Globalization and Firm Dynamics - Lirias@Lessius

FACULTEIT ECONOMISCHE EN 

TOEGEPASTE ECONOMISCHE 

WETENSCHAPPEN 

KATHOLIEKE 

UNIVERSITEIT 

LEUVEN 

Globalization and Firm Dynamics 

Proefschrift voorgedragen tot 

het behalen van de graad van 

Doctor in de Economische 

Wetenschappen 

door 

Ilke VAN BEVEREN 

Number 293 2008

Doctoral committee 

Advisor: 

prof. dr. Jozef Konings Katholieke Universiteit Leuven 

Co-advisor: 

prof. dr. Ysabel Nauwelaerts Lessius, Antwerpen 

Members: 

prof. dr. Bruno Cassiman IESE Business School, Barcelona 

prof. dr. Beata Smarzynska Javorcik University of Oxford 

prof. dr. Jo Swinnen Katholieke Universiteit Leuven 

prof. dr. Frank Verboven Katholieke Universiteit Leuven 

prof. dr. Gerald Willmann Katholieke Universiteit Leuven 

Daar de proefschriften in de reeks van de Faculteit Economische en Toegepaste 

Economische Wetenschappen het persoonlijk werk zijn van hun auteurs, zijn 

alleen deze laatsten daarvoor verantwoordelijk 

i

Acknowledgements 

I would like to begin this word of thanks by thanking two people, without 

whom I might have never embarked on this journey: Guido Van Rompuy, 

professor in economics at Lessius and supervisor of my undergraduate thesis 

and Mathew Tharakan, with whom I have had the pleasure to work for three 

years at the University of Antwerp. Guido, thank you for guiding me through 

my first steps into quantitative research and for stimulating my interest in 

economics in general. Mathew, I would like to thank you for your endless 

patience, for the countless hours spent discussing our research project, for 

your kindness and most of all for allowing me to benefit from your enormous 

expertise during my years at the University of Antwerp. 

This project could not have been realized without the support of Joep 

Konings, who has been my supervisor these past four years. Joep, thank 

you for always pointing me in the right direction whenever I felt lost. It 

continues to amaze me how, every time I was on the verge of giving up on 

some research idea because of some difficulty encountered along the way, you 

could put me back on track within fifteen minutes. Most of all, I would like 

to thank you for providing me with a tool kit that will allow me to continue 

to do good scientific research in the coming years. This tool kit includes not 

only knowledge, but also a strong sense of self-confidence and independence, 

which is perhaps even more important. 

I would also like to express my gratitude to my co-supervisor, Ysabel 

Nauwelaerts. Ysabel, thank you for your support these last few years. Bruno 

Cassiman, Beata Smarzynska Javorcik, Jo Swinnen, Frank Verboven and 

Gerald Willmann, who were part of my scientific committee, have been instrumental 

in shaping the final version of my Ph.D. Bruno, I would like 

iii

iv 

thank you particularly for your input on chapter 4. Beata and Gerald are 

thanked for providing critical comments on all chapters, which have proven 

very helpful. Jo, thank you for many critical and insightful comments given 

at various LICOS seminars, they have greatly improved the end product. Finally, 

Frank, thank you for your methodological and econometric suggestions, 

particularly in relation to chapters 1 and 4. 

Without my two co-authors, Veerle Slootmaekers and Marialuz Moreno 

Badia, chapter 2 could not have been written. Veerle and Marialuz, I would 

like to thank you for a wonderful cooperation and learning experience! I very 

much hope we can continue to work together in the future. 

The firm-level data used in chapter 2 were kindly provided by the Estonian 

Business Registry. Specifically, Larissa Merlukova and Kadri Rohulaid of the 

Centre of Registers and Infosystems are thanked for providing the data and 

some clarifications on the database. The trade data used in this chapter were 

obtained through the IMF. The Community Innovation Survey data used in 

chapter 4 were kindly provided by the Belgian Science Policy (Belspo). I 

am especially grateful to Manu Monard, Peter Teirlinck and the CFS/STAT 

Commission of Belspo for allowing me to access the data for Belgium; for 

answering questions related to the data and for their hospitality during my 

visits there. 

All chapters in this thesis have benefited greatly from comments made by 

participants of the LICOS International Trade and Development Seminars. 

Carmine Ornaghi provided useful comments on chapter 1. Chapter 1 has also 

benefited greatly from numerous discussions with Veerle Slootmaekers and 

Damiaan Persyn. An anonymous referee is thanked for the suggestions made 

on chapter 3. Leo Sleuwaegen, Hylke Vandenbussche and Filip De Beule are 

thanked for their insights on chapter 4. Participants at the Spring Meeting 

for Young Economists in Hamburg (2007) and at the EARIE conference in 

Toulouse (2008) provided useful comments on chapters 1 and 4 respectively. 

The department of business studies at Lessius and particularly Paul Verheyen 

and Flora Carrijn, are gratefully acknowledged for providing me with

the resources to work on my Ph.D. and for allowing me to spend most of my 

time on my research these past few years. LICOS, Centre for Institutions 

and Economic Performance, and particularly Jo Swinnen, is thanked for providing 

me with an office and most importantly, for allowing me to benefit 

from a highly stimulating research environment. 

I would also like to express my deepest gratitude to my friends and colleagues 

at both Lessius and LICOS. To Filip De Beule, for allowing me to 

benefit from his expertise both in terms of research and teaching and for his 

comments on my work; but also for his general support. Filip, I hope we 

can continue to work together over the coming years. To my fellow graduate 

students and other colleagues at Lessius and particularly to An, Anneleen, 

Bert, Carlos, Frederick, Frederiek, Karen, Kelly, Lien, Nico, Raf, Steve, Sofie, 

Sophie, Tim, Tom; thank you all very much for your continuous support, but 

also for many shared moments of joy, laughter and friendship, which have 

given me a home away from home. I look forward to continue to be a part 

of our growing and dynamic team! 

I have also greatly benefited from the expertise and help of many of my 

fellow Ph.D. students at LICOS. Damiaan, I think you must realize that your 

Stata expertise has saved more than one graduate student at LICOS from 

ultimate disaster, but still: thank you so much for all your help, countless 

discussions and everlasting patience. Stijn, thank you for introducing me to 

Belfirst and for sharing research ideas and insights. Emilia, thank you for 

your support and especially for your sense of humor, which has brightened my 

day more than once. Ziga, thank you for your critical comments at various 

seminars and many useful discussions. Veerle, working together with you has 

truly exceeded all my expectations! Italo, apart from your insights, I have 

tremendously enjoyed your optimistic view of the world and your openness 

towards people. I wish you all the best in your future careers! Finally, I 

would also like to thank Tom Van Ourti, whom I had the pleasure to work 

with when I was still at the University of Antwerp, and who introduced me 

to Stata. 

v

vi 

Last but certainly not least, a word of gratitude is in order for my friends 

and family. I would like to thank my parents, for allowing me to make my 

own choices in life. Dad, thank you for coming over for my public defense, 

that means a lot to me! Special thanks goes to my sister, Jo, who never 

forgot to enquire about my progress and who was always just a phone call 

away. I would also like to thank Michel’s family, whom we can always rely on, 

and who have always been very supportive throughout these last few years. 

My biggest critic throughout this venture has always been Michel, while 

at the same time, he has been my greatest fan. Michel, thank you for continuously 

reminding me of what really matters in life. Your critical appraisal 

of the world in general and the sense and nonsense of academic research in 

particular forces me to always question the relevance to society of everything 

I do, which is a good thing. Most of all, I simply want to thank you for being 

a part of my life! 

Ilke Van Beveren 

Antwerp, November 19 2008.

Contents 

Doctoral committee i 

Acknowledgements iii 

Table of contents ix 

List of Figures xiii 

List of Tables xv 

General introduction 1 

1 Total Factor Productivity Estimation: A Practical Review 9 

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 

1.2 Total factor productivity: Methodological issues . . . . . . . . 12 

1.2.1 Some preliminaries on the production function . . . . . 12 

1.2.2 Endogeneity of attrition or selection bias . . . . . . . . 14 

1.2.3 Endogeneity of input choice or simultaneity bias . . . . 15 

1.2.4 Omitted output price bias . . . . . . . . . . . . . . . . 16 

1.2.5 Omitted input price bias . . . . . . . . . . . . . . . . . 17 

1.2.6 Endogeneity of the product mix (multi-product firms) . 18 

1.2.7 Summary of methodological issues . . . . . . . . . . . . 19 

1.3 Total factor productivity estimation . . . . . . . . . . . . . . . 21 

1.3.1 Fixed effects estimation . . . . . . . . . . . . . . . . . 21 

1.3.2 Instrumental variables (IV) and GMM . . . . . . . . . 22 

1.3.3 Olley-Pakes estimation algorithm . . . . . . . . . . . . 23 

1.3.4 Levinsohn-Petrin estimation algorithm . . . . . . . . . 26 

1.3.5 Olley-Pakes versus Levinsohn-Petrin . . . . . . . . . . 28 

ix

x Contents 

1.3.6 Extensions of the Olley-Pakes methodology . . . . . . . 30 

1.3.7 Summary of estimation algorithms . . . . . . . . . . . 33 

1.4 Empirical application: Food and beverages industry in Belgium 34 

1.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 

2 Globalization Drives Strategic Product Switching 53 

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 

2.2 Industry dynamics in Estonia . . . . . . . . . . . . . . . . . . 56 

2.3 Determinants of firm dynamics . . . . . . . . . . . . . . . . . 63 

2.3.1 Firm characteristics . . . . . . . . . . . . . . . . . . . . 64 

2.3.2 Product market characteristics: Domestic . . . . . . . 67 

2.3.3 Product market characteristics: International . . . . . 68 

2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 

2.4.1 Baseline results . . . . . . . . . . . . . . . . . . . . . . 72 

2.4.2 Self-selection into new markets . . . . . . . . . . . . . 77 

2.5 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . 84 

2.5.1 Results by size class . . . . . . . . . . . . . . . . . . . 84 

2.5.2 Results by time period . . . . . . . . . . . . . . . . . . 87 

2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 

2.A Data appendix . . . . . . . . . . . . . . . . . . . . . . . . . . 91 

2.A.1 Data and sample selection . . . . . . . . . . . . . . . . 91 

2.A.2 Definitions of variables . . . . . . . . . . . . . . . . . . 93 

3 Footloose Multinationals in Belgium? 101 

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 

3.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . 103 

3.3 Data and preliminary facts . . . . . . . . . . . . . . . . . . . . 106 

3.4 Empirical model . . . . . . . . . . . . . . . . . . . . . . . . . . 114 

3.5 Empirical results . . . . . . . . . . . . . . . . . . . . . . . . . 117 


3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 


3.A.1 Sample selection . . . . . . . . . . . . . . . . . . . . . 125 

3.A.2 Definition of variables . . . . . . . . . . . . . . . . . . 126

Contents xi 

4 Multinational firms, Research Effort and Innovative Output: 

An Integrated Approach 129 

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 

4.2 The innovation production function . . . . . . . . . . . . . . . 133 

4.3 Data and empirical facts . . . . . . . . . . . . . . . . . . . . . 141 

4.4 Empirical results . . . . . . . . . . . . . . . . . . . . . . . . . 146 

4.4.1 Preliminary evidence . . . . . . . . . . . . . . . . . . . 147 

4.4.2 The innovation production function: baseline results . 149 

4.4.3 High-tech versus low-tech sectors . . . . . . . . . . . . 152 

4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 


4.A.1 Sample selection . . . . . . . . . . . . . . . . . . . . . 159 

4.A.2 Definitions of variables . . . . . . . . . . . . . . . . . . 159 

Bibliography 163 

Doctoral Dissertations Faculty of Business and Economics 177

List of Figures 

1.1 Weighted productivity index: Comparison estimation methods 45 

2.1 Sample size distribution . . . . . . . . . . . . . . . . . . . . . 85 

3.1 Foreign multinationals by country of origin . . . . . . . . . . . 108 

3.2 Kaplan-Meier survival functions by nationality of ownership . 111 

xiii

List of Tables 

1.1 TFP estimation: Summary of methodological issues . . . . . . 20 

1.2 TFP estimation: Summary of estimation algorithms . . . . . . 35 

1.3 Summary statistics of key variables . . . . . . . . . . . . . . . 38 

1.4 Production function estimates . . . . . . . . . . . . . . . . . . 39 

1.5 Production function estimates: Variety-specific demand . . . . 43 

1.6 Decomposition aggregate TFP: De Loecker methodology . . . 47 

1.7 Comparison of decomposition results . . . . . . . . . . . . . . 49 

2.1 Exits and industry switches, 1997-2004 . . . . . . . . . . . . . 58 

2.2 Sector distribution . . . . . . . . . . . . . . . . . . . . . . . . 59 

2.3 Four-digit product switches decomposed . . . . . . . . . . . . 61 

2.4 Destination of product switches by technology class . . . . . . 62 

2.5 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . 71 

2.6 Baseline specification . . . . . . . . . . . . . . . . . . . . . . . 74 

2.7 Product switching versus industry switching . . . . . . . . . . 78 

2.8 Industry switching: Manufacturing versus services . . . . . . . 80 

2.9 Unit value difference between industry of origin and destination 83 

2.10 Determinants of firm dynamics across size categories . . . . . 86 

2.11 Determinants of firm dynamics across time . . . . . . . . . . . 89 

2.A.1Sector classification: Manufacturing . . . . . . . . . . . . . . . 97 

2.A.2Sector classification: Services . . . . . . . . . . . . . . . . . . 98 

3.1 Sector distribution of firms . . . . . . . . . . . . . . . . . . . . 109 

3.2 Summary statistics by ownership type (1996-2001) . . . . . . . 112 

3.3 Regression results: Cox proportional hazard model (1996-2001) 118 


xv

xvi List of Tables 

4.1 Sector distribution of firms by firm type . . . . . . . . . . . . 142 

4.2 Sample distribution according to global engagement status . . 144 

4.3 Summary statistics by firm type . . . . . . . . . . . . . . . . . 145 

4.4 Innovation production function: Preliminary evidence . . . . . 148 

4.5 Innovation production function: Two-stage estimation . . . . . 150 

4.6 Innovation production function: High-technology sectors . . . 153 

4.7 Innovation production function: Low-technology sectors . . . . 155 

4.A.1Sector classification: High-tech versus low-tech . . . . . . . . . 161

General Introduction 

Globalization is traditionally associated with increasing integration of the 

world economy, both in terms of international trade and foreign direct investment. 

Technological advances, combined with the continuing liberalization of 

capital and goods markets, as well as the opening up of a number of “new” 

markets such as India or China, have led to increased trade and financial 

flows, both within the EU and the world (EU, 2005). 

Several trends can illustrate this evolution. First, in 2004, international 

trade flows increased by 9 percent, compared to a rise of 4 percent for world 

GDP (WTO, 2005). Second, FDI stocks have, in spite of the slowdown in 

flows since 2000, more than doubled since the 1980s and accounted for more 

than 20 percent of global GDP in 2005 (UNCTAD, 2004). Moreover, between 

1990 and 2003, sales and exports of foreign affiliates have tripled, while employment 

has more than doubled. By comparison, GDP has only grown by 

about 60 percent overall over the same period. According to UNCTAD estimates, 

around 61,000 multinational companies worldwide, together with 

their 900,000 foreign affiliates are responsible for this international production. 

Global international production is estimated to account for about 10 

percent of GDP and 30 percent of world exports (UNCTAD, 2004). 

International economic integration brings gains to the world economy through 

lower prices for consumers and firms, increased variety and potential efficiency 

gains in production (Bernard, Jensen, Redding and Schott, 2007c). 

According to the EU (2005), about 20 percent of the increase in living standards 

over the past 50 years in the EU is attributable to its deeper integration 

in the world economy. These aggregate data mask important heterogeneity 

among groups in the economy however, and specifically, among workers and 

1

2 General Introduction 

firms. As a consequence, while economists generally agree that the overall 

impact of globalization on welfare is positive, public concerns largely remain. 

While high-skilled labor in developed countries has been generally shown 

to benefit from the increased internationalization of economic activity, the 

lower-skilled have been faced with lower relative wages (in the United States) 

and fewer employment opportunities (in Europe) 1 . Similarly, while trade 

liberalization has been shown to increase aggregate industry productivity 

growth, this masks important differences between (heterogeneous) firms. 

Empirical analyses of trade liberalization at the firm level provide evidence 

that this aggregate productivity growth is driven by the expansion and entry 

into export markets of high-productivity firms, as well as by the contraction 

and exit of low-productivity firms. 

This thesis links in with a recent and growing literature dealing with what 

has been labeled by Greenaway (2004) as firm-level adjustment to globalization. 

This literature has been motivated by a wealth of empirical work 

documenting firm heterogeneity in international trade, starting out with the 

seminal contribution of Bernard and Jensen (1995), who were among the 

first to note the important differences between exporters and non-exporters 

in US manufacturing sectors. In response to these empirical findings, several 

theoretical contributions have emerged in recent years taking into account 

firm-level heterogeneity in open-economy models (Bernard et al., 2007c). Examples 

of such models include Melitz (2003), Melitz and Ottaviano (2008) 

and Costantini and Melitz (2007). The growing availability of firm- and 

worker-level data sets, as well as the development of more powerful econometric 

tools have also been instrumental as driving forces in this growing 

strand of literature (Greenaway, 2004). 

As noted by Greenaway (2004), there are several dimensions to firm-level 

adjustment to globalization. A first dimension is concerned with the link between 

globalization and firm-level performance, i.e. its productivity. Melitz 

(2003) demonstrates that, in the presence of imperfect competition and het- 

1 For a comprehensive survey on the impact of international trade on wages, with a 

focus on the US case, I refer to Feenstra (2000).

erogeneous firms, trade liberalization will lead to an increase in a country’s 

imports and will hence erode domestic firms’ market shares. This will cause 

firms at the high end of the productivity distribution to expand into export 

markets (by more than their reduction in domestic sales), while low productivity 

firms will be forced to contract or exit. 

An important issue in the vast number of studies dealing with the impact 

of international trade on firm-level performance concerns the measurement 

of total factor productivity (TFP) at the firm level. Chapter 1 in this thesis 

aims to contribute to this literature by providing a practical overview of the 

methodological issues that arise when estimating total factor productivity at 

the establishment level, as well as of the existing (parametric and semiparametric) 

techniques designed to overcome them. Apart from the well-known 

simultaneity and selection bias; attention is given to methodological issues 

that have emerged more recently and that are related to the use of deflated 

values of inputs and outputs (as opposed to quantities) in estimating productivity 

at the firm level, as well as to the endogeneity of product choice. Using 

data on single-product firms active in the Belgian food and beverages sector, 

the biases introduced in traditional TFP estimates are illustrated and the 

performance of a number of alternative estimators that have been proposed 

in the literature is discussed. 

Recent theoretical and empirical work has uncovered a different dimension 

of firm-level adjustment to globalization. In two influential papers; Bernard, 

Redding and Schott (2005, 2006b) point to the importance of taking firms’ 

product-level decisions explicitly into account when analyzing firm dynamics. 

Bernard, Jensen and Schott (2006a) and Greenaway, Gullstrand and 

Kneller (2008) build on this work to further explore the determinants of 

firms’ product-level strategies. A central result emerging from this literature 

is that, apart from the choice between continuing its production or exiting, 

firms have other strategies available to them to respond to increasing globalization 

pressures. Chapter 2 of this thesis links in with this small but growing 

literature investigating the impact of trade liberalization on firm strategies 

and specifically, on the choice between continuing its production, switching 

products or exiting from the market. 

3


Specifically, in Chapter 2, which is joint work with Marialuz Moreno Badia 

(IMF) and Veerle Slootmaekers (OECD), we use firm-level data for Estonia 

between 1997 and 2005 to analyze the impact of international competition 

on firm dynamics, considering both firm closedown and product switching. 

The analysis contributes to the literature in two important ways: (1) this 

is the first paper to study the determinants of exit and product switching 

in an emerging market; and (2) we consider explicitly the role of export 

opportunities. Results indicate that globalization does not affect firm exit 

significantly while it is an important factor explaining product switching. 

Previous studies on industrial countries have shown that product switching 

has been a defensive strategy against low-cost imports. In contrast, results 

obtained in Chapter 2 suggest that Estonian firms have switched products as 

an offensive strategy to take advantage of the export opportunities created 

by trade liberalization. 

Perhaps the most widely researched strand of the literature dealing with 

firm-level adjustment to globalization concerns the activities of multinational 

firms and the consequences of their presence in industries and countries. 

While most of the literature in this line of research has focused on the identification 

of potential spillovers of these firms to the domestic economy, a small 

literature has focused on these firms’ potential “footloose” nature. Given the 

vast amounts of resources spent by governments worldwide to attract multinational 

firms (Greenaway, 2004), this question is particularly relevant from 

a policy perspective. Chapter 3 in my thesis links in with this literature. 

In Chapter 3, firm-level panel data for the years 1996-2001, covering all 

sectors of the economy, is used to estimate the impact of multinational ownership 

on the exit decisions of firms located in Belgium. In the analysis, I 

clearly distinguish for nationality of ownership, allowing for differences between 

firms that are foreign-owned and multinationals rooted in the domestic 

economy. Controlling for various firm- and industry-specific factors, it is 

found that while foreign multinationals are more likely to shut down operations 

compared to national firms in both manufacturing and service sectors, 

domestic multinationals only exhibit significantly higher exit rates in the 

manufacturing industries. The analysis has important policy implications,

especially in terms of the desirability of the large impact of multinational 

firms on employment and output generation in Belgium. 

While the literature on the relationship between firms’ international activities 

and their productivity abounds 2 , it remains largely silent on the sources 

of the productivity advantages associated with firms’ global integration. One 

way for firms to achieve productivity increases, is through the accumulation 

of knowledge or innovative output (Castellani and Zanfei, 2007). Chapter 4 

of this volume links in with this recent body of literature. 

Specifically, Chapter 4 investigates whether firms’ innovative output differs 

according to their global engagement status. The chapter relies on the innovation 

production function framework introduced by Mairesse and Mohnen 

(2002) and takes the endogeneity of research inputs specifically into account 

in the innovation output function. As a consequence, I am able to distinguish 

between the indirect (through higher research spending) and direct effect of 

firms’ global integration on its innovative output. 

From a policy perspective, the analysis in Chapter 4 yields several important 

insights. Overall, the empirical results indicate that (part of) the 

innovation premium associated with firms’ global engagement status in Belgium 

can be traced back to these firms’ different spending patterns, i.e. their 

higher likelihood to engage in continuous R&D spending. However, within 

the high technology sectors, firms with exposure on international markets 

(whether through exporting or FDI) are found to be more likely to generate 

innovative sales, conditional on their R&D spending patterns; while in low 

technology sectors, these firms are found to be less likely to be innovative. 

This result suggests that government policies aimed at the attraction of 

(foreign) multinational firms in order to stimulate innovative output (or inputs) 

are more likely to be successful (i.e. will generate a higher rate of 

return) when they are specifically targeted at high technology sectors. For 

the low technology sectors, the high returns to R&D obtained in these sectors, 

2 For reviews of this extensive literature, I refer to Greenaway and Kneller (2007) and 

Wagner (2007). Important theoretical contributions in this field include Melitz (2003) and 

Helpman, Melitz and Yeaple (2004). 

5


especially for the selection stage of the model, suggest that policies targeted 

towards these sectors might be more successful when aimed at stimulating 

research efforts in general, indiscriminate of the global engagement status of 

the firm.

Chapter 1 

Total Factor Productivity 

Estimation: A Practical Review 

1.1 Introduction 

While the origins of total factor productivity analysis can be traced back 

to the seminal paper by Solow (Solow, 1957); recent years have seen a surge 

in both theoretical and empirical studies on total factor productivity (TFP). 

This renewed interest has been driven both by the increasing availability of 

firm-level data, allowing for estimation of TFP at the level of the individual 

establishment (Bartelsman and Doms, 2000); as well as by a number of 

methodological improvements that have emerged from the literature since the 

mid-1990s (Ackerberg, Benkard, Berry and Pakes, 2007, henceforth ABBP). 

Typically, establishment-level productivity studies assume output (usually 

measured as deflated sales or value added) to be a function of the inputs the 

firm employs and its productivity (Katayama, Lu and Tybout, 2005). The 

measure of TFP obtained as the residual in this functional relationship is then 

used to evaluate the impact of various policy measures, such as the extent of 

foreign ownership (eg. Smarzynska Javorcik, 2004), trade liberalization (eg. 

Pavcnik, 2002; Amiti and Konings, 2007; De Loecker, 2007) and antidumping 

protection (eg. Konings, 2008). 

∗ Published as LICOS Discussion Paper 182/2007. 

9

10 Total factor productivity estimation 

However, several methodological issues emerge when TFP is estimated using 

traditional methods, i.e. by applying Ordinary Least Squares (OLS) to 

a balanced panel of (continuing) firms. First, since productivity and input 

choices are likely to be correlated, OLS estimation of firm-level production 

functions introduces a simultaneity or endogeneity problem. Moreover, by 

using a balanced panel, no allowance is made for entry and exit, resulting in a 

selection bias. Although the simultaneity and selection bias are well-known 1 ; 

several other methodological issues have emerged more recently. Specifically, 

the typical practice of proxying for firm-level prices using industry-level deflators 

has been challenged (see for instance Katayama et al., 2005). Moreover, 

Bernard, Redding and Schott (2005) note that firms’ product choices are 

likely to be related to their productivity. 

In response to these methodological issues, several (parametric and semiparametric) 

estimators have been proposed in the literature. However, traditional 

estimators used to overcome endogeneity issues (i.e. fixed effects, 

instrumental variables and Generalized Method of Moments or GMM) have 

not proved satisfactory for the case of production functions. Likely causes 

for these estimators’ poor performance are related to their underlying assumptions. 

Therefore, a number of semiparametric alternatives have been 

proposed. Both Olley and Pakes (1996, henceforth OP) and Levinsohn and 

Petrin (2003, henceforth LP) have developed a semiparametric estimator that 

addresses the simultaneity bias (and the selection bias in the case of the OP 

estimator). Several extensions to their model have already been introduced 

(eg. De Loecker, 2007). 

The present paper aims to provide empirical researchers with an overview 

of the methodological issues that arise when estimating TFP at the establishment 

level, as well as of the existing techniques designed to overcome them. 

Using data on single-product firms active in the Belgian food and beverages 

sector, I illustrate the biases introduced in traditional TFP estimates and discuss 

the performance of a number of alternative estimators that have been 

used in the literature. The food and beverages industry in Belgium presents 

1 They have been documented at least since Marschak and Andrews (1944) and Wed- 

ervang (1965) respectively.

Introduction 11 

an interesting case, since the sector underwent significant restructuring at 

the end of the 1990s following the dioxin crisis 2 . 

The production function coefficients obtained using various estimation 

techniques (i.e. OLS, fixed effects, GMM, Olley-Pakes, Levinsohn-Petrin 

and De Loecker) are generally in line with theoretical predictions. Aggregate 

productivity growth in the food and beverages industry increases significantly 

after 1999, consistent with the period of restructuring and increasing 

investments in the sector in response to the dioxin scandal (VRWB, 2003). 

Decomposing aggregate productivity into a within productivity component 

and a reallocation share on the basis of firms’ turnover shares shows that 

this increase is mainly due to the average firm becoming more productive; 

while reallocation of market shares explains only a minor part. Applying 

the same decomposition using employment rather than output shares yields 

similar results. 

The rest of the paper is structured as follows. Section 1.2 provides an 

overview of the methodological issues arising when estimating firm-level TFP. 

In section 1.3, several estimation methods are reviewed, with specific attention 

for their advantages and drawbacks. Section 1.4 illustrates the different 

methodologies for the Belgian food and beverages industry (NACE 15). Section 

1.5 concludes. 

Given the vast amount of papers that continue to emerge in this field, a 

number of choices have to be made at the outset. First, since primary interest 

is in consistent estimation of TFP, attention will mostly be limited to recent 

papers, i.e. from 1990 onwards. Second, only parametric and semiparametric 

estimators applied to TFP estimation will be discussed here. Van 

Biesebroeck (2007) provides an excellent review of several non-parametric 

methods (specifically, index numbers and data envelopment analysis) 3 used 

2 The Economist (1999). Excessive concentrations of dioxin were found in eggs, chicken, 

pork and milk, caused by contaminated animal food. 

3 Van Biesebroeck (2007) compares the robustness of five commonly used techniques 

to estimate TFP: index numbers, data envelopment analysis, stochastic frontiers, GMM 

and semiparametric estimation; to the presence of measurement error and to differences 

in production technology.


to estimate firm-level productivity. Finally, given the multitude of papers 

dealing with the impact of some policy measure on TFP, it is beyond the 

scope of the present paper to provide a complete review of all the empirical 

work done in this area. Selection of which references to include is therefore 

based on their methodological and econometric contributions to the field. 

1.2 Total factor productivity: Methodologi- 

cal issues 

1.2.1 Some preliminaries on the production function 

I start by assuming that production takes the form of a Cobb-Douglas 

production function. However, as shown by ABBP (2007); estimation methods 

discussed in the next section carry over to other types of production 

functions, provided some basic requirements are met 4 . Specifically, the production 

function looks as follows: 

Yit = AitK βk 

it Lβl it Mβm it 

(1.1) 

where Yit represents physical output of firm i in period t; Kit, Lit and 

Mit are inputs of capital, labor and materials respectively and Ait is the 

Hicksian neutral efficiency level of firm i in period t. 

While Yit, Kit, Lit and Mit are all observed by the econometrician (although 

usually in value terms rather than in quantities, see below), Ait is 

unobservable to the researcher. Taking natural logs of (1.1) results in a linear 

production function, 

yit = β0 + βkkit + βllit + βmmit + εit 

where lower-case letters refer to natural logarithms and 

ln (Ait) = β0 + εit 

4 Variable inputs need to have positive cross-partials with productivity and the value 

of the firm has to be increasing in the amount of fixed inputs used (ABBP, 2007).

Total factor productivity: Methodological issues 13 

. 

While β0 measures the mean efficiency level across firms and over time; 

εit is the time- and producer-specific deviation from that mean, which can 

then be further decomposed into an observable (or at least predictable) and 

unobservable component. This results in the following equation, which will 

serve as the starting point for the rest of this and the next section: 

yit = β0 + βkkit + βllit + βmmit + ωit + u q 

it 

(1.2) 

where ωit represents firm-level productivity5 and u q 

it is an i.i.d. component, 

representing unexpected deviations from the mean due to measurement 

error, unexpected delays or other external circumstances. 

Typically, empirical researchers estimate (1.2) and solve for ωit. Estimated 

productivity can then be calculated as follows: 

ˆωit = yit − ˆ βkkit − ˆ βllit − ˆ βmmit 

(1.3) 

and productivity in levels can be obtained as the exponential of ˆωit, i.e. 

ˆΩit = exp (ˆωit). The productivity measure resulting from (1.3) can be used to 

evaluate the influence and impact of various policy variables directly at the 

firm level; or alternatively, firm-level TFP can be aggregated to the industry 

level by calculating the share-weighted average of ˆ Ωit. 

Weights used to aggregate firm-level TFP can be either firm-level output 

shares, as in OP; or employment shares, as in De Loecker and Konings 

(2006). As will be illustrated in section 1.4, normalized industry productivity 

can then be further decomposed into an unweighted average and a (crosssectional) 

sample covariance term. While differences in the unweighted average 

over time refer to within-firm changes in TFP; changes in the sample 

covariance term signal reallocation of market shares as the driver of productivity 

shifts (Olley and Pakes, 1996; De Loecker and Konings, 2006). 

5 The productivity term is identified through the assumption that ωit is a state variable 

in the firm’s decision problem (i.e. it is a determinant of both firm selection and input 

demand decisions), while u q 

it is either measurement error or a non-predictable productivity 

shock (Olley and Pakes, 1996).


In what follows, it will be shown that estimating (1.2) using OLS leads to 

biased productivity estimates, caused by the endogeneity of input choices and 

selection bias. Moreover, in the presence of imperfect competition in output 

and/or input markets, an omitted variable bias will arise in standard TFP 

estimation if data on physical inputs and output and their corresponding 

firm-level prices are unavailable. Finally, if firms produce multiple products, 

potentially differing in their production technology; failure to estimate the 

production function at the appropriate product level, rather than at the firm 

level, will also introduce a bias in standard TFP measures. I will discuss 

each of these problems in turn. 

1.2.2 Endogeneity of attrition or selection bias 

Traditionally, entry and exit of firms is accounted for in TFP estimation 

by constructing a balanced panel; i.e. by omitting all firms that enter or exit 

over the sample period (Olley and Pakes, 1996). However, several theoretical 

models (eg. Jovanovic, 1982; Hopenhayn, 1992) predict that the growth and 

exit of firms is motivated to a large extent by productivity differences at 

the firm level. Empirically, Fariñas and Ruano (2005) find, for a sample of 

Spanish manufacturing firms, that entry and exit decisions are systematically 

related to differences in productivity. They show that firms’ exit patterns 

reflect initial productivity differences, leading to the prediction that higher 

productivity will lower the exit probability at the firm level. 

Moreover, Dunne et al. (1988) report exit rates in excess of 30 percent 

between two census years in US manufacturing and a strong correlation between 

entry and exit rates in the data. Since low productivity firms have a 

stronger tendency to exit than their more productive counterparts, omitting 

all firms subject to entry or exit is likely to lead to biased results. If firms 

have some knowledge about their productivity level ωit prior to their exit, 

this will generate correlation between εit and the fixed input capital, conditional 

on being in the data set (ABBP, 2007). This correlation has its origin 

in the fact that firms with a higher capital supply will (ceteris paribus) be 

able to withstand lower ωit without exiting.


In sum, the selection bias or “endogeneity of attrition”- problem will generate 

a negative correlation between εit and Kit, causing the capital coefficient 

to be biased downwards in a balanced sample (i.e. not taking entry and exit 

into account). While this selection problem has been discussed in the literature 

at least since the work of Wedervang (1965), the estimation algorithm 

introduced by Olley and Pakes (1996) was the first to take it explicitly 6 into 

account. 

1.2.3 Endogeneity of input choice or simultaneity bias 

Although (1.2) can be estimated using Ordinary Least Squares (OLS), 

this method requires that the inputs in the production function are exogenous 

or, in other words, determined independently from the firm’s efficiency 

level. Marschak and Andrews (1944) already noted that inputs in the production 

function are not independently chosen, but rather determined by 

the characteristics of the firm, including its efficiency. This “endogeneity of 

inputs” or simultaneity bias is defined as the correlation between the level of 

inputs chosen and unobserved productivity shocks (De Loecker, 2007). 

If the firm has prior knowledge of ωit at the time input decisions are made, 

endogeneity arises since input quantities will be (partly) determined by prior 

beliefs about its productivity (Olley and Pakes, 1996; ABBP, 2007). Hence, 

if there is serial correlation in ωit, a positive productivity shock will lead to 

increased variable input usage; i.e. E (xitωit) > 0 , where xit = (lit, mit); 

introducing an upward bias in the input coefficients for labor and materials 

(De Loecker, 2007). In the presence of many inputs and simultaneity 

issues, it is generally impossible to determine the direction of the bias in the 

capital coefficient. Levinsohn and Petrin (2003) illustrate, for a two-input 

production function where labor is the only freely variable input and capital 

is quasi-fixed, that the capital coefficient will be biased downward if a positive 

correlation exists between labor and capital (which is the most likely 

setup). 

6 It is possible to correct implicitly for the selection bias by using an unbalanced panel of 

firms. But, as will be shown in section 1.3, Olley and Pakes (1996) introduce an additional 

(explicit) correction in their estimation algorithm, i.e. they take the firm-level survival 

probability into account.


Traditional methods to deal with heterogeneity and endogeneity issues include 

fixed effects and instrumental variables estimation (Wooldridge, 2005). 

However, as I will discuss below, both alternatives to OLS are plagued by 

a number of problems. Both the estimation algorithm introduced by Olley 

and Pakes (1996) and Levinsohn and Petrin (2003) provide a more adequate 

solution to the simultaneity problem discussed here. 

1.2.4 Omitted output price bias 

In the absence of information on firm-level prices, which are typically unavailable 

to the researcher, industry-level price indices are usually applied 

to deflate firm-level sales (and hence obtain a measure of the firm’s output) 

in traditional production function estimates (De Loecker, 2007). However, if 

firm-level price variation is correlated with input choice; this will result in biased 

input coefficients. The problem can be illustrated as follows. Replacing 

output in quantities by deflated sales in (1.2) results in the following model: 

rit = pit + yit − p it 

= β0 + βkkit + βllit + βmmit + (pit − p it) + ωit + u q 

it 

(1.4) 

where rit represents deflated sales, pit are firm-level prices and p it is 

the industry-level price deflator, all in logarithmic form. For now, inputs 

are still assumed to be available in quantities. From (1.4) it is clear that 

if input choice is correlated with unobserved firm-level price differences, i.e. 

E (xit (pit − p it)) = 0, where xit = (lit, mit, kit); a bias is introduced in the 

input coefficients. 

Assuming inputs and output are positively correlated and output and price 

are negatively correlated (as in a standard demand and supply framework); 

the correlation between (variable) inputs and firm-level prices will be negative; 

resulting in a negative bias for the coefficients on labor and materials 

(De Loecker, 2007). Hence, the bias resulting from using industry-level price 

deflators rather than firm-level prices to deflate sales, will generally be opposite 

to the bias introduced by simultaneity of input choice and productivity 

discussed in the previous section.


Since the omitted output price bias will only arise if industry-level price 

deflators are used and if firm-level prices deviate from these deflators (i.e. in 

the presence of imperfect competition), it can be avoided by using quantities 

of output rather than deflated sales. However, since this requires information 

on actual firm level prices, it is a very rare setup. Exceptions include 

Dunne and Roberts (1992), Eslava, Haltiwanger, Kugler and Kugler (2004), 

Foster, Haltiwanger and Syverson (2008), Jaumandreu and Mairesse (2004) 

and Mairesse and Jaumandreu (2005). Alternatively, it is possible (in the 

absence of information on prices) to introduce demand for output into the 

system and solve for firm-level prices 7 , as suggested by Klette and Griliches 

(1996), Levinsohn and Melitz (2002) and, in the context of the Olley-Pakes 

semiparametric estimator, De Loecker (2007). The specifics of the latter 

estimation algorithm will be discussed in section 1.3. 

1.2.5 Omitted input price bias 

In the presence of imperfect competition in input markets, input prices 

are likely to be firm-specific. However, since input prices (like output prices) 

are typically unavailable, quantities of inputs are usually proxied by deflated 

values of inputs for capital and materials (the amount of labor used tends to 

be available in annual accounts data commonly used to estimate production 

function relationships). Assuming that quantities of output are given, this 

leads to the following relationship: 

yit = β0 + βk kit + βllit + βm mit + ωit + u q 

it 

 

kit + p k it − pk 

it + βllit 

+ βm (mit + p m it − pmit ) + ωit + u q 

it 

yit = β0 + βk 

(1.5) 

where kit and mit are deflated values of capital and material inputs 

respectively, pk it and pm it represent firm-level prices of these inputs and pk it and 

pm it refer to their industry-level price indices. From (1.5) it is clear that in 

the presence of unobserved firm-level input price differences, coefficients on 

7 Ornaghi (2006) criticizes this approach however, see section 1.3 below.


kit and mit will be biased. 

De Loecker (2007) argues that if imperfect output markets are treated explicitly, 

this can partly take care of the omitted input price bias, to the extent 

that higher input prices are reflected in higher output prices; which in turn 

depends on the relevant firm-level mark-up. However, Levinsohn and Melitz 

(2002) argue that even with a competitive factor market, adjustment costs 

will lead to differences in the shadow price of the input index across firms, 

induced by differences in current levels of the quasi-fixed factors (capital). 

Katayama et al. (2005) similarly argue that factor prices are important to 

take into account in TFP estimation procedures. 

Similar to the situation of imperfect competition in output markets, a 

number of studies are able to exploit information on input prices and quantities 

to resolve the omitted price bias, examples include Eslava et al. (2004) 

and Ornaghi (2006). A formal solution for the bias induced by firm-specific 

input prices has yet to be introduced. 

1.2.6 Endogeneity of the product mix (multi-product 

firms) 

Bernard, Redding and Schott (2005, henceforth BRS) argue that firms’ 

decisions on which goods to produce, are typically made at a more disaggregated 

level than is recorded in manufacturing data sets (either using census 

data or annual accounts data). If firms produce multiple products within 

the same industry and if these products differ in their production technology 

or in the demand they face, this will lead to biased TFP estimates, since 

the production function assumes identical production techniques and final 

demand (through the use of common output price deflators) across products 

manufactured by a single firm. 

BRS (2006b) have examined the pervasiveness and determinants of product 

switching among US manufacturing firms for the period 1972-1992. They 

find that two-thirds of firms alter their mix of five-digit SIC codes every five 

years and they further demonstrate that product adding and dropping by


surviving firms accounts for nearly one-third of the aggregate growth in real 

US manufacturing output between 1992 and 1997. 

In principle, consistent estimation of TFP in the presence of multi-product 

firms requires information on the product mix, product-level output, inputs, 

as well as prices. Given these high requirements in terms of data, BRS (2005) 

suggest several (partial) solutions to circumvent the bias introduced by multiproduct 

firms. First, in the absence of information on inputs and outputs at 

the product level, it is possible to sort firms into groups that make a single 

product, which will eliminate the bias introduced by endogenous product 

choice. Alternatively, if the researcher has knowledge of the number and 

type of products produced by each firm, consistent estimates of productivity 

can be obtained by allowing the parameters of the production technology to 

vary across firms making different products. De Loecker (2007) is among the 

first to take the number of products as well as product-specific demand into 

account when estimating TFP for the Belgian textiles sector. However, his 

estimation procedure provides only a partial solution to the bias introduced 

by endogenous product choice (cfr. section 1.3). 

1.2.7 Summary of methodological issues 

Traditional productivity estimates, obtained as the residual from a balanced 

OLS regression of deflated output on deflated inputs and a constant, 

are plagued by a number of econometric and specification issues. Table 1.1 

provides an overview. 

First, given the prevalence of entry and exit in manufacturing populations, 

the use of a balanced panel introduces a selection bias in the sample, causing 

the capital coefficient to be biased downward. Second, if firms have some 

prior knowledge or expectations concerning their efficiency, current input 

choice will be correlated with productivity. Coefficients on variable inputs 

will be biased upward as a result of this endogeneity or simultaneity problem, 

while the coefficient on capital will be biased downward provided the 

correlation between labor and capital is positive.


Table 1.1: TFP estimation: Summary of methodological issues 

Multi-product firms Endogenous product choice: undetermined Bernard, Redding, Schott (2005) 

Differences in production technologies across Bernard, Redding, Schott (2006b) 

products produced by single firm. De Loecker (2007) 

Omitted input Imperfect competition in input markets: downward bias in βl Levinsohn and Melitz (2002) 

price bias Correlation between firm-level deviation of downward bias in βm 

Katayama et al. (2005) 

input price deflators 

and inputs xit. upward bias in βk De Loecker (2007) 

p k,m 

it 

− pk,m it 

Selection bias Endogeneity of attrition: downward bias in βk Wedervang (1965) 

Correlation between εit and Kit (the quasi-fixed Olley and Pakes (1996) 

input), conditional on being in the data set. ABBP (2007) 

Simultaneity bias Endogeneity of inputs: upward bias in βl Marschak and Andrews (1944) 

Correlation between εit and inputs xit if firms’ upward bias in βm Olley and Pakes (1996) 

prior beliefs about εit influence its choice of inputs. downward bias in βk Levinsohn and Petrin (2003) 

ABBP (2007) 

Ackerberg et al. (2006) 

Omitted output Imperfect competition in output markets: downward bias in βl Klette and Griliches (1996) 

price bias Correlation between firm-level deviation of downward bias in βm Levinsohn and Melitz (2002) 

output price deflator (pit − pit) and inputs xit. upward bias in βk De Loecker (2007) 

Origin of the bias Definition Direction of the bias References

Total factor productivity estimation 21 

Third, in the presence of imperfect competition in input and/or output 

markets, the failure to take firm-level deviations from the industry-level price 

deflator into account will result in an omitted output and/or input price bias. 

The resulting bias(es) will, in a standard demand/supply framework, work 

in the opposite direction as the simultaneity bias, rendering any prior on the 

overall direction of the bias hard. Finally, if firms produce multiple products, 

which potentially differ in terms of their production technology and demand, 

an additional bias will be introduced in traditional TFP estimates. I now 

turn to the various estimators that have been introduced in the literature on 

consistent estimation of total factor productivity. 

1.3 Total factor productivity estimation 

1.3.1 Fixed effects estimation 

By assuming that ωit is plant-specific, but time-invariant; it is possible to 

estimate (1.2) using a fixed effects estimator (Pavcnik, 2002; Levinsohn and 

Petrin, 2003). The estimating equation then becomes: 

yit = β0 + βkkit + βllit + βmmit + ωi + u q 

it 

(1.6) 

Equation (1.6) can be estimated in levels using a Least Square Dummy 

Variable Estimator (LSDV, i.e. including firm-specific effects) or in first (or 

mean) differences. Provided unobserved productivity ωit does not vary over 

time, estimation of (1.6) will result in consistent coefficients on labor, capital 

and materials. 

Fixed effects or within estimators have a long tradition in the production 

function literature, in fact they were introduced to economics in this context 

(Mundlak, 1961; Hoch, 1962). By using only the within-firm variation in the 

sample, the fixed effects estimator overcomes the simultaneity bias discussed 

in the previous section (ABBP, 2007). Moreover, to the extent that exit 

decisions are determined by the time-invariant, firm-specific effects ωi, and 

, the within estimator also eliminates the selection bias, caused by 

endogenous exit in the sample. As a result, estimation of (1.6) using either 

not by u q 

it


the balanced or unbalanced (i.e. allowing for entry and exit) sample should 

result in similar estimates for the coefficients. 

In spite of the attractive properties of the fixed effects estimator, it does 

not perform well in practice (ABBP, 2007). Estimation of (1.6) often leads 

to unreasonably low estimates of the capital coefficient. Moreover, Olley and 

Pakes (1996) perform fixed effects on the balanced and unbalanced sample 

and find large differences between the two sets of coefficients, suggesting the 

assumptions underlying the model are invalid. The time-invariant nature of 

ωi in the fixed effects model has been relaxed by Blundell and Bond (1999) 

in the context of production functions, by allowing productivity to be decomposed 

into a fixed effect and an autoregressive AR(1)-component. 

1.3.2 Instrumental variables (IV) and GMM 

An alternative method to achieve consistency of coefficients in the production 

function is by instrumenting the independent variables that cause 

the endogeneity problems (i.e. the inputs in the production function) by 

regressors that are correlated with these inputs, but uncorrelated with unobserved 

productivity. To achieve consistency of this IV estimator, three 

requirements have to be met (ABBP, 2007). First, instruments need to be 

correlated with the endogenous regressors (inputs). Second, the instruments 

can not enter the production function directly and finally, instruments need 

to be uncorrelated with the error term. 

Assuming input and output markets operate perfectly competitive, input 

and output prices are natural choices of instruments for the production function 

(ABBP, 2007). Other examples of instruments include variables that 

shift the demand for output or the supply of inputs. Like the fixed effects 

estimator, the IV estimator has not been particularly successful in practice. 

One of the obvious shortcomings of the technique is the lack of appropriate 

instruments in many data sets. Input and output prices are usually not 

reported in typical plant or firm level data sets and if they are reported, frequently 

not enough variation exists in the data in order to identify coefficients 

of the production function (ABBP, 2007). Moreover, while estimation using 

IV techniques overcomes the simultaneity bias (provided the instruments are


appropriate), it does not provide a solution for the selection issues. If input 

prices are used as instruments for input quantities and if exit decisions are 

driven (in part) by changes in these input prices, results will remain biased. 

In response to these unsatisfactory results, Blundell and Bond (1999) propose 

an extended GMM estimator. They attribute the bad performance of 

standard IV estimators to the weak instruments used for identification, i.e. 

lagged levels of variables are often used as instruments in the estimation 

in first differences. They propose an extended GMM estimator using lagged 

first-differences of the variables as instruments in the level equations and find 

that this estimator yields more reasonable parameter estimates. As already 

noted above, they also stress the importance of allowing for an autoregressive 

component in ωit. 

1.3.3 Olley-Pakes estimation algorithm 

As an alternative to the methods discussed above; Olley and Pakes (1996) 

have developed a consistent semiparametric estimator. This estimator solves 

the simultaneity problem by using the firm’s investment decision to proxy for 

unobserved productivity shocks. Selection issues are addressed by incorporating 

an exit rule into the model. In what follows, the proposed methodology 

will be discussed in somewhat more detail. It should be noted here however, 

that the focus in this section is on the estimation methodology. For the more 

technical aspects (and related proofs), the interested reader is referred to 

Ericson and Pakes (1995) and Olley and Pakes (1996). 

Olley and Pakes (1996) were the first to introduce an estimation algorithm 

that takes both the selection and simultaneity problem explicitly into 

account. They develop a dynamic model of firm behavior that allows for 

idiosyncratic productivity shocks, as well as for entry and exit. At the start 

of each period, each incumbent firm decides whether to exit or to continue 

its operations. If it exits, it receives a particular sell-off value and it never 

re-enters. If it continues, it chooses an appropriate level of variable inputs 

and investment. The firm is assumed to maximize the expected discounted 

value of net cash flows and investment and exit decisions will depend on the 

firm’s perceptions about the distribution of future market structure, given


the information currently available. Both the lower bound to productivity 

(i.e. the cut-off value below which the firm chooses to exit) and the investment 

decision are determined as part of a Markov perfect Nash equilibrium 

and will hence depend on all parameters determining equilibrium behavior. 

In order to achieve consistency, a number of assumptions need to be made. 

First, the model assumes there is only one unobserved state variable at the 

firm level, i.e. its productivity. Second, the model imposes monotonicity 

on the investment variable, in order to ensure invertibility of the investment 

demand function. This implies that investment has to be increasing in productivity, 

conditional on the values of all state variables. As a consequence, 

only non-negative values of investment can be used in the analysis. This 

condition needs to hold for at least some known subset of the sample (see 

below). Finally, if industry-wide price indices are used to deflate inputs and 

output in value terms to proxy for their respective quantities , it is implicitly 

assumed that all firms in the industry face common input and output prices 

(Ackerberg, Benkard, Berry and Pakes, 2007). 

Starting out from the basic Cobb-Douglas production function 8 given by 

(1.2), the estimation procedure can be described as follows. Capital is a state 

variable, only affected by current and past levels of ωit. Investment can be 

calculated as: 

Iit = Kit+1 − (1 − δ) Kit 

Hence, investment decisions at the firm level can be shown to depend on 

capital and productivity or iit = it (kit, ωit), where lower-case notation refers 

to logarithmic transformation of variables, as above. Provided investment 

is strictly increasing in productivity, conditional on capital, this investment 

decision can be inverted, allowing us to express unobserved productivity as 

a function of observables: 

ωit = ht (kit, iit) 

where ht (.) = i −1 

t (.). Using this information, (1.2) can be rewritten as: 

8 The production function in (1.2) differs from that employed by OP in two respects. 

First, OP include age as an additional state variable, which is omitted here. Second, OP 

start out from a value added production function, i.e. including only labor and capital as 

production factors.


yit = β0 + βkkit + βllit + βmmit + ht (kit, iit) + u q 

it 

Next, define the function ϕ (iit, kit) as follows: 

ϕ (iit, kit) = β0 + βkkit + ht (iit, kit) 

(1.7) 

Estimation of (1.7) proceeds in two steps (OP, 1996). In the first stage of 

the estimation algorithm, the following equation is estimated using OLS: 

yit = βllit + βmmit + ϕ (iit, kit) + u q 

it 

(1.8) 

where ϕ (iit, kit) is approximated by a higher order polynomial in iit and 

kit (including a constant term). Estimation of (1.8) results in a consistent 

estimate of the coefficients on labor and materials (the variable factors of 

production). 

In order to recover the coefficient on the capital variable, it is necessary 

to exploit information on firm dynamics. Productivity is assumed to follow 

a first order Markov process, i.e. ωit+1 = E (ωit+1|ωit) + ξit+1, where 

ξit+1 represents the news component and is assumed to be uncorrelated with 

productivity and capital in period t+1. As noted above, firms will continue 

to operate provided their productivity level exceeds the lower bound, i.e. 

χit+1 = 1 if ωit+1 ≥ ω it+1, where χit+1 is a survival indicator variable. Since 

the news component ξit+1 is correlated with the variable inputs; labor and 

material inputs are subtracted from the log of output. Considering the expectation 

of E (yit+1 − βllit+1 − βmmit+1), conditional on the survival of the 

firm results in the following expression: 

E [yit+1 − βllit+1 − βmmit+1|kit+1, χit+1 = 1] 

= β0 + βkkit+1 + E [ωit+1|ωit, χit+1 = 1] 

The second stage of the estimation algorithm can then be derived as follows:


yit+1 −βllit+1 − βmmit+1 

= β0 + βkkit+1 + E (ωit+1|ωit, χit+1) + ξit+1 + u q 

it+1 

= β0 + βkkit+1 + g (Pit, ϕt − βkkit) + ξit+1 + u q 

it+1 

(1.9) 

where E (ωit+1|ωit, χit+1) = g (Pit, ϕit − βkkit) follows from the law of 

motion for the productivity shocks and Pit is the probability of survival of 

firm i in the next period 9 , i.e. Pit = Pr {χit+1 = 1}. A consistent estimate 

of the coefficient on capital is obtained by substituting the estimated coefficients 

on labor and materials from the first stage, as well as the estimated 

survival probability in (1.8). As in the first stage of the estimation procedure, 

the function g (Pit, ϕit − βkkit) is approximated using a higher order polynomial 

expansion in Pit and ϕit − βkkit. Finally, this results in the following 

estimating equation: 

yit+1 −βllit+1 − βmmit+1 

 

= β0 + βkkit+1 + g Pit, ϕt − 

βkkit 

+ ξit+1 + u q 

it+1 

(1.10) 

The coefficient on capital can then be obtained by applying Non-Linear 

Least Squares on (1.10). 

1.3.4 Levinsohn-Petrin estimation algorithm 

While Olley and Pakes (1996) use the investment decision to proxy for 

unobserved productivity; Levinsohn and Petrin (2003) rely on intermediate 

inputs as a proxy. The monotonicity condition of OP requires that investment 

is strictly increasing in productivity. Since this implies that only observations 

with positive investment can be used when estimating (1.8) and (1.10), this 

can result in a significant loss in efficiency, depending on the data at hand. 

9 An estimate of Pit can be obtained by estimating a probit model, where the dependent 

variable is a survival dummy (i.e. dummy equal to one if the firm survives in a particular 

period). Left-hand side variables are the same polynomial terms used in the first stage 

of the estimation procedure, i.e. a higher-order polynomial in investment and capital, 

including a constant term. Pit can then be obtained as the predicted survival probability 

from this regression.


Moreover, if firms report zero investment in a significant number of cases, this 

casts doubt on the validity of the monotonicity condition. Hence, Levinsohn 

and Petrin (2003) use intermediate inputs rather than investment as a proxy. 

Since firms typically report positive use of materials and energy in each 

year, it is possible to retain most observations; which also implies that the 

monotonicity condition is more likely to hold. 

Their estimation algorithm differs from that introduced by OP in two important 

respects. First, they use intermediate inputs to proxy for unobserved 

productivity, rather than investment. This implies that intermediate inputs 

(materials in this case) are expressed as a function of capital and productivity, 

i.e. mit = mt (kit, ωit). Provided the monotonicity condition is met and 

materials inputs are strictly increasing in ωit, this function can be inverted, 

again allowing us to express unobserved productivity as a function of observables, 

i.e. ωit = st (kit, mit), where st (.) = m −1 

t (.). Using this expression, it 

is possible to rewrite (1.2), analogous to the OP-approach described above. 

yit = β0 + βkkit + βllit + βmmit + st (kit, mit) + u q 

it 

(1.11) 

It should be noted that the coefficient on the proxy variable, i.e. materials; 

is now only recovered in the second stage of the estimation algorithm, rather 

than in the first as for the OP approach. The second difference between 

the approach used by OP and LP is in the correction for the selection bias. 

While OP allow for both an unbalanced panel as well as the incorporation 

of the survival probability in the second stage of the estimation algorithm, 

LP do not incorporate the survival probability in the second stage; since the 

efficiency gains associated with it in the empirical results presented by OP 

were very small provided an unbalanced panel was used. Apart from using 

materials instead of investment as a proxy and omitting the survival correction 

in the second stage 10 , estimation is fully analogous to the approach used 

by OP and summarized above. Moreover, Petrin, Levinsohn and Poi (2003) 

have developed a Stata program implementing the LP approach (levpet). For 

further details on the LP approach, I refer to LP and Petrin et al. (2003). 

10 In principle it is possible to implement the explicit correction for firm survival in the 

LP estimation algorithm.


1.3.5 Olley-Pakes versus Levinsohn-Petrin 

As was noted above, the OP and LP estimation algorithms are analogous 

apart from the use of different proxies and the in- or exclusion of the survival 

probability to correct for the selection bias. How then, is one to choose 

among the two estimators? I will briefly discuss some of the results emerging 

from the literature here. 

It is useful to start with the most obvious shortcoming of the OP estimation 

algorithm, i.e. the invertibility condition, which implies that only firms with 

positive investment can be included in the analysis. Although consistent 

production function coefficients can be obtained by estimating (1.10) for the 

subset in the sample with recorded positive investment; this implies a loss 

in efficiency and, particularly if there are few firms with positive investment 

flows in the industry, can cast doubt on the monotonicity condition (see 

above). 

Moreover, according to Ackerberg, Caves and Frazer (2006), collinearity 

between labor and the non-parametric terms (i.e. the polynomial in materials 

and capital for LP and in investment and capital for OP) in the first stage of 

the estimation algorithm can cause the labor coefficient to be unidentified. 

This collinearity arises from the fact that labor, like materials and capital, 

needs to be allocated in some way by the firm, at some point in time. While 

this problem can arise in the context of the OP and LP estimator, it is 

particularly problematic for the LP estimator. 

For the LP estimator, since labor and materials are both chosen simultaneously, 

a natural assumption could be that they are allocated in similar 

ways. However, this would imply that labor and materials are both chosen 

as a function of productivity and capital: 

mit = ft (ωit, kit) 

lit = gt (ωit, kit) 

Hence, both labor and materials depend on the same state variables. Using 

the invertibility condition of LP, i.e. ωit = f −1 

t (mit, kit), this leads to the


following result (Ackerberg et al., 2006): 

−1 

lit = gt ft (mit, kit) , kit = ht (mit, kit) 

Since it is not possible to simultaneously estimate a non-parametric function 

of ωit and kit together with the coefficient on the labor variable, which 

is also a function of those same variables; the labor coefficient will not be 

identified in the first stage. Hence collinearity between the labor variable and 

the non-parametric function in the first stage can cause the labor coefficient 

to be unidentified. Ackerberg, Caves and Frazer further investigate to what 

extent plausible assumptions can be made about the data generating process 

for labor in order to “save” the LP first stage estimation, with little success. 

As noted above, this collinearity problem can also arise in the context of the 

OP estimation procedure. However, for the OP estimator, identification of 

the labor coefficient can be achieved by assuming that labor is not a perfectly 

variable input and that firms decide on the allocation of labor without perfect 

information about their future productivity (i.e. investment and labor are 

determined by different information sets). If this assumption holds for the 

data at hand, the labor coefficient can be identified in the first stage of 

the estimation algorithm in the case of OP. For LP, this assumption does 

not solve the collinearity problem, since choosing labor prior to choosing 

material inputs will make the choice of the latter directly dependent on the 

choice of labor inputs, again preventing identification of the labor coefficient 

in the first stage. This difference between the two estimators stems from 

the fact that investment, unlike materials, is not directly linked to period 

t outcomes, so that a firm’s allocation of labor will not directly affect its 

investment decisions (Ackerberg et al., 2006). 

Ackerberg, Caves and Frazer suggest an alternative estimation procedure, 

where the coefficient on labor (in a value added production function) is no 

longer estimated in the first stage of the algorithm. All input coefficients are 

obtained in the second stage, while the first stage only serves to net out the 

error component in the production function.


Moreover, in the presence of imperfect competition in input or output 

markets, consistency of either the OP or LP estimator is likely to break down, 

as an omitted price variable will bias results. Therefore, the OP algorithm has 

been augmented to take imperfect competition in output markets explicitly 

into account (De Loecker, 2007, see below). For LP however, De Loecker 

(2007, Appendix C) shows that imperfect competition in output markets is 

likely to invalidate the invertibility condition, while it has no effect on the 

monotonicity condition of OP. Therefore, even if the LP estimation algorithm 

is augmented with the correction for imperfect competition (discussed below), 

coefficients are likely to be biased. Hence, I will focus on the OP algorithm 

in what follows. 

1.3.6 Extensions of the Olley-Pakes methodology 

Many of the extensions and alternatives that emerge from the literature are 

still work in progress, making it particularly hard to choose among the many 

candidates. For a recent technical review of a number of extensions to the OP 

methodology, I refer to ABBP 11 (2007). Alternatives to the semiparametric 

estimators of OP and LP are proposed by (among others) Katayama et al. 

(2005). However, a full discussion of these works lies beyond the scope of the 

present paper. 

As was noted in section 1.2, De Loecker (2007) implements the correction 

for the omitted output price bias, introduced by Klette and Griliches 

(1996) in the OP estimation algorithm. In what follows, the specifics of his 

model will be discussed. While De Loecker (2007) also introduces a correction 

for multi-product firms, I have elected not to discuss this extension here 

for two reasons. First, in the absence of product-level data on inputs and 

outputs, consistent estimation of TFP can only be obtained by either focusing 

on single-product firms or by allowing the parameters of the production 

technology to vary across firms making different products (BRS, 2005). Although 

De Loecker (2007) is able to exploit information on which products a 

11 ABBP focus on the assumptions underlying the semiparametric estimators introduced 

by OP and LP and show how to test their validity and how to relax some of them; they do 

not treat the bias introduced by endogenous product choice or by imperfect competition 

in input and output markets explicitly.


firm produces, allowing him to introduce product level demand rather than 

industry level demand as well as to control for the number of products a 

firm produces; the production technology is still (necessarily) assumed to be 

identical across products in an industry. 

Moreover, BRS (2006b) find that more than 60 percent of US firms alter 

their product mix every five years. This implies that any information on the 

product space firms are active in, would have to be dynamic in nature 12 . Since 

typical annual accounts data usually provide no or very limited information 

at the relevant product level and given the remaining biases in the resulting 

production function coefficients in the absence of (dynamic) product-level 

data on inputs and outputs, I will restrict attention to single-product firms. 

The relevant model to start from in the presence of imperfect competition 

in the output market is given by (1.4). In order to estimate (1.4) consistently 

without information on establishment-level prices, it is necessary to impose 

some structure on the demand system, which will be used to implicitly solve 

for the firm-level prices. Following De Loecker (2007), I start out from a 

simple conditional (Dixit-Stiglitz) demand system 13 : 

Qit = QJt 

η Pit 

PJt 

exp u d 

it 

where Qit represents demand for the firm’s product, QJt is industry output 

at time t, Pit 

PJt 

is the price of firm i relative to the average price in industry 

J, u d it is an idiosyncratic firm-specific demand shock and η is the elasticity of 

substitution (demand) between differentiated goods in the industry (−∞ < 

η < −1). 

Taking natural logarithms results in the following expression for the 

demand system. 

12 Although De Loecker has very detailed information on which firms are active in which 

sectors, the data are only available for 2001. Hence the firm-level product mix is necessarily 

assumed to be constant over the sample period in his analysis. 

13 The industry is assumed to be characterized by product differentiation. A key characteristic 

of Dixit-Stiglitz demand is that the price (substitution) elasticities are constant 

over time and independent of the number of varieties.


qit = qJt + ηpit − ηpJt + u d it 

(1.12) 

It is possible to derive an expression for pit from (1.12) and substitute the 

result into (1.4). 

pit = 1 

qit − qJt − u 

η 

d 

it + pJt 

rit = pit + yit − pit = 1 

qit − qJt − u 

η 

d 

it + pJt + yit − pit Using the fact that changes in the industry-wide price index p it can be 

considered as a weighted average of the changes in firm-specific prices, i.e. 

p it = pJt , results in the following relationship: 

rit = 

η + 1 

η 

(β0 + βkkit + βllit + βmmit + ωit + u q 1 

it ) − 

η qJt − 1 

η udit (1.13) 

where ωit will be proxied by the investment decision as in section 1.3.3. 

Hence, it is clear from (1.13) that consistent estimation in the presence of 

imperfectly competitive output markets requires adding a term to the production 

function. By putting structure on the demand system, it is possible 

to proxy for unobserved firm-level prices by adding industry output as an 

additional regressor in the production function 14 . Specifically, the final estimating 

equation looks as follows: 

rit = α0 + αkkit + αllit + αmmit + ω ′ 

it + u′ q 

it + αηqJt − u ′ d 

it 

(1.14) 

where αh = ((η + 1) /η) βh for h = 0, l, m, k; ω ′ 

it = ((η + 1) /η) ωit and 

αη = (−1/η). The final production function coefficients can be obtained by 

multiplying the coefficients obtained in (1.14) with the relevant mark-up, i.e. 

η/ (η + 1). Similarly, firm-level productivity is now obtained as follows: 

14 Ornaghi (2006) invalidates the correction suggested by Klette and Griliches by confirming 

the existence of asymmetric biases among the input coefficients introduced by the 

use of deflated values of inputs and outputs rather than observed quantities. Given this 

asymmetric bias, multiplying all input coefficients with an identical upward correction 

term (i.e. the mark-up) as illustrated in (1.15) can not yield unbiased input coefficients.


ˆωit = 

 

ˆη 

ˆω 

ˆη + 1 

′ 

 

ˆη 

it = (˜rit − ˆαkkit − ˆαllit − ˆαmmit − ˆαηqJt) (1.15) 

ˆη + 1 

Hence, for the OP estimator including the correction for market power, 

productivity as obtained in (1.3) additionally needs to be multiplied by the 

relevant mark-up; as shown in (1.15). Although this correction simply implies 

a rescaling of firm-level productivity in this particular case, it is straightforward 

to interact industry output at a more disaggregated level with sector 

dummies at an equal level of aggregation to allow the demand elasticity and 

relevant mark-up to vary across sub-sectors 15 . Allowing the demand elasticity 

to vary across sub-sectors in (1.14) leads to the following estimating 

equation (De Loecker, 2007): 

rit = α0 + αllit + αkkit + αmmit + ω ′ 

it + u′ q 

it + 

M 

s=1 

αηsqJtsIis − u ′ d 

it 

(1.16) 

where s represents the sub-sector and M equals the total number of 

sub-sectors. Iis is a dummy variable equal to 1 if a firm is active in a given 

sub-sector and qJts is the relevant industry demand shifter, proxied by output 

in the different sub-sectors. The number of estimated elasticities ηs equals the 

number of sub-sectors in the industry. Industry output is simply replaced in 

the estimation by M 

αηsqJtsIis. It should be noted that if demand parameters 

i=1 

are allowed to vary across sub-sectors; the resulting production coefficients 

βh will also be specific to those sub-sectors, since the estimates obtained 

from estimating (1.16) have to be transformed using the relevant (sub-sector) 

mark-up. 

1.3.7 Summary of estimation algorithms 

Table 1.2 summarizes the different estimation algorithms discussed in this 

section. While fixed effects and instrumental variables methods are theoretically 

able to solve the simultaneity bias introduced when estimating (1.2) 

15 De Loecker additionally includes product dummies in the first stage of the estimation 

algorithm to control for unobserved product quality differences.


using OLS; their application has not been entirely successful. Likely causes 

for the failure of both techniques to produce sensible and unbiased results 

are the lack of time-invariance of ωit in the case of fixed effects and the lack 

of good instruments in the case of IV estimation. Blundell and Bond (1999) 

have developed an extended GMM estimator, taking some of these issues 

into account. 

Both semiparametric estimators (OP and LP) are able to resolve simultaneity 

issues by using a proxy variable to substitute for unobserved productivity; 

assuming a strict monotonicity condition holds and ωit is the only 

unobserved firm-level variable (i.e. the scalar unobservable). While it is 

possible to take selection issues into account by using an unbalanced panel 

for both estimators, only the OP estimation algorithm explicitly takes the 

survival probability at the firm level into account in the second stage of the 

estimation algorithm. Extensions have been developed, mainly in the context 

of the OP procedure, to take imperfect competition in output markets, 

as well as multi-product firms into account (De Loecker, 2007). 

1.4 Empirical application: Food and bever- 

ages industry in Belgium 

In what follows, I will illustrate the different methodologies introduced 

in the previous section, using firm-level data on the Belgian food and beverages 

industry. The data set is constructed on the basis of the Belfirst 

database, which groups annual accounts data on the entire population of 

limited-liability firms located in Belgium. The database is commercialized 

by BvDEP (2006). Firms are uniquely defined by their VAT number and 

data on employment, net value added, total fixed assets etc. are available 

for the years 1996-2005. Firms are classified into sectors according to the 

NACE-Bel nomenclature, i.e. a five-digit extension of the NACE (Revision 

1) classification commonly used for European statistics 16 . Producer price 

indices used to deflate firm-level output are available from Eurostat (2007) 

16 The NACE Rev. 1 classification can be downloaded from the Eurostat Ramon server: 

http://europa.eu.int/comm/eurostat/ramon/.

Empirical application: Food and beverages industry in Belgium 35 

Estimation algorithm Assumptions Resolved issues References 

Fixed effects ωit is plant-specific, but time- Simultaneity Mundlak (1961) 

invariant. Selection if ωit = ωi, ∀i Hoch (1962) 

ABBP (2007) 

Instrumental variables Correlation between instruments and Simultaneity Blundell and Bond (1999) 

& GMM endogenous regressors. Selection (unbalanced panel) ABBP (2007) 

No correlation between instruments 

and error term. 

Semiparametric estimator: Invertibility condition: investment Simultaneity Olley and Pakes (1996) 

Olley & Pakes has to be strictly increasing in ωit. Selection (unbalanced panel) ABBP (2007) 

Scalar unobservable assumption: Selection (survival probability) Ackerberg et al. (2006) 

ωit is only unobserved state variable. 

Semiparametric estimator: Invertibility condition: Simultaneity Levinsohn and Petrin (2003) 

Levinsohn & Petrin mit has to be strictly increasing in ωit. Selection (unbalanced panel) Petrin et al. (2003) 

Scalar unobservable assumption: Ackerberg et al. (2006) 

ωit is only unobserved state variable. 

OP with imperfect Assumptions OP. Simultaneity Klette and Griliches (1996) 

competition in output Selection (unbalanced panel) Levinsohn and Melitz (2002) 

markets Selection (survival probability) De Loecker (2007) 

Omitted output price bias 

Extended OP Assumptions OP. Simultaneity Klette and Griliches (1996) 

including correction Common production technology Selection (unbalanced panel) Levinsohn and Melitz (2002) 

for multi-product firms for all products of a firm. Selection (survival probability) De Loecker (2007) 

Demand elasticity is common Omitted output price bias 

across products and constant. Endogenous product choice 

Table 1.2: TFP estimation: Summary of estimation algorithms


at the three-digit Nace level. Deflators for material inputs and investment 

were obtained from Belgostat (2007). 

Following Mata and Portugal (1994); Mata et al. (1995) and Van Beveren 

(2007b); entry and exit in the sample are defined as economic exit and entry 

17 , implying that exit occurs if a firm’s employment drops to zero in a 

particular year and entry takes place if there was no previous employment 

recorded. Firms exhibiting irregular exit or entry patterns are omitted from 

the sample. Similarly, in order to verify that no re-entry occurs after a firm 

exits, the last two years in the sample are dropped. 

There are several reasons why the evolution of TFP in the food and beverages 

sector in Belgium is of interest. First, the sector represents a significant 

share of industrial employment in Belgium, accounting for 14.2 percent of the 

total (CRB, 2004), second only to the metals industry (16 percent). Moreover, 

the outbreak of the dioxin crisis in 1999, when excessive concentrations 

of dioxin were found in eggs, chickens, milk and pork; resulting from contaminated 

animal food (The Economist, 1999); led to a period of significant 

restructuring and increasing investments in the sector; reflected in the sample 

by high entry and exit rates (see below). Given these preliminaries, it 

can be expected that some of these events will be reflected in the industry’s 

TFP performance. 

Using the Belfirst database, I was able to collect information on all firms 

active in the food and beverages sector (NACE 15). Firms with no recorded 

data on one of the variables used in the empirical analysis are omitted 18 , as 

well as firms producing multiple products. To identify multi-product firms, 

I rely on the number of five-digit NACE-Bel codes a firm lists, i.e. the most 

17 Although the Belfirst database reports firms’ legal status and hence also legal exit; 

I do not rely on this measure for two reasons. First, inspection of the data reveals that 

the official date associated with the legal status in the database often does not concur 

with the actual time the firm exits the market. Second, communications with Bureau Van 

Dijk made clear that although the legal status is correctly reported whenever available, 

many companies fail to report their annual accounts after ending their activities. For the 

specifics associated with the exit and entry variables, I refer to Van Beveren (2007a). 

18 Belgian accounting rules only require firms to report full annual accounts (including 

data on turnover) once a certain threshold in terms of employment, total assets or turnover 

is reached. Therefore, the sample necessarily excludes smaller firms.


detailed level available in the database. If a firm is active in more than one 

five-digit sector, it is omitted from the analysis. Finally, the data are checked 

for outliers and gaps. Firms exhibiting variable input growth of more than 

200 percent (employment and materials inputs) in one year or output growth 

of more than 500 percent are excluded from the sample. 

This results in a final sample of 1,025 firms (5,551 observations). Table 1.3 

reports summary statistics for the sample for the period 1996-2003. From the 

table it is clear that the average firm in the sample is relatively large (average 

employment amounts to 54.61 employees). By comparison, in the full sample 

of firms active in sector 15, the average firm employs about 30 people. As 

noted above, the period considered here involved significant restructuring in 

the sector, translated in high entry and exit rates. Specifically, 184 firms 

(18 percent) enter the sample between 1996 and 2003; while 131 firms (13 

percent) exit over the same period. 

Table 1.4 reports the production function coefficients obtained using the 

different methodologies introduced in section 1.3. All reported estimates 

are obtained for the unbalanced panel of firms (allowing for implicit entry 

and exit); apart from the fixed effects estimator, where I report both the 

unbalanced and balanced sample result. The first column in the table reports 

the number of observations associated with each specific estimator and clearly 

shows one of the main advantages of the LP estimator compared to OP. Since 

material inputs are used to proxy for unobservable productivity; I am able 

to retain the full sample of firms in the first estimation stage; while for OP, 

only those observations with positive investment can be retained in the first 

stage. In the second stage, one year of observations is lost due to the dynamic 

nature of the model, both for OP and LP. 

All estimations reported in table 1.4 are performed in Stata 10. For the 

OLS and fixed effects estimators, built-in commands reg and xtreg are used. 

The GMM estimator is obtained using the xtabond2 command, due to Roodman 

(2006). No built-in or user-developed command exists to date to implement 

the OP estimator 19 ; but Arnold (2005) provides some practical tips, 

19 A user-developed command, opreg, has recently been made available in Stata, due to 

Yasar, Raciborski and Poi (2008). I have not relied on this command for the empirical


Table 1.3: Summary statistics of key variables 

Real values are obtained by deflating monetary values using three-digit producer price indices obtained 

from Eurostat. Output is defined as turnover of the firm. Employment is measured as the number of 

employees (full-time equivalents). The materials variable includes raw materials, consumables, services 

and other goods. Capital is defined as total fixed tangible assets. Investment is calculated on the basis 

of firm-level capital, using a standard depreciation rate of 15 percent. Data pertain to the Food and 

Beverages sector (NACE 15) in Belgium, for the years 1996 to 2003. 

Real output ( Rjt,ex 1,000) 5,551 19,454.79 61,007.18 0.97 950,812.10 

Employment (Ljt) 5,551 54.61 181.91 1 3,443.00 

Real materials ( Mjt,ex 1,000) 5,551 16,600.85 49,454.16 1.14 807,434.90 

Real capital ( Kjt,ex 1,000) 5,551 3,036.16 15,605.24 0.99 447,185.80 

Real (pos.) investment ( Ijt,ex 1,000) 3,588 662.46 2,653.91 0.01 61,377.32 

Standard 

Variable N Mean Deviation Minimum Maximum


Labor Materials Capital 

Method N βl SE βm SE β SE 

OLS 5,551 0.2113*** [0.0152] 0.7700*** [0.0138] 0.0266*** [0.0072] 

Fixed Effects (balanced) 3,568 0.1696*** [0.0192] 0.6474*** [0.0419] 0.0277*** [0.0063] 

Fixed Effects (unbalanced) 5,551 0.1685*** [0.0166] 0.6814*** [0.0379] 0.0248*** [0.0052] 

GMM 5,551 0.1520*** [0.0368] 0.7890*** [0.0434] 0.0372** [0.0173] 

OP (no survival correction) 3,588 0.1925*** [0.0153] 0.7722*** [0.0150] 0.0445** [0.0195] 

OP (survival correction) 3,588 0.1925*** [0.0153] 0.7722*** [0.0150] 0.0453*** [0.0167] 

Levinsohn-Petrin 5,551 0.2139*** [0.0148] 0.7915*** [0.0802] 0.0484** [0.0205] 

De Loecker (1) 3,588 αl = 0.1947*** [0.0153] αm = 0.7686*** [0.0151] αk = 0.0461* [0.0240] 

Transformed coefficients DL αq = 0.2926*** 0.2707*** [0.0223] 1.0837*** [0.0426] 0.0654** [0.0338] 

[0.0199] 

Values are coefficients, standard errors reported between brackets. (1) The coefficients for the DL estimator are obtained by 

multiplying the alpha’s with the relevant mark-up. The elasticity of substitution η equals (−1/αq) or -3.42. The relevant mark-up 

therefore equals η/ (η + 1) = 1.41. 

Table 1.4: Production function estimates


particularly on the implementation of the nonlinear second stage. The LP 

estimator was implemented using the levpet command, due to Petrin et al. 

(2003). 

In order to interpret the estimated coefficients, it is useful to briefly go back 

to table 1.1. In the third column of this table, the general direction of the 

biases introduced by the different endogeneity issues are given. Theoretically, 

the fixed effects estimator corrects for both the simultaneity and selection 

bias, hence the coefficients on the variable inputs (labor and materials) are 

expected to be lower compared to the OLS result; while the coefficient on 

capital is expected to be higher. While the coefficients on the variable inputs 

in table 1.4 are in line with expectations (βl and βm are lower compared to 

the first row); the capital coefficient is still very low, both for the balanced 

and unbalanced sample. 

Moreover, as was discussed in section 1.3, comparing the results of the 

balanced and unbalanced sample for the FE estimator enables us to determine 

whether the FE estimator adequately corrects for the selection bias; 

i.e. whether exit decisions at the firm level are only determined by the timeinvariant, 

firm-specific effects ωi. Given the small differences between the 

coefficients obtained for the balanced and unbalanced sample; results in table 

1.4 suggest that the FE estimator is able to correct for the selection bias 

in the sample. 

Since the GMM estimator is theoretically able to correct for the simultaneity 

bias, βl and βm in row 4 of table 1.4 are expected to be lower, while 

βk should increase compared to their OLS counterparts; similarly to the FE 

estimator. Results in row 4 show a lower labor coefficient and higher capital 

coefficient (in line with expectations); but lower coefficient on materials (not 

in line with expectations). 

The last four rows in table 1.4 display the production function coefficients 

for the semiparametric estimators of OP (both with and without explicit correction 

for firms’ survival probability), LP and De Loecker. Comparing OP 

estimations.


estimates to the OLS estimates in the first row, shows that the coefficients 

on both labor and materials are lower compared to OLS results, while the 

capital coefficient is significantly higher; which is in line with expectations. 

Including the estimated survival probability in the second stage of the estimation 

algorithm has virtually no impact on the capital coefficient. This 

result is in line with the findings of OP, who similarly found no significant 

improvement in the capital coefficient from the explicit correction for survival 

when an unbalanced panel is used. Although the LP coefficient on capital 

is higher than its OLS counterpart, the labor and materials coefficients are 

somewhat higher than the OLS estimates. 

The final row of table 1.4 summarizes the results of estimating (1.14) using 

the estimation algorithm introduced 20 by De Loecker (2007). Essentially, this 

amounts to the inclusion of industry output in the first stage of estimation 

and subtracting the resulting coefficient times output from the left-hand-side 

in (1.10). Industry output is calculated at the three-digit level in each year 

as the share-weighted average of firm-level outputs, where shares are based 

on deflated revenues. This comes from the observation that the industry 

price index (which is available at the three-digit level) represents a shareweighted 

average of firm-level prices, where weights are output shares (De 

Loecker, 2007). For now, the elasticity of demand (substitution) is assumed 

to be identical across the different subsectors within the food and beverages 

industry. 

As was shown in section 3, the coefficient on industry output αq relates 

to the elasticity of demand in the following way: αq = (−1/η). Moreover, 

using the demand elasticity, which amounts to -3.42; it is possible to calculate 

the relevant mark-up at the industry level η/ (η + 1), equal to 1.41. 

This estimate is somewhat higher than the result found by Konings (2001), 

who find a mark-up of 1.30 for the food and beverages industry in Belgium 

in the period 1992-1996. The last row in table 1.4 further reports both the 

20 Although the correction for market power in output markets was originally suggested 

by Klette and Griliches (1996), De Loecker was the first to implement this correction 

into the semiparametric estimation framework introduced by Olley and Pakes (1996). 

Abraham, Konings and Slootmaekers (2007a) report results of the DL estimator as a 

robustness check in their paper on FDI spillovers in China.


estimated coefficients and the true production coefficients βh = (η/η + 1) αh. 

Consistent with the theoretically predicted biases in table 1.1, the coefficients 

on labor and materials are significantly higher compared to the OP coefficients 

without including industry output. However, the coefficient on capital 

is somewhat higher compared to the basic OP results, which is not in line 

with expectations. 

As was indicated in section 1.3, it is straightforward to allow the demand 

elasticity to vary over the different three-digit industries by interacting industry 

output with the respective industry dummies in (1.16). Since this results 

both in different demand elasticities and associated mark-ups; production 

function coefficients also become specific for each separate three-digit industry 

in this case. However, note that while production coefficients become 

variety-specific in this case, the production technology is still assumed to be 

constant for all three-digit industries within the food and beverages sector. 

Table 1.5 reports the results of estimating (1.16) for the sample of singleproduct 

firms in the food and beverages industry. The first row in table 1.5 

shows the estimated coefficients αh. Compared to the estimated coefficient 

for the constant-elasticity estimator reported in the last row of table 1.4, the 

labor and materials coefficients are very similar, while the capital coefficient 

is somewhat higher. Turning to the industry-specific output coefficients, it 

is clear that large variation exists between the different three-digit subsectors 

of the food and beverages industry. Calculated demand elasticities vary 

between -2.8 and -3.6; associated mark-ups range between 1.39 and 1.56. 

These differences point to the importance of allowing the demand (substitution) 

elasticity to vary across different sub-sectors of a particular industry. As 

a consequence, variety-specific production coefficients also vary considerably 

across the different three-digit industries. 

Two caveats should be noted here. First, I have continued to assume 

throughout that input prices for materials (capital) at the firm level are 

adequately captured by the materials (investment) deflator. To the extent 

that input price differences are translated into output price deviations, which 

are taken into account using industry output, this should partly take care of


Three-digit industry Output Demand Labor Materials Capital 

NACE Description Coefficient Elasticity Mark-up Coefficient Coefficient Coefficient 

- αh(h = l, m, k) - - - 0.1948*** 0.7685*** 0.0569*** 

[0.0153] [0.0151] [0.0215] 

151 Meat (products) 0.3348*** -2.9869 1.5033 0.2896*** 1.1592*** 0.0863*** 

[0.0224] [0.0251] [0.0517] [0.0335] 

152 Fish(products) 0.3552*** -2.8154 1.5508 0.2981*** 1.1931*** 0.0888*** 

[0.0239] [0.0265] [0.0577] [0.0347] 

153 Fruit and vegetables 0.3145*** -3.1799 1.4587 0.2802*** 1.1215*** 0.0835*** 

[0.0218] [0.0237] [0.0477] [0.0323] 

154 Oils and fats 0.3587*** -2.7881 1.5593 0.2999*** 1.2003*** 0.0894*** 

[0.0243] [0.0270] [0.0588] [0.0347] 

155 Dairy products 0.2951*** -3.3888 1.4186 0.2728*** 1.0919*** 0.0813*** 

[0.0197] [0.0227] [0.0416] [0.0314] 

156 Grain mill products 0.3026*** -3.3046 1.4339 0.2780*** 1.1126*** 0.0828*** 

[0.0224] [0.0235] [0.0446] [0.0320] 

157 Prepared animal feeds 0.3103*** -3.2226 1.4499 0.2786*** 1.1151*** 0.0830*** 

[0.0206] [0.0238] [0.0438] [0.0321] 

158 Other food products 0.2871*** -3.4831 1.4027 0.2692*** 1.0774*** 0.0802*** 

[0.0194] [0.0220] [0.0407] [0.0309] 

159 Beverages 0.2784*** -3.592 1.3858 0.2656*** 1.0631*** 0.0791*** 

[0.0184] [0.0216] [0.0377] [0.0304] 

Values are coefficients, standard errors reported between brackets. The variety-specific production function 

coefficients are obtained by multiplying the alpha’s (given in the first row) with the relevant mark-up. The 

elasticity of substitution (demand) η is obtained as the inverse and negative of the output coefficient. The 

relevant mark-up equals η/(η + 1). 

Table 1.5: Production function estimates: Variety-specific demand


the omitted input price bias (De Loecker, 2007). However, as was already 

noted in section 1.2, a formal solution to this bias (in the absence of firm-level 

data on input prices) has yet to be introduced. 

Second, the selection of single-product firms in the sample is obtained by 

resorting to the NACE-Bel codes reported by firms in their annual accounts, 

where the codes typically relate to the latest year available. Hence, the 

selection of firms is made in a particular year, whereas it is quite possible 

that some of these firms produced multiple products in any of the previous 

years. 

The production function coefficients obtained in tables 1.4 and 1.5 can be 

used to calculate firm-level productivity for each of the sample years. By 

imposing coefficient stability on the model, it is possible to retain the full 

sample of firms for all estimators, even in the absence of positive investment 

(as for the OP estimators). Firm-level productivity for the OLS, fixed effects, 

GMM, OP (with and without survival correction) and LP estimators 

is obtained on the basis of (1.3). For the OP estimator including the correction 

for market power (De Loecker, with or without variety-specific demand), 

productivity as obtained in (1.3) additionally needs to be multiplied by the 

relevant mark-up; as was shown in (1.15). 

Finally, using the estimates of firm-level productivity obtained from applying 

(1.3) and (1.15) to the sample using the production function coefficients 

from tables 1.4 and 1.5, it is possible to calculate aggregate industry productivity 

for each year in the sample as a weighted average of firm-level TFP: 

Nt 

ˆPJt = sit ˆ Ωit 

i=1 

(1.17) 

where sit is a firm-specific weight, equal to (Sit/ ( 

i Sit)) and S represents 

either turnover or employment (De Loecker and Konings, 2006). Normalizing 

this index to 1 in 1996 allows us to compare the evolution of aggregate TFP 

in the food and beverages industry for the different estimators discussed here.


TFP (1996 = 1.00) 

.95 1 1.05 1.1 1.15 1.2 

1996 1997 1998 1999 2000 2001 2002 2003 

year 

ols fe_b 

fe gmm 

lp opbasic 

opsurv dl 

dl_var 

(Weights are turnover shares, 1996 = 1) 

Figure 1.1: Weighted productivity index: Comparison estimation methods 

Figure 1.1 shows the evolution of industry productivity between 1996 and 

2003, using turnover shares as weights. From the figure, it is clear that TFP 

in the food and beverages industry exhibits a clear upward trend in the period 

following the dioxin crisis of 1999. However, whereas TFP continues to increase 

until 2002 when imperfect competition in output markets is not taken 

into account; TFP estimated using the DL methodology increases sharply 

between 1999 and 2000 and exhibits a more of less stable pattern after that. 

For the DL estimator with variety-specific demand, this pattern is even more 

apparent. Moreover, compared to the other estimators shown in figure 1.1, 

TFP calculated using the coefficients of table 1.5 declines more sharply prior 

to 1999 and grows less strongly after 1999. These results suggest that imperfect 

competition in output markets, when not taken into account in the 

production function estimation, may yield misleading results concerning the 

timing and magnitude of productivity shocks. The different growth pattern 

observed for the DL estimator with and without variety-specific demand further 

suggests that it is important to take the demand structure into account 

at the appropriate level of aggregation 21 . 

21 Ideally, this would be at the product level. However, this would require not only information 

on aggregate product output, but also on product-level price evolutions (indices).


To assess whether the evolution of aggregate TFP in the food and beverages 

industry is due to firm-level improvements in TFP or rather to the 

reallocation of market shares between firms, various decompositions can be 

used (De Loecker and Konings, 2006). I will rely on the decomposition 22 introduced 

by Olley and Pakes (1996), who decompose aggregate productivity 

into a within component and a covariance term in the following way: 

Nt 

ˆPJt = 

ˆPJt = 

i=1 

Nt 

ˆPJt = sit ˆ Ωit 

i=1 

 

¯ˆΩt 

(¯st + ∆sit) + ∆ˆ 

Ωit 

 

Nt¯st ¯ 

Ωt 

ˆ + 

ˆPJt = ¯ ˆ 

Pit + 

Nt 

i=1 

Nt 

i=1 

 

∆sit∆ˆ 

Ωit 

 

∆sit∆ˆ 

Ωit 

where ¯ Pit 

ˆ is the unweighted average of plant-level total factor productivity 

and Nt 

∆sit∆ˆ 

Ωit refers to the sample covariance between TFP and output 

i=1 

(or employment) shares. The results of applying this decomposition using either 

turnover (left-hand side) or employment shares (right-hand side) for the 

TFP measure of De Loecker allowing for three-digit industry-specific demand 

elasticities, are displayed in table 1.6. The first column for each type of share 

consists of the share-weighted average productivity measured calculated on 

the basis of (1.17), normalized to 1 for 1996. The second and third column 

show the percentage contribution of the within productivity component and 

the reallocation share to aggregate weighted TFP respectively. 

One might also argue that in such a case, it is preferable to allow not only the industry 

output coefficient, but also the input coefficients to vary across products, i.e. to estimate 

a separate production function for each of the products (or sub-sectors in the absence of 

product-level information). 

22 An alternative to the OP decomposition is provided by Foster et al. (2006) . In 

addition to a within firm and reallocation term, they allow for a separate net-entry and 

interaction term. Given the complexity of their decomposition, it is beyond the scope of 

the present paper to apply it here.


Year Turnover shares Employment shares 

Weighted Mean Reallo- Weighted Mean Reallo- 

Index TFP cation Index TFP cation 

(1996 = 1) (%) (%) (1996 = 1) (%) (%) 

1996 1.000 102.71 -2.71 1.000 107.18 -7.18 

1997 0.9260 101.95 -1.95 0.9420 104.58 -4.58 

1998 0.9166 101.09 -1.09 0.9338 103.54 -3.54 

1999 0.935 100.96 -0.96 0.9637 102.22 -2.22 

2000 1.0506 101.14 -1.14 1.0791 102.76 -2.76 

2001 1.0451 100.84 -0.84 1.0748 102.32 -2.32 

2002 1.0297 100.51 -0.51 1.0687 101.06 -1.06 

2003 1.0323 99.95 0.05 1.0855 99.19 0.81 

Weighted average productivity is calculated according to (1.17), 

weights are firm-level turnover or employment shares. 

Table 1.6: Decomposition aggregate TFP: De Loecker methodology


From table 1.6, it is clear that most of the productivity improvements 

realized in the food and beverages sector since 1996 have been associated 

with within firm productivity growth. When employment shares rather than 

turnover shares are used (right-hand side of the table), the reallocation share 

is somewhat larger than for the case of turnover shares. Hence, I conclude 

that most of the productivity increases realized in the food and beverages 

industry in Belgium following the dioxin scandal in 1999 were due to the 

average firm becoming more productive, while reallocation of market share 

(either in terms of employment or turnover) has only played a minor role. 

Reallocation shares are consistently negative throughout the sample period, 

both using turnover and employment shares, with the exception of 2003, 

when it becomes positive in both cases. 

For comparison purposes, table 1.7 summarizes the results of the OP decomposition 

for each of the different estimators listed in table 1.2. The table 

shows, apart from weighted normalized TFP in 2003 for each of the estimators, 

the average shares of unweighted average TFP and the sample covariance 

term in aggregate weighted industry productivity. Values reported are 

eight-year averages. Although the within firm growth component dominates 

regardless of the estimators applied to calculate industry productivity, there 

are some important differences worth noting. 

Of the eight decompositions summarized in table 1.7, five yield similar 

results. Specifically, for the OLS, GMM, OP and De Loecker estimators 

the sample covariance terms (both for turnover and employment) are small 

and positive. For both fixed effects estimators however, reallocation shares 

are much larger, although still positive. The De Loecker estimator allowing 

for variety-specific demand, as well as the LP estimator yield a small but 

consistently negative sample covariance term between productivity and either 

output or employment. 

1.5 Conclusions 

This paper has reviewed the methodological issues arising when total factor 

productivity or TFP is estimated at the establishment level. The traditional

Conclusions 49 

Turnover shares Employment shares 

Weighted Mean Reallo- Weighted Mean Reallo- 

Method Index TFP (2) cation Index TFP (2) cation 

(1996 = 1)(1) (%) (%) (2) (1996 = 1)(1) (%) (%) (2) 

OLS 1.1393 99.24 0.76 1.1606 99.39 0.61 

Fixed Effects (balanced) 1.1619 92.18 7.82 1.1773 84.58 15.42 

Fixed Effects (unbalanced) 1.1657 90.42 9.58 1.1787 81.53 18.47 

GMM 1.1453 98.52 1.48 1.1686 96.44 3.56 

OP (no survival correction) 1.1377 99.45 0.55 1.1582 99.57 0.43 

OP (survival correction) 1.1375 99.48 0.52 1.1580 99.64 0.36 

Levinsohn-Petrin 1.1293 101.68 -1.68 1.1504 105.08 -5.08 

De Loecker 1.1148 97.74 2.26 1.1613 96.62 3.38 

DL (variety-specific) 1.0323 101.15 -1.15 1.0855 102.86 -2.86 

Weighted average productivity is calculated as in equation 17, weights are firm-level turnover of employment 

shares. (1) Weighted normalized TFP in 2003 (1996 = 1). (2) Values reported are eight-year 

averages of the shares of unweighted average TFP and the sample covariance term. 

Table 1.7: Comparison of decomposition results


biases introduced by the simultaneity of input choice and endogeneity of attrition 

have been discussed; as well as a number of issues that have emerged 

more recently, i.e. related to imperfect competition in input and/or output 

markets and endogeneity of product choice. Various estimators have been 

introduced in the literature attempting to overcome some of these issues. 

Given the relatively poor performance and shortcomings of the more traditional 

estimators, i.e. fixed effects and GMM; a number of semiparametric 

estimators have been introduced, which have been briefly reviewed here. A 

recent extension to these estimators taking the omitted output price bias into 

account; in addition to dealing adequately with simultaneity and selection 

issues has also been discussed. 

I have illustrated the performance of these estimators using data on the 

food and beverages industry in Belgium in the period 1996 to 2003, when 

the sector was undergoing significant changes and restructuring, especially 

following the outbreak of the dioxin crisis in 1999. Findings confirm the 

theoretically expected biases in traditional production function estimates, 

obtained using OLS. Moreover, the evolution of industry TFP over the sample 

period shows a clear upward trend in aggregate productivity following the 

dioxin scandal in 1999. 

Which estimator would researchers ideally want to use then? In light of the 

traditionally poor performance of both the GMM and fixed effects estimators, 

it would seem that the semiparametric estimators are to be preferred, and 

specifically the Olley-Pakes methodology. Moreover, comparing aggregate 

industry productivity growth patterns for the different estimators shows that 

a failure to take imperfect competition in output markets into account may 

yield misleading results concerning the timing and magnitude of observed 

industry growth patterns, hence favoring the estimator of De Loecker. 

However, the choice of which estimator to use will essentially also depend 

on the data at hand. Reliable industry output measures are not always 

available to the researcher. Similarly, positive investment data are not always 

available for a sufficiently large sample of firms within an industry or 

might not be trustworthy. Data can also be prone to measurement error


or production technology may differ widely within an industry, invalidating 

some of the parametric methods discussed here. 

Van Biesebroeck (2007) compares the sensitivity of different estimators 

(index numbers, data envelopment analysis or DEA, stochastic frontiers, IV 

(GMM) and semiparametric estimation. He finds that the GMM-SYS estimator 

is the most robust technique when there is a lot of measurement error 

or some technological heterogeneity. However, for the GMM-SYS estimator 

to be reliable, at least some of the productivity differences have to be 

constant over time. He further notes that the GMM estimator might lead 

to downwardly biased input coefficients when measurement error becomes 

severe. When measurement error is small, technology is heterogeneous and 

returns to scale are not constant, non-parametric techniques such as DEA or 

index numbers should be preferred. 

In spite of the multitude of estimators that have been developed in recent 

years in order to achieve consistent estimates of total factor productivity, a 

number of issues remain to be resolved. In particular, both the lack of a 

formal correction for the omitted input price bias in the presence of imperfect 

competition in input markets, as well as the implications of endogenous 

product choice following from BRS (2005, 2006b) offer ample scope for future 

research.

Chapter 2 

Globalization Drives Strategic 

Product Switching 


Globalization has led to increasingly integrated markets across the world, 

changing the competitive environment in which firms operate. In the face of 

international competition in domestic and foreign markets, the least productive 

firms may be forced into bankruptcy while the most productive ones will 

take advantage of new business opportunities in foreign markets. Moreover, 

incumbent firms may respond by increasing their productivity or, since this 

may prove difficult in mature industries, by diversifying into a different industry 

or product variety. The importance of firm shutdown and changes in 

the product mix as a response to pressures from international trade has been 

highlighted in recent empirical studies (see, for example, Bernard, Jensen 

and Schott, 2006a; and Greenaway, Gullstrand and Kneller, 2008). However, 

this literature has focused on only one aspect of trade -import competitionignoring 

firm dynamics induced by profitable opportunities in export markets. 

This second aspect is potentially very important for emerging markets, 

particularly in the aftermath of trade liberalization. 

∗ This is joint work with Marialuz Moreno Badia (IMF) and Veerle Slootmaekers (OECD 

and KU Leuven, LICOS). Published as LICOS Discussion Paper 209/2008 and IMF Working 

Paper 08/246. The views expressed herein are those of the authors and should not be 

held to represent those of the institutions of affiliation. 

53

54 Globalization drives strategic product switching 

The purpose of this paper is to analyze the impact of international trade 

on firm dynamics, focusing on how production patterns are adjusted in response 

to import competition and changing conditions in export markets. 

Our focus is Estonian manufacturing firms from 1997 to 2005. Estonia is 

a particularly interesting case because of the firm restructuring and trade 

liberalization that took place in the aftermath of the transition process and 

in the run-up to European Union (EU) membership. Buoyed partly by the 

Association Agreement with the EU, Estonian exports of goods increased 

by 240 percent 1 . This extraordinary performance was accompanied by an 

increase in product variety (Kandogan, 2006) and a shift toward exports of 

higher quality and technological intensity (Fabrizio et al., 2007). This seems 

to suggest that Estonian firms may not have merely responded defensively to 

increased competition from importers, but also reacted offensively by taking 

advantage of the opportunities created by trade liberalization. 

To identify these effects, we use a longitudinal data set of Estonian manufacturing 

firms and consider three potential firm strategies: continue its business, 

switch products, or close down. To model these strategic alternatives, we 

estimate a multinomial logit model in which the firm decision is a function of 

firm-level and product-market characteristics. Following the previous literature, 

we include the value of imports, type of trade (intra- or interindustry), 

and revealed comparative advantage as measures of trade. To identify the 

effect of export opportunities, we also include in our estimation the value 

of exports, the degree of competition in export markets, and the quality of 

exports relative to direct competitors. 

Overall, we find that firm exit is mainly determined by firm characteristics, 

whereas product switching also depends on conditions in export markets. 

In particular, firms are more likely to switch if they are in sectors without 

revealed comparative advantage, with less exports, or with lower product 

quality relative to export competitors. One interpretation of this result is 

that firms are more willing to incur the sunk costs of breaking into a new 

1 The Association Agreement with the EU was signed in 1995 and entered into force 

in 1998. The agreement replaced previous treaties with the EU (an Agreement on Trade 

and Commercial Cooperation, signed in 1992, which was converted into a Free Trade 

Agreement in 1994). For a more detailed description, see Weber and Taube (1999).


product line or industry when the long-term prospects of the current export 

market for their products are weak, particularly early in the sample when 

trade flows were increasing rapidly. These results are in contrast with previous 

studies on industrial countries that find that firms switch products as a 

defense against low-cost imports. Interestingly, we find that the conditions 

on export markets matter predominantly for switches within the same industry 

or what we call product switches 2 . Switches across industries on the other 

hand are determined by firm-level characteristics. These results suggest that 

product diversification was a major strategy of Estonian firms faced with 

increasing trade openness. Finally, we find a positive link between a firm’s 

capital intensity and quality upgrading. However, moving up the quality 

ladder is not necessarily related to technology upgrading; it occurs mainly 

within the medium-high-tech sector. 

The literature on the relationship between globalization and firm dynamics 

has expanded rapidly in recent years (see, for example, Melitz, 2003; Bernard, 

Eaton, Jensen and Kortum, 2003; and Helpman, Melitz and Yeaple, 2004). 

This literature builds on Melitz’s (2003) dynamic industry model with heterogeneous 

firms, where sunk costs of market entry result in self-selection 

into export markets. More recent theoretical work models how these dynamics 

interact with a country’s industry characteristics (Bernard, Redding and 

Schott, 2007b and Melitz and Ottaviano, 2008). Rigorous empirical work, 

triggered by the work of Bernard and Jensen (1995), has further nourished 

the understanding of firm adjustment to trade liberalization and falling trade 

costs 3 . In short, these studies document substantial variation in productivity 

across firms, frequent firm entries and exits, sizeable sunk costs of entry into 

export markets, and better performance among exporting firms. 

2 Product switches are defined as changes in industry at the four-digit level henceforth. 

Although four-digit NACE codes are not true products in the strictest sense of the word, 

this is the most detailed classification we have in our data. This notation has also been 

used in Bernard et al. (2006a) and Greenaway et al. (2008). 

3 For an overview of the empirical literature, see Tybout (2003) and Bernard, Jensen, 

Redding and Schott (2007c).


Our paper is more closely related in spirit to the work of Bernard et al. 

(2006a) and Greenaway et al. (2008) 4 . These papers reveal a new dimension 

of adjustment to increased international competition by illustrating that 

firms are more likely to change their product mix than to shut down in response 

to globalization. Controlling for a number of firm and industry characteristics, 

they find that firms are more likely to switch away from industries 

where exposure to low-wage countries is high. Bernard et al. (2006a) note 

that U.S. firms shift towards industries facing less competitive pressure, but 

with greater capital and skill intensity than the industry of origin. Greenaway 

et al. (2008) broaden the analysis, and consider mergers and acquisitions as 

a third exit strategy. They report that closure is the least likely exit strategy 

in Sweden, as most firms merge with or acquire another firm in response 

to higher levels of international competition. A primary contribution of our 

paper relative to these studies is that we consider explicitly the role of export 

opportunities in determining firm dynamics. Also, ours is the first paper to 

examine the impact of globalization on firm dynamics in an emerging market 

context 5 . Since countries at different stages of development exhibit large 

differences in terms of firm size distribution, efficiency, and cost structure 

it is important to explore whether enterprises in emerging markets respond 

differently to globalization than enterprises in more advanced countries. 

The remainder of this paper is organized as follows. Section 2.2 describes 

the industry dynamics in Estonia. Section 2.3 outlines the estimation strategy 

used to analyze the impact of international trade on firm dynamics in 

Estonia. Section 2.4 presents the results. Section 2.5 discusses the robustness 

checks. Section 2.6 concludes. 

2.2 Industry dynamics in Estonia 

The data used in this paper are provided by the Estonian Business Registry 

and cover the years 1997-2005. The data set is an unbalanced panel 

4 These papers build on the work of Bartelsman and Doms (2000), Foster, Haltiwanger 

and Krizan (2006) and Bernard and Jensen (2007a). 

5 Goldberg, Khandelwal, Pavcnik and Topalova (2008) analyze the response of Indian 

firms to trade liberalization but they focus on product margins rather than looking into 

the firm dynamics per se.

Industry dynamics in Estonia 57 

containing detailed information on balance sheets and income statements of 

all registered firms in Estonia. The unit of observation is the firm, which 

can be tracked over time using a unique registration code 6 . As all business 

entities in Estonia are required to file their annual accounts with the registry, 

the data set comprises firms from all size classes, including microenterprises 

with less than 10 employees. 

Our primary interest lies in identifying the impact of international competition 

on firms’ strategic choices. Therefore, we focus on the sector for which we 

observe trade flows at a disaggregated level, namely, the manufacturing sector. 

Nonetheless, since companies active in sectors other than manufacturing 

are equally obliged to report to the registry, we are able to identify not only 

industry switches within the manufacturing sector, but also switches to other 

sectors of the economy 7 . The sample used in the empirical analysis consists 

of 4,844 firms and 16,117 observations (see Table 2.1). Entry and exit are 

observed, and the number of manufacturing entities in the registry increased 

significantly over the sample period, from 1,196 firms in 1997 to 2,767 firms 

in 2004 8 . For each firm we observe its primary sector of activity at the fourdigit 

NACE (General Industrial Classification of Economic Activities within 

the European Community) level 9 . Unfortunately we do not have information 

on the total number of different products per firm, but we do observe changes 

in its main product line over time. 

Firm exit (Exitit+1) is identified using the firm’s official liquidation date, 

available from the registry. In addition to exit, we are interested in industry 

switches at both the two-digit (Switch2d,it+1) and four-digit (Switch4d,it+1) 

6 For a detailed description of the data, see section 2.A. More information on the 

Estonian Business Registry can be found in Masso et al. (2004) 

7 Firms switching to other sectors of the economy are retained until their last year of 

activity within the manufacturing sector. 

8 Since we cannot observe switching for 2005 (no data for 2006 are available), this year 

is omitted from the analysis. 

9 The classification used by the Registry is the EMTAK Classification of Economic Activities 

of Estonia. EMTAK is a five-digit extension of the four-digit European NACE 

classification system, the official statistical classification system of economic activities in 

the European Community. The first four digits of the EMTAK codes are therefore equivalent 

to the NACE (Rev. 1.1.) codes. The NACE classification system can be downloaded 

from the Eurostat Ramon server (http://europa.eu.int/comm/eurostat/ramon/).


Number of Industry Product 

Year observations switch (1) switch (1) Exit (1) 

Panel A: Distribution of the full sample 

1997 1,196 14.8 21.7 5.1 

1998 1,398 9.5 14.5 4.9 

1999 1,621 11.0 16.5 4.4 

2000 1,930 8.9 12.8 3.1 

2001 2,175 6.6 11.4 3.3 

2002 2,396 6.2 9.9 2.2 

2003 2,634 6.0 9.3 1.8 

2004 2,767 4.8 7.3 0.7 

Total (observations) 16,117 1,244 1,912 452 

Total (firms) 4,844 1,090 1,566 452 

Panel B: Distribution for microenterprises (2) 

1997 515 19.6 26.8 6.8 

1998 616 12.2 16.6 6.3 

1999 765 12.7 17.9 5.2 

2000 944 11.2 14.4 3.2 

2001 1,100 8.6 13.5 4.5 

2002 1,232 8.4 11.5 2.5 

2003 1,423 7.6 11.1 2.0 

2004 1,553 6.2 8.4 0.8 

Total (observations) 8,148 782 1,092 267 

Total (firms) 3,214 709 949 267 

Notes: (1) Industry (product) switches are identified at the two-digit (four-digit) 

level. Values are percentages. (2) Firms with less than 10 employees. 

Table 2.1: Exits and industry switches, 1997-2004 

NACE level. For the remainder of the paper, we refer to two-digit switches 

as industry switches and to four-digit switches as product switches. Whereas 

Bernard et al. (2006a) study switches at the product level, Greenaway et al. 

(2008) study switches at the industry level because they want to focus on big 

changes in production and to minimize the possibility of classification problems. 

Nevertheless, the latter analyze the product switches as a robustness 

check and report substantial differences in the determinants of both types of 

switching. In our analysis, we focus on the product-level switches, but we 

also discuss the differences with the industry switches.


Observations Industry switches Product switches Exits 

NACE Two-digit industry Number Percent Number Percent Number Percent Number Percent 

15 Food and beverages 1,314 8.2 83 6.3 131 10.0 58 4.4 

17 Textiles 689 4.3 44 6.4 61 8.9 21 3.0 

18 Clothing 1,396 8.7 59 4.2 111 8.0 48 3.4 

19 Leather (products) 270 1.7 8 3.0 10 3.7 6 2.2 

20 Wood (products) 3,529 21.9 255 7.2 410 11.6 120 3.4 

21 Pulp, paper (products) 198 1.2 13 6.6 16 8.1 3 1.5 

22 Publishing and printing 1,455 9.0 74 5.1 139 9.6 33 2.3 

23 Coke, petroleum products 5 0.0 1 20.0 1 20.0 0 0.0 

24 Chemicals 356 2.2 41 11.5 48 13.5 8 2.2 

25 Rubber and plastic 601 3.7 48 8.0 74 12.3 12 2.0 

26 Non-metallic mineral products 509 3.2 43 8.5 63 12.4 13 2.6 

27 Basic metals 16 0.1 5 31.3 5 31.3 0 0.0 

28 Fabricated metal products 1,819 11.3 150 8.3 273 15.0 41 2.3 

29 Machinery and equipment 812 5.0 125 15.4 160 19.7 18 2.2 

30 Office machinery and computers 59 0.4 17 28.8 17 28.8 0 0.0 

31 Electrical machinery 340 2.1 40 11.8 51 15.0 9 2.6 

32 Communications equipment 290 1.8 45 15.5 53 18.3 5 1.7 

33 Medical, precision, optical instruments 413 2.6 38 9.2 44 10.7 6 1.5 

34 Motor vehicles 121 0.8 12 9.9 14 11.6 2 1.7 

35 Other transport equipment 295 1.8 34 11.5 37 12.5 8 2.7 

36 Furniture 1,630 10.1 109 6.7 194 11.9 41 2.5 

Total 16,117 100.0 1,244 7.7 1912 11.9 452 2.8 

Industry (product) switches are defined at the two-digit (four-digit) NACE level. All industry switches are also observed at the 

product level; out of 1,912 product switches in the sample, 1,244 are observed at both the product and industry level. 

Table 2.2: Sector distribution


Table 2.1 reports the distribution of the sample over time. Out of 4,844 

firms, 452 firms exited between 1997 and 2004, and 1,566 firms switched 

products. Of the latter group, 1,090 changed to a different industry and 476 

firms only switched products within the same industry. Industry and product 

switches were more frequent at the end of the 1990s, but the rates declined 

steadily toward the end of the sample period. On average, the industry 

switching rate is 7.7 percent, compared with a switching rate of 11.9 percent 

at the product level. The product switching rate is only slightly higher than 

the figures reported by Bernard et al. (2006a) and Greenaway et al. (2008), 

who find a product switching rate of 7-8 percent for the U.S. and Sweden, 

respectively. 

Tables 2.2, 2.3 and 2.4 give a preliminary indication that there are broad 

differences in firm dynamics across sectors. First, it is clear from Table 2.2 

that sectors facing a high number of exits also tend to undergo more industry 

changes. This is, however, related to the size of the sectors, and switching 

and exit rates are not necessarily correlated. For example, the switching rates 

of ”basic metals,” ”office machinery and computers,” and ”coke, petroleum 

products” are very high but the exit rates in these sectors are zero. Rather 

than permanently exiting the market, companies in those sectors seem to 

look for other opportunities by switching to a different industry. Second, 

about 35 percent of all product switches occur within the same two-digit 

sector; 23 percent to other two-digit manufacturing sectors; 4 percent to the 

primary sector; and 38 percent to services (Table 2.3 10 ). 

Third, the majority of firms in our sample are active in low-tech manufacturing 

(65 percent) and most of the product switches take place among firms in 

this group (Table 2.4 11 ). However, product switches from medium-high-tech 

10 The primary sector comprises NACE 1-14 (agriculture and mining activities); 

manufacturing sector or secondary sector comprises NACE 15-37; and services or the 

tertiary sector comprises NACE 40-99. Most firms that switch to services end up in 

industries like wholesale trade and retail trade. We explore further the nature of these 

switches in Section 2.4. 

11 Using the Eurostat classification, which is available at 

http://europa.eu.int/comm/eurostat/ in the section “Science and Technology”, 

we classify the manufacturing sectors according to technology intensity and services 

according to knowledge intensity. Appendix tables 2.A.1 and 2.A.2 provides a list of

Product Within same 

Switches Industry To To To 

NACE Two-digit industry (N) (two-digit) primary manufacturing services 


15 Food and beverages 131 36.6 14.5 2.3 46.6 

17 Textiles 61 27.9 0.0 39.3 32.8 

18 Clothing 111 46.9 0.0 23.4 29.7 

19 Leather (products) 10 20.0 0.0 50.0 30.0 

20 Wood (products) 410 37.8 11.5 15.9 34.9 

21 Pulp, paper (products) 16 18.8 0.0 31.3 50.0 

22 Publishing and printing 139 46.8 1.4 10.1 41.7 

23 Coke, petroleum products 1 0.0 0.0 0.0 100.0 

24 Chemicals 48 14.6 4.2 22.9 58.3 

25 Rubber and plastic 74 35.1 0.0 32.4 32.4 

26 Non-metallic mineral products 63 31.8 9.5 25.4 33.3 

27 Basic metals 5 0.0 0.0 100.0 0.0 

28 Fabricated metal products 273 45.1 0.0 24.5 30.4 

29 Machinery and equipment 160 21.9 1.9 30.0 46.3 

30 Office machinery and computers 17 0.0 0.0 5.9 94.1 

31 Electrical machinery 51 21.6 0.0 29.4 49.0 

32 Communications equipment 53 15.1 0.0 39.6 45.3 

33 Medical and optical instruments 44 13.6 0.0 38.6 47.7 

34 Motor vehicles 14 14.3 7.1 28.6 50.0 

35 Other transport equipment 37 8.1 0.0 46.0 46.0 

36 Furniture 194 43.8 1.6 23.7 30.9 

Total 1,912 34.9 4.3 22.7 38.0 

The last four columns in the table decompose product switches into four different categories: product switches 

that occur within the same two-digit manufacturing sector; to the primary sector; to other two-digit manufacturing 

sectors; and to services. The row total of these four categories equals 100 percent for each two-digit industry. Values 

are percentages, unless otherwise indicated. 

Table 2.3: Four-digit product switches decomposed


Table 2.4: Destination of product switches by technology class 

Notes: Sectors are classified according to technology intensity, based on the classification of industry according to technology 

level from Eurostat. The list of NACE (Rev. 1.1) codes assigned to each particular technology class is given in appendix 

tables 2.A.1 and 2.A.2. 

N 16,117 452 1,912 52 122 298 630 174 445 191 

High tech 5.1 2.7 6.4 25.2 12.2 4.9 5.7 17.1 32.5 2.4 

Medium-high tech 9.8 8.0 13.9 4.5 21.1 20.7 2.3 7.5 32.3 11.7 

Medium-low tech 20.0 16.4 23.6 1.3 10.4 47.0 7.5 7.5 15.5 10.6 

Low tech 65.0 73.0 56.1 0.3 0.4 2.3 54.4 9.2 23.2 10.2 

Total 100.0 100.0 100.0 2.7 6.4 15.6 32.9 9.1 23.3 10.0 

Destination industry by technology class 

Number Number Number Manufacturing Services 

Origin of of of product High Medium- Medium- Low Knowl. Less-knowl. 

industry firms exits switches tech high-tech low-tech tech intensive intensive other

Determinants of firm dynamics 63 

and medium-low-tech industries account for 13.9 percent and 23.6 percent of 

switches, respectively, which is more than the proportion of observations in 

these groups (10 percent and 20 percent, respectively). Meanwhile, about 

three-fourths of the total number of bankruptcies take place in low-tech 

manufacturing sectors. Within the manufacturing sector, firms switch mostly 

to products with similar technology content. This is especially the case for 

the medium-low-tech to low-tech firms, where half of all firms switch to other 

sectors in the same technology class. Table 2.4 also shows that a significant 

proportion of manufacturing firms are moving into the less-knowledgeintensive 

services sector, independently of the technology intensity of the 

origin industry. 

2.3 Determinants of firm dynamics 

At the end of each observed period, a firm decides whether to continue its 

activities in the same product line or to exit. It can exit from a particular 

product line and enter a new product line or even a new industry in the 

next year, or it can exit altogether (i.e., “die”), in which case the firm drops 

permanently out of the sample. To model these strategic alternatives, we 

estimate a multinomial logit model, as follows (Greene, 2008, p. 844): 

Pr Yit+1 = j 

Xt ¯ ′ 

= exp β jXit 

 

1 + 

2 

k=1 

exp (β ′ kXit) 

 

(2.1) 

where j equals 0 for continuing firms, 1 for firms that switch products 

(industries), and 2 for closing enterprises. The vector of covariates (Xit) 

contains a number of one-year lagged firm- and product-level variables, in 

addition to a constant, year dummies, and two-digit industry dummies: 

Firm level: 

ln(Sizeit), ln(Ageit), ln(Capitalit), ln(Wageit), ln(TFPit), Foreignit 

NACE (Rev. 1.1) codes assigned to each particular technology class.


Domestic market: 

International market: 

Sunkjt, Herfjt 

ln(Importsjt), IITjt, CAjt, ln(Exportsjt), Herfexjt, ln(UV Rjt) 

Subscript i refers to firms, j to products and t to time. All productlevel 

variables are defined at the four-digit NACE level. For the complete 

definitions of these variables, we refer to section 2.A.2. 

Whereas the determinants of firm and plant death have been an active area 

of empirical and theoretical research, the literature on product switching is 

still in its infancy. Our motivation for deciding which variables to include in 

the multinomial logit model is therefore grounded in the relevant literature 

and complemented by the salient facts emerging from the summary statistics 

presented in Table 2.5. Our goals are (1) to identify groups of variables–firm 

and product level–that may matter for the exit/switching decision; and (2) 

to check whether they differ according to the exit strategy. 

2.3.1 Firm characteristics 

A common feature of the theoretical models on heterogeneous firms is the 

negative relation between failure rates and a firm’s age and size (Jovanovic, 

1982; Hopenhayn, 1992 and Ericson and Pakes, 1995). This selection effect is 

driven by the interaction of economies of scale and an idiosyncratic learning 

process, and is confirmed by a large empirical literature 12 . To capture this 

scale effect, we include a firm-size variable–measured by employment in year t 

(Sizeit). Additionally, Dunne, Klimek and Roberts (2005) demonstrate that 

a firm’s past experience positively affects its survival rate. We control for this 

experience effect by including the age of the firm (Ageit), which is calculated 

12 The existence of economies of scale implies, on the one hand, sunk costs for larger 

firms, which discourage or delay exit, and, on the other hand, strong cost disadvantages 

for smaller companies. Moreover, larger firms are in general more diversified, which makes 

them less vulnerable to negative productivity shocks. See Caves (1998) for an overview of 

the empirical literature on firm turnover.


as the number of years since its official registration date. A priori, we expect 

exit to be negatively related to size and age. However, the effect of these 

variables on product switching is ambiguous: small and young firms have 

greater flexibility to switch but larger and older firms have more experience 

and resources to switch 13 . 

A further implication of the theoretical models on firm dynamics is the 

link between the productivity of a firm and its survival. Upon entering the 

market, firms pay a sunk cost of entry, after which they discover their true 

productivity. If firm productivity is below the zero profitability cutoff, the 

firm will immediately exit the market. Hence, from these models, a negative 

relationship between firm-level productivity and exit is predicted, a finding 

which is confirmed by the empirical literature. Fariñas and Ruano (2005) 

explore the relationships among entry, exit, and firm-level productivity for 

a sample of Spanish manufacturing firms. They find that entry and exit 

decisions by firms are systematically related to productivity differences and 

that the productivity distribution of exiting firms is stochastically dominated 

by that of continuing firms. We therefore expect firm productivity to be 

negatively related to firm death. 

Switching products could be seen as a form of exit, where unproductive 

companies that cannot face competition turn to a market with a lower degree 

of competition. Yet, entering a new industry or product market implies that 

a firm has to incur additional adjustment costs, related to the production of a 

new good. If these sunk costs of entry in a new product market are substantial, 

the same reasoning as above applies, that is, only the more productive 

firms will be able to enter this product market and start production. We 

therefore expect firm-level productivity to be negatively or positively related 

to product switching –depending on the importance of product market entry 

barriers. Productivity is defined as total factor productivity (TFPit). We 

estimate TFP at the two-digit industry level, while allowing firms to switch 

industries over time, and apply the methodology of Levinsohn and Petrin 

13 In fact, Greenaway et al. (2008) do not find significant effects of age and size on 

industry switching. Bernard et al. (2006a), on the other hand, find a positive coefficient 

for size and a negative coefficient for age (both significant).


(2003) to account for selectivity and simultaneity 14 . 

The labor cost variable (Wageit) is defined as total real labor costs per 

employee at the firm level. The expected effect of this variable on firm dynamics 

is ambiguous. To the extent that higher labor costs reflect higher 

skill intensities and associated sunk costs at the firm level (related to investments 

in firm-specific human capital), higher wages can, ceteris paribus, act 

as a barrier to exit from a particular product market. Audretsch and Mahmood 

(1995) find a negative relationship between wages and the propensity 

to exit in a study on the performance of 12,000 U.S. manufacturing plants. 

However, it is not clear whether this relationship will continue to hold when 

productivity at the firm level is taken into account. If firms pay higher wages 

for a given level of productivity, this could signal lower competitiveness and, 

hence, increase their exit probability, ceteris paribus (Konings, 2005). 

Capital intensity (Capitalit) is defined as total fixed assets per employee. 

Firms with higher capital intensity are expected to face higher sunk costs, 

which will act as an exit barrier both at the product and firm level. Firms 

with larger capital stock can also expect larger future returns for a given level 

of current productivity and, hence, will continue operating at lower productivity 

levels (Olley and Pakes, 1996). Therefore, we expect capital intensity to 

be negatively related to firm exit. In the case of product switching, the overall 

effect is less clear-cut. Even though capital intensity can be interpreted 

as a form of sunk cost, an opposite force may be at work. Instead of being 

passive or defensive when faced with increasing competitive pressures, a firm 

can actively look for the new opportunities offered by globalization. The 

production of these new goods requires additional investment to enter the 

new product market, which can be more easily incurred by capital-intensive 

firms that liquidate or transfer their assets into the new sector. Hence, the 

sign of the capital intensity variable for product switching is ambiguous. 

Several empirical studies have provided evidence that foreign multinational 

enterprises are more footloose than domestic firms, that is, they are more 

14 For a detailed description of the methodology employed to estimate TFP using the 

current data set, we refer to Moreno Badia and Slootmaekers (2008).


likely to exit the market than domestic firms of comparable size, productivity, 

and wages (Görg and Strobl, 2003; Bernard and Sjöholm, 2003; and 

Van Beveren, 2007a). However, foreign multinationals typically lack in-depth 

knowledge of the host market and need to overcome substantial disadvantages 

when entering foreign markets, causing them to incur higher sunk costs and 

hence reducing their exit probabilities. Moreover, since they usually have 

more diversified sources of income, they can withstand larger shocks before 

being forced to exit the market. Hence, the effect of foreign ownership 

(Foreignit) on exit is ambiguous. Nevertheless, since multinationals are, by 

their very nature, more flexible than purely domestic firms, they can respond 

more quickly to adverse shocks in the host country and reinvent themselves 

through product switches. Hence, we expect to find a positive relationship 

between foreign ownership and product switching. 

2.3.2 Product market characteristics: Domestic 

Besides firm structure, the characteristics of the product market in which 

a firm operates also affect its evolution. While international competition 

may exert a strong influence on firm dynamics, conditions in the domestic 

market can also play an important role. This is especially true in a country 

with a history of a planned economic system, and where the domestic market 

may still be undergoing substantial changes as state monopolies are broken 

up and a new private sector emerges. Hence, in our empirical model on 

exit strategies, we incorporate the product market characteristics of both 

the home and foreign markets in which Estonian firms are active. 

Hopenhayn (1992) illustrates how an increase in sunk entry costs lowers 

the entry rate and, hence, also the probability of exit since incumbent firms 

face less competition through new entry. Intuitively, the argument is as 

follows. High initial investment costs to enter a product market will act as a 

natural deterrent to entry, since only the most promising firms will be able to 

start production. A lower entry rate implies less competition for incumbent 

producers and will induce fewer firms to exit. In addition, the high initial 

investment cost will act as an exit barrier, forcing inefficient firms to stay in 

the market to recover at least some of the initial sunk cost. We define sunk 

costs (Sunkjt) as the natural logarithm of the median of real sales in each


particular four-digit industry j at time t 15 . We expect to find a negative 

relationship between sunk costs at the product level and both types of exit 

in our empirical analysis, since higher sunk costs are associated with both 

higher entry and exit barriers at the product level. 

A central prediction of the stochastic dynamic model in Asplund and Nocke 

(2006) is that the level of firm turnover is positively related to the size of the 

domestic market. The smaller average markup in larger markets, resulting 

from tougher competition, implies that the marginal surviving firm has to be 

more efficient in larger than in smaller markets. To capture this competition 

effect, we include the Herfindahl-Hirschman index for the domestic market 

(Herfjt). It is defined as the sum of the squares of the market shares of each 

individual firm. As such, it ranges from 0 to 1 as it moves from a very large 

amount of very small firms to a single monopolistic producer. 

As noted by Görg and Strobl (2003), the impact of industry concentration 

on firm turnover is ambiguous. On the one hand, higher concentration is 

associated with wider price-cost margins, which will increase the survival 

chances of firms. However, behavior by aggressive rivals in a concentrated 

market can actually raise exit probabilities. Görg and Strobl (2003) in fact 

find a positive relationship between industry concentration and exit, using 

data on the Irish manufacturing sector between 1973 and 1996. Hence, the 

impact of concentration on firm exit (whether through firm death or product 

switching) can be negative or positive, depending on the behavior of the 

firm’s rivals in the domestic market. 

2.3.3 Product market characteristics: International 

Increased international trade implies higher competitive pressure in the 

domestic market. This pressure may force firms to improve their efficiency or 

to switch to products in which they have a comparative advantage. However, 

greater economic openness also creates new business opportunities in foreign 

markets. With improved access to the international markets, firms can raise 

15 This measure is known in the literature as the minimum efficiency scale (MES). We 

prefer the MES based on sales to a MES based on median employment in the sector, since 

the latter is not able to capture the fixed costs of capital-intensive industries adequately.


their sales and expand their production capacity to benefit from economies of 

scale, or explore new product markets with better prospects. To analyze the 

impact of increasing globalization, we include a number of variables capturing 

various aspects of international trade. 

In theory, the impact of trade on exit could be driven by two aspects: import 

competition, through smaller markups (Melitz and Ottaviano, 2008) and 

export intensity (Melitz, 2003). The empirical literature so far has mainly 

focused on the first aspect, import competition, which is associated with 

higher exit and switching rates (Bernard et al., 2006a; Coucke and Sleuwaegen, 

2006; and Greenaway et al., 2008). A recent study by Colantone and 

Sleuwaegen (2007) draws attention to the significance of the second aspect of 

international competition: export intensity. In this case, the most successful 

firms self-select into the export market and continue to grow by capturing 

new market opportunities abroad. This raises the average efficiency level and 

increases the pressure on factor prices in the home market, thereby crowding 

out the least efficient firms. At the same time, firms in sectors with promising 

exporting markets will be less likely to switch sectors. To capture both aspects 

of increasing international competition, we include imports (Importsjt) 

and exports (Exportsjt), both defined at the four-digit product level, in our 

empirical model 16 . We expect imports to have a positive effect on exit and 

product switching. Exports, however, are expected to have a positive effect 

on exit but a negative one on product switching. 

Rising intra-industry trade due to falling trade costs raises the number of 

products supplied in a market. This, in turn, raises competition and cuts 

into firms’ markups. Only the more productive firms survive this pressure, 

while others are forced out of business. To capture this effect, we include 

the Grubel-Lloyd intra-industry trade index (IITjt) in our empirical model, 

defined as [1 − (|Xjt − Mjt|/(Xjt + Mjt))]. The effect of IIT on exit and 

switching is however ambiguous since this index captures both the effect of 

16 Ideally, we would have preferred to weight imports and exports at the product level 

by domestic production and include measures of import penetration and export intensity 

in the empirical model. However, no reliable data exist on domestic production by sector 

(either at the two-digit industry or four-digit product level) for Estonia. Hence, we include 

the value of imports and exports in our empirical model.


import competition and export opportunities. 

In product markets where Estonia has a revealed comparative advantage 

and, therefore, good export opportunities, we expect fewer exits irrespective 

of its form (either firm closedown or product switch). The comparative 

advantage dummy (CAjt) takes the value of one if exports are larger than 

imports for a given four-digit product. 

Finally, we want to check how firms’ strategies are related to changes in the 

quality embedded in Estonian exports. A substantial amount of theoretical 

work predicts that quality systematically affects the direction of international 

trade, a finding that is confirmed by some recent empirical papers (eg. Hallak, 

2006) 17 . As a measure of product quality, we use a composite index of the unit 

value of Estonia’s exports in a given geographic market relative to the unit 

value of all exporters in that market (UV Rjt). On the premise that a higher 

relative price reflects higher quality than direct competitors’, UV Rjt acts as 

a proxy for product quality (Hallak and Schott, 2008). However, concerns 

remain that this measure could be picking up factors other than quality. 

This is especially the case if local monopolies exist and competition does not 

arbitrage away differences in quality-adjusted prices. To control for this effect 

of markups on the unit value of exports, we include the Herfindahl-Hirschman 

index of the export market (Herfexjt). This measure of competitiveness is 

a composite index of the weighted sum of the Herfindahl-Hirschman indices 

for each of the relevant geographical export markets. 

Table 2.5 gives a first indication of the differences in characteristics across 

the exit strategies by summarizing a number of firm- and product-level characteristics 

for three groups separately: (1) all firms in the sample; (2) firms 

undergoing either an industry switch (two-digit) or product switch (fourdigit); 

and (3) exiting firms. 

17 For a theoretical background on trade and quality, see, among others, Falvey and 

Kierzkowski (1987); Flam and Helpman (1987); Stokey (1991) and Murphy and Shleifer 

(1997). Empirical papers on this topic include Schott (2004); Schott (2004); Hummels and 

Klenow (2005) and Hallak and Schott (2008).


Industry Product 

Variable All switch switch Exit 

Number of observations 16,117 1,244 1,912 452 

(Percentage of total) (100.0 ) (7.7) (11.9 ) (2.8) 

Size 28.1 21.6*** 26.9 20.7*** 

(Number of employees) (100.4) (73.9) (131.4) (46.2) 

Age 6.9 6.1*** 6.3*** 5.7*** 

(Years) (3.5) (3.3) (3.3) (3.1) 

Capital 104.7 119.7** 109.7 91.4 

(Thousands of Krooni) (270.8) (307.6) (266.6) (455.6) 

Wage 60.3 60.3 58.4* 50.7** 


TFP 61.2 60.4 58.3** 41.3*** 


Foreign 0.1 0.1 0.1** 0.1** 

(Ownership dummy) (0.3) (0.3) (0.3) (0.3) 

Sunk 14.7 14.7* 14.8 14.7 

(Minimum Efficient Scale) (0.7) (0.7) (0.7) (0.6) 

Herf 0.1 0.1*** 0.1*** 0.1** 

(HHI domestic market) (0.2) (0.2) (0.2) (0.2) 

Imports 417,270 489,774*** 448,386** 337,168*** 

(Imports, thousands of Krooni) (583,623) (707,494) (644,580) (349,766) 

IIT 0.5 0.5*** 0.5*** 0.5** 

(Intra-industry trade) (0.3) (0.3) (0.3) (0.2) 

CA 0.6 0.5*** 0.5*** 0.6*** 

(Comparative advantage) (0.5) (0.5) (0.5) (0.5) 

Exports 766,759 715,993** 692,268*** 807,780 

(Exports, thousands of Krooni) (1,046,466) (1,050,674) (1,019,043) (973,382) 

Herfex 0.2 0.2* 0.2 0.2 

(HHI export market) (0.1) (0.1) (0.1) (0.1) 

UVR 1.3 1.2 1.3 1.3 

(Relative unit values) (3.3) (2.7) (4.1) (3.8) 

Notes: Reported values are means (standard deviations), with the exception of the first 

row. Significance levels (*** p


Compared with continuing firms, enterprises that exit or switch industries 

are significantly smaller and younger. Firms that switch industries are on 

average more capital intensive than continuing firms, while both exiting firms 

and product switching firms have a significantly lower labor cost and productivity. 

Turning to product characteristics, we find that industry switchers 

tend to be active in industries with a higher level of sunk costs, while the opposite 

is true for exiting firms. Switchers also tend to come from sectors with 

significantly higher market power, as indicated by the Herfindahl-Hirschman 

index for the domestic market. Sectors characterized by higher imports, less 

intra-industry trade and lower exports display a higher rate of industry and 

product switching than other sectors. Prior to switching, these firms tend 

to be in sectors with relatively lower revealed comparative advantage. Enterprises 

that permanently exit the market, however, are active in sectors 

with lower imports and higher exports. They also tend to be more present 

in industries with revealed comparative advantage than do continuing firms. 

2.4 Results 

2.4.1 Baseline results 

In this section we report the results from a multinomial logit regression in 

which we analyze the determinants of three alternative strategies at the firm 

level: (1) stay active in the same product market (the baseline category); 

(2) change its main product line or (3) exit entirely from the market. Table 

6 reports the coefficients and standard errors as well as the marginal effects 

of the variables on the probability of exit (whether in the form of a product 

switch or a true exit). These marginal effects are calculated at the mean 

of the independent variables, to provide some guidance on the magnitude 

of the effect (reported in italics in the tables). We pool the observations 

across years for all firms in the sample, and we include year and two-digit 

industry fixed effects to control for aggregate variation in industry dynamics. 

Standard errors are clustered at the firm level. 

The results on the firm characteristics confirm our priors on firm dynamics 

as discussed in Section 2.3. Controlling for size at the firm level, exiting firms

Results 73 

are on average younger, less productive and have lower capital intensity 18 . 

These results suggest that the performance of these firms is not good enough 

to keep up with the dynamics in the market. The probability of product 

switching, on the other hand, is significantly decreasing with plant size and 

age, while significantly increasing with firm productivity and capital intensity. 

In other words, smaller and younger firms are more likely to change their 

main product line than their older and larger counterparts. Moreover, only 

the more productive and capital-intensive firms switch to different product 

markets. This is a first indication that switches are not necessarily driven by 

a lack of competitiveness but are rather the outcome of firms’ own choices. 

Controlling for the other firm-level characteristics discussed above, foreign 

ownership of the firm has no significant impact on either product switching 

behavior or exit. A possible explanation for this can be found in the motives 

of foreign firms to invest in central and eastern European countries. Although 

cost advantage plays a role, various studies illustrate the importance of high 

market potential as an incentive for foreign firms to enter these markets 19 . 

This suggests that these investments are partly strategic and forward looking 

and that foreign firms will not necessarily exit the market more rapidly than 

domestic firms when faced with short-run adverse shocks. 

With respect to the conditions in the domestic market, the sunk cost variable 

is never significant, whereas the coefficient on the Herfindahl-Hirschman 

index only has a positive and significant sign for product switching. As noted 

by Caves (1998), concentration and sunk costs of a particular industry are simultaneously 

determined, since the forces of exit and entry will influence the 

equilibrium number of firms in an industry. Hence, including both variables 

together as independent variables may be the reason for the insignificant coefficients 

on sunk costs. As a robustness check, we ran the empirical model 

including either only the Herfindahl-Hirschman index or only the sunk cost 

measure in our basic specification, but the results are equivalent to those 

18 The marginal probabilities for exit are quite small in magnitude because the likelihood 

of exit in any period is relatively small. 

19 See, for instance Bevan and Estrin (2004) and Carstensen and Toubal (2004).


Variables Product switch Exit 

log(Employment) -0.143*** -0.014 -0.036 0.000 

[0.026] [0.054] 

log(Age) -0.178*** -0.017 -0.300*** -0.005 

[0.044] [0.072] 

log(TFP) 0.092** 0.010 -0.367*** -0.007 

[0.041] [0.066] 

log(Capital) 0.051** 0.005 -0.220*** -0.004 

[0.020] [0.036] 

log(Wages) -0.080 -0.008 0.169 0.003 

[0.049] [0.108] 

Foreign (d) -0.088 -0.008 0.044 0.001 

[0.094] [0.185] 

Herfindahl domestic market 0.253* 0.025 -0.230 -0.004 

[0.144] [0.334] 

Sunk 0.008 0.000 0.139 0.002 

[0.057] [0.093] 

log(Exports) -0.064* -0.006 0.063 0.001 

[0.038] [0.068] 

Herfindahl export market 0.007 0.002 -0.739 -0.013 

[0.386] [0.679] 

log(UVR) -0.102* -0.010 0.07 0.001 

[0.055] [0.105] 

log(Imports) 0.014 0.001 -0.011 0.000 

[0.049] [0.091] 

IIT -0.212* -0.020 -0.135 -0.002 

[0.128] [0.235] 

CA (d) -0.218** -0.021 -0.011 0.000 

[0.106] [0.207] 

Number of observations: 16,043 

Pseudo R-square: 0.054 

Notes: This table reports the results from a multinomial logit (0=continuing; 

1=switching products; 2=closing). Robust standard errors are in brackets 

below coefficient estimates. The numbers in italics next to the coefficient 

estimates represent the marginal probability change at the mean of the independent 

variable or the discrete change of a dummy variable (d) from 0 to 

1. Though not reported, all regressions include a constant, two-digit industry, 

and time fixed effects. Standard errors are clustered at the firm level. 

Significance levels: *** p

Results 75 

shown in Table 2.6 20 . The positive coefficient on the Herfindahl-Hirschman 

index for product switching tells us that firms are more likely to switch away 

from the more concentrated industries. A marginal increase in average sector 

concentration pushes the probability of switching up with 2.5 percentage 

points. This result suggests that firms confronted with the aggressive behavior 

of their rivals may be persuaded to exploit opportunities in a different 

product market. 

Contrary to our initial expectations we do not find that international competition 

is driving firm exit in Estonia. Conversely, switching is strongly 

influenced by the product market evolution, both in the domestic and international 

scene. As a rule, Estonian firms active in a sector with more exports 

should be able to compete in global markets. Export-intensive sectors are 

therefore more dynamic and promising for companies. Unfortunately, the 

data do not allow us to distinguish between exporters and nonexporters at 

the firm level. However, following Melitz’s (2003) arguments, developments 

in the export market will have repercussions on domestic market structure. 

In this spirit, we expect enterprises to react to changes in global markets to 

remain viable in an increasingly competitive environment. This idea is reflected 

in all export variables listed in Table 2.6. The comparative advantage 

variable indicates that firms have a 2.1 percentage point lower probability 

to switch away from products in which Estonia has a revealed comparative 

advantage, that is, products for which exports are larger than imports. On 

the other hand, the total value of exports decreases the likelihood of product 

switching but does not have a significant impact on the likelihood of exit. 

This is in contrast with Melitz’s argument that the least productive firms, 

confronted with international competition (i.e. exports), will be forced to 

close down. 

A similar idea is reflected in the result on intra-industry trade for product 

switching. Higher rates of intra-industry trade denote that Estonia is 

simultaneously exporting and importing a particular product, and thus that 

the number of varieties supplied in that product market is higher. This in- 

20 Results are not reported here for brevity but can be obtained from the authors upon 

request.


crease in intra-industry trade could either represent stiffer competition from 

imports or better export opportunities. Whereas inefficient firms could be 

driven into bankruptcy due to this increased pressure, the more productive 

firms may take advantage of these opportunities. Indeed, controlling for firm 

and domestic market characteristics, firms are less likely to leave industries 

in which export opportunities abound. A rise of intra-industry trade leads 

to a decrease in the probability of product switching. It does not generate, 

however, a destructive competition effect on domestic firms, as reflected in 

the insignificant coefficient for exit. Turning to the unit value variable, we 

find that, to the extent that price differences reflect differences in quality, 

there is a negative relation between the quality of Estonia’s exports and the 

probability of switching. In other words, the lower the quality of the products, 

the more likely firms will change their products. Thus, firms tend to 

exit low-quality exporting sectors. 

Existing empirical work, in particular the paper of Bernard et al. (2006a), 

focuses on the impact of imports coming from low-wage countries. The authors 

provide evidence, using data on the U.S. manufacturing sector, that 

a higher degree of import penetration from low-wage countries is associated 

with a higher probability of exiting and switching products. However, in the 

case of Estonia, the origin of imports does not seem to play a role in the 

strategic decisions taken by firms 21 . 

Overall, we notice that the determinants of product switching are very different 

from the determinants of firm death. On the one hand, the insignificant 

coefficients for all domestic and international product market characteristics 

for firm exit show that bankruptcy in Estonia is entirely driven by the firm’s 

own behavior, rather than by a reaction to external factors. On the other 

hand, the results for our trade variables suggest that Estonian firms are exploiting 

opportunities in global markets. These findings imply that product 

switching might be more than just an alternative way of escaping from increasing 

competition. Both firm and industry characteristics point toward 

an active policy of looking for new and better opportunities. Rather than 

21 Results are not reported here for brevity, but can be obtained from the authors upon 

request.

Results 77 

switch products out of defense, firms seem to be changing their product lines 

out of choice. In the following sections, we will investigate this hypothesis in 

more detail. 

2.4.2 Self-selection into new markets 

Our results suggest that firms systematically self-select into new product 

markets on the basis of their performance in and knowledge about the 

market. This result is a first new insight in firm dynamics and an important 

contribution to the existing empirical literature which restricted its attention 

to the import side of international competition. Whereas these studies find 

that firms in industrial countries change their product lines out of defense 

against low-cost competition, we find that Estonian firms follow an offensive 

strategy by actively exploring the market. In this subsection we want to 

dig deeper into this new facet of product switching. We start by exploring 

potential differences between product switching and industry switching. Afterward, 

we look more closely at the characteristics of the product markets or 

industries to which firms switch. In particular, we investigate the differences 

in export unit value ratio and technology intensity between the origin and 

destination industry. 

Industry versus product switching 

In order to compare the determinants of industry (two-digit) and product 

(four-digit) switches, we split the group of product switches into (1) product 

switches that are not observed at the two-digit level, that is, firms that 

change their main four-digit product line but stay within the same two-digit 

industry; and (2) industry switchers, that is, those firms that switch to a 

new product line in a different two-digit industry. Using this distinction, 

we estimate the same multinomial logit model as before, except that the 

dependent variable now takes on four different values (rather than three): 

0 for firms that stay in the same product market, 1 for firms that switch 

products but not industries, 2 for firms that change two-digit industries, and 

3 for exit.


Variable Product switch Industry switch Exit 

log(Employment) 0.070* 0.003 -0.258*** -0.016 -0.036 0.000 

[0.040] [0.033] [0.054] 

log(Age) -0.201*** -0.006 -0.166*** -0.010 -0.300*** -0.005 

[0.071] [0.053] [0.072] 

log(TFP) 0.058 0.002 0.104** 0.007 -0.368*** -0.007 

[0.063] [0.050] [0.066] 

log(Capital) 0.01 0.000 0.070*** 0.005 -0.220*** -0.004 

[0.035] [0.024] [0.036] 

log(Wages) -0.091 -0.003 -0.069 -0.004 0.169 0.003 

[0.078] [0.059] [0.108] 

Foreign (d) -0.169 -0.005 -0.045 -0.003 0.045 0.001 

[0.151] [0.117] [0.185] 

Herfindahl domestic market 0.194 0.005 0.275* 0.017 -0.229 -0.004 

[0.251] [0.165] [0.335] 

Sunk costs 0.003 0.000 0.003 0.000 0.14 0.002 

[0.109] [0.062] [0.094] 

log(Exports) -0.156*** -0.005 0.002 0.000 0.064 0.001 

[0.058] [0.046] [0.068] 

Herfindahl export market -0.738 -0.023 0.426 0.029 -0.743 -0.013 

[0.697] [0.436] [0.679] 

log(UVR) -0.277** -0.008 -0.009 0.000 0.071 0.001 

[0.111] [0.060] [0.105] 

log(Imports) -0.026 -0.001 0.042 0.003 -0.011 0.000 

[0.074] [0.060] [0.091] 

IIT -0.149 -0.004 -0.262 -0.016 -0.136 -0.002 

[0.191] [0.160] [0.236] 

CA (d) -0.268* -0.008 -0.209 -0.013 -0.011 0.000 

[0.162] [0.129] [0.208] 

Number of observations 16,043 

Pseudo R-square 0.059 

Notes: This table reports the results from a multinomial logit regression (0=continuing; 

1=product switching within the same industry; 2=industry switching; 3=closing). Robust 

standard errors are in brackets below coefficient estimates. The numbers in italics next 

to the coefficient estimates represent the marginal probability change at the mean of the 

independent variable or the discrete change of a dummy variable (d) from 0 to 1. Though 

not reported, all regressions include a constant, two-digit industry, and time fixed effects. 

Standard errors are clustered at the firm level. Significance levels: *** p

Results 79 

Table 2.7 shows a striking difference between the determinants of industry 

switching and those of product switching. International trade does not seem 

to play a significant role in the dynamics behind industry switching. Firms 

change to a different two-digit industry in response to changes in their own 

performance and the domestic market, but not on account of international 

trade aspects. We also find that, while more productive and more capitalintensive 

firms are more likely to switch industries, the effect of these variables 

on product switches is insignificant. This implies that the results obtained 

earlier on these variables are driven by industry switches rather than product 

switches. The insignificant effect of the trade variables on industry switches, 

along with the negative link between exports and unit values, on the one 

hand, and product switching, on the other hand, further suggests that the 

switching pattern in Estonia is determined by product differentiation. Enterprises 

observe changes within their sector and respond by modifying their 

products to take part in the growth process. 

As was already noted in section 2.2, firms that switch industries towards 

the services sector are retained in the sample until their last year of activity 

in the manufacturing sector. This allows us to delve deeper into the 

determinants of industry switching by comparing firms that switch within 

the manufacturing sector to firms that switch towards services. In our sample, 

7 percent of the 1,244 industry switches are to the primary sector, 35 

percent stay within the manufacturing sector, and 58 percent of industry 

switches are to the services sector. Each of these dynamics is likely to be 

driven by diverging underlying causes. Given the growing importance of the 

services sector in Estonia’s domestic economy, the switches to services are of 

particular interest to us. 

We therefore define a new dependent variable that equals zero for twodigit 

switches within manufacturing and 1 for switches to services, and run 

a logit regression (Table 2.8). In this case, we do not compare stayers with 

switchers, but instead explore dissimilarities among the switchers. Similar 

to our previous regressions, we include time and two-digit industry fixed 

effects, and cluster the standard errors at the firm level. This regression 

reveals that larger firms and foreign firms are more likely to switch within


Variable Coefficient Marginal effect 

log(Employment) -0.461*** -0.111 

[0.063] 

log(Age) 0.072 0.017 

[0.103] 

log(TFP) 0.182** 0.044 

[0.083] 

log(Capital) -0.021 -0.005 

[0.045] 

log(Wages) 0.135 0.033 

[0.107] 

Foreign (d) -0.401* -0.098 

[0.208] 

Herfindahl domestic market 0.244 0.059 

[0.312] 

Sunk 0.133 0.032 

[0.118] 

log(Exports) 0.058 0.014 

[0.096] 

Herfindahl export market 2.380** 0.572 

[1.026] 

log(UVR) -0.009 -0.002 

[0.124] 

log(Imports) -0.221* -0.053 

[0.123] 

IIT -0.173 -0.042 

[0.308] 

CA (d) 0.312 0.075 

[0.266] 

Number of observations: 1,244 

Pseudo R-square: 0.059 

Notes: Results reported are from a logit estimation comparing (two-digit) industry 

switches to other manufacturing sectors (dependent variable equal to 0) with 

industry switches to services (dependent variable equals 1). Robust standard errors 

are in brackets below coefficient estimates, the numbers in italics next to the 

coefficient estimates represent the marginal probability change at the mean of 

the independent variable or the discrete change of a dummy variable (d) from 0 

to 1. Though not reported, all regressions include a constant, two-digit industry 

dummies, and time dummies. Standard errors are clustered at the firm level. 


Results 81 

the manufacturing sector. Conversely, the more productive firms tend to 

leave the manufacturing sector and enter the services sector. A closer look 

at the destinations reveals that the majority of the firms enter the wholesale 

trade and retail trade sectors 22 . 

Turning to the trade variables, we see that a higher concentration in exporting 

markets drives firms into the services sector while import growth is 

associated with industry switches within manufacturing. Further data analysis 

reveals that the increase in switching to services is entirely driven by 

switches to the distribution sector 23 . These results suggest that firms tend 

to move from the production side to the distribution sector if the export market 

is highly concentrated. This does not necessarily mean, however, that a 

firm stops producing the product. For example, an enterprise that realizes 

the potential in the market for its products, could continue production but 

shift its attention toward the distribution of its products. Unfortunately, our 

data set provides only firms’ main sector of activity, and not the global picture 

of their activities. We also find that, if imports generate strong pressures 

there is less incentive for the firm to move into the distribution sector, given 

the lack of competitiveness of its product. To minimize the loss of sunk investments, 

it is more efficient to enter into a comparable market where firms 

can continue employing their physical and human capital without much additional 

cost or effort. 

Hence, Estonian firms that are switching in response to changes in the 

international trade environment either move into the distribution of goods 

(some of which they may have produced in the past) or switch toward other 

manufacturing industries. Which switching strategy they adopt depends 

on whether the changes in the global environment are manifested through 

imports or rather driven by the concentration in export markets. These 

results suggest that Estonian companies are aware of the prospects in the 

global market and are trying to exploit these opportunities by proactively 

changing their business plans. 

22 Out of the 727 industry switches to the services sector observed in the sample, about 

54 percent goes to the retail and wholesale sectors (NACE codes 50, 51, and 52). 

23 If we exclude the wholesale and retail sectors from the analysis, no significant results 

are obtained for the export variables.


Quality upgrading versus technology upgrading 

In our baseline results, we observed that Estonian companies tend to leave 

low-quality exporting sectors (as measured by the relative export unit value). 

The question now is to which sectors these firms are moving. Do they switch 

to products with an even lower relative unit value because they are not 

able to compete within the price quality range of the export market, or do 

they switch to sectors with a higher relative unit value because they see 

opportunities at the higher end of the quality array? Fabrizio et al. (2007) 

document an impressive shift in product quality and technology intensity of 

exports for Estonia and other central and eastern European countries over 

the past decade. With these facts in our mind, we now explore the direction 

of switches and accompanying firm characteristics in detail. 

Of the total number of product switches within manufacturing, 483 switches 

occur to markets with a lower relative export unit value, whereas 429 switches 

go in the opposite direction. In other words, almost half of the product 

switches result in quality upgrading. In the first column of Table 2.9, we 

check the firm characteristics behind this shift up the quality ladder. To 

do so, we calculate for each firm that switches within manufacturing the 

log difference in the export unit value ratio between its origin and destination 

industry, at the four-digit level. More specifically, a positive value for 

the log difference stands for a switch to a product with a higher unit value, 

and thus of higher quality. These log differences in export unit value ratios 

are then regressed on a number of firm characteristics using ordinary least 

squares (OLS), while controlling for market power in the export market, as 

well as for industry and time fixed effects. The results indicate a positive link 

between a firm’s capital intensity and quality upgrading. Among the firms 

that alter their product line, only the more capital-intensive firms are able 

to move into higher-quality product markets. Controlling for other characteristics, 

these firms tend to be smaller than the average Estonian switching 

firm. 

To explore whether this quality upgrading is related to technology upgrading, 

we define two dummies to capture the change in technological intensity. 

The dummy Technology upgrading equals 1 if a firm moves towards a sector

Results 83 

Variables (1) (2) (3) 

log(Employment) -0.032* -0.027 0.005 

[0.018] [0.018] [0.020] 

log(Age) 0.028 0.024 0.001 

[0.031] [0.031] [0.034] 

log(TFP) 0.052 0.052 -0.044 

[0.042] [0.043] [0.029] 

log(Capital) 0.048*** 0.048*** 0.043** 

[0.018] [0.018] [0.019] 

log(Wages) -0.017 -0.025 0.016 

[0.046] [0.047] [0.044] 

Foreign (d) 0.022 0.007 0.022 

[0.070] [0.070] [0.069] 

Herfindahl export market -0.453 -0.473 -0.625** 

[0.33] [0.330] [0.250] 

Technology upgrading ... 0.085 ... 

[0.094] 

Technology downgrading ... -0.186** ... 

[0.079] 

Destination High tech ... ... -0.119 

[0.150] 

Destination Medium-high-tech ... ... 0.202** 

[0.091] 

Destination Medium-low-tech ... ... 0.069 

[0.049] 

Industry fixed effects (two-digit) Yes Yes No 

Year fixed effects Yes Yes Yes 

Number of observations: 1,097 1,097 1,097 

R-square 0.097 0.103 0.036 

Notes: The dependent variable is the log difference in export unit 

value ratio between the origin industry and destination industry, at 

the four-digit level. The dummy technology up- (down-) grading 

equals 1 if the firm moves up (down) one category of technology intensity. 

The dummies H-tech, MH-tech, and ML-tech destination equal 

1 if a firm moves to respectively a high-tech, medium-high-tech or 

medium-low-tech sector. The regressions are estimated using OLS, 

and robust standard errors are in brackets below the coefficient estimates. 

Coefficients for the constant and industry and year dummies 

are suppressed. Standard errors are clustered at the firm level. Significance 

levels: *** p


with a higher intensity of technology (this is the case for 97 observations), 

whereas the dummy Technology downgrading equals 1 if a firm moves down 

the technology ladder (123 observations). As can be seen in the second column 

of Table 2.9, technology downgrading is negatively related to the log 

difference in export unit value ratio, whereas the coefficient on technology 

upgrading is positive but statistically insignificant. These results suggest 

that quality upgrading is not necessarily related to technology upgrading. 

Yet this finding is not completely unexpected, as the majority of the product 

changes happen along the same level of technology (877 observations). 

To understand at which technological level this quality upgrading is taking 

place, we return to our original specification, used in column 1 of Table 2.9, 

while adding three dummies identifying the technological intensity of the 

industry of destination for product switches. The results in column 3 of 

Table 2.9 should be interpreted relative to the base category -the low tech 

industry. The positive and significant coefficient on the medium-high-tech 

industry dummy shows that Estonian companies are moving up the quality 

ladder mainly within the medium-high-tech sectors. 

2.5 Robustness checks 

2.5.1 Results by size class 

Because of data constraints, the empirical literature so far has been able to 

look only at the switching behavior of relatively large firms. The U.S. Longitudinal 

Research Database used by Bernard et al. (2006a) includes only 

firms with a minimum of 10 employees, while Greenaway et al. (2008) study 

Swedish manufacturing firms with at least 50 employees. An important advantage 

of our data is the absence of any size thresholds. This allows us 

to draw conclusions for the entire population of manufacturing firms in Estonia 

and also reveals important insights in the dynamics of the smallest 

among them, namely, microenterprises employing fewer than 10 employees. 

This feature is particularly important for a transition country: while a small 

number of large enterprises dominated the economic landscape of Estonia 

during the Soviet era, the transition period has been characterized by the

Robustness checks 85 

1−9 10−19 

20−49 50−99 

100−249 250−499 

500 or more 

Figure 2.1: Sample size distribution 

emergence of many small and medium-sized enterprises. Masso et al. (2004) 

compare the average size of Estonian enterprises to the Organization for Economic 

Cooperation and Development (OECD) average and find that, while 

the mean size is very close to this average, the standard deviation is much 

smaller due to the small number of very large enterprises in Estonia. In fact, 

about 50 percent of manufacturing firms in our sample are microenterprises 

(Figure 1), and more than three fourths of the firms employ fewer than 50 

employees. These microbusinesses are also more dynamic (Table 2.1, Panel 

B). Compared with the sample average (Table 2.1, Panel A), we notice that 

microbusinesses modify their product lines more frequently (13.4 percent in 

the micro sample versus 11.9 percent in the full sample) and have a higher 

exit rate than the average Estonian firm (3.3 percent versus 2.8 percent). 

To analyze differences in determinants across size categories, we run our 

baseline specification for small and big firms separately. Table 2.10 shows 

the results for firms with fewer than 10 employees (product switching in 

column 1 and exiting in column 2) versus the rest of the sample (columns 3 

and 4). The results for the firm-level characteristics are very similar to our 

baseline results in Table 2.6, except for labor cost which turns out to be a


Table 2.10: Determinants of firm dynamics across size categories 

Notes: This table reports the results from a multinomial logit regression (0=continuing; 1=switching products; 2=closing), for two groups: firms with fewer 

than 10 employees (columns 1 and 2) and firms with 10 employees or more (columns 3 and 4). Robust standard errors are in brackets below coefficient 

estimates. The numbers in italics next to the coefficient estimates represent the marginal probability change at the mean of the independent variable or 

the discrete change of a dummy variable (d) from 0 to 1. Though not reported, all regressions include a constant, two-digit industry and time fixed effects. 

Standard errors are clustered at the firm level. Significance levels: *** p


significant determinant of both product switching and exiting for firms with 

10 employees or more. In particular, larger firms with relatively high wages 

are, on the one hand, more likely to go bankrupt and, on the other hand, 

less likely to change their product line. Remember from Section 2.3 that 

high labor cost can either reflect inefficient use of labor or high skill intensity 

of the work force. If the labor costs of a firm are too high compared to its 

competitors, for a given level of productivity, the firm will not be able to 

compete in the market and will face bankruptcy. For firms with a large pool 

of employees the efficient use of its labor force becomes more crucial than 

for small firms where labor costs are not a determinant of its strategies. Yet, 

to the extent that higher wages reflect higher skill intensities, investments 

in product-specific human capital become a sunk cost acting as a barrier to 

change to a different product market. These sunk costs are substantially 

lower for firms with few employees since fewer employees will have to be 

re-trained for the production of the new products. This is reflected in the 

insignificance of the coefficient on labor cost for the smallest firms, whereas 

a marginal change in this variable implies a decrease in the probability of 

product switching of 2.4 percentage points for larger firms. 

Another major difference between small and larger firms is the effect of 

international openness on their switching behavior. Product switching among 

microenterprises does not seem to be driven by the level of exports per se, 

but is fairly sensitive to changes in the relative unit values of Estonia’s export 

products, with an elasticity of around 0.13. The opposite is true for product 

switching among larger firms, which seems to be affected only by the level of 

exports. 

2.5.2 Results by time period 

Now we consider whether the determinants behind firm dynamics have 

changed over time by splitting the sample into two periods, 1997-2000 and 

2001-04 (Table 2.11). Several interesting insights come out of this exercise. 

The firm determinants of product switching are roughly alike across both 

periods. Remarkably, trade emerges as an important driver of the productswitching 

behavior of firms in the first half of the sample period (1997-2000), 

while acting solely as a driver of firm death in the second half of the pe-


riod. This seems to reflect a Melitz-type effect: as the quality of Estonia’s 

exported products increases relative to its competitors in a certain sector, 

the most successful firms will probably self-select into the export market. 

Over time, this raises the average efficiency level in the sector and firms that 

cannot cope with the pace of quality upgrading are forced to exit. On the 

other hand, globalization is no longer encouraging firms to switch products 

or industries in the most recent period. Changes in a firm’s product mix 

are rather driven by evolutions in the domestic market, as reflected by the 

Herfindahl-Hirschman index. 

From a pessimistic point of view, one could argue that the driving forces 

behind the self-selection process of moving into more promising markets have 

tapered off in Estonia. Whereas Estonia significantly expanded its market 

share in the 1990s, this process has decelerated since the beginning of the new 

century. This could indicate that some firms might be losing market share 

because they cannot withstand competition and are unable to proactively 

search for better opportunities. However, from a positive point of view, this 

finding could also be interpreted as a sign that Estonia has reached a steady 

state after the wide-ranging restructuring process of the 1990s. Switching 

and exit rates were substantial at the end of the 1990s before gradually 

declining toward the end of the sample period in 2004. Estonia’s transition 

from a planned economic system was accompanied by extensive privatization 

and restructuring, in conjunction with the dismantling of trade barriers and 

the inflow of foreign investment. This catching-up process has now slowed, 

and changes in the market are mainly serving to keep the industrial sectors 

healthy by forcing the least efficient firms to exit. 


This paper provides new evidence on the link between globalization and 

firm dynamics, focusing on the case of Estonia. We contribute to the literature 

in two important respects. First, this is the first paper to study the 

determinants of exit and product switching in an emerging market. Second, 

we consider explicitly the effect of export market conditions on firm 

dynamics. For that purpose, we include three product-level measures in our


1997 - 2000 (N = 6,090) 2001 - 2004 (N = 9,953) 

Product switch (N = 979)) Exit (N = 262) Product switch (N = 933)) Exit (N = 190) 

log(Employment) -0.130*** -0.017 0.007 0.001 -0.122*** -0.010 -0.032 0.000 

[0.035] [0.069] [0.038] [0.080] 

log(Age) -0.167** -0.019 -0.471*** -0.013 -0.217*** -0.017 -0.152 -0.001 

[0.065] [0.100] [0.055] [0.096] 

log(TFP) 0.114** 0.016 -0.308*** -0.009 0.092* 0.008 -0.385*** -0.004 

[0.058] [0.102] [0.055] [0.083] 

log(Capital) 0.057** 0.008 -0.180*** -0.005 0.040 0.003 -0.271*** -0.003 

[0.029] [0.053] [0.027] [0.046] 

log(Wage) -0.072 -0.010 0.123 0.004 -0.179*** -0.014 -0.014 0.000 

[0.068] [0.156] [0.066] [0.129] 

Foreign (d) -0.196 -0.023 -0.192 -0.004 0.019 0.001 0.342 0.004 

[0.126] [0.255] [0.126] [0.264] 

Herfindahl domestic market -0.132 -0.014 -0.651 -0.018 0.724*** 0.057 0.622 0.006 

[0.195] [0.415] [0.210] [0.560] 

Sunk 0.108 0.014 0.110 0.003 -0.089 -0.007 0.166 0.002 

[0.074] [0.126] [0.080] [0.159] 

log(Exports) -0.085* -0.012 0.114 0.004 -0.031 -0.002 0.003 0.000 

[0.048] [0.090] [0.058] [0.120] 

Herfindahl export market -0.397 -0.053 0.363 0.012 0.595 0.049 -1.918 -0.021 

[0.557] [0.856] [0.547] [1.185] 

log(UVR) -0.156* -0.019 -0.303* -0.008 -0.071 -0.006 0.241** 0.003 

[0.085] [0.169] [0.072] [0.116] 

log(Imports) -0.058 -0.007 -0.010 0.000 0.038 0.003 0.000 0.000 

[0.065] [0.127] [0.074] [0.147] 

IIT -0.233 -0.029 -0.227 -0.006 -0.108 -0.009 0.112 0.001 

[0.171] [0.324] [0.191] [0.401] 

CA (d) -0.310** -0.039 -0.220 -0.005 -0.216 -0.018 0.100 0.001 

[0.155] [0.287] [0.144] [0.322] 

Pseudo R-square 0.039 0.035 

Notes: This table reports the results from a multinomial logit regression (0=continuing; 1=switching products; 2=closing), for two periods: 1997-2000 

(columns 1 and 2) and 2001-04 (columns 3 and 4). Robust standard errors are in brackets below coefficient estimates. The numbers in italics next to the 

coefficient estimates represent the marginal probability change at the mean of the independent variable or the discrete change of a dummy variable (d) 

from 0 to 1. Though not reported, all regressions include a constant, and two-digit industry fixed effects. Standard errors are clustered at the firm level. 



estimation: (1) the value of exports; (2) the degree of competition in export 

markets; and (3) the quality of exports relative to direct competitors. 

Our results indicate that globalization is generally not an important driver 

of firm exit, while it emerges as an important factor explaining product 

switching. What matters for exit are firm characteristics: younger firms 

and those with lower productivity and capital intensity are more likely to 

exit. Meanwhile, product switching is also affected by conditions in export 

markets. In particular, firms are more likely to switch if they are in sectors 

with relatively small revealed comparative advantage, and where the total 

value and quality of exports are relatively low. However, this effect only 

matters for product switches and for the early sample period when trade flows 

were increasing rapidly. A possible interpretation of our results is that firms 

initially moved out of products for which prospects in the export markets, 

which were increasingly opening up, were not good. This result is in contrast 

with previous studies on industrial countries which have found that firms 

change their product line as a defensive strategy against low-cost imports. 

Finally, we find that firms switching to relatively higher quality products are 

more capital intensive; however, these switches are not related to technology 

upgrading. 

Our findings raise a number of questions worthy of further research. First, 

it would be interesting to know whether the effect of intra-industry trade 

on product switching is related to trade in different products (vertical intraindustry 

trade) or similar products (horizontal intra-industry trade). Second, 

additional theoretical models need to be developed to gain further insights 

into the determinants of product switching versus industry switching, occurring 

both within the manufacturing sector and to services. Finally, it 

could be important to explore whether the quality of exports matters for the 

product switches irrespective of the exporting status of the firm.

Data appendix 91 

2.A Data appendix 

2.A.1 Data and sample selection 

Trade data 

The trade data are from the UN Comtrade (commodity trade statistics) 

database and consist of the trade values and quantities of import flows. We 

include all goods (not just manufacturing) and use import flows because reporting 

of imports is generally more reliable than that of exports. This means 

that Estonia’s exports are calculated by looking at the imports of its trading 

partners. The import data are at the six-digit product level, according 

to the Harmonized System (HS) classification 1988/92. For each product, 

an observation consists of the reporter, country of origin, time, trade value 

in dollars, quantity, and units in which the quantity is expressed. In order 

to create our sample, we focus on the top geographic destinations of Estonia’s 

imports/exports that account for at least 90 percent of imports/exports 

during the period 1997-2005. 

Firm data 

The firm-level data used in this paper are provided by the Estonian Business 

Registry and cover the period 1995-2005. Due to missing information 

on employment for the years 1995 and 1996, these years are omitted from 

the sample. Since the main purpose of this paper is to identify the impact 

of international trade on firm dynamics, we can work only with those sectors 

for which we observe imports and exports. This implies that those firms 

that are not active in the manufacturing sector have to be excluded from the 

sample 24 . However, firms that switch to the primary or services sector are retained 

until their last year of activity within the manufacturing sector. This 

allows us to distinguish between switches occurring within the manufacturing 

sector and those to other sectors of the economy. Several cleaning procedures 

are applied to the sample: 

24 Trade data from Comtrade are available only for 3 out of 28 two-digit service sectors 

(computers and related activities, business services, and other service activities). Given 

this limited coverage of services trade, we exclude these sectors from the analysis and focus 

on the manufacturing sector.


• We construct a longitudinal panel using registration codes and apply 

several corrections to take into account changes in firms’ registration 

codes (i.e., firms’ identification number): (1) firms that change registration 

codes because of a transfer from the Enterprise Registry to the 

Business Registry are considered the same firm; (2) in case of acquisitions, 

the acquiring and acquired firms are considered to be a single 

entity for the entire sample period; and (3) for all other transactions 

(e.g., merger, breakup and divesture), firms are treated as separate 

entities both before and after the transaction. 

• Observations with extreme values on one of the variables used in the 

empirical analysis are dropped, as well as all observations with missing 

information on some variables. This leaves us with 38 percent of the 

registered firms, accounting on average for about 60 percent of aggregated 

value added in the manufacturing sector 25 . 

• Because state-owned firms are not necessarily profit-maximizing agents, 

it is not certain whether their behavior can be captured accurately by 

our empirical model. Hence, we exclude state-owned firms from the 

sample (eight firms). 

• All observations for which no matching trade data could be obtained 

are omitted. While some of these missing trade data are related to 

incorrect EMTAK 26 (NACE) codes reported by firms, others are related 

to the concordance used to convert the trade data from Harmonized 

System (HS, six-digit, 1992) to NACE (Rev.1.1., four-digit). The trade 

data are converted using a concordance table provided by Eurostat. 

For the majority of the codes, the concordance gives a many-to-one 

mapping, that is, multiple HS 6 digit codes are mapped into one NACE 

code. However, there are a number of cases where one HS code maps 

25 Comparability between the micro and macro data is limited, however, owing to 

methodological inconsistencies. Value added at the macro level is a broader concept since 

it covers not only the activities of enterprises but also of other economic units. According 

to the Statistical Office of Estonia, all enterprises registered in Estonia accounted for about 

70 percent of aggregate value added in 2005. 

26 The industry classification used in the Estonian Business Registry, EMTAK, is a 

five-digit extension of the NACE (Rev.1.1.) classification, that is, the official statistical 

classification of economic activities in the European Community.


into more than one NACE code. To minimize potential errors, we 

have omitted these codes from our concordance table and dropped the 

resulting observations from our sample. Firms active in sectors that do 

not appear in our concordance table are also dropped. 

2.A.2 Definitions of variables 

Firm-level variables 

All monetary variables are expressed in real terms. Output and intermediate 

input deflators, as well as the gross capital formation price index, were 

obtained from the Statistical Office of Estonia. Deflators are available for 16 

sectors corresponding to the International Standards Industrial Classification 

(ISIC Rev.3.1) at the one-digit level. 

Exitit+1 

Dummy variable, equal to 1 in period t if the firm exits in period t+1. Firm 

exit is defined based on the official date of liquidation from the Commercial 

Register. If a firm disappears from the data set and it has an official liquidation 

date, it is considered to exit in the last year of observation. Liquidation 

due to a merger, acquisition or reregistration in the registry is not considered 

an exit. 

Switch2d,it+1 

Dummy variable, equal to 1 in period t if the firm switches two-digit NACE 

industries in period t + 1. An industry switch is defined as a change in the 

firm’s primary sector of activity. 

Switch4d,it+1 

Dummy variable, equal to 1 in period t if the firm changes four-digit NACE 

products in period t+1. A product switch is defined as a change in the firm’s 

main product line. 

Ageit 

Age of the firm in period t, defined as the number of years the firm has been 

in the registry (using the registry entry date).


Sizeit 

Firm size, measured by the number of employees in period t. 

Wageit 

Average real labor cost, defined as total firm-level labor costs divided by the 

number of employees. 

Capitalit 

Capital intensity, measured as real capital per employee. Capital is defined 

as the sum of tangible and intangible assets, net of goodwill at the firm level. 

TFPit 

Total factor productivity at the firm level, estimated at the two-digit industry 

level using the methodology of Levinsohn and Petrin (2003) while taking 

into account industry switches over time. 

Foreignit 

Foreign ownership dummy, equal to 1 if at least 50 percent of the firm’s 

shares are foreign owned. 

Product-level variables 

All product-level variables are defined at the four-digit NACE level. 

Sunkjt 

Sunk cost variable, defined as the natural logarithm of the median of real 

sales in each particular four-digit industry j at time t. 

Herfjt 

Herfindahl-Hirschman index for the domestic market, defined as the sum of 

squared market shares. Market shares are defined as firm-level real sales over 

product-level total real sales. 

Importsjt


Total imports of product j at time t, measured in Estonian Krooni (EEK). 

IITjt 

Intra-industry trade variable, defined as the Grubel-Lloyd index, that is, 

[1 − (|Xjt − Mjt|/(Xjt + Mjt))], where Xjt and Mjt are, respectively, exports 

and imports of product j at time t. 

CAjt 

Revealed comparative advantage dummy, equal to 1 if Xjt > Mjt, i.e. if 

exports are larger than imports for product j at time t. 

Exportsjt 

Total exports of product j at time t, measured in Estonian Krooni (EEK). 

Herfexjt 

The Herfindahl-Hirschman index for the export market is calculated at the 

Harmonized System (HS) six-digit level. Conversion to NACE Rev. 1.1. at 

the four-digit level is achieved using a concordance table provided by Eurostat 

(cfr. section 2.A.1). We define a market as a pair consisting of a geographic 

destination and a product. We calculate the index of concentration in market 

m as 

H j 

m,t = 

∀Exporter 

where sn,t is the market share of exporter n in market m at period t. Aggregating 

across all export markets for product j, we obtain the overall index 

of market concentration for exports of product j: 

Herfex jt = 

∀m 

s 2 n,t 

H j 

m,t ∗ β j 

m,t 

where β j 

m,t is the share of market m in total exports of product j in period t. 

UV Rjt 

Average relative unit values index for Estonian export products, taking into 

account Estonia’s main competitors. In particular, we compute the unit value 

for product j in market p by dividing the export value of each exporter by


the export quantity. We consider only those markets in which quantities are 

expressed in the same unit across the sample of exporters for that market. 

Relative unit values for product j in market p are then calculated dividing 

the unit value of Estonia by the weighted average of the unit values of its 

competitors in that market. The overall relative unit value for product j is 

the weighted sum of the relative unit values across all markets, with weights 

equal to export shares.


Manufacturing 

NACE Description 

High-technology manufacturing 

244 Pharmaceuticals 

30 Office machinery - computers 

32 Radio, TV, communication equipment 

33 Medical, precision, optical instruments 

353 Aircraft - spacecraft 

Medium-high-technology manufacturing 

24 Chemicals, excl. pharmaceuticals 

29 Machinery and equipment 

31 Electrical machinery 

34 Motor vehicles 

35 Other transport equipment (excl. 351 & 353) 

Medium-low-technology manufacturing 

23 Coke, refined petroleum products 

25 Rubber and plastic 

26 Nonmetallic mineral products 

27 Basic metals 

28 Fabricated metal products 

351 Building/repairing of ships and boats 

Low-technology manufacturing 

15 Food and beverages 

16 Tobacco 

17 Textiles 

18 Clothing 

19 Leather (products) 

20 Wood (products) 

21 Pulp, paper (products) 

22 Publishing and printing 

36 Furniture 

37 Recycling 

Source: Eurostat 

Table 2.A.1: Sector classification: Manufacturing


Services 

NACE Description 

Knowledge-intensive services 

61 Water transport 

62 Air transport 

64 Post and telecommunications 

65 Financial intermediation 

66 Insurance and pension funding 

67 Ancilliary financial activities 

70 Real estate activities 

71 Renting activities 

72 Computer and related activities 

73 Research and development 

74 Other business activities 

80 Education 

85 Health and social work 

92 Recreational activities 

Less-knowledge-intensive services 

50 Wholesale/retail trade of motor vehicles 

51 Wholesale trade 

52 Retail trade 

55 Hotels and restaurants 

60 Land transport 

63 Supporting transport activities 

75 Public administration, defense 

90 Sewage and refuse disposal 

91 Activities of membership organizations 

93 Other service activities 

95 Activities of households 

99 Extraterritorial organizations and bodies 

Source: Eurostat 

Table 2.A.2: Sector classification: Services

Chapter 3 

Footloose Multinationals in 

Belgium? 


The popular claim that production abroad has a detrimental impact on 

employment and exports at home is generally not confirmed by the available 

empirical evidence (see Lipsey (2002) for a review of this extensive literature). 

Within host countries, multinational firms tend to pay higher wages for given 

productivity levels and they are also generally found to be more efficient than 

local firms. Although evidence on the existence of spillovers of these effects to 

the domestic economy is mixed, foreign direct investment (FDI) has played a 

clear role in the transformation of some economies from exporters of agricultural 

goods and raw materials to exporters of manufacturing goods (Lipsey, 

2002). However, in spite of all the evidence pointing to the beneficial impact 

of multinational firms and their operations on home and host countries’ 

economic performance, public concerns largely remain. 

One reason for these concerns is that multinationals, due to their ability 

to set up production abroad in a profitable way, are often considered to be 

footloose; causing them to react more swiftly to shocks and hence possibly to 

relocate production more rapidly than national 1 firms. This naturally raises 

∗ Published in Review of World Economics, 143(3), 2007, 483-507. 

1 The term national firms will be used to refer to firms without access to a global 

network. 

101

102 Footloose multinationals in Belgium? 

policy concerns, particularly with respect to foreign multinationals, as the 

recent turmoil surrounding the takeover of the European steel group Arcelor 

by Mittal Steel has illustrated 2 . 

The main objective of the present paper is to determine the impact of 

multinational ownership on the exit decisions of firms located in Belgium. 

The empirical analysis links in with a small and recent literature 3 dealing 

with the impact of multinational ownership on exit patterns of manufacturing 

establishments and contributes to this literature in two important respects. 

First, to my knowledge, the role of ownership structure in shutdown decisions 

at the firm level has thus far only been considered for the manufacturing 

sector. Due to the extensive coverage of the database employed, I am able to 

extend the analysis to cover all sectors of the economy 4 . Second, this study is 

the first to control for the (possibly differing) impact of foreign and domestic 

multinational ownership on shutdown decisions at the firm level. 

From a policy perspective, the analysis could yield several important insights. 

The reduction of the corporate tax rate in Belgium in 2003, as well as 

the introduction of the ”notional interest deduction” scheme 5 in June 2005, 

although both non-discriminatory between domestic and foreign companies, 

are primarily aimed at the attraction of new, capital-intensive investments 

by foreign multinationals. Moreover, Vandenbussche and Tan (2005) find 

2 The Economist (2006, February 2 nd : 11-12). Several government officials in France 

and Luxembourg were strongly opposed to the takeover of Arcelor, Europe’s largest steel 

company, by Mittal Steel, in spite of the weak strategic importance of the steel industry 

today. Similarly, the French government raised objections against possible plans of 

US-based PepsiCo to take over the French dairy group Danone (The Economist, 2005, 

September 1 st : 56). 

3 Both Bernard and Sjöholm (2003) and Görg and Strobl (2003) provide evidence that 

foreign multinationals, in Indonesia and Ireland respectively, are more footloose than domestic 

companies, i.e. they are more likely to exit the market, after controlling for various 

firm and industry characteristics. Bernard and Jensen (2007a) provide similar evidence 

for US multinationals in their home market. For a more detailed discussion of the related 

literature, I refer to section 3.2. 

4 Since the primary sector represents only a small share of the sample total (less than 

2 percent of all firms are active in this sector, accounting for 0.66 of total employment 

and 1.37 percent of net value added in 2001), it is excluded from the analysis. However, 

results do not change when firms active in agriculture and mining are taken into account 

in the empirical estimations. 

5 A tax deduction for equity financing, implemented since January 2006.

Literature review 103 

evidence of substantial tax discrimination in favor of foreign multinationals 

in Belgium compared to domestic companies. 

If, as was shown for a number of other countries, multinationals (either 

foreign or home-based) are found to be more footloose than national firms, 

this information could certainly be of relevance in fine-tuning current policies, 

especially given the impact of multinational firms on the Belgian economy, 

both in terms of employment and sales. Although less than 2 percent of the 

population of firms in Belgium are foreign-owned and an equal percentage 

are domestic multinationals; foreign multinationals accounted for about 25 

percent of total employment and 31 percent of net value added in 2001. 

Domestic multinationals accounted for another 18 percent of employment 

and 21 percent of total net value added 6 . 

The rest of the paper is organized as follows. In section 3.2 I present a 

selective review of the related literature. Section 3.3 introduces the data set 

and presents some summary statistics, while section 3.4 and 3.5 discuss the 

empirical model and results respectively. In section 3.6 the robustness of the 

results is verified on the basis of a number of sensitivity analyses applied to 

the model. Section 3.7 concludes. 

3.2 Literature review 

The focus in this section is on the literature investigating the footloose 

nature of multinational enterprises (MNEs) and specifically dealing with the 

impact of multinational ownership on firm turnover. For a more complete review 

of the main theoretical and empirical contributions in the industrial organization 

literature on firm dynamics and turnover, I refer to Caves (1998), 

although I will come back to some of the relevant issues emerging from this 

literature when I discuss the empirical model in section 3.4. 

6 Data are calculated by the author on the basis of the Belfirst database (BvDEP, 

2004), which groups the annual accounts of virtually all limited-liability companies active 

in Belgium. A more detailed description of the database will be given in section refdata


Flamm (1984) looks into the footloose image of US multinationals abroad, 

by modeling the volatility of US FDI in the semiconductor industry in response 

to changes in the host country environment. His findings indicate that 

while US FDI is only moderately sensitive to wage levels in the host countries, 

adjustment of the investment portfolio in response to adverse shocks 

(such as changes in wages or other costs) occurs extremely rapid. 

Rodrik (1997) further notes the asymmetry between groups in the economy 

that are free to take their resources wherever they are most in demand, 

such as multinational firms; and those that can not, including semi-skilled 

and unskilled workers in the home country. He argues that the increased 

substitutability of the services of these immobile groups is likely to lead to 

a higher elasticity of demand for their services. Konings and Murphy (2006) 

find support for this hypothesis using firm level data on 1,067 multinational 

firms and their 2,078 affiliates located in the European Union (EU) and 

Central and Eastern European Countries (CEECs). Their results indicate 

that some substitution in employment takes place between the parent firm 

and its North EU affiliates, i.e. the elasticity of parent employment with 

respect to affiliate wages is positive and statistically significant; while no 

substitution effects are found for the CEEC and South EU affiliates, contrary 

to the popular belief that major reallocation of employment towards the 

CEECs is taking place. Stronger substitution effects are found when the 

sectors of activity of the parent and affiliate are different. 

Similarly, Vandenbussche and Tan (2005) argue that since corporate tax 

policy is an important instrument for national governments wanting to attract 

foreign direct investment and given the fact that multinationals’ mobility 

increases their bargaining power, it is likely that foreign multinationals 7 

pay lower (effective) taxes. Their empirical results, for Belgium, suggest 

that there is considerable tax discrimination in favor of foreign firms, again 

lending support to the footloose image of MNEs. 

7 They do not distinguish between domestic firms with and without access to a global 

network.

Literature review 105 

Focusing on the extensive margin, several studies have taken up the issue 

whether multinationals are more footloose. Theoretically, the expected 

impact of multinational ownership on firm turnover is ambiguous. On the 

one hand, it can be argued that MNEs have certain specific characteristics 

that enable them to set up production abroad in a profitable way; allowing 

them in turn to respond more swiftly to negative shocks than is possible for 

national firms. On the other hand, given the fact that multinationals are on 

average more skill- and capital-intensive than incumbent firms, they might 

face higher sunk costs when setting up production, causing them to exit the 

market less rapidly, all else equal (Lipsey, 2002). Moreover, since domestic 

MNEs are likely to be more strongly rooted in their home country; it is not 

certain that domestic and foreign multinationals will react in the same way 

to changes in the economy. 

Bernard and Sjöholm (2003), Görg and Strobl (2003) and Gibson and 

Harris (1996) provide some of the first evidence on the footloose nature of 

multinationals in the host country; i.e. they assess the effect of foreign ownership 

on plant-level exit patterns in the manufacturing sector in Indonesia, 

Ireland and New Zealand respectively. Both Bernard and Sjöholm (2003) 

and Görg and Strobl (2003) find that while foreign-owned plants face lower 

hazard rates than domestic plants, this difference is caused by the different 

characteristics of these plants, rather than by their foreign character as such. 

Foreign plants are on average older, larger and more productive, causing 

them to exhibit lower exit rates. After controlling for these plant-specific 

differences, foreign plants become more likely to exit the market. Gibson 

and Harris (1996) on the other hand, find evidence that foreign MNEs are 

less likely to exit the market than incumbent firms, after controlling for plant 

and industry characteristics; although it is possible these results are partly 

driven by the increased trade liberalization that was taking place in New 

Zealand at the time. 

Bernard and Jensen (2007a) perform an analysis of US manufacturing 

plants’ deaths in order to determine the impact of the multinational character 

of US firms on exit patterns in the home country (they are not able to identify 

foreign MNEs in the data). Their model also takes into account whether an


establishment is part of a single- or multi-plant firm. Their findings indicate 

that multinational plants, as well as plants belonging to a larger group, are 

more likely to exit the market, after controlling for various other plant- and 

industry-specific variables. Unconditionally, both plants owned by multiplant 

firms and plants owned by US multinationals exhibit lower exit rates 

than incumbent plants. 

For Belgium, both Van den Cruyce (1999) and Pennings and Sleuwaegen 

(2000, 2002) have searched for the determinants of different modes of 

downsizing at the firm level (exit, partial relocation or downsizing) in the 

manufacturing sector. They find a positive impact of multinational ownership 

on relocation decisions of firms, but no significant impact on exit 

patterns. However, a serious drawback of their multinational ownership variable 

is that it does not distinguish for nationality of ownership; foreign and 

domestic multinationals firms are taken together in the empirical analysis. 

This paper differs from previous studies in two important respects. First, 

due to the extensive coverage of the database employed, I am able to investigate 

the impact of multinational ownership on exit patterns separately 

for firms active in manufacturing and services. Second, by distinguishing 

between foreign firms and domestic MNEs in the empirical analysis, it is 

possible to investigate whether nationality of ownership plays a role in shaping 

the exit patterns of multinational firms. 

3.3 Data and preliminary facts 

The data set employed in this paper is constructed on the basis of the 

Belfirst database, which groups the annual accounts reported by firms located 

in Belgium to the National Bank of Belgium 8 . The database is com- 

8 “Most enterprises in which the liability of the shareholders or members is limited to 

their contribution to the company, plus some other enterprises, have to file their annual 

accounts and/or consolidated accounts with the Central Balance Sheet Office at the National 

Bank every year” (http://www.nbb.be). This implies that the database consists 

of the complete population of (limited-liability) firms in Belgium (318,316 companies). 

However, about 160,000 firms in the database are very small, i.e. they have no recorded 

employment over the sample period. Since a number of variables used in the empirical 

analysis (such as firm size and wages) are not available for these companies and given

Data and preliminary facts 107 

mercialized by BvDEP (2004) and has been used in a growing number of 

academic papers in recent years 9 . Firms in the database are uniquely defined 

by their VAT number and data on employment, wages, net value added etc. 

are available for the years 1996-2003. Sectors in the database are classified 

according to the NACE-Bel nomenclature, i.e. a five-digit extension of the 

NACE (Revision 1) Classification commonly used for European statistics 10 . 

Furthermore, the database includes information on the ownership structure 

of firms, in terms of both subsidiaries and shareholders, either foreign or 

domestic. 

To identify entry and exit, I follow the procedure used by Mata and Portugal 

(1994) and Mata, Portugal and Guimarães (1995). For the specifics 

associated with the exit variable and the selection of the sample used in the 

empirical analysis, I refer to the data appendix (section 3.A). Omitting all 

observations that do not fit the definition of exit, as well as all firms for 

which data needed in the empirical analysis are incomplete, results in an 

unbalanced sample of 26,046 companies 11 . 

A firm is considered to be foreign-owned in the data set if it has some 

foreign ownership. Likewise, a firm is identified as a domestic multinational 

if it has subsidiaries in countries other than Belgium and it is not foreignowned. 

Ideally, I would have liked to identify multinationals on the basis 

of some minimum share of direct ownership, but unfortunately these data 

the otherwise limited reporting requirements for these firms, they are necessarily excluded 

from the analysis. 

9 Examples include Vandenbussche, Crabbé and Janssen (2005) in a study on the determinants 

of effective corporate tax rates of large Belgian firms and De Loecker (2007) 

who estimates the impact of increased trade liberalization on productivity in the Belgian 

textile sector. 

10 The NACE Rev. 1 classification can be downloaded from the Eurostat Ramon server: 

http://europa.eu.int/comm/eurostat/ramon/. 

11 Omitting all companies with no recorded employment over the sample period, as well 

as firms that exhibit irregular entry and exit patterns or were subject to an ownership 

change over the sample period and observations with missing data for any of the independent 

variables results in a sample of 98,100 enterprises. Following Mata et al. (1994, 1995), 

only firms that employ 10 employees or more in at least one sample year are included (see 

appendix 3.A). Some of these restrictions will be relaxed in section 3.6 however.


Number of firms 

0 200 400 600 800 

864 

Netherlands 

621 

France 

252 

US 

172 

Germany 

145 

UK 

95 94 81 

Others EU 

Sweden 

Switzerland 

52 45 45 37 

Figure 3.1: Foreign multinationals by country of origin 

are missing for the majority of cases 12 . Another drawback of the ownership 

measure is that it is time-invariant; the information refers to the latest year 

available. Figure 3.1 displays the country distribution of the foreign firms in 

the sample. 

As can be seen in figure 3.1, most ownership of foreign firms is concentrated 

in the EU (83 percent); the US accounts for about 10 percent and 

only 7 percent of foreign owners originate in other countries of the world. 

Germany, the Netherlands, the U.K. and France, which are all direct neighbors 

of Belgium, together account for over 70 percent of foreign ownership. 

Table 3.1 shows the sector distribution of firms in the sample, distinguished 

by ownership type. 

About 25 percent of all firms in the sample are active in manufacturing 

(6,524 firms), while 75 percent or 19,522 firms are active in services. Within 

manufacturing, the largest share is taken up by the metal industries (19 

percent), followed closely by the food and textile industries (shares of 14 

12 Similarly, information regarding indirect ownership was missing for the majority of 

cases. 

Denmark 

Luxembourg 

Japan 

ROW


National Foreign Domestic Total 

Sector of activity Firms MNEs MNEs N 

D. Manufacturing 5,202 704 618 6,524 

DA. Food, beverages and tobacco 756 75 93 924 

DB. Textiles 646 23 64 733 

DC. Leather (products) 33 1 2 36 

DD. Wood (products) 231 7 13 251 

DE. Pulp and paper 583 51 59 693 

DF. Coke and petroleum (products) 7 9 4 20 

DG. Chemicals 170 127 58 355 

DH. Rubber and plastics 201 48 43 292 

DI. Other non-metallic products 337 53 30 420 

DJ. Metallic products 1,032 111 80 1,223 

DK. Machinery and equipment n.e.c. 329 72 61 462 

DL. Electrical and optical equipment 259 73 63 395 

DM. Transport equipment 130 32 18 180 

DN. Manufacturing n.e.c. 488 22 30 540 

E-Q. Services and related activities 16,828 1,799 895 19,522 

E. Electricity, gas and water supply 28 10 4 42 

F. Construction 3,670 145 97 3,912 

G. Wholesale and retail trade 6,212 785 343 7,340 

H. Hotels and restaurants 1,165 39 13 1,217 

I. Transport and communication 2,017 196 112 2,325 

J. Financial intermediation 269 57 37 363 

K. Other business activities 2,504 516 249 3,269 

L. Public services 12 0 0 12 

M. Education 44 4 4 52 

N. Health and social work 380 2 7 389 

O. Other service activities 524 45 29 598 

P. Private household act. 1 0 0 1 

Q. Extra-territorial org. 2 0 0 2 

Total 22,030 2,503 1,513 26,046 

Table 3.1: Sector distribution of firms


and 11 percent respectively). Compared to the sample average, domestic 

multinationals are somewhat over-represented in the manufacturing sector; 

41 percent of all domestic MNEs are active in this sector, compared to 25 

percent for the overall sample. Within the service sector, most firms are 

active in wholesale and retail activities (38 percent) and in construction (20 

percent). 

In order to examine whether multinational firms, either foreign or domestic, 

are unconditionally more or less likely to exit than national firms, 

I calculate (nonparametric) Kaplan-Meier survival estimates separately for 

each group in Own (Own = 0 for national firms; 1 for foreign MNEs and 2 

for domestic MNEs): 

ˆS (a) = 

j|aj≤a 

 

nj − dj 

nj 

(3.1) 

where nj indicates the number of plants that have survived up to aj 

years of age; dj is the number of plants that die at age aj and ˆ S(a) gives the 

probability of surviving up to age aj (Greene, 2003, 789-789). The survival 

functions corresponding to (3.1) are plotted in figure 3.2. Analysis time 

represents firms’ age. 

From the graph, it is clear that multinationals (foreign or domestic, Own 

equals 1 or 2) have a higher survival rate than national firms (Own equals 

0). A log-rank test of equality of the survival functions yields similar results; 

the hypothesis that survival functions are equal for multinational firms and 

national firms is rejected decisively (with probabilities of 0.00 and associated 

Chi-square values higher than 10) for both foreign and domestic MNEs. This 

result holds when stratifying over two-digit sectors and years. The difference 

between foreign and domestic MNEs is less clear-cut, both graphically and 

mathematically. If survival is not stratified by sector and year, their survival 

functions are significantly different at the 5 percent level (probability of 0.03). 

This result no longer holds when stratifying over sector and years, the logrank 

test of equality of the survival functions does not reject the hypothesis 

of equality in this case (probability of 0.13).


Survival probability 

0.00 0.25 0.50 0.75 1.00 

Kaplan−Meier survival estimates 

0 50 100 150 

Analysis time 

Own = Domestic Own = Foreign MNE 

Own = Domestic MNE 

Figure 3.2: Kaplan-Meier survival functions by nationality of ownership 

The finding that multinationals exhibit lower exit rates can also be seen in 

the second row of table 3.2. Between 1996 and 2001, about 9 percent of national 

firms exit the sample, compared to 6 percent for domestic MNEs and 

7 percent for foreign firms. Hence, it is clear that multinationals, either foreign 

or domestic are unconditionally less likely to exit the market compared 

to national firms. This finding is consistent with the results obtained by 

Bernard and Jensen (2007a); Bernard and Sjöholm (2003); Görg and Strobl 

(2003); who also report lower exit rates for multinational firms. 

Table 3.2 further reports summary statistics for the pooled sample, distinguished 

by ownership type. All reported measures, with the exception of 

the number of firms and exits, represent averages over the sample period; 

standard deviations are reported in brackets. For a detailed description of 

the data and variables used, I refer to the data appendix (section 3.A). 

The first row in table 3.2 shows the number of firms distinguished by ownership 

type. National firms account for about 85 percent of the sample total, 

while domestic and foreign MNEs account for about 5 and 10 percent respectively. 

Although the number of multinationals in the sample seems rather


Domestic Foreign Domestic 

Variable firms MNEs MNEs 

Number of firms 22,030 2,503 1,513 

(% of total firms) (84.58) (9.61) (5.81) 

Number of exits 2,040 187 90 

(Total % over sample period) (9.26) (7.47) (5.95) 

Age (1) 18.99 22.82 23.3 

(years) (14.83) (18.96) (18.98) 

Size (1) 29.97 165.22 204.51 

(number of employees) (120.83) (577.67) (1,453.59) 

Wage per employee (1) 34.14 52.2 45.58 

(x 1,000e) (33.95) (28.56) (38.40) 

Productivity per worker (1) 54.19 95.08 100.37 

(x 1,000e) (91.71) (208.76) (518.70) 

Labor costs relative to productivity (2) 0.83 0.89 0.78 

(8.79) (4.79) (2.03) 

Industry concentration (1) 503.71 811.97 823.07 

(Herfindahl index three-digit NACE) (870.59) (1,107.58) (1,165.77) 

Notes: (1) Values represent sample means. Standard deviations are reported in 

brackets. (2) Wage over productivity. Value represents sample mean, standard 

deviation reported in brackets. 

Table 3.2: Summary statistics by ownership type (1996-2001)


limited, their importance in terms of employment and turnover should not 

be underestimated. Multinational firms account for over 50 percent of total 

employment and more than 60 percent of net value added in the sample. This 

is caused by the fact that multinationals, both domestic and foreign, are on 

average much larger than national firms. The average foreign multinational 

employs more than 5 people for each person employed in a national firm (165 

versus 30); domestic MNEs employ (on average) more than 6 people for each 

employee in a national firm (205 versus 30). The average multinational is 

also older than national firms (23 versus 19 years). 

Turning to the figures on wages and productivity per employee in table 3.2, 

it is clear that multinationals, either foreign or domestic, are on average more 

productive and pay higher wages compared to national firms. Since economic 

theory suggests a close link between labor costs and productivity, table 3.2 

also displays the sample average of labor costs relative to productivity. The 

figures indicate that the average foreign firm pays 89 cents per euro of net 

value added, compared to 83 cents per euro for national firms and only 78 

cents per euro for domestic MNEs. If these numbers are interpreted as a 

measure of competitiveness at the firm level, they suggest that foreign firms 

are, on average, less competitive than their domestic counterparts (either 

national or multinational) given that they pay higher wages after controlling 

for productivity 13 (Konings, 2005). The last row of table 3.2 shows that 

multinationals tend to be active in sectors characterized by a higher industry 

concentration, as measured by the Herfindahl concentration index (see 

appendix 3.A for the specifics of this variable). 

Since the simple summary measures presented here do not allow us to 

disentangle the impact of firm- and industry-specific effects from that of 

multinational ownership (foreign or domestic) on firms’ propensity to exit, I 

now turn my attention to the full empirical model. 

13 If value added is considered a proxy for total output, the ratio of the total wage bill to 

value added gives an indication of the relative competitiveness of a firm, since it measures 

to what extent value added of the firm covers the wage bill. Lipsey (2002) sums up a 

number of motives for foreign firms to pay higher wages for labor of a given quality.


3.4 Empirical model 

In order to assess the impact of multinational ownership on exit patterns, 

it is necessary to estimate a model that accounts for the differences in size, 

productivity and wages between multinational and national firms. Given the 

nature of the data 14 and following the related literature (eg. Bernard and 

Sjöholm (2003); Chen (2002); Disney et al. (2003); Görg and Strobl (2003); 

Mata and Portugal (1994); Mata et al. (1995)), I estimate a proportional 

hazard model: 

λ (a, x, β, λ0) = ϕ (x, β) λ0 (a) (3.2) 

where the hazard function λ (.) depends multiplicatively on the vector 

of explanatory variables x with unknown coefficients β and the baseline hazard 

λ0 (a) (corresponding to ϕ (.) equal to 1). For the special case where 

ϕ (x, β) = exp (x ′ β), estimation of β does not require specification of the 

baseline hazard λ0 (a) (Kiefer, 1988, 666). 

Since the fundamental interest is in the effect of the covariates on the 

hazard (and not in the shape of the baseline hazard), a natural choice is 

to normalize the baseline hazard to 1 and estimate (3.2) using the partial 

likelihood approach first suggested by Cox (1972). Specifically, the empirical 

model looks as follows: 

h (a) = h0 (a) exp [αOwn + βX] 

X = [ln (Sizeit) , ln (Wageit) , ln (Prodit),ln (Herfjt)] 

Own = [Fori, Domi] 

(3.3) 

where h (a) is the hazard rate; i.e. the rate at which plants exit at age a, 

conditional upon having survived up to a-1; h0 (a) is baseline hazard, allowed 

to vary over two-digit sectors and years; X is the vector of control variables, 

including firm-level employment (Sizeit), wages (Wageit), labor productivity 

(Prodit) and industry-level concentration, measured using a Herfindahl 

14 The data are characterized by left truncation (delayed entry), i.e. firms may exist for 

some time before the sample period starts; and right censoring occurs, i.e. it is possible 

that a firm has not failed at the end of the sample period. Kiefer (1988) presents an 

overview of the different concepts related to duration data. Unlike conventional regression 

methods, duration models are specifically designed to take these issues into account.

Empirical model 115 

index (Herfjt). Fori and Domi are the foreign and domestic multinational 

ownership dummies respectively. 

Estimations of (3.3) are stratified by two-digit sector and year, allowing for 

equal coefficients across strata, but baseline hazards specific to each sector 

and year. In what follows, I will briefly discuss the definitions and expected 

signs of the variables included in (3.3). 

An empirical implication of models dealing with firm dynamics and allowing 

for entry, exit and heterogeneity across firms (eg. Hopenhayn (1992); 

Ericson and Pakes (1995)); is that firms’ hazard rate is decreasing in size, 

conditional on age. Dunne et al. (1988, 1989) already established a positive 

relationship between firms’ age and size and their survival probabilities and 

many empirical studies find support for this hypothesis (eg. Audretsch and 

Mahmood, 1995). Consequently, I include firms’ current size 15 in the model, 

defined in terms of employment at time t (expressed in logarithm form). 

Firm size is expected to have a negative effect on the hazard rate (positive 

effect on firm survival). 

The wage variable is defined as firms’ wages 16 , divided by employment in 

year t. To the extent that higher wages reflect higher skill intensities and 

associated sunk costs at the firm level (related to training and investment in 

firm-specific human capital), higher wages can be expected to have a negative 

effect on firm failure. Audretsch and Mahmood (1995) find support for the 

negative relationship between wages and the propensity to exit in their study 

on the post-entry performance of 12,000 US manufacturing plants. However, 

it is not certain whether this result will hold when taking into account productivity 

at the firm level. As was already alluded to in section 3; if firms pay 

higher wages for given levels of productivity, this could signal lower competitiveness 

and hence increase their exit probability, ceteris paribus (Konings, 

2005). 

15 The effect of age on the hazard rate is incorporated directly into the model, since 

duration is a function of the firm’s age. Consequently, multicollinearity issues prevent 

inclusion of the firm’s age as a separate regressor in the model. 

16 The wage variable covers both remunerations and social security costs (including pensions). 

For an overview of the definitions of the independent variables, I refer to section 3.A.


Several theoretical models (eg. Jovanovic, 1982; Hopenhayn, 1992) predict 

that the growth and exit of firms is motivated to a large extent by productivity 

differences at the firm level. Empirically, Fariñas and Ruano (2005) find, 

for a sample of Spanish manufacturing firms, that entry and exit decisions 

are systematically negatively related to differences in productivity at the firm 

level. A negative effect of the labor productivity variable on firms’ hazard 

rates is therefore expected. 

In addition to the firm-specific variables, an industry-specific variable is 

introduced in the model. Economic theory predicts a negative relationship 

between industry concentration and turnover. Higher industry concentration, 

as measured by the Herfindahl index, is expected to lead to increased 

price-cost margins and, through increased profitability, reduce firm turnover 

(Mueller, 1991). However, Caves (1998) notes that as concentration is determined 

by the extent of entry barriers in an industry, as well as the degree of 

cooperation among producers, it is not clear what the effect of including this 

variable to explain hazard rates will be. Results obtained in earlier studies 

are similarly ambiguous. Görg and Strobl (2003) find a significantly positive 

impact of concentration on hazard rates at the firm level, while Mata 

and Portugal (1994) find a negative, though insignificant effect of concentration 

on exit patterns. The Herfindahl index (Herfjt) is calculated at the 

three-digit industry level using all firms with reported turnover in the Belfirst 

database (see section 3.A). The market shares are calculated on the basis of 

firm and industry turnover in each specific year. 

Finally, I come to the two ownership variables, Domi and Fori. As was 

noted previously, there is no clear theoretical indication about the impact 

of multinational ownership on exit patterns at the firm level. It can be 

argued that multinationals’ ability to shift production around the world in a 

profitable way enables them to respond more swiftly to adverse shocks in the 

home or host country, causing them to exit the market more rapidly, all else 

equal. However, it is also possible that multinationals face higher sunk costs 

when setting up production, due to their higher capital and skill intensity 

(on average), causing them to exit the market less rapidly, ceteris paribus. In 

addition, it is not inconceivable that foreign multinationals react differently

Empirical results 117 

to adverse shocks compared to domestic multinationals that are more firmly 

rooted in the local economy. I now turn to the discussion of the empirical 

results. 

3.5 Empirical results 

Table 3.3 reports the results of the Cox proportional hazard model applied 

to the unbalanced sample of 26,046 firms over 6 years (1996-2001). Correlations 

between the independent variables are generally low, with the exception 

of ln(Wagesit) and ln(Prodit) for which the correlation amounts to 0.70. For 

each regression, coefficients and associated robust standard errors, adjusted 

for clustering at the firm level, are reported. The first two columns in the 

table show the results of the model applied to all sectors (NACE 15-99), 

while columns III-IV and V-VI display results separately for manufacturing 

and services respectively. 

Columns I, III and V in table 3.3 show the results for the basic model in 

(3.3). From figure 3.2 it is clear that multinationals (either foreign or domestic) 

exhibit lower exit rates than national firms. Estimation of the full model 

however, reveals that after controlling for the fact that multinationals are on 

average larger, more productive and pay higher wages, foreign firms are more 

likely to exit the market than national firms; the coefficient on Fori is positive 

and significant both for the manufacturing and service sectors. Results 

for the domestic ownership variable (Domi) are less clear-cut. While the domestic 

ownership dummy is found to have a significantly positive coefficient 

in the manufacturing sector, its coefficient is very small and insignificant for 

services. 

In order to interpret the magnitude of these effects, it is useful to calculate 

the hazard ratio corresponding to these coefficients by taking its exponential. 

For a dummy variable, the hazard ratio can be interpreted as the increase in 

the overall hazard rate facing the firm, corresponding to Fori or Domi equal 

to 1. Calculation of the exponential of the coefficients on Fori yields hazard 

ratios between 1.27 (for services) and 1.70 (for manufacturing), indicating 

that the hazard rate is between 1.3 and 1.7 times higher for foreign firms than


All Manufacturing Services 

Sectors (NACE 15-99) (NACE 15-37) (NACE 40-99) 

Variables I II III IV V VI 

Domestic MNE 0.10 0.09 0.50** 0.50** -0.04 -0.05 

[0.12] [0.12] [0.20] [0.20] [0.15] [0.15] 

Foreign MNE 0.30*** - 0.53*** 0.24** - 

[0.09] [0.18] [0.10] 

EU Parent - 0.32*** - 0.45** - 0.30*** 

[0.09] [0.19] [0.10] 

US Parent - 0.35 - 0.78* - 0.16 

[0.28] [0.45] [0.38] 

ROW Parent - -0.24 - 1.57*** - 0.88 

[0.46] [0.50] [0.67] 

log(Employment) -0.60*** -0.60*** -0.61*** -0.61*** -0.61*** -0.61*** 

[0.03] [0.03] [0.06] [0.06] [0.04] [0.04] 

log(Wages) 0.11 0.11 0.55*** 0.55*** 0.04 0.05 

[0.07] [0.07] [0.19] [0.19] [0.08] [0.08] 

log(Productivity) -0.76*** -0.76*** -0.95*** -0.95*** -0.73*** -0.73*** 

[0.04] [0.04] [0.10] [0.10] [0.05] [0.05] 

log(Herfindahl) 0.00 0.00 -0.10 -0.10 0.02 0.02 

[0.03] [0.03] [0.08] [0.08] [0.03] [0.03] 

Wald test 764.6*** 769.6*** 219.7*** 225.0*** 578.7*** 581.3*** 

N 143,010 36,721 106,289 

Number of firms 26,046 6,524 19,522 

Number of failures 2,317 520 1797 

% of total 8.9 8.0 9.2 

Notes: Robust standard errors, adjusted for clustering at the plant level (reported 

in brackets). Baseline hazard stratified by two-digit sector and year. Efron option 

for handling ties. Significance levels: *** p


for national firms, depending on their sector of activity. Similarly, taking the 

exponential of the coefficient on Domi yields hazard ratios between 0.96 for 

services (insignificant) and 1.65 for firms active in manufacturing, indicating 

that domestic MNEs active in the production sector are 1.65 times as likely 

to exit than national firms. 

The results obtained for the foreign ownership variable are in line with 

the findings of Bernard and Sjöholm (2003) and Görg and Strobl (2003) 

for Indonesia and Ireland respectively. Moreover, the results for the domestic 

ownership dummy are similarly in line with those obtained (for the 

manufacturing sector) by Bernard and Jensen (2007a) for the US. These 

findings lend support to the hypothesis that MNEs, either domestic or foreign, 

are indeed more footloose than national firms. Furthermore, the fact 

that domestic multinationals engaged in service activities do not exhibit significantly 

higher exit rates than national firms, after controlling for their 

differing characteristics, has important policy implications. 

Results obtained on the firm- and industry-specific independent variables 

in table 3.3 are in accordance with expectations. The coefficient on firm size 

is significantly negative, supporting the hypothesis that larger firms tend 

to exhibit lower exit rates. Wages have a positive impact on firms’ hazard 

rate after controlling for productivity at the firm level, in line with the hypothesis 

that firms are relatively less competitive if they pay higher wages 

for given productivity levels. However, if productivity is omitted from the 

regression (unreported), I find a significantly negative sign for the wage variable, 

supporting the hypothesis that higher wages reflect a higher relative 

skill intensity, leading to higher sunk costs and hence a lower probability of 

exit. The coefficient on industry concentration as measured by the Herfindahl 

index is never statistically significant. 

Columns II, IV and VI in table 3.3 present the results of a refinement to 

the model. Specifically, I introduce three separate variables representing the 

home countries of the foreign firms to replace the foreign ownership dummy. 

Discussion of results here is limited to the ownership variables, since all other 

variables perform similarly to the basic model. The variable EUi groups all


firms with home countries in the European Union (EU), while USi represents 

firms originating in the US or Canada. ROWi groups all other home 

countries. 

As can be seen in the table, a positive and significant effect is found for all 

three variables in the manufacturing sector, while for services only firms originating 

in the EU exhibit significantly higher exit rates than national firms. 

For manufacturing, corresponding hazard ratios for the coefficients are 2.18 

(for USi) and 4.81 (for ROWi), compared to 1.57 for firms in the European 

Union. For services, the hazard ratio corresponding to EUi amounts to 1.35; 

while for firms originating in the US or the rest of the world, results are insignificant. 

These results suggest that manufacturing firms, originating from 

outside the EU, are more likely to exit the market, not only compared to 

national firms, but also compared to foreign MNEs originating in the EU. 

In summary, after controlling for the fact that multinational firms tend to 

be larger, more productive and pay higher wages than domestic firms (cfr. 

table 3.2), both foreign and domestic multinationals show a higher propensity 

to exit than national firms active in manufacturing. For services, only foreign 

multinationals exhibit significantly higher hazard ratios. Furthermore, within 

the manufacturing sector, firms originating from outside the EU are found 

to exhibit higher hazard ratios, both compared to foreign firms from within 

the EU and compared to national firms. The magnitude of these effects is in 

line with results obtained in earlier studies. Görg and Strobl (2003) find that 

foreign firms’ hazard rates are 1.4 times as high as domestic companies and 

Bernard and Sjöholm (2003) associate foreign ownership with an increase in 

the exit probability of 20 percent. 

3.6 Robustness checks 

In order to verify the robustness of the results presented in the previous 

section, a number of sensitivity analyses have been applied to the original 

model, both in terms of sample selection and methodology used. First, as 

is explained in detail in the data appendix (section 3.A), the sample used 

to estimate (3.3) in the previous section consists only of firms employing 10


or more people in at least one sample year. However, a common finding in 

papers dealing with exit rates at the firm level is that small plants tend to 

exhibit (proportionately) higher exit rates than their larger counterparts (eg. 

Bernard and Jensen, 2007a; Disney et al., 2003). Bernard and Sjöholm (2003) 

further note that given the larger average size of MNEs, exclusion of small 

firms from an empirical analysis dealing with the impact of multinationality 

on exit patterns is likely to seriously bias the results . 

In order to verify the sensitivity of the results displayed in table 3.3 to the 

inclusion of small firms in the sample, I have re-estimated (3.3) using the full 

sample of firms with positive employment in at least one sample year (98,100 

firms). The results are displayed separately for manufacturing and services in 

the first two columns of table 3.4. The most notable differences compared to 

estimation of the restricted sample include the increase in the size effect for 

the full sample (from about 0.60 to more than 1.00); the significantly negative 

coefficient for wages; the significantly positive effect of industry concentration 

for services and the lower coefficient for the productivity variable, which is 

only half as large as for the restricted sample. For the ownership variables, 

results are very similar to those obtained in table 3.3, although the coefficient 

on the foreign ownership variable is somewhat larger in both sectors (0.83 for 

manufacturing and 0.33 for services, compared to 0.53 and 0.24 respectively). 

Hence, it has been shown that the results obtained in the previous section 

are robust to the inclusion of small firms in the analysis. 

Second, although the Belfirst database provides information on whether 

firms have been subject to a change in ownership (eg. a merger or takeover), 

it is not possible to track these firms following the ownership change. It is 

possible that a takeover results in an exit, followed by the start-up of a new 

firm; in which case only the exit of the firm can be observed. Since the exit 

rate of those firms that have been subject to some form of ownership change 

amounts to 85 percent, compared to 8.90 percent for the sample; it seems 

more than likely that a number of these exits are in fact misclassified. Moreover, 

no information is available on the nationality of ownership before and 

after the change. Therefore, all firms that have been subject to a takeover, a 

merger or a scission over the sample period, were initially omitted from the


Cox PH Model (1) Cox PH Model (1) 

Model Incl. small firms (2) Incl. takeovers (3) Probit (4) 

Variables I II III IV V VI 

Sector of activity Industry Services Industry Services Industry Services 

Constant - - - - -1.07*** -1.19*** 

[0.35] [0.32] 

Domestic MNE 0.59*** -0.02 0.50*** 0.00 0.14* 0.01 

[0.22] [0.09] [0.19] [0.14] [0.08] [0.06] 

Foreign MNE 0.83*** 0.33*** 0.43** 0.20** 0.25*** 0.11*** 

[0.19] [0.08] [0.17] [0.09] [0.07] [0.04] 

Merger dummy - - 3.09*** 2.61*** - - 

[0.20] [0.07] 

log(Age) - - - - 0.01 -0.03*** 

[0.02] [0.01] 

log(Employment) -1.12*** -1.26*** -0.57*** -0.55*** -0.25*** -0.23*** 

[0.04] [0.01] [0.06] [0.04] [0.03] [0.02] 

log(Wages) -0.47*** -0.46*** 0.52*** 0.02 0.28*** 0.06*** 

[0.05] [0.01] [0.19] [0.08] [0.09] [0.04] 

log(Productivity) -0.46*** -0.33*** -0.91*** -0.67*** -0.44*** -0.35*** 

[0.04] [0.01] [0.10] [0.05] [0.05] [0.03] 

log(Herfindahl) 0.00 0.06*** -0.08 0.03 -0.04 0.01 

[0.04] [0.01] [0.07] [0.03] [0.03] [0.01] 

Wald test 1,434.2*** 13,600.6*** 382.4*** 1,807.6*** 509.0*** 1,738.4*** 

N 71,226 367,747 37,079 107,897 36,721 106,289 

Number of firms 14,017 84,083 6,594 19,826 6,524 19,522 

Number of failures 1,760 19,172 583 2,049 520 1,797 

% of total 12.6 22.8 8.8 10.3 8.0 9.2 

Notes: Robust standard errors, adjusted for clustering at the plant level (reported in brackets). 



sample. 

However, it is possible to re-estimate 3.3 on the sample including firms 

that have been subject to an ownership change, taking their different nature 

into account by introducing a dummy variable indicating whether the firm 

has been subject to some form of merger or takeover (Mergeri). Results 

for this alternative model are displayed in columns III and IV of table 3.4. 

Again, results are in line with those obtained in table 3.3. Results for the 

sample including all firms subject to an ownership change, but excluding the 

merger dummy (unreported), are also in line with the results obtained using 

the restricted sample. 

Finally, in the last two columns of table 3.4 the sensitivity of the results 

to the estimation technique used is verified. Following Bernard and Jensen 

(2007a), a probit model is used to assess the determinants of exit behavior at 

the firm level. Focusing on the ownership variables, the associated marginal 

effects for the coefficients on Fori and Domi are between 0.003 and 0.007, 

implying that multinational firms are between 20 (for domestic MNEs active 

in manufacturing) and 50 percent (for foreign MNEs active in manufacturing) 

more likely to exit compared to national firms . Again, overall, results of the 

probit model are very similar to those obtained using the Cox proportional 

hazard model. 


This paper has investigated the hypothesis that multinational companies 

in Belgium are more footloose than national firms. In the empirical analysis 

I have distinguished for nationality of ownership, using a different ownership 

variable for foreign and domestic multinationals. Using an unbalanced sample 

of 26,046 firms located in Belgium between 1996 and 2001, I find that 

multinational firms, both foreign and domestic, have a lower propensity to 

exit unconditionally; that is, they exhibit lower exit rates compared to national 

firms. However, multinationals also tend to be larger, more productive 

and pay higher wages than national firms. Since all these characteristics have 

been found to have significant effects on the probability of firm failure, a Cox


proportional hazard model is estimated to explicitly test for the determinants 

of exit in the sample. 

Consistent with previous studies, I find a significantly negative influence 

of size and productivity at the firm level on the exit probability of firms. 

Results further indicate that firms that pay higher wages for given levels of 

productivity, are more likely to exit. For the ownership variables, the results 

are mixed. The analysis clearly shows that foreign multinationals are more 

footloose than national firms of comparable size, age and productivity, both 

in the manufacturing and service industries. Domestic MNEs on the other 

hand, while more likely to exit in the manufacturing sector, do not exhibit 

significantly higher exit rates than national firms in the service sectors, after 

controlling for firm- and industry-specific variables. I have presented a 

number of refinements and sensitivity analyses to the model to verify the 

robustness of these results. 

These findings have clear policy implications, especially in terms of the 

desirability of the large impact of multinational firms on employment and 

output generation in Belgium. The fact that domestic multinationals, while 

sharing most of the beneficial characteristics of foreign firms in terms of 

employment generation and average productivity, do not seem to share the 

footloose nature of their foreign counterparts in the service sector sheds a 

new light on recent government measures in Belgium, primarily aimed at 

attracting foreign investments; as well as on the recent liberalization of many 

service industries, such as telecommunications and electricity. Naturally, a 

drawback of the present analysis is that it only focuses on the extensive 

margin, i.e. on the shutdown decision of the firm.



3.A.1 Sample selection 

All data are obtained from the Belfirst database (BvDEP, 2004). Exit 17 

occurs if a firm’s employment drops to zero in a particular year. Likewise, a 

firm enters the market in a certain year if there was no previous employment 

recorded. In order to ensure that a failure to report in a given year or a 

sudden drop in employment for some other reason is not misclassified as an 

exit, the data are subject to a number of robustness checks. First, to ensure 

that an entering plant is indeed “new”, I impose the condition that a plant’s 

year of incorporation can not differ by more than two years from the year the 

firm enters the market according to the entry variable 18 . Second, to avoid 

misclassification of “temporary exits”, all firms that re-enter within two years 

after having exited are dropped from the sample 19 . For similar reasons, I omit 

the last two years for which data are available (2002 and 2003), since I can 

not reliably identify entry and exit for these years. Third, I omit all firms that 

have been subject to takeovers, mergers or acquisitions from the sample 20 . 

Finally, since it is more plausible for small 21 firms to reduce employment to 

zero without actually exiting the market, I further limit attention to those 

firms employing at least 10 employees in any particular sample year. 

17 Although the Belfirst database reports firms’ legal status and hence also legal exit, 

I have chosen not to rely on this measure for two reasons. First, inspection of the data 

reveals that the official date associated with the legal status in the database often does 

not concur with the actual time the firm exits the market. Second, communications with 

Bureau Van Dijk made clear that although the legal status is correctly reported whenever 

available; many companies fail to report to the National Bank after ending their activities. 

18 I allow for this two year lag, since it is possible that a firm already exists as a legal 

entity for some time before actually starting its activities. The official year of incorporation 

is replaced by the actual date of entry in these cases. 

19 This procedure differs somewhat from that employed by Mata and Portugal (1994) 

who only omit firms with a temporary exit of at least two years. If a firm exits only one 

year and then enters again, it is considered alive in their sample, while in this case it is 

omitted from the sample. 

20 Cases are identified using the legal status variable. The categories omitted are: 

“Merger with another company to form a third one”, “Absorption by another company” 

and “Scission into several companies”. 

21 I consider firms employing less than 10 employees to be “small”.


3.A.2 Definition of variables 

Ageit 

Difference between year t and the official year of incorporation of the firm 

(replaced by the actual date of entry where necessary). 

Sizeit 

Average number of employees (full-time equivalents, fte) at the firm level in 

year t. 

Wageit 

“Remunerations, social security costs and pensions” per employee (fte) of 

firm i in year t, measured in thousands of euros. The wage variable is expressed 

in real terms, using three-digit producer price indices to deflate the 

monetary values. 

Prodit 

“Net Value Added” per employee (fte) for firm i in year t, measured in 

thousands of euros. Labor productivity is expressed in real terms, using 

three-digit producer price indices to deflate the monetary values. 

Herfjt 

Herfindahl concentration index, measured at three-digit NACE level and calculated 

for the full sample of firms in the Belfirst database with available data 

on turnover. Market shares are calculated on the basis of firm and industry 

turnover. 

Fori 

Foreign multinational ownership variable. A firm is considered to be foreignowned 

if it has some foreign ownership. 

Domi 

Domestic multinational ownership variable. A firm is identified as a domestic 

MNE if it has subsidiaries in countries other than Belgium and it is not 

foreign-owned.

Chapter 4 

Multinational firms, Research 

Effort and Innovative Output: 

An Integrated Approach 


While the literature on the relationship between firms’ international activities 

and their productivity abounds 1 , it remains largely silent on the sources 

of the productivity advantages associated with firms’ global integration. One 

way for firms to achieve productivity increases, is through the accumulation 

of knowledge or innovative output (Castellani and Zanfei, 2007). 

The central purpose of this paper is to investigate whether firms’ innovative 

output differs according to their global engagement status. The analysis is 

carried out using cross-section data from the fourth Community Innovation 

Survey for Belgium (CIS4), pertaining to the years 2002-2004. To estimate 

the firm-level innovation production function, I rely on the methodology 

implemented by Mairesse and Mohnen (2004). 

The present analysis is most closely related to the works of Castellani and 

Zanfei (2006); Criscuolo, Haskel and Slaughter (2005) and Frenz and Ietto- 

1 For reviews of this extensive literature, I refer to Greenaway and Kneller (2007) and 

Wagner (2007). Important theoretical contributions in this field include Melitz (2003) and 

Helpman et al. (2004). 

129

130 Multinationals, research effort and innovative output 

Gillies (2007). While the first use CIS2 data for Italy, the latter two both 

use CIS data for the UK (waves 2 and 3). Controlling for firm size and 

sector of activity, these papers generally find a positive correlation between 

firms’ global engagement status and their innovative output. Criscuolo et 

al. (2005) additionally provide evidence of a positive relationship between 

firm-level research inputs and their innovative output. 

As noted by Mairesse and Mohnen (2002), analyzing innovative output 

is not very different from analyzing firm-level production, since innovative 

output is simply a function of innovative inputs (research effort) at the firm 

level. Moreover, drawing the analogy with the traditional production function 

literature further, simultaneity of research inputs and output introduces 

endogeneity issues. Mairesse and Mohnen (2004) take this endogeneity of research 

inputs in the innovation output production function into account by 

instrumenting for research inputs. Their approach follows that of Crépon, 

Duguet and Mairesse (1998), who introduced an empirical model taking into 

account both the selection issues inherent in the CIS data, as well as the endogeneity 

of research inputs in the innovative output function. Since similar 

selection issues apply to both innovative inputs and outputs, instrumenting 

for research inputs is achieved by once again estimating a Heckman selection 

model. 

The present analysis contributes to the literature in two important ways. 

First, from a methodological point of view, the empirical analysis offers an 

integrated approach. Potential endogeneity of research inputs is explicitly 

taken into account in the innovation output function. Moreover, by applying 

a two-step estimation procedure as in Mairesse and Mohnen (2004), I am 

able to distinguish between the indirect (through higher R&D spending) and 

direct effect of firms’ global integration on innovative output. 

Second, from a policy perspective, the analysis may yield important insights. 

The Belgian economy is characterized by a high dependence on multinational 

firms, not only in terms of employment and sales, but also in terms 

of innovative performance. Cincera et al. (2006) find that seventy percent of 

all patents invented in Belgium are owned by foreign firms, either directly


(through foreign assignees) or indirectly (through a Belgian subsidiary of a 

foreign firm). 

Teirlinck (2005a) further shows that the majority of total R&D spending in 

Belgium originated in foreign-owned firms in 2001. He then goes on to show 

that this dominance of foreign firms in total R&D spending is related to their 

size and sector distribution. Furthermore, his findings indicate that, once the 

technology intensity of the sector is taken into account, location motives for 

R&D are not different for foreign and domestic firms. These results suggest 

that government policy aimed at stimulating or attracting R&D does not 

necessarily need to target foreign firms specifically. Rather, policy should 

be aimed at fostering entrepreneurship in certain high-technology sectors. 

However, one could argue that what matters to long-term growth is not 

research inputs, but rather innovative output; as a driver of total factor 

productivity growth. 

The empirical results can be summarized as follows. Exporters and multinational 

firms (both foreign and home-based) are found to be more likely 

to generate innovative sales than national firms, both unconditionally (not 

taking R&D inputs into account) and conditional upon their engagement in 

R&D investment. However, taking into account that globally engaged firms 

exhibit different R&D spending patterns, i.e. conditional on research effort; 

only exporters are found to be significantly more likely to generate innovative 

sales. Foreign affiliates of multinational firms are not found to be more 

innovative than purely domestic (national) firms. Home-based multinationals 

are significantly less likely to innovate, after controlling for their higher 

average inputs. In terms of innovation intensity (the share of innovative sales 

in total sales), globally active firms are not found to behave differently from 

national firms, either conditionally or unconditionally. Distinguishing between 

high-technology and low-technology sectors further shows that there 

is an innovation premium associated with firms’ international activities in 

high-technology sectors, conditional on their research intensity; while in lowtechnology 

sectors, a negative innovation premium emerges.


A caveat is worth noting here. Aw, Roberts and Winston (2007) provide 

evidence showing that firms jointly make their decisions to innovate and to 

engage in international markets. Costantini and Melitz (2007) build a dynamic 

model that captures the self-selection of more productive firms into 

international markets, the joint export and innovation decisions and the continuing 

innovation of firms following entry into global markets. From this 

point of view, innovation can be linked to exports and productivity in two 

ways. Firms can innovate prior to entry on international markets, enabling 

them to gain the productivity advantage needed to overcome the sunk cost 

associated with global engagement. On the other hand, firms’ international 

experiences may induce further innovative activities (for instance stimulated 

by contacts with foreign buyers), hence reinforcing their productivity advantages. 

Since the data used in this paper are cross-sectional in nature, I am not able 

to investigate to what extent results obtained here are driven by selection 

effects. Several papers investigate the selection hypothesis specifically for 

innovative output, with mixed results. Wakelin (1998), Basile (2001) and 

Cassiman and Golovko (2007) study the impact of innovative output (product 

or process innovation) on both the export status and intensity of firms in 

the UK, Italy and Spain respectively. Their empirical findings support the 

selection hypothesis. Damijan, Kostevc and Polanec (2008) apply matching 

techniques to Slovenian firm-level data to investigate the dual relationship 

between firms’ innovation and export decisions. Their results lend support 

to the learning hypothesis (although only for process innovations), while 

they find no evidence of selection effects. Given these mixed results, results 

obtained here should be interpreted as correlations and can not be interpreted 

as causal effects of firms’ global activities on innovation output. 

The rest of the paper is organized as follows. Section 4.2 formalizes the 

concept of the innovation production function, which will serve as the estimating 

framework in the empirical analysis. Section 4.3 discusses the data 

and relevant empirical facts, while section 4.4 presents the results. Section 4.5 

concludes.

The innovation production function 133 

4.2 The innovation production function 

As noted by Mairesse and Mohnen (2002, 2004), analyzing firm-level innovative 

output is not very different from analyzing regular production output. 

Firm-level innovative output, like regular production output, results from 

the transformation of certain inputs into output; a relationship that can be 

analyzed in the context of a production function. In its most general form, 

this production function looks as follows: 

Yi = Aif(Xi) (4.1) 

where Yi represents physical (innovative) output of firm i and Xi are 

firm-level (research) inputs, measured in quantities. Ai is the unobservable 

firm-level productivity term, labeled “Innovativity” or “Innovativeness” by 

Mairesse and Mohnen (2002). Since firm-level quantities of inputs and outputs 

are generally unavailable, estimation of (4.1) in the traditional production 

function literature is usually achieved by deflating inputs and outputs 

using appropriate (industry-level) price indices. 

However, estimating equation (4.1) for a balanced panel of firms, using Ordinary 

Least Squares (OLS) raises a number of methodological issues, comparable 

to the issues arising when estimating traditional production functions 2 . 

First, to the extent that less innovative firms are more likely to exit, the 

use of a balanced panel of firms introduces a selection bias in the sample. 

Second, since firms simultaneously decide on the allocation of inputs and 

production of output, endogeneity issues arise when (4.1) is estimated using 

OLS. Third, if input or output markets are characterized by imperfect 

competition, the use of industry-wide deflators rather than firm-level prices 

will result in an omitted output and input price bias. Finally, if firms produce 

multiple products (or different products), which potentially differ in 

terms of their production technology and demand, an additional bias will be 

introduced in traditional TFP estimates. 

2 For a recent review of the different methodological issues arising when estimating 

output production functions, and the existing parametric and semiparametric techniques 

designed to overcome them, I refer to Van Beveren (2007b).


Several parametric and semiparametric estimators have been introduced in 

the production function literature to take some of these issues into account. 

Apart from the more conventional techniques of Instrumental Variable (IV) 

or Generalized Method of Moments (GMM) estimation, several estimators 

have been specifically developed for the estimation of production functions. 

Olley and Pakes (1996) and Levinsohn and Petrin (2003) introduce a semiparametric 

estimator. They are able to resolve the simultaneity bias by 

using a proxy variable to substitute for unobserved productivity (investment 

in Olley and Pakes (1996) and materials in Levinsohn and Petrin (2003)). 

Both estimators are also able to correct for the selection bias, by taking 

firms’ survival probability into account in the estimation procedure. De 

Loecker (2007) extends the algorithm developed by Olley and Pakes to take 

imperfect competition in output markets, as well as multi-product firms into 

account. 

Unfortunately, while data on firm-level output and inputs tend to be available 

in relatively long time-series, often for the full population of firms in a 

particular country; innovation data generally suffer from a number of important 

drawbacks, rendering consistent estimation of (4.1) even more problematic. 

The discussion here will focus specifically on the Community Innovation 

Survey data. While other firm-level data sets on innovation exist 3 , 

the CIS data have the important advantage of being uniform (with some 

exceptions) across European countries. Moreover, the CIS data have been 

used in a growing number of contributions, some recent examples include 

Cassiman and Veugelers (2006) and Griffith, Huergo, Mairesse and Peters 

(2006) 4 . The Community Innovation Survey is carried out every two years 

in all European Union (EU) member countries, under the supervision of the 

authorized national agencies. Firm-level data (anonymous or not) can only 

be obtained through the various national statistical agencies collecting the 

data 5 . 

3 See for instance Cassiman and Golovko (2007) for Spain. 

4 For an overview of the general characteristics of the CIS data and its advantages and 

drawbacks, I refer to Smith (2005). Hall and Mairesse (2006) provide a (selective) review 

of a number of recent empirical papers applying the CIS data. 

5 Eurostat only provides the data in micro-aggregated form.


When using the CIS data to estimate (4.1), several difficulties emerge. 

First, the CIS data are cross-sectional in nature. While data of different 

waves can be merged, this results in limited sample sizes, since sampling is 

performed independently for each wave (i.e. firms are not always observed 

in each wave). Moreover, while some researchers have been able to obtain 

firm-level data for different waves (eg. Frenz and Ietto-Gillies, 2007; Criscuolo 

et al., 2005), most studies use cross-section data for a particular wave. This 

limits the available estimators to those that do not require information on 

firm dynamics and hence rules out the use of GMM and the semi-parametric 

estimators of Olley and Pakes, Levinsohn and Petrin and De Loecker. 

The use of cross-section data has an important additional drawback. By 

definition cross-section data contain no information on firm entry and exit. 

In fact, in the Belgian case, only firms that continued their activities throughout 

the period 2002-2004 were selected for the CIS4 population (Teirlinck, 

2005b). Using different waves of the questionnaire would still require additional 

external information sources on entry and exit of firms to identify firm 

dynamics, since firms are not tracked across different waves. Hence, potential 

selection effects can generally not be accounted for. 

Another issue researchers encounter when estimating (4.1) specifically for 

innovative output is the lack of suitable price deflators (or ideally, firm-level 

prices) of innovative inputs and outputs. On the output side, using industrylevel 

output price indices to deflate innovate sales imposes the assumption 

that prices of new products evolve in the same way as regular output prices. 

On the input side, finding appropriate deflators is even more problematic, 

specifically when use is made of the CIS data. Firms in the CIS questionnaire 

are asked to report their total internal R&D expenditures in a particular 

year. These expenditures include personnel costs, material costs and capital 

expenditures on materials and equipment specifically for R&D. Apart from 

the fact that no price indices exist specifically for R&D personnel, material 

costs and investment; no information is available in the CIS data on the shares 

of each of these expenditures in firms’ total research spending, rendering any 

weighting of these price indices to obtain a composite index arbitrary.


Moreover, since firms’ expenditures reported in the CIS questionnaire, include 

information on both variable inputs (personnel and materials costs) 

as well as fixed inputs (capital expenditures), little can be said about the 

direction of the biases introduced by the methodological issues listed above. 

For instance, Levinsohn and Petrin (2003) show, in a general production 

function setup, where labor and capital are positively correlated, that the 

simultaneity bias will lead to an upward bias in the coefficient of the variable 

input (labor) and a downward bias for the fixed input coefficient (capital). 

Since variable and fixed inputs are not separable when use is made of the 

CIS-variable on internal research spending, it is not possible to make any 

predictions on the direction of the biases introduced. 

Apart from the drawbacks of the data as compared to regularly available 

production function data, the CIS data possess another important characteristic 

that needs to be taken into account. Only firms that had ongoing, 

successful or abandoned innovation activities during the period considered 

are asked to fill out the full questionnaire. Non-innovators on the other hand, 

only need to answer particular general questions 6 . As a consequence, the CIS 

data are characterized by selection issues, unrelated to the selection issues 

introduced by the use of a sample of continuing firms. 

Given these important data constraints, researchers interested in estimating 

(4.1) can essentially only rely on IV estimation to correct for the simultaneity 

bias. A natural choice of model to take the selection issues inherent in 

the questionnaire into account is the Heckman selection model. Crépon et al. 

(1998) were the first to propose such a framework, linking research inputs to 

innovative output (and innovative output to total factor productivity). Their 

approach corrects for simultaneity by instrumenting for R&D inputs in the 

innovation output production function and tackles selection issues through 

estimation of a generalized tobit (Heckman) model. 

Many recent studies implement the approach of Crépon et al. (1998), examples 

include Benavente (2006); Griffith et al. (2006); Lööf and Heshmati 

6 For Belgium, these are mostly related to general firm characteristics; such as employment, 

turnover and group membership.


(2004) and Van Leeuwen and Klomp (2006). The description of the model 

given below follows most closely that of Mairesse and Mohnen (2004), who 

implement the approach of Crépon et al. (1998) using French CIS3 data. 

The basic model proposed by Crépon et al. (1998) and implemented by 

Mairesse and Mohnen (2004) consists of a two-stage model. In the first 

stage, firms in the sample decide on whether or not they invest in R&D and, 

if they invest, on the scale of their investment. Estimation of the first stage is 

achieved using a generalized tobit (Heckman) model. In the second stage of 

the model, innovative output is related to firms’ research effort (instrumented 

from the first stage) and other factors. Since the present analysis focuses on 

innovative sales to measure innovative output 7 , selection issues apply also to 

the second stage of the model. Hence, similar to the first stage, a generalized 

tobit model is estimated; allowing for a selection and innovation intensity 

equation. 

Formally, stage 1 of the model can be summarized as follows (Greene, 2008; 

Verbeek, 2000): 

⎧ 

⎪⎨ 

⎪⎩ 

s ∗ rd,i = x′ 1 

rd,i β 1 rd,i + ε1 rd,i 

y ∗ rd,i = x′ 2 

rd,i β 2 rd,i + ε2 rd,i 

y ∗ rd,i = yrd,i; srd,i = 1 if s ∗ rd,i 

> 0 

yrd,i not observed; srd,i = 0 if s ∗ rd,i 

≤ 0 

(4.2) 

where yrd,i refers to firm i’s R&D intensity, which is only observed for those 

firms that continuously engaged in R&D between 2002 and 2004 and srd,i 

is the selection variable, i.e. a dummy variable equal to one for continuous 

are their corresponding latent variables, xrd,i 

R&D performers. y∗ rd,i and s∗rd,i refers to the explanatory variables, βrd,i to the coefficients of the model and 

εrd,i are the (correlated) error terms. 

7 The CIS data contain a number of other innovation output indicators, such as whether 

a firm has introduced a product or process innovation or has applied for a patent. I have 

chosen to rely only on the innovative sales variable here because of its close relation to 

regular production function output. Mairesse and Mohnen (2004) implement the approach 

outlined here for the different output measures available in the CIS data.


Similarly, stage 2 of the model looks as follows: 

⎧ 

⎪⎨ 

⎪⎩ 

s ∗ inn,i = x′ 3 

inn,i β 3 inn,i + ε3 inn,i 

y ∗ inn,i = x′ 4 

inn,i β 4 inn,i + ε4 inn,i 

y ∗ inn,i = yinn,i; sinn,i = 1 if s ∗ inn,i 

> 0 

yinn,i not observed; sinn,i = 0 if s ∗ inn,i 

≤ 0 

(4.3) 

where yinn,i refers to firm i’s innovative sales intensity, which is only observed 

for those firms that generated innovative sales between 2002 and 2004, 

and sinn,i is the selection variable, i.e. a dummy variable equal to one for 

innovators. y ∗ inn,i and s ∗ inn,i are their corresponding latent variables, xinn,i 

refers to the explanatory variables, βinn,i to the coefficients of the model and 

εinn,i are the (correlated) error terms. 

Since only firms with ongoing, successful or abandoned innovation activities 

are required to fill out the full CIS questionnaire, the choice of which 

variables to include in x ′ 1 

rd,i is limited to those variables available for all firms. 

Variables included are firm size (measured as the natural logarithm of employment), 

two-digit sector dummies, three region dummies and firms’ global 

engagement status. Four (exclusive) groups of firms are distinguished: (1) 

firms that are only active on the domestic market (the baseline category, 

captured in the regression constant), (2) firms that export but are not part 

of a multinational firm, (3) foreign affiliates of multinational firms and (4) 

home-based multinationals. 

Since firms with reported R&D expenditures are by definition innovators, 

i.e. they are required to fill out the full questionnaire, more variables are available 

to explain the scale of investment in research at the firm level. Hence, 

x ′ 2 

rd,i includes, apart from the sector, region and global engagement dummies, 

a dummy indicating whether the firm engaged in research cooperation activities 

between 2002 and 2004, as well as a funding dummy, equal to one if the 

firm has acquired regional, national or EU funding during the sample period. 

Employment is excluded from x ′ 2 

rd,i for identification purposes 8 . 

8 If employment is included in the second stage of the model, its coefficient is not 

significant. This is not surprising, since research effort is specified relative to total sales.


Similarly to the first stage, x ′ 3 

rd,i and x ′ 4 

rd,i include two-digit sector dummies, 

three region dummies and firms’ global engagement status. In addition, the 

predicted value of the logarithm of firms’ R&D intensity from the first stage, 

y∗ rd,i , enters both equations of the generalized tobit model. As in the first 

stage, employment only enters the selection equation of the model. No other 

explanatory variables are included in the second stage. This implies that the 

cooperation and funding dummies act as exclusion restrictions in the model 

and are essentially assumed to affect innovative sales only indirectly through 

their effect on research effort (Mairesse and Mohnen, 2004). 

Whether these are reasonable assumptions is open for debate. Both cooperation 

and funding can be considered as sources of external finance to 

the firm for its research activities. From this point of view, they are likely 

to enhance firms’ own research efforts, but will not necessarily contribute 

to the success or failure of a particular research project over and beyond 

their impact on research spending. On the other hand, to the extent that 

cooperation generates knowledge spillovers or other externalities, its impact 

on innovative output may not be limited to its effect on research spending. 

Due to the innovation-centric nature of the CIS questionnaire and the crosssectional 

nature of the data, finding good instruments for research effort is a 

difficult task, that certainly merits more attention in future work. 

Apart from the global engagement dummies, which are defined as in Castellani 

and Zanfei (2006) and Criscuolo et al. (2005), the specification of the 

model outlined above closely follows that of Mairesse and Mohnen (2004). 

However, a number of important differences with respect to their model 

are worth noting here. First, in the R&D selection equation, Mairesse and 

Mohnen (2004) additionally include demand pull and cost push indicators. 

However, these variables are not available for all firms in the CIS4 data for 

Belgium. Mairesse and Mohnen (2004) additionally include firms’ weighted 

market shares and a diversification index, calculated using data from external 

sources, in the first stage of the model. They further include a number 

of firm-specific knowledge sources, available in the CIS data, in the R&Dintensity 

equation.


Further, the measure of R&D-intensity used by Mairesse and Mohnen 

(2004) is defined as the natural logarithm of R&D per employee. I have 

chosen to define research effort relative to firm turnover. As noted above, 

no suitable (industry-level) price indices exist specific to research inputs and 

outputs. By weighting innovative inputs and outputs by total turnover at 

the firm-level, I am able to take out at least that part of firm-level price 

variation that is common to regular output as well as research inputs and 

outputs. Weighting the input variable by the number of employees, while 

innovative output is measured relative to sales; as is done in Mairesse and 

Mohnen (2004) implies that this general price variation is taken out on the 

output side, but not on the input side, introducing an additional bias in the 

estimation results. I therefore rely on firm-level sales to weight both research 

inputs and innovative outputs in the empirical specification. 

The approach of Crépon et al. (1998) and Mairesse and Mohnen (2004) 

offers a number of important advantages over existing work analyzing the 

impact of firms’ global activities on their innovative performance (eg. Castellani 

and Zanfei, 2006; Criscuolo, Haskel and Slaughter, 2005 and Frenz and 

Ietto-Gillies, 2007). First, from a methodological point of view, the empirical 

model offers an integrated approach, taking both the selection issues inherent 

in the CIS data, as well as potential endogeneity of research inputs in the 

innovation output function into account. While Criscuolo et al. (2005) take 

the endogeneity of R&D inputs in the innovative production function into 

account 9 , they do not take the selection issues noted above explicitly into account. 

Second, by applying a two-step estimation procedure, it is possible to 

distinguish between the indirect (through higher R&D spending) and direct 

effect of firms’ global integration on innovative output. 

9 Criscuolo et al. (2005) instrument for R&D personnel (their measure of research inputs) 

using the four-digit (NACE) industry averages of R&D personnel (or alternatively, 

the share of R&D personnel in the total employee base), constructed using CIS2-data, 

excluding those firms that reappeared in CIS3. They obtain similar results as in their 

baseline specification when instrumenting for research inputs.

Data and empirical facts 141 

4.3 Data and empirical facts 

The innovation data used in the empirical analysis are taken from the 

Community Innovation Survey (CIS4) for Belgium and are obtained through 

the Belgian Science Policy 10 . The CIS data contain limited information on 

ownership. The data allow for identification of all firms that are part of a 

group and also lists the country of origin of the headquarters of that group. 

Additionally, the database contains information on firm-level exports in 2004. 

This enables me to identify exporters and foreign affiliates of multinational 

firms. To verify whether firms are part of a home-based multinational (i.e. a 

Belgian firm with subsidiaries abroad), I use subsidiary information obtained 

from Belfirst 11 (BvDEP, 2006). For a detailed description of the sample selection 

and definitions of variables, I refer to the data appendix (section 4.A). 

The CIS4 questionnaire pertains to the years 2002 to 2004. However, qualitative 

questions are asked only once for the entire period, while quantitative 

measures apply to the year 2004. Only employment and turnover are reported 

for 2002 and 2004. Hence, all innovation data are cross-sectional in 

nature. The sample used in the empirical analysis consists of 2,988 firms. 

Table 4.1 gives the sector distribution of the sample by ownership type. The 

last column in the table lists total R&D spending by sector. Sectors are 

classified according to the NACE (Rev. 1.1) classification 12 . 

There are important differences in the sector distribution of firms in the 

manufacturing sector according to the degree of internationalization. While 

domestic firms and exporters dominate in Metallic products and in Food, beverages 

and tobacco, foreign affiliates of multinational firms feature predominantly 

in Chemicals and Metallic products. Home-based multinationals, like 

domestic firms, tend to be active in Food, beverages and tobacco, but also in 

10 I would like to thank Manu Monard, Peter Teirlinck and the CFS-STAT Commission 

for allowing me to access the firm-level data; for answering questions related to the data 

used and for their hospitality during my visits there. 

11 The Belfirst database contains ownership and accounting information on the population 

of Belgian firms. Other recent papers using this database include Abraham, Konings 

and Vanormelingen (2007b); Konings (2008) and Monfort, Vandenbussche and Forlani 

(2008). 

12 The NACE classification can be downloaded from the Eurostat Ramon server at 

http://ec.europa.eu/eurostat/ramon.

Table 4.1: Sector distribution of firms by firm type 


Notes: Four firm types are distinguished: (1) National firms: do not export and are not part of a 

multinational firm; (2) Exporters: firms that export, but are not part of a multinational; (3) Foreign 

affiliates of multinational firms; (4) Home-based (domestic) multinational firms. Total internal R&D 

expenditures in 2004 of all continuous R&D performers between 2002-2004. Variables are defined in the 

data appendix (section 4.A.2). 

National Foreign Domestic Total 

Firms Exporters MNEs MNEs Int. R&D 

C. Mining and quarrying 4 8 6 2 538 

CA. Of energy producing materials 2 0 0 0 0 

CB. Excluding energy producing materials 2 8 6 2 538 

D. Manufacturing 272 546 321 99 731,510 

DA. Food, beverages and tobacco 62 71 32 13 20,343 

DB. Textiles 25 55 13 10 11,878 

DC. Leather (products) 1 3 1 0 2,636 

DD. Wood (products) 8 26 2 4 589 

DE. Pulp and paper 28 67 26 9 5,942 

DF. Coke and petroleum (products) 2 2 2 0 2,400 

DG. Chemicals 5 27 48 10 121,069 

DH. Rubber and plastics 7 23 20 6 43,848 

DI. Other non-metallic products 9 30 23 7 9,604 

DJ. Metallic products 56 90 51 9 34,587 

DK. Machinery and equipment n.e.c. 15 44 32 12 67,968 

DL. Electrical and optical equipment 24 39 28 12 336,433 

DM. Transport equipment 11 25 32 4 71,334 

DN. Manufacturing n.e.c. 19 44 11 3 2,880 

E-Q. Services and related activities 786 432 438 74 1,065,743 

E. Electricity, gas and water supply 5 0 2 0 14,716 

F. Construction 141 31 17 11 175 

G. Wholesale and retail trade 262 191 216 20 32,424 

H. Transport, storage and communication 162 95 62 9 113,445 

J. Financial intermediation 36 12 32 12 52,282 

K. Other business activities 180 103 109 22 852,702 

Total 1,062 986 765 175 1,797,791


Machinery and Electrical and optical equipment. The majority of domestic 

multinationals and exporters are active in the manufacturing sector, while 

foreign firms and national firms dominate in services. Within the tertiary 

sector, Wholesale and retail trade and Other business activities are the most 

populated sectors. 

In terms of R&D spending, the two largest sectors are Other business 

activities, which includes the IT sector and Electrical and optical equipment, 

together they account for more than 60 percent of R&D spending in the 

sample. Chemicals, Transport and communication and Transport equipment 

are also included in the top five. Finally, from the table, it is clear that 

R&D spending is highly concentrated in a few sectors; the top five sectors 

together account for more than eighty percent of total research spending in 

the sample. 

Table 4.2 summarizes the sample distribution of firms according to their 

global engagement status. In terms of the number of firms, domestic firms 

and firms that only export (but are not part of a multinational firm) dominate, 

together these firm account for almost 70 percent of the total number 

of firms. However, in terms of employment and output generation, and also 

for innovative inputs (internal R&D expenditures) and output (innovative 

sales), multinational firms clearly dominate. Taken together, foreign and 

home-based multinationals generated 67 percent of total employment, 80 

percent of turnover, 62 percent of internal R&D expenditures and 87 percent 

of total innovative sales in the sample. 

From table 4.2 it is clear that globally active firms dominate the sample, 

both in terms of total research efforts as well as total innovative sales. 

These preliminary facts suggest that the innovation premium attributed to 

globally active firms (whether exporters or multinationals) in previous work 

(eg. Castellani and Zanfei, 2006; Criscuolo et al., 2005; and Frenz and Ietto- 

Gillies, 2007) might be (partly) related to these firms’ different spending patterns 

on R&D. By estimating the innovation production function along the 

lines suggested by Mairesse and Mohnen (2004) and described in section 4.2, 

I am able to distinguish between the indirect (through R&D spending) and


National Foreign Domestic 

Firm type firms Exporters MNEs MNEs 

Number of firms 1,062 986 765 175 

(% of total) (35.54) (33.00) (25.60) (5.86) 

Employment (fte) 77,456 51,045 157,316 103,914 

(% of total) (19.87) (13.10) (40.37) (26.66) 

Turnover (xe1,000) 16,300,000 12,000,000 78,800,000 40,200,000 

(% of total) (11.08) (8.15) (53.48) (27.30) 

Internal R&D (xe1,000) 577,191 93,899 632,698 494,003 

(% of total) (32.11) (5.22) (35.19) (27.48) 

Innovative sales (xe1,000) 1,119,372 1,235,825 11,900,000 4,456,683 

(% of total) (5.98) (6.60) (63.61) (23.81) 

Notes: Four firm types are distinguished: (1) National firms: do not export and 

are not part of a multinational firm; (2) Exporters: firms that export, but are not 

part of multinational; (3) Foreign affiliates of multinational firms; (4) Home-based 

multinational firms. Variables are defined in the data appendix (section 4.A.2). 

All data pertain to the year 2004 and represent sample totals (percentage of 

total). 

Table 4.2: Sample distribution according to global engagement status 

direct impact of firms’ global engagement status on innovative output. However, 

before proceeding to the empirical results, table 4.3 presents summary 

statistics for the key variables used in the empirical analysis, again distinguished 

according to firms’ global engagement status. 

A number of striking features emerge from table 4.3. Contrary to expectations, 

firms that only export are (significantly) smaller than national firms 

(on average). Multinational firms on the other hand, are on average much 

larger than both domestic firms and exporters, which also explains their 

large impact on total employment and output generation in the sample (cfr. 

table 4.2). Moreover, both exporters and multinational firms (foreign and 

home-based) show a higher propensity to engage continuously in R&D and 

to generate innovative sales, compared to national firms. In relative terms, 

home-based multinationals show the highest proportion of R&D spending 

firms (43 percent) as well as the highest proportion of firms generating innovative 

sales (49 percent).


National Foreign Domestic 

Variables Firms Exporters MNEs MNEs 

All firms 

N 1,062 986 765 175 

Employment 72.93 51.77** 205.64*** 593.79*** 

Cont. R&D 0.07 0.21*** 0.23*** 0.43*** 

Innovators 0.13 0.32*** 0.37*** 0.49*** 

Continuous R&D performers 

N (cont. R&D) 75 205 178 75 

R&D intensity 0.06 0.07 0.08 0.06 

Cooperation 0.48 0.48 0.70 0.68*** 

Funding 0.29 0.44** 0.38* 0.56*** 

Innovators 

N (innovator) 140 311 283 86 

Innovative sales intensity 0.24 0.25 0.23 0.24 

Notes: Four firm types are distinguished: (1) National firms: do not export 

and are not part of a multinational firm; (2) Exporters: firms that export, but 

are not part of multinational; (3) Foreign affiliates of multinational firms; (4) 

Home-based (domestic) multinational firms. Values are sample means (except 

when the number of firms is reported). Variables are defined in the data 

appendix (section 4.A.2). Significance levels (* p < 0.10 ; ** p < 0.05 ; *** p 

< 0.01) refer to a one-tailed test on the difference with respect to the baseline 

category (national firms). 

Table 4.3: Summary statistics by firm type


The second part of table 4.3 summarizes the key variables relevant to the 

first stage of the empirical model, i.e. estimation of firms’ research intensity. 

In spite of the large contribution of multinationals to total R&D spending 

in the sample (cfr. table 4.2), sample averages of the R&D over sales ratio 

(R&D-intensity) are not found to be different according to firms’ global engagement 

status. This lends support to the findings of Teirlinck (2005a) that 

the dominance of foreign firms in R&D spending is at least partly attributable 

to their larger size. Domestic multinationals are more likely to engage in cooperation 

activities compared to national firms. Finally, exporters, foreign 

affiliates and home-based multinationals are more likely than domestic firms 

to have obtained funding from regional, national or EU authorities between 

2002 and 2004. 

The last part of table 4.3 shows (apart from the number of firms that 

generate innovative sales) the average innovative sales intensity for the subsample 

of firms with innovative sales, which is the relevant sample in the final 

stage of the estimation procedure. Similar to the R&D-intensity variable, 

no differences between the firms according to their global activities can be 

discerned. Again, this suggests that multinational firms’ dominance in terms 

of total innovative inputs and output is, at least partly, attributable to their 

larger average size. 

4.4 Empirical results 

Before proceeding to the estimation results of the full empirical model 

summarized in (4.2) and (4.3), it is useful to investigate whether there is in 

fact an innovation premium associated with firms’ global engagement status 

in Belgium, taking into account firm size and sector distribution. As 

was noted above, Castellani and Zanfei (2006), Criscuolo et al. (2005) and 

Frenz and Ietto-Gillies (2007) provide evidence that firms that are active on 

international markets are more innovative than their national counterparts, 

controlling for size and sector distribution.


4.4.1 Preliminary evidence 

Table 4.4 presents preliminary evidence on the existence of an innovation 

premium for globally engaged firms in Belgium. Both parts of the table 

(models I and II) estimate an innovation production function. In the first 

part of the table (model I), only the global engagement dummies, firm size 

(in the selection equation) and sector and region dummies are taken into 

account. The second part of the table (model II) controls for R&D inputs 

by including a dummy equal to one for firms that continuously engage in 

R&D 13 . 

Results obtained in table 4.4 suggest that exporters and multinational 

firms (whether foreign and home-based) are more likely to generate innovative 

sales compared to national firms. However, conditional on selection, firms’ 

global engagement status does not have a significant impact on the scale of 

innovative output (i.e. the innovative sales share). Comparing the results 

of the two models in table 4.4 further points to the importance of taking 

R&D inputs into account in the innovation output function. Controlling for 

continuous R&D engagement of the firm lowers the innovation premium of 

globally active firms and also significantly reduces the coefficient obtained on 

firm size. 

Overall, these preliminary results support the existence of an innovation 

premium associated with firms’ global engagement status. Table 4.4 shows 

that globally active firms are more likely to generate innovative sales than 

their national counterparts. However, conditional on their higher selection 

probability, they do not generate more innovative sales (in relative terms) 

than domestically oriented firms. 

A final point is worth noting about these preliminary results. The different 

results obtained for the selection and continuous part of the model, suggest 

that it is important to estimate a flexible model, such as the generalized 

13 Since the R&D-intensity variable in (4.3) is specified in logarithmic form and is hence 

undefined for firms with no reported R&D expenditures, it is not possible to implement 

this variable here. I therefore rely on a dummy variable indicating firms’ engagement in 

research effort.


(I) (II) 

Selection Innovative Selection Innovative 

Dependent variable equation sales equation equation sales equation 

Exporter (d) 0.16*** 0.16 0.12*** 0.15 

[0.02] [0.10] [0.02] [0.10] 

Foreign MNE (d) 0.14*** 0.00 0.13*** -0.03 

[0.03] [0.11] [0.03] [0.11] 

Home-based MNE (d) 0.19*** 0.07 0.11** 0.04 

[0.05] [0.14] [0.05] [0.14] 

log(Employment) 0.07*** - 0.03*** - 

[0.01] [0.01] 

R&D (d) - - 0.51*** 0.05 

[0.03] [0.08] 

Sector dummies yes yes 

Region dummies yes yes 

N 2,988 2,988 

left-censored N 2,168 2,168 

Results of heckman selection model. Dependent variable: log of innovative 

sales intensity. Reported values represent marginal effects evaluated at the 

mean of the independent variable or the discrete change of a dummy variable 

(d) from 0 to 1, standard errors are reported in brackets. For the selection 

equation, marginal effects refer to the marginal probability change. For the 

intensity equation, marginal effects refer to the change in intensity, conditional 

upon being selected. The group of exporters comprises of firms that export 

but are not part of a multinational firm. R&D dummy is equal to one for 

continuous R&D performers. Significance levels: * p < 0.10 ; ** p < 0.05 ; 

*** p < 0.01. Variables are defined in the data appendix (section 4.A.2). 

Table 4.4: Innovation production function: Preliminary evidence


tobit setting applied here, hence allowing coefficients to differ in the two 

estimation stages. Both Castellani and Zanfei (2006) and Criscuolo et al. 

(2005) estimate a tobit model 14 , hence restricting the coefficients of the two 

estimation stages to be equal. 

4.4.2 The innovation production function: baseline re- 

sults 

Table 4.5 shows the results of estimating the full empirical model, as specified 

in (4.2) and (4.3). In the first stage, a generalized tobit model is estimated, 

where R&D-intensity (specified in logarithmic form) is the dependent 

variable. As was noted in section 4.2, selection issues inherent in the CIS 

data limit the choice of variables that can be included in the selection equation 

of this model to those questions answered by all firms in the CIS sample. 

Hence, while the selection and intensity equation of stage 1 both include firm 

size, global engagement status, sector and region dummies; the R&D intensity 

equation further includes two dummies, representing whether the firm 

has engaged in cooperation activities (Cooperation) or has obtained funding 

from the regional, national or European level during the period 2002-2004 

(Funding). 

Results for the first stage of the estimation algorithm (firm-level R&D intensity) 

suggest that exporters and multinational firms (both home-based 

and foreign) are more likely to engage in R&D investment, controlling for 

their size and distribution across sectors and regions. However, conditional 

on their higher selection probability, only domestic multinationals are shown 

to spend significantly more on R&D than purely domestic firms. Firms that 

have acquired funding from the regional, national or EU level spend significantly 

more on internal research effort, conditional on selection. Cooperation 

is not found to be a significant determinant of internal R&D intensity. 

The right-hand side of table 4.5 shows the results of estimating (4.3), 

while instrumenting for research effort using the predicted logarithm of R&D- 

14 Frenz and Ietto-Gillies (2007) investigate innovative output using a probit model, 

where the dependent variable is a dummy variable equal to one if the firm has introduced 

a (product or process) innovation.


Stage 1 Stage 2 

Selection R&D intensity Selection Inn. Sales 

Exporter (d) 0.12*** 0.15 0.08*** 0.07 

[0.02] [0.24] [0.03] [0.11] 

Foreign MNE (d) 0.05** 0.24 0.04 -0.09 

[0.02] [0.24] [0.03] [0.12] 

Home-based MNE (d) 0.17*** 0.58** -0.09** -0.21 

[0.04] [0.26] [0.04] [0.19] 

log(Employment) 0.07*** - 0.14*** 

[0.01] [0.01] 

Cooperation (d) - -0.16 - - 

[0.15] 

Funding (d) - 0.54*** - - 

[0.15] 

log(R&D/sales)* (1) 0.41*** 0.42** 

[0.05] [0.19] 



N 2,988 2,988 


Results of two-step heckman selection model. Reported values represent 

marginal effects evaluated at the mean of the independent variable or the discrete 

change of a dummy variable (d) from 0 to 1, standard errors are reported 

in brackets. For the selection equation, marginal effects refer to the marginal 

probability change. For the intensity equation, marginal effects refer to the 

change in intensity, conditional upon being selected. The group of exporters 

comprises of firms that export but are not part of a multinational firm. R&D 

dummy is equal to one for continuous R&D performers. Significance levels: * p 

< 0.10 ; ** p < 0.05 ; *** p < 0.01. Variables are defined in the data appendix 

(section 4.A.2). (1) Predicted R&D intensity obtained from first stage of the 

estimation procedure. 

Table 4.5: Innovation production function: Two-stage estimation


intensity obtained from the first stage. As was noted in section 4.2, the 

cooperation and funding variable act as exclusion restrictions in the second 

stage, i.e. they are assumed to affect the innovative sales share only through 

their impact on internal R&D (Mairesse and Mohnen, 2004). Results shown 

in the right-hand panel of table 4.5 show some striking differences compared 

to the results in table 4.4. 

Conditional on their higher R&D spending patterns, multinational firms 

are no longer found to be significantly more likely to generate innovative sales, 

compared to their uni-national counterparts. Only exporters are found to be 

significantly more likely to be innovative, conditional on their R&D spending, 

although the marginal effect is lower than in table 4.4 (0.08 instead of 0.16 in 

the left-hand side panel of table 4.4). Home-based multinationals are found 

to be significantly less likely to generate innovative sales, once their higher 

probability to engage in R&D and their higher spending patterns are taken 

into account. 

The marginal effect on the predicted R&D intensity (in logs) in table 4.5 

amounts to 0.41 for the selection stage, suggesting that a ten percent increase 

in firm-level research effort is associated with an increase of 4.1 percentage 

points in the probability that the firm will generate innovative sales. In light 

of the overall probability that a firm in the sample generates innovative sales, 

which amounts to 27.44 percent 15 ; this suggests that a ten percent increase 

in firm-level research effort is associated with a 15 percent increase in the 

probability that the firm will generate innovative output 16 . For innovative 

sales intensity, a ten percent increase in research spending is associated with 

a 4.2 percent increase in innovative sales intensity 17 . The magnitude of these 

returns to R&D effects is much lower than the results obtained by Mairesse 

and Mohnen (2004) for France, who report elasticities up to ten times larger 

than those obtained here. 

15 Out of the 2,988 firms in the sample, 820 or 27.44 percent generated innovative sales 

during the period 2002-2004. 

16 (27.44 + 4.1) / 27.44 = 14.94. 

17 The marginal effect amounts to 4.2. Since both variables are expressed in logarithmic 

form, this marginal effect can be interpreted as an elasticity.


4.4.3 High-tech versus low-tech sectors 

Cassiman and Mendi (2008) compare the performance of subsidiaries of foreign 

and domestic firms in Spain, using a data set comparable to the CIS data 

and find that foreign firms are less innovative than subsidiaries of Spanish 

firms, both in terms of innovative inputs and output. However, they further 

show that this result is mainly driven by the low-technology manufacturing 

and service sectors, while the difference in performance between foreign and 

domestic affiliates is insignificant for high- and medium-high tech industries. 

Furthermore, Mairesse and Mohnen (2004) report sizeable differences in the 

returns to R&D (both in the selection and intensity equation of stage two) 

between high-technology and low-technology sectors. To investigate to what 

extent results obtained in table 4.5 differ according to the technology intensity 

of the sector, tables 4.6 and 4.7 show the results of estimating (4.2) and 

(4.3) separately for (medium-)high technology and (medium-)low technology 

sectors respectively. 

Firms are classified into high- versus low-tech sectors based on the classification 

of industry according to technology from Eurostat. The different 

NACE (Rev. 1.1.) codes assigned to each category are listed in appendix 

table 4.A.1. Comparing the results of tables 4.6 and 4.7, several important 

differences are worth noting. 

I will begin by discussing the left-hand side of tables 4.6 and 4.7, i.e. the 

R&D intensity part of the model shown in (4.2). Exporters and domestic 

multinational firms are more likely to engage in continuous R&D spending, 

both in high- and low-technology sectors. However, the magnitude of the 

marginal effects (i.e. the change in the probability to engage in research 

effort that is associated with a change from 0 to 1 for the dummy variables) is 

much larger in the high- than in the low-technology sectors. Foreign affiliates 

of multinational firms are found to be more likely to engage in continuous 

research effort in high-tech sectors, while for low-tech sectors, the foreign 

ownership dummy has an insignificant marginal effect. 

Turning to the results on the R&D intensity equation in the first stage, 

both foreign and domestic multinational ownership are positively and signifi-




Exporter (d) 0.28*** 0.07 0.23*** 0.12 

[0.05] [0.35] [0.05] [0.16] 

Foreign MNE (d) 0.13** -0.09 0.15*** 0.23 

[0.05] [0.36] [0.05] [0.17] 

Home-based MNE (d) 0.34*** 0.26 0.20*** 0.13 

[0.08] [0.38] [0.08] [0.25] 


[0.01] [0.02] 

Cooperation - -0.45** - - 

[0.21] 

Funding - 0.77*** - - 

[0.21] 

log(R&D/sales)* (1) - - 0.10 0.32 

[0.06] [0.21] 



N 2,988 2,988 













Table 4.6: Innovation production function: High-technology sectors


cantly related to firm-level R&D intensity, conditional on selection. However, 

this result only holds in the low-technology sectors. In high-technology sectors, 

firms’ global engagement status is not significantly related to research 

intensity, conditional on selection. Furthermore, firm-level cooperation activities 

and funding opportunities are only found to be significantly related 

to the intensity of research effort in high-technology sectors, with opposite 

signs (negative and positive respectively). 

The right-hand sides of tables 4.6 and 4.7 list the results of the second 

estimation stage (4.3) separately for high-tech and low-tech sectors respectively. 

Again, there are some important differences between the sectors that 

are worth noting. While firms’ global engagement status (whether through 

exporting or FDI) is positively and significantly related to the probability 

that firms generate innovative sales in high-tech sectors, the opposite applies 

for low-technology sectors. These results suggest that exporters and 

multinational firms are more (less) likely to innovate in high- (low-) tech 

sectors. Also striking is the result on the R&D intensity variable, which is 

insignificant for high-technology sectors and highly significant for low-tech 

activities. 

Finally, firms’ global engagement status is not significantly related to the 

innovative sales intensity for the high-tech sectors (table 4.6), while multinational 

ownership is significantly negatively related to this intensity in the 

low-tech sectors. The R&D elasticity (i.e. the marginal effect obtained for 

the predicted R&D intensity from the first stage) is insignificant in both 

sectors. 

These results suggest that globally engaged firms (whether through exporting 

or FDI) active in high-technology sectors are able to benefit from 

their access to a global network, allowing them to be more innovative, even 

when taking into account that these firms tend to spend more (on average) 

on R&D compared to national firms. The fact that globally integrated firms 

are not found to be more innovative in low-tech sectors, conditional on their 

R&D spending patterns might suggest that, within the low-technology sectors, 

firms’ international activities serve as a means to achieve economies of




Exporter (d) 0.07*** 0.15 -0.051* 0.02 

[0.02] [0.31] [0.027] [0.15] 

Foreign MNE (d) 0.04 0.52* -0.259*** -0.47** 

[0.02] [0.31] [0.026] [0.23] 

Home-based MNE (d) 0.11** 0.83** -0.216*** -0.63 

[0.05] [0.35] [0.012] [0.41] 


[0.01] [0.021] 

Cooperation - 0.17 - - 

[0.22] 

Funding - 0.36 - - 

[0.23] 

log(R&D/sales)* (1) - - 1.02*** 0.55 

[0.08] [0.36] 



N 2,988 2,988 













Table 4.7: Innovation production function: Low-technology sectors


scale, rather than as a learning opportunity allowing the firm to continuously 

improve its existing product lines and expand into new ones. However, in 

the absence of any dynamic information on firms’ global engagement and 

innovation activities, it might equally be argued that results in the high-tech 

sectors are driven by a learning effect (access to the global market as a driver 

for innovation), while results in the low-tech sectors can be explained by selection 

effects (innovation as a driver of the firms’ international expansion). 

Hence, these conclusions should be treated with caution. 


This paper has investigated whether firms’ innovative output differs according 

to their global engagement status. Specifically, by relying on the innovation 

production function framework introduced by Mairesse and Mohnen 

(2002) and by taking the endogeneity of research inputs specifically into account 

in the innovation output function, following the model of Crépon et 

al. (1998), I am able to distinguish between the indirect (through higher research 

spending) and direct effect of firms’ global integration on its innovative 

output. 

Unconditionally, exporters and multinational firms in Belgium are found to 

be significantly more likely to generate innovative output than their domestic 

counterparts, pointing to the existence of an innovation premium, not unlike 

the productivity premium associated with firms’ international activities. In 

terms of innovative sales intensity, no significant differences emerge between 

the different types of firms. 

However, taking into account that internationally engaged firms typically 

spend more on R&D than national firms (i.e. conditional on research spending 

patterns), only exporters are found to be more likely to generate innovative 

sales, while home-based multinationals are significantly less likely to innovate. 

Distinguishing between high-technology and low-technology sectors 

further shows that there is an innovation premium associated with firms’ international 

activities in high-technology sectors, conditional on their research 

intensity; while in low-technology sectors, a negative innovation premium


emerges. 

From a policy perspective, the analysis yields several important insights. 

Overall, the empirical analysis has shown that (part of) the innovation premium 

associated with firms’ global engagement status in Belgium can be 

traced back to these firms’ different spending patterns, i.e. their higher likelihood 

to engage in continuous R&D spending. However, within the hightechnology 

sectors, firms with exposure to international markets (whether 

through exporting or FDI) are found to be more likely to generate innovative 

sales, conditional on their R&D spending patterns; while in low-technology 

sectors, these firms are found to be less likely to be innovative. This result 

suggests that government policies aimed at the attraction of (foreign) multinational 

firms in order to stimulate innovative output (or inputs) are more 

likely to be successful (i.e. will generate a higher rate of return) when they 

are specifically targeted at high-technology sectors. 

For the low-technology sectors, the high returns to R&D obtained in these 

sectors, especially for the selection stage of the model, suggests that policies 

targeted towards these sectors might be more successful when aimed at stimulating 

research efforts in general, indiscriminate of the global engagement 

status of the firm. However, as already noted above, these conclusions should 

be treated with caution, since, in the absence of any dynamic information 

on the timing of firms’ innovation and internationalization decisions, I am 

not able to provide any insights into the direction of causality between firms’ 

international exposure and its innovation activities. 

From a methodological point of view, the analysis clearly points to the 

importance of taking exporters’ and multinational firms’ different research 

spending patterns into account when investigating their impact on (correlation 

with) innovative output. Results clearly show that failure to take these 

firms’ higher average innovative effort into account, might lead to misleading 

results concerning the innovation premium associated with firms’ international 

activities. 

Finally, as highlighted in section 4.2, drawing the analogy between the 

innovation production function and the traditional production function liter-


ature further, offers numerous challenges to researchers. While the framework 

of Mairesse and Mohnen (2002), Mairesse and Mohnen (2004) and Crépon et 

al. (1998) serves as a useful starting point, many issues remain unresolved. 

This is related to the fact that while many of the existing techniques applied 

in the production function literature (eg. GMM or semiparametric 

estimation) can easily be carried over to the innovation production function, 

choices of estimation techniques are often constrained by the lack of dynamics 

in many firm-level innovation data sets.



4.A.1 Sample selection 

The population of CIS4-firms was selected on the basis of the full population 

of Belgian firms, registered at the National Office for Social Security 

at the end of 2004 (Teirlinck, 2005b). Of these, all firms with at least ten 

employees were selected. After applying a number of corrections 18 , a population 

of about 25,000 firms was retained. Sampling was performed on the 

population after stratifying according to sector (NACE two-digit, three-digit 

in some cases), size (three size classes) and region (either at the two-digit 

Nuts or provincial level). The final questionnaire was sent to 1,814 firms in 

Brussels; 2,118 in the Walloon region and 3,000 in Flanders. The full sample 

of CIS4-firms consists of 3,322 firms. 

Firms with missing identification number are omitted (one firm), as well 

as firms with exports amounting to more than 100 percent of total sales in 

2004 (one firm). Similarly, firms reporting unrealistically high R&D to sales 

ratios are omitted (14 firms). All firms for which no matching information 

could be obtained from Belfirst are similarly omitted (100 firms). Finally, 

218 firms were subject to a merger, acquisition or absorption by another 

company between 2002 and 2005. These firms are not taken into account 

in the empirical analysis. The final sample of firms used in the empirical 

analysis consists of 2,988 firms and accounts for about 65 percent of turnover 

and employment generation in the full CIS sample. 

4.A.2 Definitions of variables 

Innovation input and output measures 

Innovative effort 

Intramural R&D over sales ratio. Specific question from CIS-survey: “Please 

estimate the amount of expenditure on intramural R&D (including personnel 

and related costs as well as capital expenditures on buildings and equipment 

18 Firms that were no longer active were omitted, as well as firms that appeared with 

a double identification number. For a detailed overview of the population selection and 

sampling process, I refer to Teirlinck (2005b).


specifically for R&D) in 2004 only.”. 

Share of new products in turnover 

Please give the percentage of your total turnover in 2004 from: goods and 

service innovations introduced during 2002-2004 that were new to your market, 

or only new to the firm . 

Control variables 

Employment 

Number of employees in 2004, full-time equivalents. 

Exporter dummy 

Equal to one if the firm has reported positive exports in 2004, but is not part 

of a multinational group, i.e. the firm is not part of a foreign- or home-based 

multinational firm. 

Foreign ownership dummy 

Equal to one if the firm is part of a Belgian group with foreign headquarters. 

Domestic multinational ownership dummy 

Equal to one if the firm is part of a Belgian group and has foreign subsidiaries. 

Cooperation 

Dummy equal to one if the firm has engaged in cooperation activities between 

2002 and 2004. 

Funding 

Dummy equal to one if the firm has obtained funding from regional, national 

or EU authorities between 2002 and 2004.


High-technology Low-technology 

24 Chemicals 15 Food and beverages 

29 Machinery and equipment 16 Tobacco 

30 Office machinery (computers) 17 Textiles 

31 Electrical machinery 18 Clothing 

32 Radio, TV, communication 19 Leather (products) 

33 Medical and optical instruments 20 Wood (products) 

34 Motor vehicles 21 Pulp and paper (products) 

35 Other transport equipment (excl. 351) 22 Publishing and printing 

61 Water transport 23 Petroleum products 

62 Air transport 25 Rubber and plastics 

64 Post and telecommunications 26 Nonmetallic mineral products 

65 Financial intermediation 27 Basic metals 

66 Insurance and pension funding 28 Fabricated metals 

67 Ancilliary financial activities 351 Shipbuilding and repairs 

70 Real estate activities 36 Furniture 

71 Renting activities 37 Recycling 

72 Computer and related activities 50 Motor vehicles trade 

73 Research and development 51 Wholesale trade 

74 Other business activities 52 Retail trade 

55 Hotels and restaurants 

60 Land transport 

63 Supporting transport activities 

Table 4.A.1: Sector classification: High-tech versus low-tech

Bibliography 

Abraham, Filip, Jozef Konings, and Stijn Vanormelingen, “The 

effect of globalization on union bargaining and price-cost margins of 

firms,” Licos Discussion Paper Series, 2007b, 181/2007. 141 

, , and Veerle Slootmaekers, “FDI spillovers in the Chinese 

manufacturing sector: Evidence of firm heterogeneity,” CEPR Discussion 

Paper, 2007a, 6573. 41 

Ackerberg, Daniel A., Kevin Caves, and Garth Frazer, “Structural 

identification of production functions,” Unpublished manuscript., 2006. 

20, 28, 29, 35 

Ackerberg, Daniel, C. Lanier Benkard, Steven Berry, and Ariel 

Pakes, “Econometric tools for analyzing market outcomes,” in James 

Heckman and Edward Leamer, eds., Handbook of Econometrics, Vol. 

6(1), Amsterdam: North-Holland, 2007, pp. 4171–4276. 9, 12, 14, 15, 

20, 21, 22, 24, 30, 35 

Amiti, Mary and Jozef Konings, “Trade liberalization, intermediate inputs, 

and productivity: Evidence from Indonesia,” American Economic 

Review, 2007, 97 (5), 1611–1638. 9 

Arnold, Jens Matthias, “Productivity estimation at the plant level: A 

practical guide,” Unpublished manuscript., 2005. 37 

Asplund, Marcus and Volker Nocke, “Firm turnover in imperfectly competitive 

markets,” Review of Economic Studies, 2006, 73 (2), 295–327. 

68 

Audretsch, David B. and Talat Mahmood, “New firm survival: New results 

using a hazard function,” The Review of Economics and Statistics, 

1995, 77 (1), 97–103. 66, 115 

163

164 Bibliography 

Aw, Bee Yan, Mark J. Roberts, and Tor Winston, “Export market 

participation, investments in R&D and worker training, and the evolution 

of firm productivity,” The World Economy, 2007, 30 (1), 83–104. 

132 

Bartelsman, Eric J. and Mark Doms, “Understanding productivity: 

Lessons from longitudinal microdata,” Journal of Economic Literature, 

2000, 38 (3), 569–594. 9, 56 

Basile, Roberto, “Export behaviour of Italian manufacturing firms over the 

nineties: the role of innovation,” Research Policy, 2001, 30 (8), 1185– 

1201. 132 

Belgostat, Materials and Investment Price Indices., National Bank of Belgium, 

2007. 36 

Benavente, Jose Miguel, “The role of research and innovation in promoting 

productivity in chile,” Economics of Innovation and New Technology, 

2006, 15 (4), 301 – 315. 136 

Bernard, Andrew B. and Fredrik Sjöholm, “Foreign owners and plant 

survival,” National Bureau of Economic Research Working Paper Series, 

2003, 10039. 67, 102, 105, 111, 114, 119, 120, 121 

and J. Bradford Jensen, “Exporters, jobs and wages in US 

manufacturing: 1976-1987,” Brookings Papers on Economic Activity - 

Microeconomics, 1995, pp. 67–119. 2, 55 

and , “Firm structure, multinationals, and manufacturing plant 

deaths,” Review of Economics and Statistics, 2007a, 89 (2), 193–204. 

56, 102, 105, 111, 119, 121, 123 

, , and Peter K. Schott, “Survival of the best fit: Exposure 

to low-wage countries and the (uneven) growth of U.S. manufacturing 

plants,” Journal of International Economics, 2006a, 68 (1), 219–237. 3, 

53, 55, 56, 58, 60, 65, 69, 76, 84 

, , Stephen J. Redding, and Peter K. Schott, “Firms in international 

trade,” Journal of Economic Perspectives, 2007c, 2 (3), 105– 

130. 1, 2, 55 

, Jonathan Eaton, J. Bradford Jensen, and Samuel Kortum, 

“Plants and productivity in international trade,” American Economic 

Review, 2003, 93 (4), 1268–1290. 55


, Stephen J. Redding, and Peter K. Schott, “Products and productivity,” 

Unpublished manuscript, 2005. 3, 10, 18, 19, 20, 30, 51 

, , and , “Multi-product firms and product switching,” Unpublished 

manuscript, 2006b. 3, 18, 20, 31, 51 

, , and , “Comparative advantage and heterogeneous Firms,” 

Review of Economic Studies, 2007b, 74 (1), 31–66. 55 

Bevan, Alan A. and Saul Estrin, “The determinants of foreign direct investment 

into European transition economies,” Journal of Comparative 

Economics, 2004, 32 (4), 775–787. 73 

Blundell, Richard and Steve Bond, “GMM estimation with persistent 

panel data: An application to production functions,” IFS Working Paper, 

1999, W99/4. 22, 23, 34, 35 

BvDEP, Belfirst Database, DVD-rom. Bureau van Dijk Electronic Publishing., 

2004. 103, 107, 125 

, Belfirst Database, DVD-rom. Bureau van Dijk Electronic Publishing, 

2006. 34, 141 

Carstensen, Kai and Farid Toubal, “Foreign direct investment in Central 

and Eastern European countries: a dynamic panel analysis,” Journal of 

Comparative Economics, 2004, 32 (1), 3–22. 73 

Cassiman, Bruno and Elena Golovko, “Innovation and the exportproductivity 

link,” IESE Working Paper Series, 2007, 668. 132, 134 

and Pedro Mendi, “Multinationals as channels for international 

technology diffusion: Empirical evidence from Spain,” Unpublished 

manuscript, 2008. 152 

and Reinhilde Veugelers, “In search of complementarity in innovation 

strategy: Internal R&D and external knowledge acquisition,” Management 

Science, 2006, 52 (1), 68–82. 134 

Castellani, Davide and Antonello Zanfei, Multinational firms, Innovation 

and Productivity, Cheltenham: Edward Elgar Publishing, 2006. 

129, 139, 140, 143, 146, 149 

and , “Internationalisation, innovation and productivity: How do 

firms differ in Italy?,” The World Economy, 2007, 30 (1), 156–176. 5, 

129


Caves, Richard E., “Industrial organization and new findings on the 

turnover and mobility of firms,” Journal of Economic Literature, 1998, 

36 (4), 1947–1982. 64, 73, 103, 116 

Chen, M. Y., “Survival duration of plants: Evidence from the US petroleum 

refining industry,” International Journal of Industrial Organization, 

2002, 20 (4), 517–555. 114 

Cincera, Michele, Bruno van Pottelsberghe de la Potterie, and 

Veugelers Reinhilde, “Assessing the foreign control of production of 

technology: The case of a small open economy,” Scientometrics, 2006, 

66 (3), 493–512. 130 

Colantone, Italo and Leo Sleuwaegen, “Entry and exit of firms in a 

global economy: A cross-country and industry analysis,” Vlerick Leuven 

Working Paper Series, 2007, 2007/36. 69 

Costantini, James A. and Marc Melitz, “The dynamics of firm-level 

adjustment to trade liberalization,” Unpublished manuscript, 2007. 2, 

132 

Coucke, Kristien and Leo Sleuwaegen, “Exit in globalising industries: 

The role of international outsourcing,” Vlerick Leuven Working Paper 

Series, 2006, 2006/14. 69 

Cox, D. R., “Regression models and life-tables,” Journal of the Royal Statistical 

Society - Series B (Methodological), 1972, 34 (2), 187–220. 114 

CRB, Verslag over de economische ontwikkeling in de voedingsindustrie in 

2003, Centrale Raad voor het Bedrijfsleven, 2004. 36 

Crépon, Bruno, Emmanuel Duguet, and Jacques Mairesse, “Research, 

innovation and productivity: An econometric analysis at the 

firm level,” Economics of Innovation & New Technology, 1998, 7 (2), 

115. 130, 136, 137, 140, 156, 158 

Criscuolo, Chiara, Jonathan E. Haskel, and Matthew J. Slaughter, 

“Global engagement and the innovation activities of firms,” NBER 

Working Paper Series, 2005, 11479. 129, 130, 135, 139, 140, 143, 146, 

149 

Damijan, Joˇze P., Črt Kostevc, and Saˇso Polanec, “From innovation 

to exporting or vice versa? Causal link between innovation activity and 

exporting in Slovenian microdata,” Licos Discussion Paper Series, 2008, 

204/2008. 132


De Loecker, Jan, “Product differentiation, multi-product firms and estimating 

the impact of trade liberalization on productivity,” National 

Bureau of Economic Research Working Paper Series, 2007, 13155. 9, 

10, 15, 16, 17, 18, 19, 20, 30, 33, 34, 35, 41, 44, 107, 134 

and Jozef Konings, “Job reallocation and productivity growth in a 

post-socialist economy: Evidence from Slovenian manufacturing,” European 

Journal of Political Economy, 2006, 22 (2), 388–408. 13, 44, 

46 

Disney, Richard, Jonathan Haskel, and Ylva Heden, “Entry, exit and 

establishment survival in UK manufacturing,” The Journal of Industrial 

Economics, 2003, 51 (1), 91–112. 114, 121 

Dunne, Timothy and Mark J. Roberts, “Costs, demand and imperfect 

competition as determinants of plant level output prices,” CES Discussion 

Paper, 1992, 92-5. 17 

, , and Larry Samuelson, “Patterns of firm entry and exit in US 

manufacturing industries,” Rand Journal of Economics, 1988, 19 (4), 

495–515. 14, 115 

, , and , “The growth and failure of US manufacturing plants,” 

Quarterly Journal of Economics, 1989, 104 (4), 671–698. 115 

, Shawn D. Klimek, and Mark J. Roberts, “Exit from regional 

manufacturing markets: The role of entrant experience,” International 

Journal of Industrial Organization, 2005, 23 (5-6), 399–421. 64 

Ericson, Richard and Ariel Pakes, “Markov-perfect industry dynamics: 

A framework for empirical work,” The Review of Economic Studies, 

1995, 62 (1), 53–82. 23, 64, 115 

Eslava, M., J. Haltiwanger, A. Kugler, and M. Kugler, “The effects 

of structural reforms on productivity and profitability enhancing reallocation: 

evidence from Colombia,” Journal of Development Economics, 

2004, 75 (2), 333–371. 17, 18 

EU, The EU economy 2005 review: Rising international economic integration, 

opportunities and challenges, Brussels: European Commission, 

2005. 1 

Eurostat, Producer price indices, 3-digit level, Belgium: Nace 15, 2007. 34


Fabrizio, Stefania, Deniz Igan, and Ashoka Mody, “The dynamics of 

product quality and international competitiveness,” IMF Working Paper 

Series, 2007, 07/97. 54, 82 

Falvey, R. and H. Kierzkowski, “Product quality, intra-industry trade 

and (im)perfect competition,” in H. Kierzkowski, ed., Protection and 

competition in international trade, Oxford: Basil Blackwell, 1987. 70 

Fariñas, Jose C. and Sonia Ruano, “Firm productivity, heterogeneity, 

sunk costs and market selection,” International Journal of Industrial 

Organization, 2005, 23 (7-8), 505–534. 14, 65, 116 

Feenstra, Robert C., The impact of international trade on wages, Chicago: 

The University of Chicago Press, 2000. 2 

Flam, Harry and Elhanan Helpman, “Vertical product differentiation 

and North-South trade,” American Economic Review, 1987, 77 (5), 810– 

822. 70 

Flamm, Kenneth, “The volatility of offshore investment,” Journal of Development 

Economics, 1984, 16 (3), 231–248. 103 

Foster, Lucia, John Haltiwanger, and C. J. Krizan, “Market selection, 

reallocation, and restructuring in the US retail trade sector in the 1990s,” 

Review of Economics and Statistics, 2006, 88 (4), 748–758. 46, 56 

, , and Chad Syverson, “Reallocation, firm turnover, and efficiency: 

Selection on productivity or profitability?,” American Economic 

Review, 2008, 98 (1), 394–425. 17 

Frenz, Marion and Grazia Ietto-Gillies, “Does multinationality affect 

the propensity to innovate? An analysis of the Third UK Community 

Innovation Survey,” International Review of Applied Economics, 2007, 

21 (1), 99 – 117. 129, 135, 140, 143, 146, 149 

Gibson, J. K. and R. I. D. Harris, “Trade liberalisation and plant exit 

in New Zealand manufacturing,” Review of Economics and Statistics, 

1996, 78 (3), 521–529. 105 

Goldberg, Pinelopi K., Amit Khandelwal, Nina Pavcnik, and Petia 

Topalova, “Multi-product firms and product turnover in the developing 

world: Evidence from India,” National Bureau of Economic Research 

Working Paper Series, 2008, 14127. 56


Görg, Holger and Eric Strobl, “Footloose multinationals?,” The Manchester 

School, 2003, 71 (1), 1–19. 67, 68, 102, 105, 111, 114, 116, 119, 

120 

Greenaway, David, “The assessment: Firm-level adjustment to globalization,” 

Oxford Review of Economic Policy, 2004, 20 (3), 335–342. 2, 

4 

and Richard Kneller, “Firm heterogeneity, exporting and foreign 

direct investment,” The Economic Journal, 2007, 117 (517), F134–F161. 

5, 129 

, Joakim Gullstrand, and Richard Kneller, “Surviving globalisation,” 

Journal of International Economics, 2008, 74 (2), 264–277. 3, 53, 

55, 56, 58, 60, 65, 69, 84 

Greene, William H., Econometric Analysis, fifth ed., New Jersey: Prentice 

Hall, 2003. 110 

, Econometric Analysis, sixth ed., New Jersey: Prentice Hall, 2008. 63, 

137 

Griffith, Rachel, Elena Huergo, Jacques Mairesse, and Bettina Peters, 

“Innovation and productivity across four European countries,” Oxford 

Review of Economic Policy, 2006, 22 (4), 483–498. 134, 136 

Hall, Bronwyn H. and Jacques Mairesse, “Empirical studies of innovation 

in the knowledge-driven economy,” Economics of Innovation and 

New Technology, 2006, 15 (4), 289 – 299. 134 

Hallak, Juan C. and Peter K. Schott, “Estimating cross-country differences 

in product quality,” National Bureau of Economic Research Working 

Paper Series, 2008, 13807. 70 

Hallak, Juan Carlos, “Product quality and the direction of trade,” Journal 

of International Economics, 2006, 68 (1), 238–265. 70 

Helpman, Elhanan, Marc J. Melitz, and Stephen R. Yeaple, “Export 

versus FDI with heterogenous firms,” American Economic Review, 2004, 

94 (1), 300–316. 5, 55, 129 

Hoch, Irving, “Estimation of production function parameters combining 

time series and cross-section data,” Econometrica, 1962, 30 (1), 34–53. 

21, 35


Hopenhayn, Hugo A., “Entry, exit and firm dynamics in long run equilibrium,” 

Econometrica, 1992, 60 (5), 1127–1150. 14, 64, 67, 115, 116 

Hummels, David and Peter J. Klenow, “The variety and quality of a 

nation’s exports,” American Economic Review, 2005, 95 (3), 704–723. 

70 

Jaumandreu, Jordi and Jacques Mairesse, “Using price and demand 

information to identify production functions,” MPRA Discussion Paper, 

2004, 1247. 17 

Jovanovic, Boyan, “Selection and the evolution of industry,” Econometrica, 

1982, 50 (3), 649–670. 14, 64, 116 

Kandogan, Yener, “Does product Differentiation Explain the Increase in 

Exports of Transition Countries?,” Eastern European Economics, 2006, 

44 (2), 6–22. 54 

Katayama, Hajime, Shihua Lu, and James R. Tybout, “Firm-level 

productivity studies: Illusions and a solution,” Unpublished manuscript, 

2005. 9, 10, 18, 20, 30 

Kiefer, Nicholas M., “Economic duration data and hazard functions,” 

Journal of Economic Literature, 1988, 26 (2), 646. 114 

Klette, Tor Jakob and Zvi Griliches, “The inconsistency of common 

scale estimators when output prices are unobserved and endogenous,” 

Journal of Applied Econometrics, 1996, 11 (4), 343–361. 17, 20, 30, 35, 

41 

Konings, Jozef, “The effects of foreign direct investment on domestic firms: 

Evidence from firm-level panel data in emerging economies,” Economics 

of Transition, 2001, 9 (3), 619. 41 

, “Wage costs and industry (re)location in the enlarged European 

Union,” Swedish Economic Policy Review, 2005, 12, 57–81. 66, 113, 

115 

, “The impact of training on productivity and wages: Evidence from 

Belgian firm level panel data,” Licos Discussion Paper Series, 2008, 

(197/2008). 9, 141 

and Alan P. Murphy, “Do multinational enterprises relocate employment 

to low wage regions? Evidence from European multinationals,” 

Review of World Economics, 2006, 142 (2), 267–286. 104


Levinsohn, J. and A. Petrin, “Estimating production functions using 

inputs to control for unobservables,” Review of Economic Studies, 2003, 

70 (2), 317–341. 10, 15, 16, 20, 21, 26, 27, 35, 65, 134, 136 

Levinsohn, Jim and Marc J. Melitz, “Productivity in a differentiated 

product markets equilibrium,” Unpublished manuscript, 2002. 17, 18, 

20, 35 

Lipsey, Robert E., “Home and host country effects of FDI,” NBER Working 

Paper Series, 2002, No. 9293. 101, 105 

Lööf, Hans and Almas Heshmati, “The impact of public funding on 

private R&D investment. New evidence from a firm level innovation 

study,” CESIS Working Paper Series, 2004, 06. 136 

Mairesse, Jacques and Jordi Jaumandreu, “Panel-data estimates of the 

production function and the revenue function: What difference does it 

make?,” Scandinavian Journal of Economics, 2005, 107 (4), 651–672. 17 

and Pierre Mohnen, “Accounting for innovation and measuring innovativeness: 

An illustrative framework and an application,” American 

Economic Review, 2002, 92 (2), 226–230. 5, 130, 133, 156, 158 

and , “The importance of R&D for innovation: A reassessment 

using French survey data,” The Journal of Technology Transfer, 2004, 

30 (1), 183–197. 129, 130, 133, 137, 139, 140, 143, 151, 152, 158 

Marschak, Jacob and Jr. Andrews William H., “Random simultaneous 

equations and the theory of production,” Econometrica, 1944, 12 (3/4), 

143–205. 10, 15, 20 

Masso, Jaan, Raul Eamets, and Kaia Philips, “Where have all the 

jobs gone? Gross job flows in Estonia,” University of Tartu Faculty of 

Economics and Business Administration Working Paper, 2004, 2004/28. 

57, 85 

Mata, José and Pedro Portugal, “Life duration of new firms,” The Journal 

of Industrial Economics, 1994, 42 (3), 227–245. 36, 107, 114, 116, 

125 

, , and Paulo Guimarães, “The survival of new plants: Start-up 

conditions and post-entry evolution,” International Journal of Industrial 

Organization, 1995, 13 (4), 459–481. 36, 107, 114


Melitz, Marc J., “The impact of trade on intra-industry reallocations and 

aggregate industry productivity,” Econometrica, 2003, 71 (6), 1695– 

1725. 2, 5, 55, 69, 129 

and Gianmarco I. P. Ottaviano, “Market size, trade, and productivity,” 

Review of Economic Studies, 2008, 75 (1), 295–316. 2, 55, 69 

Monfort, Ph., Hylke Vandenbussche, and E. Forlani, “Chinese competition 

and skill-upgrading in European textiles: Firm-level evidence,” 

Licos Discussion Paper Series, 2008, 198/2008. 141 

Moreno Badia, Marialuz and Veerle Slootmaekers, “Productivity 

growth and financial constraints in Estonia,” Licos Discussion Paper 

Series, 2008, 208/2008. 66 

Mueller, D.C., “Entry, exit and the competitive process,” in P.A. Geroski 

and J. Schwalbach, eds., Entry and Market Contestability: An International 

Comparison, Cambridge, Massachussets: Blackwell Publishing, 

1991. 116 

Mundlak, Yair, “Empirical production function free of management bias,” 

Journal of Farm Economics, 1961, 43 (1), 44–56. 21, 35 

Murphy, Kevin M. and Andrei Shleifer, “Quality and trade,” Journal 

of Development Economics, 1997, 53 (1), 1–15. 70 

Olley, Steven G. and Ariel Pakes, “The dynamics of productivity in the 

telecommunications equipment industry,” Econometrica, 1996, 64 (6), 

1263–1297. 10, 13, 14, 15, 16, 20, 22, 23, 25, 26, 35, 41, 46, 66, 134 

Ornaghi, Carmine, “Assessing the effects of measurement errors on the 

estimation of production functions,” Journal of Applied Econometrics, 

2006, 21 (6), 879–891. 17, 18, 32 

Pavcnik, Nina, “Trade liberalization, exit, and productivity improvements: 

Evidence from Chilean plants,” Review of Economic Studies, 2002, 69 

(1), 245–276. 9, 21 

Pennings, Enrico and Leo Sleuwaegen, “International relocation: firm 

and industry determinants,” Economics Letters, 2000, 67 (2), 179–186. 

106 

and , “The reorganization decisions of troubled firms: Exit, downscale 

or relocate,” Vlerick Leuven Working Paper Series, 2002, 2002/21. 

106


Petrin, Amil, Jim Levinsohn, and Brian P. Poi, “Production function 

estimation in Stata using inputs to control for unobservables,” Stata 

Journal, 2003, 4 (2), 113–123. 27, 35, 40 

Rodrik, D., Has Globalization Gone Too Far?, Washington DC: Institute 

for International Economics, 1997. 104 

Roodman, David, “How to do xtabond2: An introduction to difference 

and system GMM in Stata,” Center for Global Development Discussion 

Paper, 2006, 103. 37 

Schott, Peter K., “Across-product versus within-product specialization in 

international trade,” Quarterly Journal of Economics, 2004, 119 (2), 

647–678. 70 

Smarzynska Javorcik, Beata, “Does foreign direct investment increase the 

productivity of domestic firms? In search of spillovers through backward 

linkages,” American Economic Review, 2004, 94 (3), 605–627. 9 

Smith, Keith, “Measuring innovation,” in Jan Fagerberg, David C. Mowery, 

and Richard R. Nelson, eds., The Oxford Handbook of Innovation, 

Oxford: Oxford University Press, 2005, pp. 148–177. 134 

Solow, Robert M., “Technical change and the aggregate production function,” 

The Review of Economics and Statistics, 1957, 39 (3), 312–320. 

9 

Stokey, Nancy L., “The volume and composition of trade between rich and 

poor countries,” The Review of Economic Studies, 1991, 58 (1), 63–80. 

70 

Teirlinck, Peter, “Location and agglomeration of foreign R&D activities in 

a small open economy,” in Peter Teirlinck and André Spithoven, eds., 

Beyond Borders: Internationalisation of R&D and Policy Implications 

for Small Open Economies, Amsterdam: Elsevier, 2005a, pp. 207–234. 

131, 146 

, “Report CIS 4,” Technical Report 2005b. 135, 159 

The Economist, “Belgium sinks in deeper,” 1999, June 10th, 1999. 11, 36 

, “Europe’s nascent merger boom,” 2005, September 1st, 2005. 102 

, “Cross-border mergers: Heavy Mittal,” 2006, February 2nd, 2006. 102


Tybout, James R., “Plant and firm-level evidence on new trade theories,” 

in K Choi and James Harrigan, eds., Handbook of International Trade, 

Oxford: Blackwell Publishing, 2003. 55 

UNCTAD, World Investment Report 2004: The Shift towards Services, New 

York: United Nations Conference on Trade and Development, 2004. 1 

Van Beveren, Ilke, “Footloose multinationals in Belgium?,” Review of 

World Economics, 2007a, 143 (3), 483–507. 36, 67 

, “Total factor productivity estimation: A practical review,” Licos Discussion 

Paper Series, 2007b, 182/2007. 36, 133 

Van Biesebroeck, Johannes, “Robustness of productivity estimates,” 

Journal of Industrial Economics, 2007, 55 (3), 529–569. 11, 51 

Van den Cruyce, Bart, “The impact of relocation, size, age, capital intensity 

and multinational group membership,” Federaal Planbureau Working 

Paper, 1999, 8-99. 106 

Van Leeuwen, George and Luuk Klomp, “On the contribution of innovation 

to multi-factor productivity growth,” Economics of Innovation 

and New Technology, 2006, 15 (4), 367 – 390. 137 

Vandenbussche, Hylke and Chang Tan, “The taxation of multinationals: 

Firm level evidence for Belgium,” Licos Discussion Paper Series, 

2005, 160/2005. 102, 104 

, Karen Crabbé, and Boudewijn Janssen, “Is there regional tax 

competition? Firm level evidence for Belgium,” De Economist, 2005, 

153 (3), 257–276. 107 

Verbeek, Marno, A Guide to Modern Econometrics, West Sussex: John 

Wiley, 2000. 137 

VRWB, De voedingsindustrie in Vlaanderen: Op weg naar 2010, Vlaamse 

Raad voor het Wetenschapsbeleid, 2003. 11 

Wagner, Joachim, “Exports and productivity: A survey of the evidence 

from firm-level data,” The World Economy, 2007, 30 (1), 60–82. 5, 129 

Wakelin, Katharine, “Innovation and export behaviour at the firm level,” 

Research Policy, 1998, 26 (7-8), 829–841. 132


Wedervang, F., Development of a Population of Industrial Firms, Oslo: 

Scandinavian University Books, 1965. 10, 15, 20 

Wooldridge, Jeffrey M., “On estimating firm-level production functions 

using proxy variables to control for unobservables,” Unpublished 

manuscript., 2005. 16 

WTO, WTO Annual Report 2005, Geneva: World Trade Organization, 2005. 

1 

Yasar, Mahmut, Rafal Raciborski, and Brian P. Poi, “Production 

function estimation in Stata using the Olley and Pakes method,” Stata 

Journal, 2008, 8 (2), 221–231. 37

Doctoral Dissertations Faculty 

of Business and Economics 

from August 1, 1971 

1. GEPTS, Stefaan. Stability and efficiency of resource allocation processes 

in discrete commodity spaces. Leuven, KUL, 1971. 86 pp. 

2. PEETERS, Theo. Determinanten van de internationale handel in fabrikaten. 

Leuven, Acco, 1971. 290 pp. 

3. VAN LOOY, Wim. Personeelsopleiding : een onderzoek naar investeringsaspekten 

van opleiding. Hasselt, Vereniging voor wetenschappelijk 

onderzoek in Limburg, 1971. VII, 238 pp. 

4. THARAKAN, Mathew. Indian exports to the European community 

: problems and prospects. Leuven, Faculty of economics and applied 

economics, 1972. X,343 pp. 

5. HERROELEN, Willy. Heuristische programmatie : methodologische 

benadering en praktische toepassing op complexe combinatorische problemen. 

Leuven, Aurelia scientifica, 1972. X, 367 pp. 

6. VANDENBULCKE, Jacques. De studie en de evaluatie van dataorganisatiemethodes 

en data-zoekmethodes. Leuven, s.n., 1973. 3 V. 

7. PENNYCUICK, Roy A. The economics of the ecological syndrome. 

Leuven, Acco, 1973. XII, 177 pp. 

8. KAWATA, T. Bualum. Formation du capital d’origine belge, dette 

publique et stratégie du développement au Zaire. Leuven, KUL, 1973. 

V, 342 pp. 

9. DONCKELS, Rik. Doelmatige oriëntering van de sectorale subsidiepolitiek 

in België : een theoretisch onderzoek met empirische toetsing. 

Leuven, K.U.Leuven, 1974. VII, 156 pp. 

177

178 Doctoral dissertations 

10. VERHELST, Maurice. Contribution to the analysis of organizational 

information systems and their financial benefits. Leuven, K.U.Leuven, 

1974. 2 V. 

11. CLEMEUR, Hugo. Enkele verzekeringstechnische vraagstukken in het 

licht van de nutstheorie. Leuven, Aurelia scientifica, 1974. 193 pp. 

12. HEYVAERT, Edward. De ontwikkeling van de moderne bank- en 

krediettechniek tijdens de zestiende en zeven-tiende eeuw in Europa 

en te Amsterdam in het bijzonder. Leuven, K.U.Leuven, 1975. 186 pp. 

13. VERTONGHEN, Robert. Investeringscriteria voor publieke investeringen 

: het uitwerken van een operationele theorie met een toepassing op 

de verkeersinfrastructuur. Leuven, Acco, 1975. 254 pp. 

14. Niet toegekend. 

15. VANOVERBEKE, Lieven. Microeconomisch onderzoek van de sectoriële 

arbeidsmobiliteit. Leuven, Acco, 1975. 205 pp. 

16. DAEMS, Herman. The holding company : essays on financial intermediation, 

concentration and capital market imperfections in the Belgian 

economy. Leuven, K.U.Leuven, 1975. XII, 268 pp. 

17. VAN ROMPUY, Eric. Groot-Brittannië en de Europese monetaire 

integratie : een onderzoek naar de gevolgen van de Britse toetreding 

op de geplande Europese monetaire unie. Leuven, Acco, 1975. XIII, 

222 pp. 

18. MOESEN, Wim. Het beheer van de staatsschuld en de termijnstructuur 

van de intrestvoeten met een toepassing voor België. Leuven, 

Vander, 1975. XVI, 250 pp. 

19. LAMBRECHT, Marc. Capacity constrained multi-facility dynamic lotsize 

problem. Leuven, KUL, 1976. 165 pp. 

20. RAYMAECKERS, Erik. De mens in de onderneming en de theorie van 

het producenten-gedrag : een bijdrage tot transdisciplinaire analyse. 

Leuven, Acco, 1976. XIII, 538 pp. 

21. TEJANO, Albert. Econometric and input-output models in development 

planning : the case of the Philippines. Leuven, KUL, 1976. XX, 

297 pp.

DOCTORAL DISSERTATIONS 179 

22. MARTENS, Bernard. Prijsbeleid en inflatie met een toepassing op 

België. Leuven, KUL, 1977. IV, 253 pp. 

23. VERHEIRSTRAETEN, Albert. Geld, krediet en intrest in de Belgische 

financiële sector. Leuven, Acco, 1977. XXII, 377 pp. 

24. GHEYSSENS, Lieven. International diversification through the government 

bond market : a risk-return analysis. Leuven, s.n., 1977. 188 

pp. 

25. LEFEBVRE, Chris. Boekhoudkundige verwerking en financiële verslaggeving 

van huurkooptransacties en ver-kopen op afbetaling bij ondernemingen 

die aan consumenten verkopen. Leuven, KUL, 1977. 228 

pp. 

26. KESENNE, Stefan. Tijdsallocatie en vrijetijdsbesteding : een econometrisch 

onderzoek. Leuven, s.n., 1978. 163 pp. 

27. VAN HERCK, Gustaaf. Aspecten van optimaal bedrijfsbeleid volgens 

het marktwaardecriterium : een risico-rendementsanalyse. Leuven, 

KUL, 1978. IV, 163 pp. 

28. VAN POECK, Andre. World price trends and price and wage development 

in Belgium : an investigation into the relevance of the Scandinavian 

model of inflation for Belgium. Leuven, s.n., 1979. XIV, 260 

pp. 

29. VOS, Herman. De industriële technologieverwerving in Brazilië : een 

analyse. Leuven, s.n., 1978. onregelmatig gepagineerd. 

30. DOMBRECHT, Michel. Financial markets, employment and prices in 

open economies. Leuven, KUL, 1979. 182 pp. 

31. DE PRIL, Nelson. Bijdrage tot de actuariële studie van het bonusmalussysteem. 

Brussel, OAB, 1979. 112 pp. 

32. CARRIN, Guy. Economic aspects of social security : a public economics 

approach. Leuven, KUL, 1979. Onregelm. gepag. 

33. REGIDOR, Baldomero. An empirical investigation of the distribution 

of stock-market prices and weak-form effi-ciency of the Brussels stock 

exchange. Leuven, KUL, 1979. 214 pp.


34. DE GROOT, Roger. Ongelijkheden voor stop loss premies gebaseerd 

op E.T. systemen in het kader van de veral-gemeende convexe analyse. 

Leuven, KUL, 1979. 155 pp. 

35. CEYSSENS, Martin. On the peak load problem in the presence of 

rationizing by waiting. Leuven, KUL, 1979. IX, 217 pp. 

36. ABDUL RAZK ABDUL. Mixed enterprise in Malaysia : the case study 

of joint venture between Malysian public corporations and foreign enterprises. 


37. DE BRUYNE, Guido. Coordination of economic policy : a gametheoretic 

approach. Leuven, KUL, 1980. 106 pp. 

38. KELLES, Gerard. Demand, supply, price change and trading volume 

on financial markets of the matching-order type. Vraag, aanbod, koersontwikkeling 

en omzet op financiële markten van het Europese type. 


39. VAN EECKHOUDT, Marc. De invloed van de looptijd, de coupon en 

de verwachte inflatie op het opbrengstverloop van vastrentende finaciële 

activa. Leuven, KUL, 1980. 294 pp. 

40. SERCU, Piet. Mean-variance asset pricing with deviations from purchasing 

power parity. Leuven, s.n., 1981. XIV, 273 pp. 

41. DEQUAE, Marie-Gemma. Inflatie, belastingsysteem en waarde van de 

onderneming. Leuven, KUL, 1981. 436 pp. 

42. BRENNAN, John. An empirical investigation of Belgian price regulation 

by prior notification : 1975 - 1979 - 1982. Leuven, KUL, 1982. 

XIII, 386 pp. 

43. COLLA, Annie. Een econometrische analyse van ziekenhuiszorgen. 


44. Niet toegekend. 

45. SCHOKKAERT, Eric. Modelling consumer preference formation. Leuven, 

KUL, 1982. VIII, 287 pp. 

46. DEGADT, Jan. Specificatie van een econometrisch model voor vervuilingsproblemen 

met proeven van toepassing op de waterverontreiniging 

in België. Leuven, s.n., 1982. 2 V.


47. LANJONG, Mohammad Nasir. A study of market efficiency and riskreturn 

relationships in the Malaysian capital market. s.l., s.n., 1983. 

XVI, 287 pp. 

48. PROOST, Stef. De allocatie van lokale publieke goederen in een economie 

met een centrale overheid en lokale overheden. Leuven, s.n., 1983. onregelmatig. 

gepagineerd. 

49. VAN HULLE, Cynthia (08/83). Shareholders’ unanimity and optimal 

corporate decision making in imperfect capital markets. s.l., s.n., 1983. 

147 pp. + appendix. 

50. VAN WOUWE, Martine (2/12/83). Ordening van risico’s met toepassing 

op de berekening van ultieme ruïnekansen. Leuven, s.n., 1983. 109 

pp. 

51. D’ALCANTARA, Gonzague (15/12/83). SERENA : a macroeconomic 

sectoral regional and national account econometric model for the Belgian 

economy. Leuven, KUL, 1983. 595 pp. 

52. D’HAVE, Piet (24/02/84). De vraag naar geld in België. Leuven, KUL, 

1984. XI, 318 pp. 

53. MAES, Ivo (16/03/84). The contribution of J.R. Hicks to macroeconomic 

and monetary theory. Leuven, KUL, 1984. V, 224 pp. 

54. SUBIANTO, Bambang (13/09/84). A study of the effects of specific 

taxes and subsidies on a firms’ R&D investment plan. s.l., s.n., 1984. 

V, 284 pp. 

55. SLEUWAEGEN, Leo (26/10/84). Location and investment decisions 

by multinational enterprises in Belgium and Europe. Leuven, KUL, 

1984. XII, 247 pp. 

56. GEYSKENS, Erik (27/03/85). Produktietheorie en dualiteit. Leuven, 

s.n., 1985. VII, 392 pp. 

57. COLE, Frank (26/06/85). Some algorithms for geometric programming. 


58. STANDAERT, Stan (26/09/86). A study in the economics of repressed 

consumption. Leuven, KUL, 1986. X, 380 pp.


59. DELBEKE, Jos (03/11/86). Trendperioden in de geldhoeveelheid van 

België 1877-1983 : een theoretische en empirische analyse van de ”Banking 

school” hypothese. Leuven, KUL, 1986. XII, 430 pp. 

60. VANTHIENEN, Jan (08/12/86). Automatiseringsaspecten van de specificatie, 

constructie en manipulatie van beslissings-tabellen. Leuven, 

s.n., 1986. XIV, 378 pp. 

61. LUYTEN, Robert (30/04/87). A systems-based approach for multiechelon 

production/inventory systems. s.l., s.n., 1987. 3V. 

62. MERCKEN, Roger (27/04/87). De invloed van de data base benadering 

op de interne controle. Leuven, s.n., 1987. XIII, 346 pp. 

63. VAN CAYSEELE, Patrick (20/05/87). Regulation and international 

innovative activities in the pharmaceutical industry. s.l., s.n., 1987. 

XI, 169 pp. 

64. FRANCOIS, Pierre (21/09/87). De empirische relevantie van de independence 

from irrelevant alternatives. Assumptie indis-crete keuzemodellen. 

Leuven, s.n., 1987. IX, 379 pp. 

65. DECOSTER, André (23/09/88). Family size, welfare and public policy. 

Leuven, KUL. Faculteit Economische en toegepaste economische 

wetenschappen, 1988. XIII, 444 pp. 

66. HEIJNEN, Bart (09/09/88). Risicowijziging onder invloed van vrijstellingen 

en herverzekeringen : een theoretische ana-lyse van optimaliteit 

en premiebepaling. Leuven, KUL. Faculteit Economische en 

toegepaste economische wetenschappen, 1988. onregelmatig gepagineerd. 

67. GEEROMS, Hans (14/10/88). Belastingvermijding. Theoretische analyse 

van de determinanten van de belastingontduiking en de belastingontwijking 

met empirische verificaties. Leuven, s.n., 1988. XIII, 409, 

5 pp. 

68. PUT, Ferdi (19/12/88). Introducing dynamic and temporal aspects in 

a conceptual (database) schema. Leuven, KUL. Faculteit Economische 

en toegepaste economische wetenschappen, 1988. XVIII, 415 pp. 

69. VAN ROMPUY, Guido (13/01/89). A supply-side approach to tax 

reform programs. Theory and empirical evidence for Belgium. Leuven, 

KUL. Faculteit Economische en toegepaste economische wetenschappen, 

1989. XVI, 189, 6 pp.


70. PEETERS, Ludo (19/06/89). Een ruimtelijk evenwichtsmodel van de 

graanmarkten in de E.G. : empirische specificatie en beleidstoepassingen. 

Leuven, K.U.Leuven. Faculteit Economische en toegepaste economische 

wetenschappen, 1989. XVI, 412 pp. 

71. PACOLET, Jozef (10/11/89). Marktstructuur en operationele efficiëntie 

in de Belgische financiële sector. Leuven, K.U.Leuven. Faculteit Economische 

en toegepaste economische wetenschappen, 1989. XXII, 547 pp. 

72. VANDEBROEK, Martina (13/12/89). Optimalisatie van verzekeringscontracten 

en premieberekeningsprincipes. Leuven, K.U.Leuven. Faculteit 

Economische en toegepaste economische wetenschappen, 1989. 95 pp. 

73. WILLEKENS, Francois. Determinance of government growth in industrialized 

countries with applications to Belgium. Leuven, K.U.Leuven. 

Faculteit Economische en toegepaste economische wetenschappen, 1990. 

VI, 332 pp. 

74. VEUGELERS, Reinhilde (02/04/90). Scope decisions of multinational 

enterprises. Leuven, K.U.Leuven. Faculteit Economische en toegepaste 

economische wetenschappen, 1990. V, 221 pp. 

75. KESTELOOT, Katrien (18/06/90). Essays on performance diagnosis 

and tacit cooperation in international oligopolies. Leuven, K.U.Leuven. 


227 pp. 

76. WU, Changqi (23/10/90). Strategic aspects of oligopolistic vertical 

integration. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 

economische wetenschappen, 1990. VIII, 222 pp. 

77. ZHANG, Zhaoyong (08/07/91). A disequilibrium model of China’s 

foreign trade behaviour. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 1991. XII, 256 pp. 

78. DHAENE, Jan (25/11/91). Verdelingsfuncties, benaderingen en foutengrenzen 

van stochastische grootheden geasso-cieerd aan verzekeringspolissen 

en -portefeuilles. Leuven, K.U.Leuven, Faculteit Economische en 

toegepaste economische wetenschappen, 1991. 146 pp. 

79. BAUWELINCKX, Thierry (07/01/92). Hierarchical credibility techniques. 

Leuven, K.U.Leuven, Faculteit Economische en toegepaste 

economische wetenschappen, 1992. 130 pp.


80. DEMEULEMEESTER, Erik (23/3/92). Optimal algorithms for various 

classes of multiple resource-constrained project scheduling problems. 

Leuven, K.U.Leuven, Faculteit Economische en toegepaste economische 

wetenschappen, 1992. 180 pp. 

81. STEENACKERS, Anna (1/10/92). Risk analysis with the classical 

actuarial risk model : theoretical extensions and applications to Reinsurance. 


economische wetenschappen, 1992. 139 pp. 

82. COCKX, Bart (24/09/92). The minimum income guarantee. Some 

views from a dynamic perspective. Leuven, K.U.Leuven, Faculteit 

Economische en toegepaste economische wetenschappen, 1992. XVII, 

401 pp. 

83. MEYERMANS, Eric (06/11/92). Econometric allocation systems for 

the foreign exchange market : Specification, estimation and testing of 

transmission mechanisms under currency substitution. Leuven, K.U.Leuven, 


XVIII, 343 pp. 

84. CHEN, Guoqing (04/12/92). Design of fuzzy relational databases based 

on fuzzy functional dependency. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 1992. 176 pp. 

85. CLAEYS, Christel (18/02/93). Vertical and horizontal category structures 

in consumer decision making : The nature of product hierarchies 

and the effect of brand typicality. Leuven, K.U.Leuven, Faculteit 


86. CHEN, Shaoxiang (25/03/93). The optimal monitoring policies for 

some stochastic and dynamic production processes. Leuven, K.U.Leuven, 


170 pp. 

87. OVERWEG, Dirk (23/04/93). Approximate parametric analysis and 

study of cost capacity management of computer configurations. Leuven, 

K.U.Leuven, Faculteit Economische en toegepaste economische 


88. DEWACHTER, Hans (22/06/93). Nonlinearities in speculative prices : 

The existence and persistence of nonlinearity in foreign exchange rates. 


wetenschappen, 1993. 151 pp.


89. LIN, Liangqi (05/07/93). Economic determinants of voluntary accounting 

choices for R & D expenditures in Belgium. Leuven, K.U.Leuven, 


192 pp. 

90. DHAENE, Geert (09/07/93). Encompassing: formulation, properties 

and testing. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 


91. LAGAE, Wim (20/09/93). Marktconforme verlichting van soevereine 

buitenlandse schuld door private crediteuren: een neo-institutionele 

analyse. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 


92. VAN DE GAER, Dirk (27/09/93). Equality of opportunity and investment 

in human capital. Leuven, K.U.Leuven, Faculteit Economische 


93. SCHROYEN, Alfred (28/02/94). Essays on redistributive taxation 

when monitoring is costly. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 1994. 203 pp.+ V. 

94. STEURS, Geert (15/07/94). Spillovers and cooperation in research 

and development. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 


95. BARAS, Johan (15/09/94). The small sample distribution of the Wald, 

Lagrange multiplier and likelihood ratio tests for homogeneity and symmetry 

in demand analysis: a Monte Carlo study. Leuven, K.U.Leuven, 


169 pp. 

96. GAEREMYNCK, Ann (08/09/94). The use of depreciation in accounting 

as a signalling device. Leuven, K.U.Leuven, Faculteit Economische 


97. BETTENDORF, Leon (22/09/94). A dynamic applied general equilibrium 

model for a small open economy. Leuven, K.U.Leuven, Faculteit 


98. TEUNEN, Marleen (10/11/94). Evaluation of interest randomness in 

actuarial quantities. Leuven, K.U.Leuven, Faculteit Economische en 

toegepaste economische wetenschappen, 1994. 214 pp.


99. VAN OOTEGEM, Luc (17/01/95). An economic theory of private 

donations. Leuven. K.U.Leuven, Faculteit Economische en toegepaste 


100. DE SCHEPPER, Ann (20/03/95). Stochastic interest rates and the 

probabilistic behaviour of actuarial functions. Leuven, K.U.Leuven, 


211 pp. 

101. LAUWERS, Luc (13/06/95). Social choice with infinite populations. 



102. WU, Guang (27/06/95). A systematic approach to object-oriented 

business modeling. Leuven, K.U.Leuven, Faculteit Economische en 


103. WU, Xueping (21/08/95). Term structures in the Belgian market : 

model estimation and pricing error analysis. Leuven, K.U.Leuven, Faculteit 

Economische en toegepaste economische wetenschappen, 1995. 

133pp. 

104. PEPERMANS, Guido (30/08/95). Four essays on retirement from the 

labor force. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 

economische wetenschappen, 1995. 128pp. 

105. ALGOED, Koen (11/09/95). Essays on insurance: a view from a dynamic 

perspective. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 


106. DEGRYSE, Hans (10/10/95). Essays on financial intermediation, product 

differentiation, and market structure. Leuven, K.U.Leuven, Faculteit 

Economische en toegepaste economische wetenschappen, 1995. 218 

pp. 

107. MEIR, Jos (05/12/95). Het strategisch groepsconcept toegepast op 

de Belgische financiële sector. Leuven, K.U.Leuven, Faculteit Economische 


108. WIJAYA, Miryam Lilian (08/01/96). Voluntary reciprocity as an informal 

social insurance mechanism: a game theoretic approach. Leuven, 

K.U.Leuven, Faculteit Economische en toegepaste economische wetenschappen, 

1996. 124 pp.


109. VANDAELE, Nico (12/02/96). The impact of lot sizing on queueing 

delays : multi product, multi machine models. Leuven, K.U.Leuven, 


243 pp. 

110. GIELENS, Geert (27/02/96). Some essays on discrete time target zones 

and their tails. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 


111. GUILLAUME, Dominique (20/03/96). Chaos, randomness and order 

in the foreign exchange markets. essays on the modelling of the markets. 



112. DEWIT, Gerda (03/06/96). Essays on export insurance subsidization. 



113. VAN DEN ACKER, Carine (08/07/96). Belief-function theory and 

its application to the modeling of uncertainty in financial statement 

auditing. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 


114. IMAM, Mahmood Osman (31/07/96). Choice of IPO Flotation Methods 

in Belgium in an Asymmetric Information Framework and Pricing 

of IPO’s in the Long-Run. Leuven, K.U.Leuven, Faculteit Economische 


115. NICAISE, Ides (06/09/96). Poverty and Human Capital. Leuven, 


1996. 209 pp. 

116. EYCKMANS, Johan (18/09/97). On the Incentives of Nations to Join 

International Environmental Agreements. Leuven, K.U.Leuven, Faculteit 

Economische en toegepaste economische wetenschappen, 1997. XV 

+ 348 pp. 

117. CRISOLOGO-MENDOZA, Lorelei (16/10/97). Essays on Decision 

Making in Rural Households: a study of three villages in the Cordillera 

Region of the Philippines. Leuven, K.U.Leuven, Faculteit Economische 


118. DE REYCK, Bert (26/01/98). Scheduling Projects with Generalized 

Precedence Relations: Exact and Heuristic Procedures. Leuven, K.U.Leuven,



XXIV+337 pp. 

119. VANDEMAELE Sigrid (30/04/98). Determinants of Issue Procedure 

Choice within the Context of the French IPO Market: Analysis within 

an Asymmetric Information Framework. Leuven, K.U.Leuven, Faculteit 

Economische en toegepaste economische wetenschappen, 1998. 241 

pp. 

120. VERGAUWEN Filip (30/04/98). Firm Efficiency and Compensation 

Schemes for the Management of Innovative Activities and Knowledge 

Transfers. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 

economische wetenschappen, 1998. VIII+175 pp. 

121. LEEMANS Herlinde (29/05/98). The Two-Class Two-Server Queueing 

Model with Nonpreemptive Heterogeneous Priority Structures. Leuven, 



122. GEYSKENS Inge (4/09/98). Trust, Satisfaction, and Equity in Marketing 

Channel Relationships. Leuven, K.U.Leuven, Faculteit Economische 


123. SWEENEY John (19/10/98). Why Hold a Job ? The Labour Market 

Choice of the Low-Skilled. Leuven, K.U.Leuven, Faculteit Economische 


124. GOEDHUYS Micheline (17/03/99). Industrial Organisation in Developing 

Countries, Evidence from Côte d’Ivoire. Leuven, K.U.Leuven, 


251 pp. 

125. POELS Geert (16/04/99). On the Formal Aspects of the Measurement 

of Object-Oriented Software Specifications. Leuven, K.U.Leuven, Faculteit 


507 pp. 

126. MAYERES Inge (25/05/99). The Control of Transport Externalities: 

A General Equilibrium Analysis. Leuven, K.U.Leuven, Faculteit 

Economische en toegepaste economische wetenschappen, 1999. XIV + 

294 pp. 

127. LEMAHIEU Wilfried (5/07/99). Improved Navigation and Maintenance 

through an Object-Oriented Approach to Hypermedia Modelling.




128. VAN PUYENBROECK Tom (8/07/99). Informational Aspects of Fiscal 

Federalism. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 


129. VAN DEN POEL Dirk (5/08/99). Response Modeling for Database 

Marketing Using Binary Classification. Leuven, K.U.Leuven, Faculteit 


130. GIELENS Katrijn (27/08/99). International Entry Decisions in the 

Retailing Industry: Antecedents and Performance Consequences. Leuven, 



131. PEETERS Aneleen (16/12/99). Labour Turnover Costs, Employment 

and Temporary Work. Leuven, K.U.Leuven, Faculteit Economische en 


132. VANHOENACKER Jurgen (17/12/99). Formalizing a Knowledge Management 

Architecture Meta-Model for Integrated Business Process Management. 



133. NUNES Paulo (12/04/2000). Contingent Valuation of the Benefits of 

Natural Areas and its Warglow Component. Leuven, K.U.Leuven, Faculteit 


XXI + 282 pp. 

134. VAN DEN CRUYCE Bart (7/04/2000). Statistische discriminatie van 

allochtonen op jobmarkten met rigide lonen. Leuven, K.U.Leuven, Faculteit 


135. REPKINE Alexandre (15/03/2000). Industrial Restructuring in Countries 

of Central and Eastern Europe : combining Branch-, Firm- and 

Product-level Data for a better Understanding of Enterprises’Behaviour 

during Transition towards Market Economy. Leuven, K.U.Leuven, Faculteit 


136. AKSOY Yunus (21/06/2000). Essays on international price rigidities 

and exchange rates. Leuven, K.U.Leuven, Faculteit Economische en 

toegepaste economische wetenschappen, 2000. IX + 236 pp.


137. RIYANTO Yohanes Eko (22/06/2000). Essays on the internal and external 

delegation of authority in firms. Leuven, K.U.Leuven, Faculteit 

Economische en toegepaste economische wetenschappen, 2000. VIII + 

280 pp. 

138. HUYGHEBAERT, Nancy (20/12/2000). The Capital Structure of 

Business Start-ups. Leuven, K.U.Leuven, Faculteit Economische en 

toegepaste economische wetenschappen, 2000, VIII + 332 pp. 

139. FRANCKX Laurent (22/01/2001). Ambient Inspections and Commitment 

in Environmental Enforcement. Leuven, K.U.Leuven, Faculteit 

Economische en toegepaste economische wetenschappen, 2001, VIII + 

286 pp. 

140. VANDILLE Guy (16/02/2001). Essays on the Impact of Income Redistribution 

on Trade. Leuven, K.U.Leuven, Faculteit Economische en 

toegepaste economische wetenschappen, 2001, VIII + 176 pp. 

141. MARQUERING Wessel (27/04/2001). Modeling and Forecasting Stock 

Market Returns and Volatility. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 2001. 

142. FAGGIO Giulia (07/05/2001). Labor Market Adjustment and Enterprise 

Behavior in Transition. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 2001, 150 pp. 

143. GOOS Peter (30/05/2001). The Optimal Design of Blocked and Splitplot 

experiments. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 

economische wetenschappen, 2001. 

144. LABRO Eva (01/06/2001). Total Cost of Ownership Supplier Selection 

based on Activity Based Costing and Mathematical Programming. 


wetenschappen, 2001. 

145. VANHOUCKE Mario (07/06/2001). Exact Algorithms for various 

Types of Project Scheduling Problems. Nonregular Objectives and 

time/cost Trade-offs. Leuven, K.U.Leuven, Faculteit Economische en 

toegepaste economische wetenschappen, 2001, 316 pp. 

146. BILSEN Valentijn (28/08/2001). Entrepreneurship and Private Sector 

Development in Central European Transition Countries. Leuven, 


2001 XVI + 188 pp.


147. NIJS Vincent (10/08/2001). Essays on the dynamic Category-level Impact 

of Price promotions. Leuven, K.U.Leuven, Faculteit Economische 


148. CHERCHYE Laurens (24/09/2001). Topics in Non-parametric Production 

and Efficiency Analysis. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 2001 VII + 169 pp. 

149. VAN DENDER Kurt (15/10/2001). Aspects of Congestion Pricing 

for Urban Transport. Leuven, K.U.Leuven, Faculteit Economische en 

toegepaste economische wetenschappen, 2001 VII + 203 pp. 

150. CAPEAU Bart (26/10/2001). In defence of the excess demand approach 

to poor peasants’economic behaviour. Theory and Empirics of 

non-recursive agricultural household modeling. Leuven, K.U.Leuven, 

Faculteit Economische en toegepaste economische wetenschappen, 2001, 

XIII + 286 pp. 

151. CALTHROP Edward (09/11/2001). Essays in urban transport economics. 



152. VANDER BAUWHEDE (03/12/2001). Earnings management in an 

Non-Anglo-Saxon environment. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 2001, 408 pp. 

153. DE BACKER Koenraad (22/01/2002). Multinational firms and industry 

dynamics in host countries : the case of Belgium. Leuven, 


2002, VII + 165 pp. 

154. BOUWEN Jan (08/02/2002). Transactive memory in operational workgroups. 

Concept elaboration and case study. Leuven, K.U.Leuven, Faculteit 

Economische en toegepaste economische wetenschappen, 2002, 

319 pp. + appendix 102 pp. 

155. VAN DEN BRANDE Inge (13/03/2002). The psychological contract 

between employer and employee : a survey among Flemish employees. 


wetenschappen, 2002, VIII + 470 pp. 

156. VEESTRAETEN Dirk (19/04/2002). Asset Price Dynamics under 

Announced Policy Switching. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 2002, 176 pp.


157. PEETERS Marc (16/05/2002). One dimensional cutting and packing 

: new problems and algorithms. Leuven, K.U.Leuven, Faculteit 


158. SKUDELNY Frauke (21/05/2002). Essays on the economic consequences 

of the European monetary union. Leuven, K.U.Leuven, Faculteit 


159. DE WEERDT Joachim (07/06/2002). Social networks, transfers and 

insurance in developing countries. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 2002, VI+129 pp. 

160. TACK Lieven (25/06/2002). Optimal run orders in design of experiments. 


economische wetenschappen, 2002, xxxi+344 pp. 

161. POELMANS Stephan (10/07/2002). Making workflow systems work. 

An investigation into the importance of task-appropriation fit, end-user 

support and other technological characteristics. Leuven, K.U.Leuven, 


237 pp. 

162. JANS Raf (26/09/2002). Capacitated Lot Sizing Problems : New Applications, 

Formulations and Algorithms. Leuven, K.U.Leuven, Faculteit 


163. VIAENE Stijn (25/10/2002). Learning to Detect Fraud from enriched 

Insurance Claims Data (Context, Theory and Applications). Leuven, 


2002, 315 pp. 

164. AYALEW Tekabe (08/11/2002). Inequality and Capital Investment in 

a Subsistence Economy.Leuven, K.U.Leuven, Faculteit Economische en 

toegepaste economische wetenschappen, 2002. V + 148 pp. 

165. MUES Christophe (12/11/2002). On the Use of Decision Tables and 

Diagrams in Knowledge Modeling and Verification. Leuven, K.U.Leuven, 


222 pp. 

166. BROCK Ellen (13/03/2003). The Impact of International Trade on 

European Labour Markets. K.U.Leuven, Faculteit Economische en toegepaste 

economische wetenschappen, 2002.


167. VERMEULEN Frederic (29/11/2002). Essays on the collective Approach 

to Household Labour Supply. Leuven, K.U.Leuven, Faculteit 

Economische en toegepaste economische wetenschappen, 2002. XIV + 

203 pp. 

168. CLUDTS Stephan (11/12/2002). Combining participation in decisionmaking 

with financial participation : theoretical and empirical perspectives. 


economische wetenschappen, 2002. XIV + 247 pp. 

169. WARZYNSKI Frederic (09/12/2003). The dynamic effect of competition 

on price cost margins and innovation. Leuven, K.U.Leuven, Faculteit 


170. VERWIMP Philip (14/01/2003). Development and genocide in Rwanda 

; a political economy analysis of peasants and power under the Habyarimana 

regime. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 


171. BIGANO Andrea (25/02/2003). Environmental regulation of the electricity 

sector in a European Market Framework. Leuven, K.U.Leuven, 


XX + 310 pp. 

172. MAES Konstantijn (24/03/2003). Modeling the Term Structure of Interest 

Rates Across Countries. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 2003, V+246 pp. 

173. VINAIMONT Tom (26/02/2003). The performance of One- versus 

Two-Factor Models of the Term Structure of Interest Rates. Leuven, 


2003. 

174. OOGHE Erwin (15/04/2003). Essays in multi-dimensional social choice. 


wetenschappen, 2003, VIII+108 pp. 

175. FORRIER Anneleen (25/04/2003). Temporary employment, employability 

and training. Leuven, K.U.Leuven, Faculteit Economische en 

toegepaste economische wetenschappen, 2003. 

176. CARDINAELS Eddy (28/04/2003). The role of cost system accuracy 

in managerial decision making. Leuven, K.U.Leuven, Faculteit Economische 

en toegepaste economische wetenschappen, 2003. 144 pp.


177. DE GOEIJ Peter (02/07/2003). Modeling Time-Varying Volatility and 

Interest Rates. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 

economische wetenschappen, 2003. VII+225 pp. 

178. LEUS Roel (19/09/2003). The generation of stable project plans. 

Complexity and exact algorithms. Leuven, K.U.Leuven, Faculteit Economische 


179. MARINHEIRO Carlos (23/09/2003). EMU and fiscal stabilisation policy 

: the case of small countries. Leuven, K.U.Leuven, Faculteit Economische 


180. BAESSENS Bart (24/09/2003). Developing intelligent systems for 

credit scoring using machine learning techniques. Leuven, K.U.Leuven, 


181. KOCZY Laszlo (18/09/2003). Solution concepts and outsider behaviour 

in coalition formation games. Leuven, K.U.Leuven, Faculteit Economische 


182. ALTOMONTE Carlo (25/09/2003). Essays on Foreign Direct Investment 

in transition countries : learning from the evidence. Leuven, 


2003. 

183. DRIES Liesbeth (10/11/2003). Transition, Globalisation and Sectoral 

Restructuring: Theory and Evidence from the Polish Agri-Food Sector. 


wetenschappen, 2003. 

184. DEVOOGHT Kurt (18/11/2003). Essays On Responsibility-Sensitive 

Egalitarianism and the Measurement of Income Inequality. Leuven, 


2003. 

185. DELEERSNYDER Barbara (28/11/2003). Marketing in Turbulent 

Times. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 


186. ALI Daniel (19/12/2003). Essays on Household Consumption and 

Production Decisions under Uncertainty in Rural Ethiopia. Leuven, 


2003.


187. WILLEMS Bert (14/01/2004). Electricity networks and generation 

market power. Leuven, K.U.Leuven, Faculteit Economische en toegepaste 


188. JANSSENS Gust (30/01/2004). Advanced Modelling of Conditional 

Volatility and Correlation in Financial Markets. Leuven, K.U.Leuven, 


189. THOEN Vincent (19/01/2004). On the valuation and disclosure practices 

implemented by venture capital providers. Leuven, K.U.Leuven, 


190. MARTENS Jurgen (16/02/2004). A fuzzy set and stochastic system 

theoretic technique to validate simulation models. Leuven, K.U.Leuven, 


191. ALTAVILLA Carlo (21/05/2004). Monetary policy implementation 

and transmission mechanisms in the Euro area. Leuven, K.U.Leuven, 


192. DE BRUYNE Karolien (07/06/2004). Essays in the location of economic 

activity. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 

Economische Wetenschappen, 2004. 

193. ADEM Jan (25/06/2004). Mathematical programming approaches for 

the supervised classification problem. Leuven, K.U.Leuven, Faculteit 

Economische en Toegepaste Economische Wetenschappen, 2004. 

194. LEROUGE Davy (08/07/2004). Predicting Product Preferences : the 

effect of internal and external cues. Leuven, K.U.Leuven, Faculteit 


195. VANDENBROECK Katleen (16/07/2004). Essays on output growth, 

social learning and land allocation in agriculture : micro-evidence from 

Ethiopia and Tanzania. Leuven, K.U.Leuven, Faculteit Economische 

en Toegepaste Economische Wetenschappen, 2004. 

196. GRIMALDI Maria (03/09/004). The exchange rate, heterogeneity of 

agents and bounded rationality. Leuven, K.U.Leuven, Faculteit Economische 


197. SMEDTS Kristien (26/10/2004). Financial integration in EMU in the 

framework of the no-arbitrage theory. Leuven, K.U.Leuven, Faculteit 

Economische en Toegepaste Economische Wetenschappen, 2004.


198. KOEVOETS Wim (12/11/2004). Essays on Unions, Wages and Employment. 

Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


199. CALLENS Marc (22/11/2004). Essays on multilevel logistic Regression. 

Leuven, K.U.Leuven, Faculteit Economische en Toegepaste Economische 

Wetenschappen, 2004. 

200. RUGGOO Arvind (13/12/2004). Two stage designs robust to model 

uncertainty. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


201. HOORELBEKE Dirk (28/01/2005). Bootstrap and Pivoting Techniques 

for Testing Multiple Hypotheses. Leuven, K.U.Leuven, Faculteit 


202. ROUSSEAU Sandra (17/02/2005). Selecting Environmental Policy Instruments 

in the Presence of Incomplete Compliance. Leuven, K.U.Leuven, 

Faculteit Economische en Toegepaste Economische Wetenschappen, 2005. 

203. VAN DER MEULEN Sofie (17/02/2005). Quality of Financial Statements 

: Impact of the external auditor and applied accounting standards. 



204. DIMOVA Ralitza (21/02/2005). Winners and Losers during Structural 

Reform and Crisis : the Bulgarian Labour Market Perspective. Leuven, 

K.U.Leuven, Faculteit Economische en Toegepaste Economische 


205. DARKIEWICZ Grzegorz (28/02/2005). Value-at-risk in Insurance and 

Finance : the Comonotonicity Approach. Leuven, K.U.Leuven, Faculteit 


206. DE MOOR Lieven (20/05/2005). The Structure of International Stock 

Returns : Size, Country and Sector Effects in Capital Asset Pricing. 



207. EVERAERT Greetje (27/06/2005). Soft Budget Constraints and Trade 

Policies : The Role of Institutional and External Constraints. Leuven, 


Wetenschappen, 2005.


208. SIMON Steven (06/07/2005). The Modeling and Valuation of complex 

Derivatives : the Impact of the Choice of the term structure 

model. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


209. MOONEN Linda (23/09/2005). Algorithms for some graph-theoretical 

optimization problems. Leuven, K.U.Leuven, Faculteit Economische en 

Toegepaste Economische Wetenschappen, 2005. 

210. COUCKE Kristien (21/09/2005). Firm and industry adjustment under 

de-industrialisation and globalization of the Belgian economy. Leuven, 



211. DECAMPS MARC (21/10/2005). Some actuarial and financial applications 

of generalized diffusion processes. Leuven, K.U.Leuven, Faculteit 


212. KIM HELENA (29/11/2005). Escalation games: an instrument to analyze 

conflicts. The strategic approach to the bargaining problem. Leuven, 



213. GERMENJI ETLEVA (06/01/2006). Essays on the economics of emigration 

from Albania. Leuven, K.U.Leuven, Faculteit Economische en 


214. BELIEN JEROEN (18/01/2006). Exact and heuristic methodologies 

for scheduling in hospitals: problems, formulations and algorithms. 



215. JOOSSENS KRISTEL (10/02/2006). Robust discriminant analysis. 



216. VRANKEN LIESBET (13/02/2006). Land markets and production efficiency 

in transition economies. Leuven, K.U.Leuven, Faculteit Economische 


217. VANSTEENKISTE ISABEL (22/02/2006). Essays on non-linear modelling 

in international macroeconomics. Leuven, K.U.Leuven, Faculteit 

Economische en Toegepaste Economische Wetenschappen, 2006.


218. WUYTS Gunther (31/03/2006). Essays on the liquidity of financial 

markets. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


219. DE BLANDER Rembert (28/04/2006). Essays on endogeneity and parameter 

heterogeneity in cross-section and panel data. Leuven, K.U.Leuven, 


220. DE LOECKER Jan (12/05/2006). Industry dynamics and productivity. 



221. LEMMENS Aurélie (12/05/2006). Advanced classification and timeseries 

methods in marketing. Leuven, K.U.Leuven, Faculteit Economische 


222. VERPOORTEN Marijke (22/05/2006). Conflict and survival: an analysis 

of shocks, coping strategies and economic mobility in Rwanda, 

1990-2002. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


223. BOSMANS Kristof (26/05/2006). Measuring economic inequality and 

inequality aversion. Leuven, K.U.Leuven, Faculteit Economische en 


224. BRENKERS Randy (29/05/2006). Policy reform in a market with 

differentiated products: applications from the car market. Leuven, 

K.U.Leuven, Faculteit Economische en Toegepaste Economische Wetenschappen, 

2006. 

225. BRUYNEEL Sabrina (02/06/2006). Self-econtrol depletion: Mechanisms 

and its effects on consumer behavior. Leuven, K.U.Leuven, Faculteit 


226. FAEMS Dries (09/06/2006). Collaboration for innovation: Processes of 

governance and learning. Leuven, K.U.Leuven, Faculteit Economische 


227. BRIERS Barbara (28/06/2006). Countering the scrooge in each of us: 

on the marketing of cooperative behavior. Leuven, K.U.Leuven, Faculteit 


228. ZANONI Patrizia (04/07/2006). Beyond demography: Essays on diversity 

in organizations. Leuven, K.U.Leuven, Faculteit Economische 

en Toegepaste Economische Wetenschappen, 2006.


229. VAN DEN ABBEELE Alexandra (11/09/2006). Management control 

of interfirm relations: the role of information. Leuven, K.U.Leuven, 


230. DEWAELHEYNS Nico (18/09/2006). Essays on internal capital markets, 

bankruptcy and bankruptcy reform. Leuven, K.U.Leuven, Faculteit 


231. RINALDI Laura (19/09/2006). Essays on card payments and household 

debt. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


232. DUTORDOIR Marie (22/09/2006). Determinants and stock price effects 

of Western European convertible debt offerings: an empirical 

analysis. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


233. LYKOGIANNI Elissavet (20/09/2006). Essays on strategic decisions of 

multinational enterprises: R&D decentralization, technology transfers 

and modes of foreign entry. Leuven, K.U.Leuven, Faculteit Economische 


234. ZOU Jianglei (03/10/2006). Inter-firm ties, plant networks, and multinational 

firms: essays on FDI and trade by Japanse firms. Leuven, 


2006. 

235. GEYSKENS Kelly (12/10/2006). The ironic effects of food temptations 

on self-control performance. Leuven, K.U.Leuven, Faculteit Economische 


236. BRUYNSEELS Liesbeth (17/10/2006). Client strategic actions, goingconcern 

audit opinions and audit reporting errors. Leuven, K.U.Leuven, 


237. KESSELS Roselinde (23/10/2006). Optimal designs for the measurement 

of consumer preferences. Leuven, K.U.Leuven, Faculteit Economische 


238. HUTCHINSON John (25/10/2006). The size distribution and growth 

of firms in transition countries. Leuven, K.U.Leuven, Faculteit Economische 

en Toegepaste Economische Wetenschappen, 2006.


239. RENDERS Annelies (26/10/2006). Corporate governance in Europe: 

The relation with accounting standards choice, performance and benefits 

of control. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


240. DE WINNE Sophie (30/10/2006). Exploring terra incognita: human 

resource management and firm performance in small and medium-sized 

businesses. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


241. KADITI Eleni (10/11/2006). Foreign direct investments in transition 

economies. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


242. ANDRIES Petra (17/11/2006). Technology-based ventures in emerging 

industries: the quest for a viable business model. Leuven, K.U.Leuven, 


243. BOUTE Robert (04/12/2006). The impact of replenishment rules 

with endogenous lead times on supply chain performance. Leuven, 


2006. 

244. MAES Johan (20/12/2006). Corporate entrepreneurship: an integrative 

analysis of a resource-based model. Evidence from Flemish enterprises. 



245. GOOSSENS Dries (20/12/2006). Exact methods for combinatorial 

auctions. Leuven, K.U.Leuven, Faculteit Economische en Toegepaste 


246. GOETHALS Frank (22/12/2006). Classifying and assessing extended 

enterprise integration approaches. Leuven, K.U.Leuven, Faculteit Economische 


247. VAN DE VONDER Stijn (22/12/2006). Proactive-reactive procedures 

for robust project scheduling. Leuven, K.U.Leuven, Faculteit Economische 


248. SAVEYN Bert (27/02/2007). Environmental policy in a federal state. 


Wetenschappen, 2007.


249. CLEEREN Kathleen (13/03/2007). Essays on competitive structure 

and product-harm crises. Leuven, K.U.Leuven, Faculteit Economische 


250. THUYSBAERT Bram (27/04/2007). Econometric essays on the measurement 

of poverty. Leuven, K.U.Leuven, Faculteit Economische en 


251. DE BACKER Manu (07/05/2007). The use of Petri net theory for business 

process verification. Leuven, K.U.Leuven, Faculteit Economische 


252. MILLET Kobe (15/05/2007). Prenatal testosterone, personality, and 

economic behavior. Leuven, K.U.Leuven, Faculteit Economische en 


253. HUYSMANS Johan (13/06/2007). Comprehensible predictive models: 

New methods and insights. Leuven, K.U.Leuven, Faculteit Economische 


254. FRANCKEN Nathalie (26/06/2007). Mass Media, Government Policies 

and Economic Development: Evidence from Madagascar. Leuven, 



255. SCHOUBBEN Frederiek (02/07/2007). The impact of a stock listing 

on the determinants of firm performance and investment policy. Leuven, 



256. DELHAYE Eef (04/07/2007). Economic analysis of traffic safety. Leuven, 



257. VAN ACHTER Mark (06/07/2007). Essays on the market microstructure 

of financial markets. Leuven, K.U.Leuven, Faculteit Economische 


258. GOUKENS Caroline (20/08/2007). Desire for variety: understanding 

consumers’ preferences for variety seeking. Leuven, K.U.Leuven, Faculteit 


259. KELCHTERMANS Stijn (12/09/2007). In pursuit of excellence: essays 

on the organization of higher education and research. Leuven,



2007. 

260. HUSSINGER Katrin (14/09/2007). Essays on internationalization, innovation 

and firm performance. Leuven, K.U.Leuven, Faculteit Economische 


261. CUMPS Bjorn (04/10/2007). Business-ICT alignment and determinants. 



262. LYRIO Marco (02/11/2007). Modeling the yield curve with macro 

factors. Leuven, K.U.Leuven, Faculteit Economie en Bedrijfswetenschappen, 

2007. 

263. VANPEE Rosanne (16/11/2007). Home bias and the implicit costs of 

investing abroad. Leuven, K.U.Leuven, Faculteit Economie en Bedrijfswetenschappen, 

2007. 

264. LAMBRECHTS Olivier (27/11/2007). Robust project scheduling subject 

to resource breakdowns. Leuven, K.U.Leuven, Faculteit Economie 

en Bedrijfswetenschappen, 2007. 

265. DE ROCK Bram (03/12/2007). Collective choice behaviour: non parametric 

characterization. Leuven, K.U.Leuven, Faculteit Economie en 

Bedrijfswetenschappen, 2007. 

266. MARTENS David (08/01/2008). Building acceptable classification 

models for financial engineering applications. Leuven, K.U.Leuven, 

Faculteit Economie en Bedrijfswetenschappen, 2008. 

267. VAN KERCKHOVEN Johan (17/01/2008). Predictive modelling: variable 

selection and classification efficiencies. Leuven, K.U.Leuven, Faculteit 

Economie en Bedrijfswetenschappen, 2008. 

268. CIAIAN Pavel (12/02/2008). Land, EU accession and market imperfections. 

Leuven, K.U.Leuven, Faculteit Economie en Bedrijfswetenschappen, 

2008. 

269. TRUYTS Tom (27/02/2008). Diamonds are a girl’s best friend: five 

essays on the economics of social status. Leuven, K.U.Leuven, Faculteit 

Economie en Bedrijfswetenschappen, 2008.


270. LEWIS Vivien (17/03/2008). Applications in dynamic general equilibrium 

macroeconomics. Leuven, K.U.Leuven, Faculteit Economie en 


271. CAPPELLEN Tineke (04/04/2008). Worldwide coordination in a transnational 

environment: An inquiry into the work and careers of global 

managers. Leuven, K.U.Leuven, Faculteit Economie en Bedrijfswetenschappen, 

2008. 

272. RODRIGUEZ Victor (18/04/2008). Material transfer agreements: research 

agenda choice, co-publication activity and visibility in biotechnology. 


2008. 

273. QUAN Qi (14/04/2008). Privatization in China: Examining the endogeneity 

of the process and its implications for the performance of 

newly privatized firms. Leuven, K.U.Leuven, Faculteit Economie en 


274. DELMOTTE Jeroen (30/04/2008). Evaluating the HR function: Empirical 

studies on HRM architecture and HRM system strength. Leuven, 

K.U.Leuven, Faculteit Economie en Bedrijfswetenschappen, 2008. 

275. ORSINI Kristian (05/05/2008). Making work pay: Insights from microsimulation 

and random utility models. Leuven, K.U.Leuven, Faculteit 


276. HOUSSA Romain (13/05/2008). Macroeconomic fluctuations in developing 

countries. Leuven, K.U.Leuven, Faculteit Economie en Bedrijfswetenschappen, 

2008. 

277. SCHAUMANS Catherine (20/05/2008). Entry, regulation and economic 

efficiency: essays on health professionals. Leuven, K.U.Leuven, 


278. CRABBE Karen (21/05/2008). Essays on corporate tax competition 

in Europe. Leuven, K.U.Leuven, Faculteit Economie en Bedrijfswetenschappen, 

2008. 

279. GELPER Sarah (30/05/2008). Economic time series analysis: Granger 

causality and robustness. Leuven, K.U.Leuven, Faculteit Economie en 

Bedrijfswetenschappen, 2008.


280. VAN HOVE Jan (20/06/2008). The impact of technological innovation 

and spillovers on the pattern and direction of international trade. Leuven, 


281. DE VILLE DE GOYET Cédric (04/07/2008). Hedging with futures 

in agricultural commodity markets. Leuven, K.U.Leuven, Faculteit 


282. FRANCK Tom (15/07/2008). Capital structure and product market 

interactions: evidence from business start-ups and private firms. Leuven, 


283. ILBAS Pelin (15/09/2008). Optimal monetary policy design in dynamic 

macroeconomics. Leuven, K.U.Leuven, Faculteit Economie en 


284. GOEDERTIER Stijn (16/09/2008). Declarative techniques for modeling 

and mining business processes. Leuven, K.U.Leuven, Faculteit 


285. LAMEY Lien (22/09/2008). The private-label nightmare: can national 

brands ever wake up? Leuven, K.U.Leuven, Faculteit Economie en 


286. VANDEKERCKHOVE Jan (23/09/2008). Essays on research and development 

with spillovers. Leuven, K.U.Leuven, Faculteit Economie en 


287. PERNOT Eli (25/09/2008). Management control system design for 

supplier relationships in manufacturing: Case study evidence from 

the automotive industry. Leuven, K.U.Leuven, Faculteit Economie en 


288. AERTS Kris (16/10/2008). Essays on the economics of evaluation: 

public policy and corporate strategies in innovation. Leuven, K.U.Leuven, 


289. BOUDT Kris (27/11/2008). Estimation of financial risk under nonnormal 

distributions. Leuven, K.U.Leuven, Faculteit Economie en 


290. MARKIEWICZ Agnieszka (01/12/2008). Essays in exchange rate economics. 


2008.


291. GLADY Nicolas (08/12/2008). Customer profitability modeling. Leuven, 

K.U.Leuven, Faculteit Economie en Bedrijfswetenschappen, 2008 

292. MIERZEJEWSKI Fernando (11/12/2008). Essays on liquidity-preference 

in markets with borrowing restrictions. Leuven, K.U.Leuven, Faculteit 


293. VAN BEVEREN Ilke (15/12/2008). Globalization and firm dynamics. 


2008.

Globalization and Firm Dynamics - Lirias@Lessius

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?