 -0.046 -0.032 0.001 4.46


2004 -0.007 -0.011 -0.47 4.1 -0.004 0.009 0.02 3.75


2005 0.017 0.048 -0.2 3.85 -0.002 0.037 0.031 3.75


2006 -0.025 -0.048 -0.33 3.95 -0.008 0.024 0.58 6.97


2007 0.004 -0.034 -0.22 3.34 0.006 0.043 0.59 4.91


2008 0.088 0.101 -0.36 6.23 0.108 0.117 0.128 6.16


All 0.008 0 -0.39 4.43 0.01 0.037 0.287 5.12


Based on the value added function (B.3)


Fixed effects Random effects


Median Mean Skewness Kurtosis Median Mean Skewness Kurtosis


2003 -0.085 -0.057 -0.44 3.74 -0.01 0.045 0.09 3.42


2004 -0.065 0.013 -0.4 4.11 0.006 0.089 0.14 3.56


2005 -0.04 0.06 -0.17 3.68 0.035 0.111 0.277 3.32


2006 -0.128 -0.111 -0.369 3.95 -0.019 0.058 0.26 3.77


2007 -0.012 -0.056 -0.16 3.07 0.046 0.113 0.37 3.54


2008 0.12 0.155 -0.223 4.66 0.25 0.27 0.17 3.89


All -0.03 0 -0.32 3.9 0.051 0.118 0.23 3.62


Source: Authors’ calculations.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 41


Table B.7


Production function estimation for other manufacturing




Robust Cluster (idf)
Fixed


effects
Random
effects


Levinsohn-
Petrin


Random Levinsohn


α0 1.916 1.916 3.960 2.370


0.022 0.026 0.027 0.013


αK 0.035 0.035 0.011 0.022 0.000


0.001 0.002 0.001 0.001 0.001


αL 0.288 0.288 0.370 0.352 0.273


0.005 0.006 0.005 0.003 0.005


αE 0.084 0.084 0.077 0.099 0.078


0.002 0.003 0.000 0.002 0.002


αM 0.671 0.671 0.445 0.597 0.755


0.004 0.005 0.003 0.002 0.007


σ u 0.630 0.410


σ e 0.280 0.280


rho 0.845 0.680


F(u=0) 5.050


R2 0.94 0.94 0.94 0.94


R2 within 0.60 0.59


R2 between 0.94 0.95


Nobs 58 747 58 747 58 747 58 747 58 743


Ngroups 24 312 24 312 24 312


Source: Authors’ calculations.






We observe a similar diminution of capital coefficient in the value-added function, although in


this case the value remains small but significant in the Levinsohn-Petrin estimate. The labour


coefficient βL is inflated in the initial regressions, but settles to a value of 0.732 similar to that observed


in the other sectors.




Table B.9 reports the yearly evolution of fixed and random effects. Our hypothesis is that


there will be no 2005 “bump” in total factor productivity; that in the textiles and apparel sectors was


associated with the opening of the European Union market once quotas were removed, and we do not


expect to see anything similar for the rest of manufactures. In fact, there is some evidence of a similar


bump in the fixed-effect terms, though not in the random-effects terms.





42 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


Table B.8


Value added function estimation for other manufactures




Robust Cluster (idf)
Fixed
effect



Random


effect


Levinsohn-
Petrin


β0 3.727 3.727 5.960 4.120


0.014 0.018 0.044 0.016


βK 0.189 0.189 0.028 0.105 0.066


0.002 0.003 0.003 0.002 0.003


βL 1.047 1.047 0.732 1.076 0.732


0.005 0.006 0.012 0.005 0.008




σ u 1.250 0.810


σ e 0.670 0.660


rho 0.780 0.590


F(u=0) 4.080




R2 0.76 0.76 0.74 0.75


R2 within 0.11 0.11


R2 between 0.77 0.78


Nobs 54 766 54 766 54 766 54 766 54 766


Ngroups 22 840 22 840 22 840


Source: Authors’ calculations.




Table B.9


Distribution of fixed and random effects for other manufactures in Turkey




Based on the production function (B.1)


Fixed effects Random effects


Median Mean Skewness Kurtosis Median Mean Skewness Kurtosis


2003 -0.004 0.001 -0.310 4.980 0.008 0.029 0.467 7.400


2004 0.005 0.002 -1.360 22.510 0.012 0.028 -1.440 39.850


2005 0.042 0.043 -0.200 5.580 0.016 0.029 0.460 9.640


2006 -0.030 -0.040 -1.350 24.800 0.000 0.011 -1.260 43.130


2007 -0.010 -0.024 -0.780 12.270 0.001 0.015 -0.425 29.220


2008 0.020 0.030 -0.700 10.850 0.032 0.047 -0.550 19.790


All 0.004 0.000 -0.860 14.910 0.010 0.026 -0.560 26.920


Based on the value added function (B.3)


Fixed effects Random effects


Median Mean Skewness Kurtosis Median Mean Skewness Kurtosis


2003 0.064 0.035 -0.410 4.630 0.109 0.114 -0.160 4.600


2004 0.057 0.036 -0.290 4.250 0.104 0.105 -0.060 4.250


2005 0.104 0.083 -0.320 4.510 0.095 0.090 -0.019 3.870


2006 -0.029 -0.107 -0.460 4.200 0.040 0.033 -0.080 4.000


2007 0.005 -0.059 -0.320 4.040 0.056 0.055 -0.020 4.590


2008 0.073 0.054 -0.250 4.580 0.127 0.138 0.037 4.580


All 0.039 0.000 -0.360 4.370 0.085 0.086 -0.037 4.310


Source: Authors’ calculations.







Turkish Enterprise-Level Response to Foreign Trade Liberalization 43




UNCTAD study series on


POLICY ISSUES IN INTERNATIONAL TRADE
AND COMMODITIES










No. 1 Erich Supper, Is there effectively a level playing field for developing country
exports?, 2001, 138 p. Sales No. E.00.II.D.22.




No. 2 Arvind Panagariya, E-commerce, WTO and developing countries, 2000, 24 p. Sales
No. E.00.II.D.23.




No. 3 Joseph Francois, Assessing the results of general equilibrium studies of multilateral
trade negotiations, 2000, 26 p. Sales No. E.00.II.D.24.




No. 4 John Whalley, What can the developing countries infer from the Uruguay Round
models for future negotiations?, 2000, 29 p. Sales No. E.00.II.D.25.




No. 5 Susan Teltscher, Tariffs, taxes and electronic commerce: Revenue implications for
developing countries, 2000, 57 p. Sales No. E.00.II.D.36.




No. 6 Bijit Bora, Peter J. Lloyd, Mari Pangestu, Industrial policy and the WTO, 2000, 47 p.
Sales No. E.00.II.D.26.




No. 7 Emilio J. Medina-Smith, Is the export-led growth hypothesis valid for developing
countries? A case study of Costa Rica, 2001, 49 p. Sales No. E.01.II.D.8.




No. 8 Christopher Findlay, Service sector reform and development strategies: Issues and
research priorities, 2001, 24 p. Sales No. E.01.II.D.7.




No. 9 Inge Nora Neufeld, Anti-dumping and countervailing procedures – Use or abuse?
Implications for developing countries, 2001, 33 p. Sales No. E.01.II.D.6.




No. 10 Robert Scollay, Regional trade agreements and developing countries: The case of
the Pacific Islands’ proposed free trade agreement, 2001, 45 p. Sales No.
E.01.II.D.16.




No. 11 Robert Scollay and John Gilbert, An integrated approach to agricultural trade and
development issues: Exploring the welfare and distribution issues, 2001, 43 p. Sales
No. E.01.II.D.15.




No. 12 Marc Bacchetta and Bijit Bora, Post-Uruguay Round market access barriers for
industrial products, 2001, 50 p. Sales No. E.01.II.D.23.




No. 13 Bijit Bora and Inge Nora Neufeld, Tariffs and the East Asian financial crisis, 2001,
30 p. Sales No. E.01.II.D.27.




No. 14 Bijit Bora, Lucian Cernat, Alessandro Turrini, Duty and quota-free access for LDCs:
Further evidence from CGE modelling, 2002, 130 p. Sales No. E.01.II.D.22.






44 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


No. 15 Bijit Bora, John Gilbert, Robert Scollay, Assessing regional trading arrangements in
the Asia-Pacific, 2001, 29 p. Sales No. E.01.II.D.21.




No. 16 Lucian Cernat, Assessing regional trade arrangements: Are South-South RTAs more
trade diverting?, 2001, 24 p. Sales No. E.01.II.D.32.




No. 17 Bijit Bora, Trade related investment measures and the WTO: 1995-2001, 2002.


No. 18 Bijit Bora, Aki Kuwahara, Sam Laird, Quantification of non-tariff measures, 2002,
42 p. Sales No. E.02.II.D.8.




No. 19 Greg McGuire, Trade in services – Market access opportunities and the benefits of
liberalization for developing economies, 2002, 45 p. Sales No. E.02.II.D.9.




No. 20 Alessandro Turrini, International trade and labour market performance: Major
findings and open questions, 2002, 30 p. Sales No. E.02.II.D.10.




No. 21 Lucian Cernat, Assessing South-South regional integration: Same issues, many
metrics, 2003, 32 p. Sales No. E.02.II.D.11.




No. 22 Kym Anderson, Agriculture, trade reform and poverty reduction: Implications for
sub-Saharan Africa, 2004, 30 p. Sales No. E.04.II.D.5.




No. 23 Ralf Peters and David Vanzetti, Shifting sands: Searching for a compromise in the
WTO negotiations on agriculture, 2004, 46 p. Sales No. E.04.II.D.4.




No. 24 Ralf Peters and David Vanzetti, User manual and handbook on Agricultural Trade
Policy Simulation Model (ATPSM), 2004, 45 p. Sales No. E.04.II.D.3.




No. 25 Khalil Rahman, Crawling out of snake pit: Special and differential treatment and
post-Cancun imperatives, 2004.




No. 26 Marco Fugazza, Export performance and its determinants: Supply and demand
constraints, 2004, 57 p. Sales No. E.04.II.D.20.




No. 27 Luis Abugattas, Swimming in the spaghetti bowl: Challenges for developing
countries under the “New Regionalism”, 2004, 30 p. Sales No. E.04.II.D.38.




No. 28 David Vanzetti, Greg McGuire and Prabowo, Trade policy at the crossroads – The
Indonesian story, 2005, 40 p. Sales No. E.04.II.D.40.




No. 29 Simonetta Zarrilli, International trade in GMOs and GM products: National and
multilateral legal frameworks, 2005, 57 p. Sales No. E.04.II.D.41.




No. 30 Sam Laird, David Vanzetti and Santiago Fernández de Córdoba, Smoke and mirrors:
Making sense of the WTO industrial tariff negotiations, 2006, Sales No.
E.05.II.D.16.




No. 31 David Vanzetti, Santiago Fernandez de Córdoba and Veronica Chau, Banana split:
How EU policies divide global producers, 2005, 27 p. Sales No. E.05.II.D.17.




No. 32 Ralf Peters, Roadblock to reform: The persistence of agricultural export subsidies,
2006, 43 p. Sales No. E.05.II.D.18.








Turkish Enterprise-Level Response to Foreign Trade Liberalization 45


No. 33 Marco Fugazza and David Vanzetti, A South-South survival strategy: The potential
for trade among developing countries, 2006, 25 p.






No. 34 Andrew Cornford, The global implementation of Basel II: Prospects and outstanding
problems, 2006, 30 p.




No. 35 Lakshmi Puri, IBSA: An emerging trinity in the new geography of international
trade, 2007, 50 p.




No. 36 Craig VanGrasstek, The challenges of trade policymaking: Analysis, communication
and representation, 2008, 45 p.




No. 37 Sudip Ranjan Basu, A new way to link development to institutions, policies and
geography, 2008, 50 p.




No. 38 Marco Fugazza and Jean-Christophe Maur, Non-tariff barriers in computable general
equilibrium modelling, 2008, 25 p.




No. 39 Alberto Portugal-Perez, The costs of rules of origin in apparel: African preferential
exports to the United States and the European Union, 2008, 35 p.




No. 40 Bailey Klinger, Is South-South trade a testing ground for structural
transformation?, 2009, 30 p.




No. 41 Sudip Ranjan Basu, Victor Ognivtsev and Miho Shirotori, Building trade-relating
institutions and WTO accession, 2009, 50 p.




No. 42 Sudip Ranjan Basu and Monica Das, Institution and development revisited: A
nonparametric approach, 2010, 26 p.




No. 43 Marco Fugazza and Norbert Fiess, Trade liberalization and informality: New stylized
facts, 2010, 45 p.




No. 44 Miho Shirotori, Bolormaa Tumurchudur and Olivier Cadot, Revealed factor intensity
indices at the product level, 2010, 55 p.




No. 45 Marco Fugazza and Patrick Conway, The impact of removal of ATC Quotas on
international trade in textiles and apparel, 2010, 50 p.




No. 46 Marco Fugazza and Ana Cristina Molina, On the determinants of exports survival,
2011, 40 p.




No. 47 Alessandro Nicita, Measuring the relative strength of preferential market access,
2011, 30 p.




No. 48 Sudip Ranjan Basu and Monica Das, Export structure and economic performance in
developing countries: Evidence from nonparametric methodology, 2011, 58 p.




No. 49 Alessandro Nicita and Bolormaa Tumurchudur-Klok, New and traditional trade flows
and the economic crisis, 2011, 22 p.




No. 50 Marco Fugazza and Alessandro Nicita, On the importance of market access for trade,
2011, 35 p.






46 POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES


No. 51 Marco Fugazza and Frédéric Robert-Nicoud, The ‘Emulator Effect’ of the Uruguay
round on United States regionalism, 2011, 45 p.




No. 52 Sudip Ranjan Basu, Hiroaki Kuwahara and Fabien Dumesnil, Evolution of non-tariff
measures: Emerging cases from selected developing countries, 2012, 38p.




No. 53 Alessandro Nicita and Julien Gourdon, A preliminary analysis on newly collected data
on non-tariff measures, 2013, 31 p.




No. 54 Alessandro Nicita, Miho Shirotori and Bolormaa Tumurchudur Klok, Survival analysis
of the exports of least developed countries: The role of comparative advantage,
2013, 25 p.




No. 55 Alessandro Nicita, Victor Ognivtsev and Miho Shirotori, Global supply chains: Trade
and Economic policies for developing countries, 2013, 33 p.




No. 56 Alessandro Nicita, Exchange rates, international trade and trade policies, 2013, 29 p.


No. 57 Marco Fugazza, The economics behind non-tariff measures: Theoretical insights and
empirical evidence, 2013, 33 p.




No. 58 Marco Fugazza and Alain McLaren, Market access, export performance and
survival: Evidence from Peruvian firms, 2013, 39 p.




No. 59 Patrick Conway, Marco Fugazza and M. Kerem Yuksel, Turkish enterprise-level
response to foreign trade liberalization: The removal of agreements on textiles and
clothing quotas, 2013, 54 p.










































Copies of UNCTAD study series on Policy Issues in International Trade and Commodities may be
obtained from the Publications Assistant, Trade Analysis Branch (TAB), Division on International
Trade in Goods and Services and Commodities (DITC), United Nations Conference on Trade and
Development, Palais des Nations, CH-1211 Geneva 10, Switzerland (Tel: +41 22 917 4644).
These studies are accessible on the website at http://unctad.org/tab.




Since 1999, the Trade Analysis Branch of the Division on International Trade in Goods and
Services, and Commodities of UNCTAD has been carrying out policy-oriented analytical work
aimed at improving the understanding of current and emerging issues in international trade and
development. In order to improve the quality of the work of the Branch, it would be useful to
receive the views of readers on this and other similar publications. It would therefore be greatly
appreciated if you could complete the following questionnaire and return to:


Trade Analysis Branch, DITC
Rm. E-8065


United Nations Conference on Trade and Development
Palais des Nations, CH-1211 Geneva 10, Switzerland


(Fax: +41 22 917 0044; E-mail: tab@unctad.org)


1. Name and address of respondent (optional):


2. Which of the following describes your area of work?


Government Public enterprise
Private enterprise institution Academic or research
International organization Media
Not-for-profitorganization Other(specify)_________________


3. Inwhichcountrydoyouwork?_________________________________________


4. DidyoufindthispublicationVeryuseful Ofsomeuse Littleuse
to your work?


5. What is your assessment of the contents of this publication?
Excellent Good Adequate Poor


6. Othercomments:


QUESTIONNAIRE


UNCTAD Study series on


POLICY ISSUES IN INTERNATIONAL TRADE
AND COMMODITIES


(Study series no. 59: Turkish enterprise-level response to foreign trade liberalization:
The removal of agreements on textiles and clothing quotas)


Readership Survey




Login