International trade and rent sharing among developed and developing countries

Economic Modelling 24 (2007) 523 – 558 www.elsevier.com/locate/econbase

International trade and rent sharing among developed and developing countries ☆ Lionel Fontagné a , Daniel Mirza b,⁎ a

b

CEPII and TEAM, University of Paris 1, France CREM, University of Rennes 1, Faculté des Sciences Economiques, 7 place Hoche, 35000 Rennes, France Accepted 30 November 2006

Abstract How are rents from openness captured and shared among employers and employees located in different countries? In this paper, we derive a theoretical equation, based on rent sharing theories, linking industry wages to market shares held in different countries and then take it to the test. We construct a dataset that provides together trade, activity and labor related data for around 29 industries and 65 countries between 1981 and 1997. We find, for OECD countries, that an increase in sales on national markets or exports to other rich countries is associated with growth in wages in roughly one third to one half of the industries. Among developing countries however, the evidence is weaker. Such phenomenon of rent-sharing can be observed in Latin America and within Mediterranean countries but only when selling to their domestic market. Producers in these group of countries and Asia, appear however, to be unable to extract rents and redistribute them whenever they export to rich countries. © 2007 Elsevier B.V. All rights reserved. JEL classification: F1; L1; J3 Keywords: International trade; Rent sharing

☆

This paper has benefited from the financial support of the Commissariat Général au Plan, the International Trade Center (Geneva) and the Leverhulme Trust (Programme Grant 114/BF). We are grateful to Rod Falvey, Keith Head, Sebastien Jean, David Margolis, Steve Matusz and Thierry Mayer for very helpful comments on an earlier draft of the work. ⁎ Corresponding author. Tel.: +33 2 23 23 33 19. E-mail address: [email protected] (D. Mirza). 0264-9993/$ - see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.econmod.2006.12.001

524

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

1. Introduction Since the 1999 WTO Seattle meeting, NGOs, unions and other representatives of the civil society from around the world have called for more equity in the globalization process. According to these claims, today's globalization would firstly benefit capital holders from developed countries. Their concern translated into a Call for Mobilisation at the World Social Forum of Porto Alegre in 2001. To quote some of the ideas expressed within that call: “…We demand the genuine recognition of the right to organise and negotiate for unions, and new rights for workers to face the globalisation strategy…Free trade is anything but free. Global trade rules ensure the accelerated accumulation of wealth and power by multinational corporations and the further marginalisation and impoverishment of small farmers, workers and local enterprises…”. While there has been a significant literature on the interactions between trade, employment and wages, the bulk of it does not address the openness issue in these terms. Indeed, most studies linking trade to labour markets do not focus much on the capture and distribution of rents from openness, but look instead at changes in the distribution of factor revenues in general equilibrium (Görg et al., 2001 or Francois and Nelson, 1998 provide examples). Besides, how rents are captured and shared between employers and employees located in different countries remains an avenue for research. Some attention has been paid however, to the role of imperfect competition, as a source of rents, in the impact of trade on labor. In a pioneering work, Oliveira Martins (1994) finds that trade's impact on wages differs among different market structures in the OECD. One common reason is that wages do not only result from pure competition mechanisms in the labor market, which is what is usually assumed by the classical view of trade and labor theorists. They could also depend on the profits gained by employers and the bargaining power of employees. Accordingly, two seminal articles by Abowd and Lemieux (1993) and Borjas and Ramey (1995), stress the importance of accounting for openness as a rent shifter from domestic to foreign firms. In the presence of unions, the loss of profits due to openness translates into a reduction in the wage premium. Borjas and Ramey propose a simple theoretical framework of a two-sector economy, in which the impact of trade on wage inequality is greater in concentrated industries: foreign firms entering the market capture rents otherwise shared with employees within the domestic firms. With regard to developing countries, Harrison and Hanson (1999) mention in a survey that openness had a small impact on wages and employment. The main reasons, the authors argue, come from both imperfections of labor and product markets consistent with rent sharing theories. This result that seems to be robust in most studies on developing countries has been reconsidered however by Goldberg and Pavcnik (2001) when studying the impact of trade policy on Colombian wage premiums. These authors do find a positive effect between protection and the wage premium when they account for both employee characteristics and industry fixed effects. Meanwhile, some issues are still puzzling. First, indirect tests of trade as rent shifters are usually applied: no structural equations linking trade variables to wages via a strict rent sharing mechanism are directly brought to the test. For instance, Abowd and Lemieux (1993) or Abowd and Allain (1996) use some variables of openness as instruments only to run instrumental variable regressions of rent sharing equations while Borjas and Ramey present some suggestive tests based on general equilibrium mechanisms. Goldberg and Pavcnik (2001) or Blom et al. (2003) study rather reduced form equations where wage premiums are regressed against instruments of protection. Here, we ask instead if there is a structural relationship between some trade measures,

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

525

rents and wages that can be identified in theory and that is easily and directly testable. Further, we discuss under which conditions that relationship arises. Second, and most importantly, the related literature does not mention the extent to which rents from trade are shifted between countries. Are rich countries gaining rents and sharing them from opening to developing countries? Symmetrically, are developing economies experiencing higher rents and distributing them as they export to developed countries? One might think that opening LDCs may be associated with the loss of large rents in industries where they are disadvantaged and usually characterized by imperfect competition while they would tend to specialize in rather competitive industries. In this paper, we tackle these two issues together. We derive a theoretical relation linking the industry wage premium to market shares that are held on each destination. We demonstrate that this firm level equation could be formally aggregated to an industry level one to be directly tested. Typically, market shares are linked to the wage premium by a precise channel of adjustment representing imperfections on the labor and product markets. In fact, when selling to a domestic or a foreign market, a national industry extracts rents as long as the industry benefits from imperfect competition. Whether or not these rents are effectively shared with employees depends then on the power of unions. Now, if either firms or unions lack of market power on the commodity and the labor markets respectively, then the channel between openness and the wage premium breaks down. In the applied part of the study we take that relation to the test. More precisely, we test the existence of such a channel between wages and domestic and/or foreign market shares at the industry level for different group of countries. We can also directly investigate which export destinations are producing rents that are distributed in a given country: are developed markets good destinations for rent-seeking LDC exporters and their employees? Conversely, are developing market destinations worthwhile for rich exporting countries? As the only data we have access to is at the industry not the firm level, and as we know that most of the literature use instead firm/employee level characteristics, we undertake some further tests of robustness and check ups to support the fact that our industry-level results are consistent with rent sharing theories. We use two UNIDO databases: the 3-digit ISIC Industrial Statistics database, as well as the Industrial Demand–Supply Balance database at the 4-digit level ISIC Code. From these sources, we construct a dataset that matches trade, activity and labor related data for around 29 industries at the 3 ISIC nomenclature (Rev.2) in 65 developed and developing countries, within the 1981–1997 period. We find, for OECD countries, that an increase in domestic market shares is associated with an increase in wages in roughly one third of the industries while half of them experienced an increase in wages when selling to industrialized markets. Among developing countries, in Latin American countries followed by Mediterranean ones, such phenomenon of rent-sharing can be observed, but only when selling to own markets not abroad. In Asian countries, rent sharing phenomena are not identified from trade. In the next section, we present the theoretical model. Section 3 highlights some stylized facts. In Section 4 we design a strategy to match theory with the data. Section 5 shows the econometric results. Section 6 assesses the relevance of our empirical evidence by drawing a comparison with prior studies. Section 7 concludes. 2. The analytical framework We begin by following an hypothesis from Sen and Dutt (1995) by considering a firm n from a country d, acting in oligopoly, where employers and unions bargain over both wages and output.

526

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

The authors do not clearly justify the intuition behind this particular type of strongly efficient bargaining as it is usually employment that is the second variable of interest in these models.1 One can actually think that from the unions' point of view, the variable behind output is actually employment as they know that these two variables are directly linked in production functions. For ease of exposition, we assume that output equals labor demand yn = ln. However, in oligopolistic markets the output of the firm stands as a strategic variable from the manager's point of view. Here, we simply add to the Sen and Dutt framework the hypothesis that the firm serves its own market d and exports to a foreign market f (i.e.: yn =xd,n +xf,n). These markets are assumed to be segmented so that firm sales on a given market depend only on that market's characteristics (see Brander and Krugman, 1983). Pure Cournot competition is considered. In Appendix B, we extend the framework to J markets and show that we still obtain the same type of theoretical relations to estimate. The Nash solution to the bargaining problem would be to choose simultaneously wages, exports and domestic sales. Hence, the objective function to maximize is ðln ½wn −wu Þk ½ pd xd;n þ pf xf ;n −wn ln 1−k

ð1Þ

where wu designates the alternative wage in the economy d, λ indicates the union's degree of market power (0 ≤ λ ≤ 1), pj,n and xj,n the price and the quantities sold by the firm on each market j, ∀j ∈ {d, f }. ln stands for labor demand of the representative firm. From the first order conditions, by deriving with respect to wn we obtain the following wage equation:
 pd xd;n þ pf xf ;n −wu ln wn ¼ k ð2Þ þ wu ln Here, firm wages are linear functions of alternative wages and quasi rents per worker (Abowd, 1989). However, as markets are assumed to be segmented, then total revenues are the sum of revenues obtained from each market. Besides, deriving with respect to xj,n and replacing wn by its corresponding function 2, we find the following quasi mark-up equation on each market j: pd −wu 1 ¼ sd;n rd pd

ð3Þ

pf −wu 1 ¼ sd;n rf pf

ð4Þ

and

1 The concept of strongly efficient bargaining (i.e. both parties negotiate over wages and employment) has been introduced by Brown and Ashenfelter (1986). It is usually opposed to the right to manage hypothesis (i.e. unions and employers bargain only over wages only) or the monopoly union model (i.e. unions choose solely the wage rate). In these models, employers settle a level of employment conditional to the wage rate accordingly determined. While Brown and Ashenfelter (1986), Card (1990) and Hosken and Margolis (1997) find a mixed support to the hypothesis of efficient bargaining, Abowd (1989) and Christofides and Oswald (1991) support completely that hypothesis. Furthermore, using data on New York State public schools Hosken and Margolis (1997) reject systematically the hypothesis that teachers' unions and school districts engage in monopoly union or right to manage style bargaining. In this article we maintain the hypothesis of strongly efficient bargaining agreements and discuss in later sections the implication of a right to manage or monopoly union assumptions on the parameters to estimate.

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

527

Unlike traditional mark-ups that express total profits per unit value, quasi mark-ups stand for the total quasi rents per unit value. Eqs. (3) and (4) are closely related to the Structure– Performance type expressions in industrial economics, since ‘quasi’ mark-ups depend on priceelasticity of demand σj and the market share sj,n = xj,n / XUj (i.e. XUj represents total sales in market j, j ∈ {d, f }). Note bd = λ / σd and bf = λ / σf. Moreover, let pn yn = pd xd,n + pf xf,n be the total revenue for firm n while ed,n and ef,n be respectively the sale rate to the domestic market and export rate to the   px foreign one (i.e. ejn ¼ pjn yj;nn ; 8jafd; ng). Replacing Eqs. (3) and (4) in 2, we end up with the following real wage function: wn wu ¼ bd ed;n sd;n þ bf ef ;n sf ;n þ pn pn

ð5Þ

Hence, the real wage equation, net from the real alternative wage, is a linear combination of the weighted export market share and the domestic share. The intuition behind this relation is that an increase in the market share in a given market j (∀j ∈ {d, f }), translates into more quasi rents for the firm, that are shared with the employees in the presence of some union power. Now, these quasi rents, and thus wage compensation gains, are the more important the more the fraction of output used to serve this market j (i.e. ej,n) is high.2 However, as we do not have access to firm-level data we present in what follows an aggregation strategy that enables us to test a variant of the above equation at the industry level (hereafter, we assume that the industry suffixP k is implicit). Actually, if we compute the real average wage at the industry level w=p ¼ ½ n wn ln =L=p using Eq. (5), we obtain a relation to test that is rather similar to the latter, except that now export rates, domestic sale rates and market shares are not specific to a firm but relative to a country in a representative industry. To see this, let Sd = Xd / X.d be the observed domestic market share and Sf = Xf / XUf the market share held on the foreign market for a given industry in an observed country. Consider Ed and Ef to be respectively the corresponding industrial domestic sale intensity and export intensity. P P x  2 Let L ¼ n ln represent total labor demand at the industry level. Moreover, define wd ¼ n Xd;ndd (resp. ψf ) to be a concentration index that informs about the degree of competition within all the firms of the observed country selling to its market d (resp. to market f ).3 We can now derive

2 Thanks to the assumption of efficient contracts, and if one has to solve completely the model, market shares, domestic and export sale variables do not depend in return on observed wages in the industry. Here, they are actually a function of alternative wages. To see this, consider Eq. (3). For more clarity, let wu,d = wu to be the alternative wage in the domestic country and wu, f that prevailing for the foreign country. When summing over all the firms in Eq. (3), we can obtain an expression for equilibrium prices pd ¼ rNrd−1 ðNd wu;d þ Nf wu; f Þ. Hence, prices only depend on alternative wages. The same kind dd of reasoning applies for Eq. (4). Replacing into expressions (3) and (4), one can easily obtain an expression for each quantity sold by a firm on its domestic or export markets that only depends on alternative rather than industry wages. This result is not surprising and has to be compared to that obtained for the equation of labor demand with efficient bargaining in the labor theory literature. In that framework, labor demand (that is a direct function of quantities which is our variables of control) is also affected by alternative rather than observed wages (see for instance Brown and Ashenfelter, 1986). As a consequence, market shares and ratio of sales to domestic and export markets are affected by alternative wages, price elasticities and the number of firms (i.e.: sj = s(wu,d, wu,f, σj, Nd, Nf ) and ej = e(wu,d, wu,f, σj, Nd, Nf ), ∀j ∈ {d, f }). 3 See Appendix for more details.

528

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

mathematically from Eq. (5), the industry relation of average wages (see Appendix A for more details): w wu ¼ bd Ed Sd þ bf Ef Sf þ p p

ð6Þ

where bd ¼

kwd rd

bf ¼

kwf rf

and

Before interpreting the β parameters, assume first that they are finite and positive. The wage relation we obtain at the industry level is comparable to that expressed at the firm level, except that now the right hand side variables are not specific to a firm but relative to an industry in the observed country. Besides, now that the relation is expressed at the industry level an additional term ψ, relative to the state of competition within the exporters in that industry, enters the equation. This equation has three main characteristics. First, it can be easily confronted to the data that could be found at the industry level when studying a country's openness to trade. Second, this relation suggests that in order to appreciate the openness impact on wages, import penetration, export market shares and export intensity indicators need to be considered together. To see how they could intervene, we call hereafter the composite variable EdSd the ‘weighted domestic market share’ and Ef Sf the ‘weighted export market share’. What is the intuition behind these weights applied on market shares? Actually, when a country's production in an industry essentially serves its domestic demand (Ed is high or inward oriented country) then rents acquired are mainly those driven by sales on the domestic market. In that case, if that country's foreign market shares happen to increase by a certain amount, neither do they greatly affect total rents nor wages in that industry. Notice also that the domestic market share variable Sd is by construction inversely related to import penetration Md as Sd = 1 − Md. This suggests that import penetration is the more painful on wages, the more the country is inward oriented in the considered industry. On the opposite, when a country is mainly outward oriented (the proportion of exports to production Ef, is high), rent extraction from exports is high. Finally, the above relation stresses explicitly the role of imperfect competition in goods and labor markets when looking at the wage response to openness. In particular, that relation between the wage premium and trade shares holds only in the presence of product and labor market imperfections. Actually, the βd and βf, express the interacted market powers of both unions and firms in determining industry real wages. Typically, the ratio of concentration to price elasticity (ψj / σj) forms an average market power indicator for firms when they export to j, ∀j ∈ {d, f }. Hence, the larger the market power, the larger the rents to be potentially shared. Whether or not these rents are effectively shared between workers or employers, depends then on the bargaining power of unions captured through λ. On the opposite, in a competitive market where price elasticity is extremely high or the concentration rate is very low, the relation between the wage premium and trade shares breaks down.

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

529

3. Stylized facts The 3-Digits Industrial Statistics Database (Indstat3) reports data on activity such as 3-digit industry total compensation (wages and benefits), employment and production (ISIC rev.2). UNIDO provides trade data with Developed and Developing countries (imports and exports) at the 4-digit industry level (ISIC rev.2 as well), easily aggregated to 3-digit. Then, by matching these two databases, we were able to construct a table of activity and trade data for 65 developed and developing countries in 29 industries between 1981 and 1997. We present in Tables 1 and 2 the number of industries where data is available in each country finally selected over the period 1981–1997. The matching of data for different countries and periods is a difficult exercise however. Table 1 summarizes available information and sheds light on the large discrepancy between countries. While information is available for 29 industries and the whole 1981–1997 period for the United States, we have information for 10 to 23 Danish industries depending on the year, or for 2 to 24 industries in Mauritius. Other countries do not provide information for the whole period: for instance, data on Germany end in 1994, Bangladesh in 1992, while Costa Rica's data start in 1984 only. On the whole, the worst information is available for El Salvador, Ethiopia, France, Ghana, Madagascar, Nepal, Nicaragua, Romania, South Africa and Tunisia. Except for France, data problems are concentrated in developing countries. We did not update the database using other sources in order to authorize the replication of our results and to stick to an homogeneous data source. This is why data for European countries was not completed. Notwithstanding such unbalanced structure, the data set entails very rich information for numerous developing countries, and this comes out as a good surprise: Chile, the Hong Kong province of China, Colombia, Costa Rica, India, Indonesia, Korea, Malaysia, Mexico, Philippines, Sri Lanka, Turkey, Uruguay and Venezuela collected complete information on a regular basis. It is to be noted that UNIDO trade data are based on the United Nations Commodity Trade tapes and thus, are expected to be exhaustive by country and industry while UN-Indstat3 database reports activity data from different sources of information. A significant proportion of this data appears to be collected from business surveys conducted by UNIDO, which suggests that wages, employment and production could be underestimated relative to their real values in national statistics. However, total compensation in the theoretical model to be tested is expressed relative to employment, and thus the related variable wi constructed from UNIDO would be a good proxy of the real one. More problematic is the production variable which is used to compute domestic and foreign market shares. However, we compared production data from the STAN-OECD database based on national accounts4 to that of UNIDO and found values that were rather similar. Let us consider the whole panel of countries and industries and tabulate the annual growth of wage per employee, labor productivity, production and exports, in all industries (see Tables 3 and 4). The best performances in terms of wage growth over the period under consideration were obtained by Lithuania, Nicaragua, Italy, Korea, Macau, Hong Kong, Slovakia, Singapore, Turkey and Spain. As pictured in Fig. 1 the ranking in terms of productivity gains closely matches the ranking in terms of wages. But more interestingly, most of these countries are well ranked in terms of export growth, with the exception of Hong Kong and Singapore that exhibit a more limited performance. 4

More rigorously, the OECD production data are estimated values from both surveys and national accounts series.

530

Table 1 Summary of available observations (values in the table design the number of industries observed) Country

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39

Argentina Australia Austria Bangladesh Bolivia Canada Chile China (Hong Kong) Colombia Costa Rica Cyprus Denmark Ecuador Egypt El Salvador Ethiopia and Eritrea Fiji Finland France Germany Germany, Western Part Ghana Greece Guatemala Honduras Iceland India Indonesia Italy Japan Jordan Korea, Republic Of Kuwait Macau Madagascar Malaysia Malta Mauritius Mexico

1981

1982

1983

1984

1985

1986

1987

1988

1989

1990

22 13 12 14 29 28 23 29

22 13 14 12 29 28 25 28

13 23 27 17 12 29 28 26 28

25 20 20 17 22 11 14 28 2

24 22 17 22 24 9 13 28 2

25 21 15 23 22 9 13 28 3

27 23 27 14 12 29 27 26 29 23 25 21 17 23 21 10 12 28 4

14 24 27 17 8 29 28 25 29 23 24 20 20 19 21 6 11 28 4

11 9 25 14 11 29 28 26 29 25 25 18 15 24

12 23 24 14 14 29 28 25 29 26 24 21 21 28

13 23 23 2 14 29 28 26 29 25 23 22 22 26

15 21 24 19 10 29 28 26 29 24 24 23 23 28

9 21 22 20 12 29 27 22 29 23 26 19 25 28

8 12 28 3

9 12 26 3

8 12 24 3

8 12 24 3

12 23 2

29 9 29 23 3 10 27 22 29 27 11 27 17 23 14 28 17 19 18

29 11 29 25 3 8

29 9 28 25 18 11 28 24 29 28 11 27 17 24 15 27 16 18 18

29 6 28 26 17 10 28 24 29 27 13 27 18 9 13 27 17 5 26

27

27

27

27

27

26

28 27 23 11 27 23 29 29 13 27

28 27 3 11 27 23 29 29 12 27 21 24 14 26 18 20 26

28 26 3 12 27 24 29 29 13 27 16 24

29 28 3 17 27 24 29 29 16 27 20 17

28

28

3 16 27 20 28 29 19 27 18 15

26 20 2 25

26 18 3 25

26 11 3 25

24 29 28 11 27 15 14 28 18 20 18

11 14 26 19 3 26

1991

1992

1993

1994

1995

1996

1997

23

22

19

18

7

23 29 29 20 29 24 26 11 28 28 17

22 28 28 20 29 25 26 10 28 28 19

23 27 29 19 29 25 26 10 28 28 23

25 27 29 20 29 25 26 10 25 17 26

25 27 28 25 29 22 26 10 23 27

26

11

11

11

29

29 25

10 21 23 17 17 29 28 23 28 24 26 19 24 28

22 21 19 24 29 29 22 29 23 26 10 28 28

12 24 3 23

12 24 3 24

28

25 16 28 25 29 29 23 27 20 14

28 26 24 14 27 25 28 29 20 28 14 16

26 18 22 25

26 18 22 25

26

26

25 18 27 27 27 29 22 29 21 15

29 27 26 17 28 26 27 29 24 29 23 53

29 29 27 17 28 26 27 28 26 29 23 13

29 26 26 19 28 27 28 25 29 22 13

26 19 23 22

26 18 21 21

24 19 22 21

25 18 24 20

19 29 27

29 13

28

28

27 22 12

27 23 15

24 17 22 27

29 24 27

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Obs

Table 2 Summary of available observations Country

1981

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Morocco Nepal Netherlands New Zealand Nicaragua Norway Pakistan Panama Peru Philippines Portugal Romania Singapore South Africa Spain Sri Lanka Sweden Thailand Trinidad and Tobago Tunisia Turkey United Kingdom United States of America Uruguay Venezuela Zimbabwe

25

26

26

26

10 22 29 13 29 23 19 27 26 26

10 16 22 16

10 6 22 7

26 9 21 7

27 11

27 11

25 11 27 11

29 22 19 27 27 25

29 25 17 27

28 25 17 27 28 26

28 27 23 28 28

28 23 18 28 26

26 27 16 27 28

22

22

23

23

18

18

19

29

18 27 26

26

26

26

25

29 22 17 26 28 26

21

29 27

22 29 19 29 28 19 28 27 26

10 7 26 29

27

22 29 13 29 27 16 26 26 27

10 7 26 29

10

22 29 17 29 28 12 26 29 27

10 8 22 29

26

22 29 22 28 25 16

10 15 22 29

20

20

19

18

18

17

17

16

15

14 15

15 15

29 27 29

29 28 28 21 22

29 28 28

29 27 28 25 21

29 27 27

29 27 27

29 27 28 24 22

29 23 26 23 22

29 24 28 26 25

24 23 28

29

29

28 29 29 26 28

28 29 29 22 29

28 29 29 25 25 26

28 29 29 25 27 26

28 29 29 25 28 26

28 29 29 27 27

28 29 29 27 27

27 29 29 27 27

28 29 29 27 29 25

28 29 29 27 28 25

14 23 28 24 27 24 22 16 28 25 29 27 28 24

24 23 28

22

29 26 27 25 6

15 15 23 28 25 26 24 23 16 28 25 29 27 28 24

25 23 28 26 29

21

29 27 27 25 22

15 15 29 29 22 28

21 16 29 28 29 27 28 23

16 29 29 29 27 28 26

16 29 29 29 27

22 27 28 29 29 26 27

21

21

23 28 28 29 27 28 25

1997 9 27

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Obs

26

531

532

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Table 3 Ranking of average industry annual changes Country

Wage/employee

Emp.

Prod

Tot. imports

Tot. exports

Lab. pty

Penet. rate

World M.S.20

Lithuania Nicaragua Italy Korea, Rep. Macau Hong Kong Slovakia Singapore Turkey Spain Germany, West. Iceland Cyprus Mauritius Germany Austria Norway Philippines Japan Finland Sweden Greece United Kingdom Denmark Peru Malta France Argentina Netherlands Uruguay Costa Rica Malaysia Chile Sri Lanka New Zealand Mexico Thailand Australia Eth. and Erit. U.S. of America Canada Gabon Tunisia Pakistan Morocco South Africa Portugal Bahamas Colombia Indonesia Myanmar Bangladesh

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52

71 67 45 35 64 70 54 49 27 36 46 56 23 11 73 51 57 26 40 62 63 52 59 28 44 33 55 60 50 41 17 5 8 19 61 42 7 53 16 43 48 68 14 12 13 34 25 66 30 1 3 2

1 2 16 6 25 49 11 30 13 19 24 47 21 9 68 22 38 17 26 50 48 39 41 34 42 14 31 66 37 33 40 5 8 10 53 44 4 52 46 51 55 57 15 18 20 63 7 23 45 3 28 12

1 68 21 4 35 7 14 23 2 9 22 53 28 12 67 27 50 11 13 41 49 16 38 42 54 33 34 29 40 8 19 15 17 31 30 3 5 48 59 36 39 66 52 55 26 56 6 45 20 18 73 47

1 72 42 31 62 49 3 17 14 30 35 28 61 18 64 39 43 33 56 54 51 29 46 50 59 57 44 19 34 25 55 7 8 26 41 11 6 37 65 36 40 71 38 60 22 32 16 20 27 2 73 24

1 2 7 4 6 5 3 15 17 14 11 20 32 31 24 8 19 30 12 22 18 28 21 40 29 13 10 47 25 27 49 33 36 26 37 34 16 35 54 38 39 51 42 43 44 56 9 67 45 41 65 63

1 72 48 26 15 31 61 54 7 10 39 43 58 29 19 37 55 14 18 33 51 17 52 53 49 47 41 4 64 9 22 40 50 42 44 3 16 35 71 27 21 20 59 66 45 34 11 65 13 56 73 69

1 72 55 27 62 44 2 13 14 32 53 20 63 18 61 37 41 35 52 51 45 29 40 43 60 56 46 25 34 23 50 6 7 30 39 12 9 49 66 36 38 71 31 64 24 33 17 19 26 3 73 28

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

533

Table 4 Ranking of average industry annual change Country

Wage/employee

Emp.

Prod

Tot. imports

Tot. exports

Lab. pty

Penet. rate

World M.S.21

Panama Kuwait India Bahrain Nepal El Salvador Senegal Fiji Zimbabwe Egypt Jordan Trin. and Tob. Honduras Bolivia Ecuador Venezuela Madagascar Romania Nigeria Guatemala Ghana

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73

39 22 21 37 10 15 69 18 31 20 4 38 9 6 29 32 24 58 65 47 72

62 54 35 64 43 58 70 36 59 60 29 65 56 27 61 67 69 72 71 32 73

43 58 24 65 44 32 62 64 10 63 60 70 51 57 25 61 69 46 71 37 72

23 66 15 63 48 58 67 4 21 10 47 53 45 12 5 13 69 68 9 52 70

50 59 46 60 48 66 53 52 55 64 62 61 69 58 57 68 71 72 70 23 73

38 70 25 57 36 24 8 63 5 60 67 62 30 23 12 32 28 2 68 46 6

22 65 16 58 42 59 68 4 21 10 54 57 47 11 5 15 69 67 8 48 70

Reciprocally, the worst performances (Fig. 2) are obtained by Ghana, Guatemala, Nigeria, Romania, Madagascar, Venezuela, Ecuador, Bolivia, Honduras, Trinidad and Tobago. What characteristics do these countries have in common? Productivity gains are generally limited, notably for Ghana, Trinidad and Tobago, Nigeria and Madagascar. But this could be the outcome of a specialization in labor intensive products, authorizing a rise in employment. This assumption is compatible with Honduras and Bolivia figures, but certainly not with the remaining countries listed here. In particular, Ghana exhibits simultaneously poor performances for exports, production, productivity and wages. Unsurprisingly, while the rank correlation between gains in wages and gains in productivity happen to be very large, the correlation with production is more difficult to establish, since exports and imports can vary at a similar pace, while the latter crowds out to some extent domestic producers. Lastly, the rank correlation between exporting and paying wages is nearly zero, notwithstanding the fact that the better performances in terms of wages are precisely obtained for those countries who successfully enter foreign markets. All this proves that the relation between exporting and distributing wages is not trivial and will appear only under certain circumstances to be identified below. In total, according to these stylized facts, the question to be addressed below is whether exporting and gaining market shares enhances wages, controlling for the expected relationship between wages and productivity in order to identify market structure related impacts. In fact, we point out below that productivity affects the alternative wage. What is left out from the difference between the observed industry wage and the alternative wage should be then function of trade shares. The stylized facts also suggest that the relationship under examination here might vary according to the region of the world considered.

534

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Fig. 1. Countries with the best wage/employee performances (variables expressed in terms of estimated annual growth).

Fig. 2. Countries with the worse wage/employee performances (variables expressed in terms of estimated annual growth).

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

535

4. Matching of data and theory Although some information is available on price indexes from different sources, the relation to be tested needs price levels at the denominator of the variables of industry and alternative wages, otherwise the parameter values (in the numerator) would be significantly overestimated. We thus approximate a vector of prices in a market by a vector of weighted mean wages, across domestic and foreign countries selling to that market.5 Besides, the alternative wage in the considered industry is not directly observable from the data. One way to model the alternative wage that is specific to a representative industry k in country i is to consider that its deviation from the mean national wage could be captured through a deviation of productivity from its mean: P

wu;ik ¼ bw P wi þ bpty ½ðPtyik Þ− Ptyi 

ð7Þ 6

where Ptyik is labor productivity (production per employee) observed in the industry, while P i and ptyi represent respectively national mean wages and productivity. The theory we develop in Section 2 is based on a homogeneous product framework sold in well defined marketplaces (i.e. domestic or foreign markets). UNIDO data does not fit directly in that framework as it relies on rather aggregate classifications, both in terms of the reported industries and market boundaries. More precisely, the data is observed at the 3-digit level (ISIC classification) and three group of destination markets can be distinguished: the domestic market, the Industrialized countries' market (Ind, hereafter) and the Developing countries' one (Dev, hereafter). Our theory extends very simply from a two to a three markets' framework (see Appendix B). Still, each industry observed describes a group of products and each market destination observed is related to a group of countries, which suggests a potential presence of product differentiation and cross-country spatial differentiation. We show in Appendix C that our wage equation is still consistent with the existence of differentiation. In that case however, the parameters on the market shares should be higher than the β ones from Eq. (14), where we have assumed a homogeneous good and perfect integration within each market considered. The reason for a higher coefficient is that differentiation increases further the degree of market power for a given firm that should be then captured through the market share parameters. Accounting for spatial and goods' differentiation, and replacing the alternative wage by its function (Eq. (7)) in the wage relation 6, gives the following specification to test7: P w

P wit wit ¼ bdom V ðEdom ⁎Sdom Þit þ bInd V ðEInd ⁎SInd Þit þ bDev V ðEDev ⁎SDev Þit þ bw ∼ ∼ pit pit P ½Ptyit − Ptyit  þ bpty þ ui þ ut þ ui;t ∼ pit

5

ð8Þ

Obviously, this variable is more a mean cost than a price proxy. This underestimation of real prices (at the denominator) would lead again to an overestimation of the parameters (at the numerators) but to a much lesser extent than if we had to choose price indexes. Indeed, following Oliveira Martins et al. (1996) and Schmalensee (1989) among others, prices are shown to be only 20–30% higher than costs on average. 6 In theory, the alternative wage should be more linked to marginal productivity. Since it is not observable however, we replace it by average labor productivity. 7 We have added country and time fixed effects in order to capture other potential components of the wage relation that are specific to country or time. The regressions are ran by industry and thus the subscript k is implicit.

536

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

The β′ parameters in a context of differentiation are defined as: 

jdom bdom V ¼ kdom wdom e rdom



and ∀j = {Ind, Dev} "

jij bjV ¼ kj wij e rj

#

Here, ∀j, σje stands for the mean price-elasticity of ‘effective’ demand (see Appendix C). In addition, an extra parameter κij, ∀j{∈ i, j′} enters the definition of the coefficients. As shown in Appendix C, this parameter is an increasing function of the degree of differentiation and  could  X take on values between 1 (homogeneous goods and perfect market integration case) and xij;nj , ∀j (perfect differentiation in products and space). Thus, the (β′)'s are expected to be always either null or positive, with values that could be very high in case of high spatial or goods' differentiation. The three β′ parameters are expected to be positive and statistically significant under two conditions: 1/ when increasing market shares are associated with rents due to imperfect competition, and 2/ when these rents are shared in return with employees (λ N 0). However, if the β j′ parameter associated to a destination j is null then this would be consistent with one of the two assumptions below: 1/ exporters from country i have no market power on that destination or 2/ unions have no bargaining power that enables them to shift rents from exporting to j. Put differently, by testing for the existence of a channel of adjustment between trade shares and wage premiums we are indirectly testing for the joint existence of imperfect competition on the domestic labor market and the destination product market. Whenever, a positive channel exists the two markets are imperfectly competitive. When either of these markets is in perfect competition, the channel breaks down and more market shares at home or on foreign destinations do not affect the wage premium. A final and very important observation has to be added, however. The β′ parameters could be estimated to be negative. This should be interpreted as a reverse causation, in which high wages lead to shrinking market shares on markets were competition is tight. But this event is unlikely to happen in the data if the alternative wage is well captured and if the contracts between unions and employers are efficient. In that case indeed, as pointed out in the theory section above, market shares are linked to alternative not observed wages. In any case, if it does happen that wages are negatively related to market shares, we try to correct for that revealed endogeneity by running General Methods of Moments (GMM) estimation methods.

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

537

5. Econometric results We first run industry-regressions by group of developed and developing countries. In a second step, we break the developing countries group into four subgroups: East Asian, Asian, Latin American as well as Mediterranean countries.8 As noted above, we deal with unbalanced panel data. We follow Davis (2002) and Wansbeek and Kapteyn (1989) who set up a method of within-type transformation which produces two way fixed effects estimators consistent with incomplete panels. We shall call this method WKD in what follows.9 The basic intuition is that the estimator of such a transformation is similar to that obtained by adding a dummy exogenous variable to the set of regressors for each missing observation. Practically however, it is far better to use their method of transformation than to add a big number of dummies that would have lead otherwise to impractical computation. Following Davis (2002), and considering matrix notation, the two way fixed effects in an unbalanced panel can be written: Y = Xβ + Δiμi + Δtμt + u where β, μi and μt are parameters in the fixed effects model.10 Assuming N the total number of observations, Ni and Nt denoting the number of observed individuals and time periods for which data is available, then Δi and Δt are two (N ⁎ Ns) matrices, s = i, t, that represent respectively the set of indicator variables denoting observations on individuals (i) and time periods (t). Davis shows that the above model could be transformed into a within-type relation: QUY = QUXβ + QUu, with Q = Q1 − P2, Q1 = I − Δt ⁎ [(Δt)′ ⁎ Δt]− ⁎ (Δt)′ and P2 = Q1 ⁎ Δi ⁎ [(Δi)′ ⁎ Q1 ⁎ Δi]− ⁎ (Δi)′ ⁎ Q1. In what follows, all our regressions shall be based on this within transformation. We also ran GMM regressions based on these transformed variables with instruments transformed in the same manner. We therefore provide WKD type estimates in the tables of results hereafter and when necessary, Instrumental variables (IV) and General Methods of Moments (GMM) on that withintype relation. Given the similarity between IV-results and GMM-ones, we preferred reporting the latter each time it was convenient. We test for exogeneity of explanatory variables as well as instruments, by running systematically Durbin–Wu–Hausman and Over-identification tests.11 When the p-value relative to DWH test exceeds 0.05, we do not reject the hypothesis that the explanatory variables are exogenous to the model and choose the Within model. However, when the DWH p-value is lower than 0.05, we choose the GMM model and present estimates where the chosen instruments are orthogonal to the residual allowing the equation to be over-identified (see p-values from over-identification test results). The instruments used are mean national Wage at (t-2) and (t-3) (i.e. lagged twice or three times), labour Productivity (t-2) and (t-3), Market Shares in Domestic, Developing and Industrialized markets at (t-2) and (t-3). Tables 6 and 7 report results for the developed and developing countries' panels. Some common observations could be derived from these two tables. First, as it is expected, the industry average wage and the productivity differential variables have significant positive effects on real 8 We preferred an industry-type specification instead of a country-type specification because from the point of view of industrial economics, market structures should be much more industry-specific than country specific (see for instance the introduction chapter in Sutton, 1991). However, we account partly for country features since we run regressions by groups with comparable characteristics. Besides, we use Within-type methods that account systematically for permanent country heterogeneity captured by the fixed effects. 9 Davis generalizes actually Wansbeek and Kapteyn's method to three-way, four or higher order error components. 10 The programs of variables transformation have been constructed using the matrix language (IML module) in SAS, and are available from the authors. We follow the method of Davis that appears to be more flexible and more general than that of Wansbeek and Kapteyn. 11 See Davidson and Mac Kinnon (1994) for more details on these tests.

538

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Table 5 Descriptive statistics on wages and market share variables for 1994 (cross industry means and standard errors) Country

Wage/employee (US $)

R. Dom. MS

R. MS in Ind

R. MS in Dev

CTY

Mean

Std

Mean

Std

Mean

Std

Mean

Std

Austria Bolivia Canada Chile Hong Kong Colombia Costa Rica Cyprus Denmark Ecuador Egypt El Salvador Finland Gabon Germany Greece Guatemala Honduras Iceland India Indonesia Italy Japan Jordan Rep. of Korea Kuwait Lithuania Macau Malaysia Malta Mauritius Mexico Morocco Netherlands New Zealand Norway Panama Peru Philippines Singapore South Africa Spain Sri Lanka Sweden Thailand Trin. and Tob. Tunisia Turkey United Kingdom U.S. America

34,982.21 2988.99 28,200.48 9137.98 15,543.18 4125.70 3638.25 12,019.72 33,264.00 2990.25 2332.70 5911.73 24,673.77 12,806.89 35,946.03 14,142.51 348.16 1809.84 27,118.41 1285.61 1033.40 32,530.36 45,615.47 3076.57 14,973.58 21,998.98 1113.66 5838.71 4631.92 10,811.24 3226.13 9087.73 3750.61 36,484.56 23,123.96 31,351.46 8890.78 5784.47 4207.46 17,788.66 9464.94 19,881.79 802.10 25,427.13 4331.79 6755.11 5575.21 8343.82 23,858.64 32,047.14

1829.53 373.29 801.28 709.79 432.26 194.99 189.93 810.77 1169.38 357.96 238.34 949.79 624.21 1966.63 690.34 610.20 36.25 141.56 621.73 82.22 67.32 791.36 1857.71 182.97 504.07 3427.27 40.72 308.66 321.49 239.90 313.66 375.44 205.33 1107.22 2486.04 1088.77 1667.13 567.94 823.21 441.61 386.71 822.87 53.49 291.95 781.79 788.11 463.92 539.32 714.41 1082.67

0.36 0.67 0.38 0.57 0.14 0.63 0.53 0.43 0.28 0.64 0.67 0.36 0.36 0.46 0.48 0.56 0.55 0.62 0.58 0.81 0.50 0.44 0.81 0.45 0.59 0.63 0.21 0.43 0.26 0.39 0.30 0.38 0.57 − 0.37 0.42 0.51 0.55 0.66 0.55 0.35 0.63 0.56 0.46 0.29 0.55 0.39 0.58 0.68 0.47 0.73

0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.10 0.03 0.03 0.05 0.03 0.06 0.03 0.12 0.03 0.03 0.03 0.01 0.03 0.03 0.02 0.04 0.03 0.06 0.43 0.04 0.04 0.03 0.05 0.05 0.04 1.83 0.04 0.06 0.04 0.03 0.04 0.02 0.07 0.03 0.04 0.03 0.03 0.04 0.04 0.02 0.04 0.02

0.001406 0.000000 0.007500 0.000443 0.000097 0.000057 0.000012 0.000011 0.002502 0.000008 0.000049 0.000001 0.002256 0.000000 0.011255 0.000351 0.000001 0.000002 0.000035 0.000083 0.000795 0.007099 0.002921 0.000000 0.001217 0.000000 0.000172 0.000003 0.001342 0.000034 0.000578 0.000952 0.000134 0.009847 0.000302 0.001198 0.000001 0.000097 0.000158 0.000012 0.000315 0.001659 0.000018 0.003098 0.000355 0.000115 0.000027 0.000139 0.005263 0.002714

0.000204 0.000000 0.001396 0.000250 0.000049 0.000018 0.000004 0.000007 0.001017 0.000002 0.000021 0.000001 0.000616 0.000000 0.001134 0.000278 0.000001 0.000001 0.000009 0.000087 0.000510 0.001595 0.000659 0.000000 0.000334 0.000000 0.000028 0.000007 0.000438 0.000025 0.000243 0.000272 0.000100 0.001611 0.000094 0.000620 0.000001 0.000065 0.000066 0.000009 0.000620 0.000392 0.000015 0.000466 0.000171 0.000057 0.000005 0.000044 0.000749 0.000384

0.000113 0.000029 0.000192 0.001080 0.001394 0.000121 0.000033 0.000007 0.000018 0.000028 0.000022 0.000053 0.000250 0.000007 0.002339 0.000086 0.000105 0.000002 0.000000 0.000211 0.002366 0.002762 0.013015 0.000112 0.004485 0.000015 0.000002 0.000008 0.004049 0.000010 0.000008 0.000121 0.000090 0.000605 0.000608 0.000099 0.000006 0.000178 0.000095 0.000366 0.000277 0.000416 0.000005 0.000336 0.000955 0.000071 0.000013 0.000309 0.001466 0.006860

0.000031 0.000012 0.000094 0.000544 0.000397 0.000046 0.000011 0.000004 0.000007 0.000006 0.000007 0.000023 0.000068 0.000005 0.000593 0.000039 0.000031 0.000001 0.000000 0.000042 0.001533 0.000820 0.002503 0.000041 0.001302 0.000017 0.000002 0.000005 0.000806 0.000014 0.000003 0.000066 0.000061 0.000119 0.000283 0.000040 0.000002 0.000135 0.000047 0.000224 0.000126 0.000078 0.000001 0.000079 0.000385 0.000022 0.000006 0.000184 0.000525 0.001196

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

539

Table 5 (continued ) Country

Wage/employee (US $)

R. Dom. MS

CTY

Mean

Mean

Uruguay Venezuela Zimbabwe

6878.32 4533.45 2758.21 39 13,381.96

Mean

Std 408.57 249.95 123.69 0.49

R. MS in Ind

R. MS in Dev

Std

Mean

Std

Mean

Std

0.55 0.64 0.54

0.03 0.03 0.03

0.000029 0.000055 0.000024

0.000024 0.000040 0.000008

0.000220 0.000170 0.000005

0.000056 0.000049 0.000002

0.00126

0.00087

R. Dom MS = Relevant Domestic Market Share; R. MS in Ind = Relevant Market Share in the Ind market. R. MS in Dev = Relevant Market Share in the Dev market.

industry wage per employee in most of the industries for the two groups of countries. However, notice that the productivity differential effects are of similar magnitude whereas the coefficients on the average wage are usually higher in the developing than in industrialized countries, even when accounting for standard errors of the estimates. Thus, the alternative wage constituted by these two variables seems to play a greater role in affecting industry wages in the less developed than in rich countries. The second important observation is that the β′ coefficients on foreign market share variables (in absolute terms), appear to be systematically higher than those on domestic market shares for all of the industries. Following our theoretical analysis, this result is consistent with spatial differentiation within the two groups of foreign destinations. For illustration, Table 5 reports some descriptive statistics among which one can observe the very small market share of each exporting country held on each destination observed in the UNIDO data (i.e. developing countries' and developed countries' destinations as a whole). However, for most of the exporting countries, the market that counts constitutes a small part of those observed markets. Put differently, perceived demand faced by a typical exporter is much smaller due to spatial differentiation, which ends up increasing significantly the estimated parameter. Nevertheless, the parameters relative to domestic shares βdom ′ , are mainly between 0 (non significant) and 0.5, which is consistent with our theory when the market is much smaller and thus much more integrated in space. In the developed country sample (i.e. Table 6), the parameters related to market shares held in rich countries βind ′ , are in a large majority of cases positive and statistically significant for about 15 industries. Thus, in half of the industries, the test suggests that there are rents captured through penetration of other rich countries' markets which are redistributed back to home employees. However, only 10 industries (one third) benefit from the same outcome when selling to the domestic market. Outside markets, within the OECD, appear then to be more profitable to exporters and their employees than domestic markets are. The evidence of an extraction and distribution of rents is more limited when selling to developing markets (in 8 industries, β Dev N 0 and is statistically significant). It is interesting to note here, that the corresponding positive impact on wages prevails in industries where the market power of rich countries' producers is potentially large (chemicals or scientific and professional equipment for instance). On the opposite, the insignificant impact appears in industries usually known to be very competitive (e.g. footwear or furniture).12 This is still perfectly consistent with our theoretical story. The correlation between our parameter estimates and the degree of market power is more illustrated however, in the next section. 12

See Oliveira Martins et al. (1996) for a classification of industries in terms of the form of market structures related to these industries.

540

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Table 6 Estimation results at the industry level for developed countries Industry

Av. wage Pty. dif

Beverages

0.213⁎⁎⁎ 0.005⁎⁎ 0.04 0.002 0.424⁎⁎⁎ 0.01⁎⁎⁎ 0.036 0.002 0.388⁎⁎⁎ – 0.014⁎⁎⁎ 0.043 0.002 0.645⁎⁎⁎ 0.017⁎⁎⁎ 0.031 0.002 0.372⁎⁎⁎ 0.002 0.031 0.002 0.879⁎⁎⁎ −0.002 0.11 0.013 1.069⁎⁎⁎ 0.015⁎⁎⁎ 0.017 0.004 0.722⁎⁎⁎ 0.009⁎⁎⁎ 0.045 0.002 0.767⁎⁎⁎ 0.012⁎⁎⁎ 0.017 0.003 0.963⁎⁎⁎ 0.029⁎⁎⁎ 0.055 0.004 1.461⁎⁎⁎ 0.025⁎⁎⁎ 0.021 0.006 0.095⁎⁎⁎ 0.011⁎⁎⁎ 0.034 0.002 0.888⁎⁎⁎ 0.003 0.043 0.002 0.338⁎⁎⁎ 0.03⁎⁎⁎ 0.085 0.007 0.883⁎⁎⁎ 0.016⁎⁎⁎ 0.023 0.004 0.477⁎⁎⁎ 0.02⁎⁎⁎ 0.133 0.005 0.591⁎⁎⁎ 0.004 0.04 0.003 1.375⁎⁎⁎ −0.001⁎ 0.057 0.0001 0.951⁎⁎⁎ 0.008 0.16 0.007 0.58⁎⁎⁎ 0.019⁎⁎⁎ 0.055 0.005 0.309⁎⁎⁎ −0.001 0.035 0.002 0.894⁎⁎⁎ 0.025⁎⁎⁎ 0.046 0.004 0.882⁎⁎⁎ 0.016⁎⁎⁎ 0.039 0.003 0.801⁎⁎⁎ 0.012⁎⁎⁎ 0.012 0.003 0.028 0.002⁎⁎⁎ 0.023 0.0001

Fab. metal prod. Food products

Footwear Furniture Glass and products Industrial chemicals Iron and steel Leather products Machinery, electric Machinery Misc. petrol. prod. Non-ferrous metals Other chemicals Other manuf. prod. Other non-metallic prod. Paper and products Petroleum refineries Plastic products Pottery/china Printing and publishing Professional and scient. Rubber products Textiles Tobacco

R. Dom share R. Share Ind R. Share Dev DWH Overid Meth βdom ′ β Ind ′ β Dev ′ P-val. P-val. 0.329⁎⁎⁎ 0.037 0.203⁎⁎⁎ 0.054 0.123 0.076 0.077⁎ 0.043 0.13⁎⁎⁎ 0.03 0.152 0.269 0.0001 0.002 0.062 0.058 −0.222⁎⁎ 0.089 0.003⁎ 0.002 −0.02 0.058 0.002 0.015 0.008⁎ 0.004 0.443⁎⁎⁎ 0.116 0.0001 0.002 0.006⁎ 0.003 0.005 0.06 0.006 0.014 −0.398⁎ 0.234 0.013 0.011 0.515⁎⁎⁎ 0.079 −0.006⁎⁎ 0.003 0.215⁎⁎⁎ 0.038 −0.074⁎ 0.039 −0.01 0.011

19.545⁎⁎ 8.047 37.58⁎⁎⁎ 9.672 21.962⁎⁎⁎

2.505 8.15 2.093 3.534 − 6.127

8.161 14.645 4.064⁎⁎⁎ 8.56 0.79 8.108 8.79⁎⁎⁎ 0.933 2.354 2.71 14.511 12.834⁎⁎ 26.895 5.352 0.518 5.792⁎⁎⁎ 0.794 1.692 5.544 0.125 3.852 1.221 −2.415 9.398⁎⁎ 3.054 3.685 −4.894⁎⁎ 6.012⁎⁎⁎ 2.102 1.644 −1.645 0.766 2.168 1.962 22.871⁎⁎⁎ − 132.231⁎⁎⁎ 6.362 21.238 1.183 1.114 1.354 0.941 20.831⁎⁎⁎ − 16.038 4.862 19.812 4.704⁎⁎⁎ − 0.831⁎ 1.595 0.457 18.629⁎ 6.021 9.755 9.228 2.605 − 1.928 2.062 6.167 − 20.286⁎⁎⁎ 14.72 4.036 14.994 − 40.14⁎ 5.57 22.562 18.445 3.229 7.449⁎⁎ 2.335 3.095 167.526⁎⁎⁎ −10.637 36.012 13.474 −2.341⁎ 3.409⁎⁎⁎ 1.209 1.045 16.919⁎⁎⁎ 18.269⁎⁎⁎ 2.134 5.368 4.795⁎ 2.722 2.566 5.236 4.502⁎⁎⁎ − 5.716 1.567 4.258

Obs

0.34

0.09

WKD 202

0.38

0.16

WKD 178

0.24

0.21

WKD 182

0.14

0.3

WKD 204

0.27

0.28

WKD 205

0.01

0.75

GMM 199

0.22

0.58

WKD 188

0.36

0.31

WKD 166

0.08

0.88

WKD 208

0.81

0.18

WKD 175

0.43

0.47

WKD 159

0.31

0.95

WKD 170

0.89

0.53

WKD 181

0.37

0.56

WKD 203

0.15

0.4

WKD 192

0.04

0.15

GMM 190

0.13

0.97

WKD 199

0.31

0.29

WKD 172

0.01

0.3

GMM 210

0.21

0.94

WKD 204

0.09

0.35

WKD 199

0.43

0.78

WKD 164

0.59

0.93

WKD 203

0.57

0.83

WKD 210

0.74

0.79

WKD 197

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

541

Table 6 (continued ) Industry

Av. wage Pty. dif

Total manufacturing Transport equipment Wearing apparel

0.614⁎⁎⁎ 0.022⁎⁎⁎ 0.09 0.033 0.003 0.064 0.699⁎⁎⁎ 0.059⁎⁎⁎ −0.393⁎⁎⁎ 0.092 0.009 0.107 0.632⁎⁎⁎ −0.005 −0.018 0.05 0.008 0.013 0.705⁎⁎⁎ 0.016⁎⁎⁎ 0.213⁎⁎⁎ 0.045 0.004 0.032

Wood products

R. Dom share R. Share Ind R. Share Dev DWH Overid Meth βdom ′ β Ind ′ β Dev ′ P-val. P-val. 16.615⁎⁎⁎ 1.59 −12.872⁎⁎⁎ 4.217 − 6.641 4.818 7.002⁎⁎⁎ 2.459

5.337 3.827 4.053 3.392 −4.03 3.937 25.127⁎⁎⁎ 7.168

Obs

0.87

0.26

WKD 165

0.75

0.47

WKD 195

0.01

0.77

GMM 180

0.9

0.81

WKD 195

Parameter estimates are in bold characters and standard errors in italics. ⁎⁎⁎, ⁎⁎ and ⁎ significant respectively at 1, 5 and 10%. Instruments used in GMM: Av. wage (t-2) and (t-3), App. Productivity (t-2) and (t-3), Market Shares in Domestic, Developing and Industrialized markets (t-2) and (t-3).

It remains that some small minority of β′ coefficients appear to be negative and significant. As explained earlier, this could be due to a reverse causality: an increase in wages should hamper competitiveness, leading to a reduction in market shares. This result is consistent with nonefficient bargaining practices between unions and employers in the corresponding industries. In that case, unions first determine wages, leaving employers determining domestic and foreign sales in a second step. High fixed wages might then lead to less competitiveness on each market. Although we have accounted for these potential endogeneity problems by conducting DWH and over-identification tests followed by GMM estimations, these methods were unable in some cases to sweep out the reverse causation and help us retrieving the expected positive sign of the coefficient. Finally, another possible explanation of that negative effect encountered in this limited number of industries could be provoked by a mismatch between the theoretical framework we use and evidence. Table 7 presents results relative to the developing countries' panel. We find eleven industries where wages are positively and significantly linked to domestic market shares. Put differently, in roughly one third of the industries, results are consistent with positive rents that are shared with employees due to an increase in domestic market shares. For the rest of the industries (two thirds), no significant link is found. Besides, only in 5 (resp. 4) industries exports to Ind (resp. Dev) seem to be shifting and redistributing rents through higher wages. All these results suggest that firms from developing countries have little market power on their own markets and especially on foreign markets that enables them to extract rents and then share them with their employees. Moreover, we must stress again a negative and statistically significant relation in the Footwear industry between foreign sales on OECD markets and wages. This outcome is also observed in five industries when selling on other developing markets. In order to better interpret the results obtained for the developing countries, we conduct more disaggregated analysis hereafter. We break the developing countries' sample into 4 subgroups (Table 8): Mediterranean economies, Latin America, Asia and East Asia.13 The results of this exercise are briefly summed up in what follows.

13 Observations related to African and Middle Eastern countries like Ghana or Kuwait used in the pooled sample regressions are not part of these 4 sub-groups. Too few observations were available for countries from central and southern Africa in order to be exploited as a group. The same case applies for Middle eastern countries.

542

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Table 7 Estimation results at the industry level for developing countries Industry

Av. wage Pty. dif

0.96⁎⁎⁎ 0.191 Fab. metal prod. 0.833⁎⁎⁎ 0.019 Food products 0.854⁎⁎⁎ 0.022 Footwear 0.676⁎⁎⁎ 0.105 Furniture 0.589⁎⁎⁎ 0.021 Glass and 1.016⁎⁎⁎ products 0.041 Industrial 1.057⁎⁎⁎ chemicals 0.01 Iron and steel 1.022⁎⁎⁎ 0.053 Leather products 0.899⁎⁎⁎ 0.004 Machinery, 1.052⁎⁎⁎ electric 0.017 Machinery 1.083⁎⁎⁎ 0.047 Misc. petrol. 0.819⁎⁎⁎ prod. 0.066 Non-ferrous 0.807⁎⁎⁎ metals 0.036 Other chemicals 1.195⁎⁎⁎ 0.06 Other manuf. 0.641⁎⁎⁎ Prod. 0.005 Other non0.722⁎⁎⁎ metallic prod. 0.032 Paper and products 0.908⁎⁎⁎ 0.029 Petroleum 0.818⁎⁎⁎ refineries 0.118 Plastic products 0.895⁎⁎⁎ 0.043 Pottery/china 0.858⁎⁎⁎ 0.035 Printing and 0.774⁎⁎⁎ publishing 0.033 Professional 1.033⁎⁎⁎ and scient. 0.027 Rubber products 0.924⁎⁎⁎ 0.089 Textiles 0.811⁎⁎⁎ 0.013 Tobacco 0.109⁎⁎⁎ 0.037

Beverages

0.013⁎⁎ 0.005 0.009⁎⁎⁎ 0.001 −0.001 0.001 0.01 0.011 0.007⁎⁎⁎ 0.001 0.018⁎⁎⁎ 0.003 0.019⁎⁎⁎ 0.007 0.009⁎⁎⁎ 0.003 0.008⁎⁎⁎ 0.001 0.007⁎⁎⁎ 0.001 −0.005 0.006 0.0001 0.001 −0.002⁎⁎ 0.001 0.015⁎⁎⁎ 0.004 −0.001 0.001 0.014⁎⁎⁎ 0.001 0.009⁎⁎⁎ 0.002 0.0001 0.0001 0.003 0.003 0.014⁎⁎⁎ 0.002 0.002⁎⁎⁎ 0.001 0.009⁎⁎⁎ 0.003 0.018⁎⁎⁎ 0.006 0.01⁎⁎⁎ 0.004 0.002⁎⁎⁎ 0.001

R. Dom share βdom′

R. Share Ind βInd′

R. Share Dev βDev′

DWH Overid Meth P-val. P-val.

− 0.007 − 153.114 34.336 0.03 0.258 161.602 23.557 0.118⁎⁎⁎ 3.068 1.761 0.43 0.029 8.479 5.029 0.061⁎ − 6.654 − 9.905 0.85 0.032 35.695 7.97 0.468⁎⁎ 4.741 − 24.66 0.01 0.187 7.089 50.727 0.076⁎⁎⁎ 8.533⁎⁎⁎ 4.158 0.07 0.024 2.996 5.613 0.0001 − 1.918 − 9.564 0.3 0.004 8.814 7.601 0.0001 − 8.834 7.296 0.01 0.0001 7.925 4.459 0.06 − 2.446 0.56 0.06 0.057 18.002 7.853 − 0.126⁎⁎⁎ − 25.268⁎⁎⁎ 1.503 0.4 0.027 6.843 1.008 0.0001 0.346⁎⁎ 0.239 0.81 0.0001 0.164 0.196 0.011⁎⁎⁎ − 3.48 − 7.704⁎ 0.01 0.004 4.932 3.933 0.034 − 16.886 − 90.826⁎⁎ 0.66 0.022 11.23 41.506 0.001 0.066 0.329 0.24 0.002 13.041 1.78 − 0.027 23.068 13.934⁎⁎⁎ 0.01 0.097 18.981 3.774 0.0001 − 2.367⁎⁎⁎ 1.448⁎⁎⁎ 0.76 0.0001 0.702 0.332 0.005⁎⁎⁎ 43.492 − 21.759⁎⁎⁎ 0.19 0.001 66.021 4.83 0.199⁎⁎⁎ 32.941⁎ 12.825⁎⁎⁎ 0.07 0.038 18.433 4.167 0.013⁎⁎⁎ 19.093 − 23.132⁎⁎ 0.72 0.004 16.563 10.45 − 0.132⁎⁎⁎ − 12.576 − 12.992 0.01 0.024 20.944 15.718 0.001 2.589 0.882 0.13 0.002 6.693 11.743 0.211⁎⁎⁎ 107.914 23.339 0.65 0.052 80.194 32.045 0.002 − 0.832 3.408 0.41 0.002 0.853 2.555 − 0.01 30.642⁎ 8.104⁎⁎ 0.01 0.055 17.217 3.437 0.01 57.404 2.059⁎ 0.01 0.014 45.065 1.225 0.085⁎⁎⁎ 49.756 − 36.427⁎⁎⁎ 0.77 0.026 36.134 10.794

Obs

0.41

GMM 371

0.3

WKD 429

0.71

WKD 445

0.46

GMM 288

0.92

WKD 380

0.25

WKD 304

0.61

GMM 347

0.35

WKD 253

0.42

WKD 329

0.96

WKD 392

0.14

GMM 354

0.41

WKD 148

0.9

WKD 236

0.84

GMM 397

0.22

WKD 388

0.74

WKD 366

0.98

WKD 382

0.98

WKD 180

0.32

GMM 358

0.59

WKD 281

0.54

WKD 405

0.21

WKD 252

0.31

GMM 341

0.99

GMM 432

0.41

WKD 236

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

543

Table 7 (continued ) Industry

Av. wage Pty. dif

Total manufacturing Transport equipment Wearing apparel

1.005⁎⁎⁎ 0.032 1.031⁎⁎⁎ 0.028 0.604⁎⁎⁎ 0.012 0.722⁎⁎⁎ 0.01

Wood products

0.018⁎⁎⁎ 0.003 0.014⁎⁎⁎ 0.002 0.009⁎⁎⁎ 0.002 0.009⁎⁎⁎ 0.001

R. Dom share βdom′ 0.005 0.029 0.097⁎⁎⁎ 0.029 0.007 0.005 0.002 0.003

R. Share Ind βInd′ − 0.392 11.543 4.959⁎⁎ 2.414 0.408 0.56 − 5.593 10.085

R. Share Dev βDev′ − 1.822 1.22 − 4.077 3.181 −0.112 4.264 − 1.509 1.474

DWH Overid Meth P-val. P-val.

Obs

0.05

0.55

GMM 430

0.12

0.52

WKD 381

0.93

0.84

WKD 270

0.87

0.58

WKD 412

Parameter estimates are in bold characters and standard errors in italics. ⁎⁎⁎, ⁎⁎ and ⁎ significant respectively at 1, 5 and 10%. Instruments used in GMM: Av. wage (t-2) and (t-3), App. Productivity (t-2) and (t-3), Market Shares in Domestic, Developing and Industrialized markets (t-2) and (t-3).

From Table 9, we see that in Mediterranean Countries, domestic market shares are associated with positive and significant effects in 11 industries. As far as foreign market shares held on OECD markets are concerned, the same result is observed in only 4 industries. A negative and significant parameter, to be interpreted as a reverse causation (absence of efficient bargaining and/ or industries characterized by tight competition), is however obtained in 8 industries. Leather, Other non metallic and Textile products which are usually considered to be traditional industries of specialization of this region, are examples of such an outcome. In Latin America (see Table 10), industry wages appear often to be positively linked to domestic market shares as well (in 15 industries) but an increase in market shares held on foreign markets is not systematically associated with higher wages. On OECD markets, and among significant parameters, four are positive while only one is negative. On developing countries' markets, seven are positive and significant and seven are negative. The Latin American case is to some extent comparable to the Mediterranean case. In total, rents gained on the domestic market for these regions can turn into higher wages, whereas there is no rent to capture by unions when selling to foreign markets. This is consistent with the traditional story stating that openness might lead to a specialization in competitive and thus non profitable industries in some developing economies. In East Asia and Asia (Tables 11 and 12), the coefficients on the domestic market share appear to be positive and significant for only one quarter of the industries. Hence, compared to other regions, the evidence of rent sharing is much more limited for Asian countries. Moreover, the number of industries where the βInd ′ and the βDev ′ are positive and significant is low (between 4

Table 8 The four group of developing countries considered Group

Related countries

Mediterranean countries Asian countries East (and South) Asian countries Latin American countries

Egypt, Morocco, Tunisia, Turkey, Cyprus, Malta Bangladesh, Madagascar, India, Indonesia, Pakistan, Sri Lanka, Nepal Macau, Hong Kong (China), Singapore, Korean Republic, Malaysia, Thailand, Philippines Bolivia, Chile, Colombia, Nicaragua, Argentina, Costa Rica, Ecuador, Salvador, Guatemala, Honduras, Mexico, Panama, Peru, Venezuela, Uruguay

544

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Table 9 Estimation results at the industry level for Mediterranean countries Industry

Av. wage Pty. dif

Beverages

0.186⁎⁎⁎ 0.011⁎⁎⁎ 0.208⁎⁎ 111.976 − 125.361 0.57 0.042 0.003 0.092 198.122 150.851 0.675⁎⁎⁎ 0.023⁎⁎⁎ 0.316⁎⁎⁎ 111.386 27.383 0.43 0.066 0.007 0.116 89.055 26.258 0.739⁎⁎⁎ 0.005 0.019 211.315 − 12.13 0.2 0.099 0.004 0.202 224.243 72.974 0.301⁎⁎⁎ 0.022⁎⁎⁎ 0.615⁎⁎⁎ 24.744⁎⁎ 43.071⁎⁎⁎ 0.17 0.039 0.005 0.088 11.384 10.62 0.413⁎⁎⁎ 0.016⁎⁎⁎ 0.479⁎⁎⁎ 50.914⁎⁎⁎ 56.449⁎⁎⁎ 0.79 0.062 0.006 0.102 14.768 18.269 0.914⁎⁎⁎ 0.02⁎⁎⁎ 0.044 11.832 − 49.633⁎⁎⁎ 0.75 0.066 0.007 0.05 25.59 11.453 1.088⁎⁎⁎ 0.0001 0.001 − 2.512 − 1.334 0.89 0.029 0.005 0.003 12.703 15.725 1.347⁎⁎⁎ 0.029⁎ 0.027 − 124.367 −10.927 0.9 0.23 0.017 0.27 102.89 16.422 0.679⁎⁎⁎ 0.012 0.248⁎ − 67.886⁎⁎ 18.826 0.79 0.053 0.01 0.14 33.817 12.01 1.103⁎⁎⁎ 0.013⁎⁎⁎ 0.001 5.561 −0.873 0.84 0.05 0.002 0.002 5.187 5.327 1.096⁎⁎⁎ 0.031⁎⁎⁎ 0.05 12.112 2.308 0.79 0.045 0.008 0.048 7.825 9.575 1.064⁎⁎⁎ − 0.001 0.08 − 12.567 −160.656 0.58 0.176 0.001 0.076 61.922 173.615 1.118⁎⁎⁎ 0.007 0.03⁎ − 1.294 2.544 0.41 0.087 0.009 0.016 35.248 4.01 1.184⁎⁎⁎ 0.025⁎⁎⁎ − 0.129 − 28.386 1.812 0.55 0.087 0.007 0.125 66.169 90.301 0.755⁎⁎⁎ 0.025⁎⁎⁎ 0.002 − 18.046⁎ 4.544 0.68 0.028 0.007 0.004 9.525 4.273 0.661⁎⁎⁎ 0.026⁎⁎⁎ 0.004 − 257.111⁎ 3.232 0.84 0.074 0.005 0.009 155.588 31.556 0.865⁎⁎⁎ 0.024⁎⁎⁎ 0.102 18.524 − 65.327⁎⁎⁎ 0.26 0.076 0.006 0.086 34.088 23.499 0.493 0.0001 0.006⁎⁎⁎ 41.956 − 6.537 0.87 0.329 0.001 0.002 98.119 102.54 0.315⁎⁎⁎ 0.032⁎⁎⁎ − 0.31 184.611 − 153.427 0.53 0.068 0.007 0.319 154.595 164.228 0.531⁎⁎⁎ 0.008 0.008 − 20.802 − 4.802 0.66 0.059 0.006 0.019 33.005 24.043 0.254⁎⁎⁎ 0.005 1.165⁎⁎⁎ 1849.714⁎⁎⁎ 178.622 0.6 0.087 0.005 0.295 644.042 162.369 1.111⁎⁎⁎ 0.018⁎ 0.004 − 12.863⁎⁎ 15.794 0.91 0.061 0.01 0.003 6.111 11.43 0.667⁎⁎⁎ 0.013 0.422⁎⁎⁎ − 131.442⁎⁎⁎ 145.798⁎⁎⁎ 0.52 0.143 0.01 0.132 30.888 41.982 0.773⁎⁎⁎ 0.0001 0.004 − 60.889⁎⁎⁎ 1.563 0.36 0.018 0.005 0.107 23.521 8.829 1.288⁎⁎⁎ 0.009⁎⁎ − 0.064 − 107.437⁎ 8.069 0.67 0.135 0.004 0.09 64.556 37.198

Fab. metal prod. Food products Footwear Furniture Glass and products Industrial chemicals Iron and steel Leather products Machinery, electric Machinery Misc. petrol. prod. Non-ferrous metals Other chemicals Other manuf. Prod. Other nonmetallic prod. Paper and products Petroleum refineries Plastic products Pottery/china Printing and publishing Professional and scient. Rubber products Textiles Tobacco

R. Dom share βdom ′

R. Share Ind β ′Ind

R. Share Dev DWH β ′Dev P-val.

Overid P-val.

Meth

Obs

0.83

WKD 53

0.46

WKD 67

0.25

WKD 75

0.22

WKD 60

0.05

WKD 63

0.77

WKD 50

0.13

WKD 60

0.23

WKD 28

0.71

WKD 49

0.59

WKD 70

0.37

WKD 70

0.62

WKD 28

0.81

WKD 31

0.61

WKD 55

0.68

WKD 66

0.54

WKD 62

0.19

WKD 51

0.19

WKD 37

0.54

WKD 42

0.8

WKD 45

0.93

WKD 61

0.7

WKD 47

0.13

WKD 45

0.73

WKD 65

0.18

WKD 43

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

545

Table 9 (continued ) Industry

Av. wage Pty. dif

Total manufacturing Transport equipment Wearing apparel

0.918⁎⁎⁎ 0.014⁎⁎⁎ 0.036 0.003 0.975⁎⁎⁎ 0.022⁎⁎⁎ 0.09 0.008 0.384⁎⁎⁎ − 0.012⁎⁎⁎ 0.027 0.002 0.737⁎⁎⁎ 0.014⁎⁎⁎ 0.038 0.002

Wood products

R. Dom share βdom ′

R. Share Ind β ′Ind

0.101⁎⁎ 0.043 0.076 0.091 0.123⁎⁎ 0.051 0.128⁎⁎ 0.052

0.504 14.433 17.678 13.042 26.276⁎⁎ 12.228 − 46.463⁎⁎ 23.386

R. Share Dev DWH β ′Dev P-val. −10.716 13.607 −10.226 14.158 − 11.411 13.441 9.463 8.84

Overid P-val.

Meth

Obs

0.42

0.74

WKD 61

0.32

0.88

WKD 69

0.6

0.54

WKD 49

0.87

0.37

WKD 63

Parameter estimates are in bold characters and standard errors in italics. ⁎⁎⁎, ⁎⁎ and ⁎ significant respectively at 1, 5 and 10%. Instruments used in GMM: Av. wage (t-2) and (t-3), App. Productivity (t-2) and (t-3), Market Shares in Domestic, Developing and Industrialized markets (t-2) and (t-3).

and 6). These countries do not usually seem to extract rents from selling to foreign Ind and Dev markets. Once again, in some industries, the estimated parameter appears to be negative. Noteworthy, the coefficients on average wage are usually higher in these two samples of countries than for other considered samples. This suggests that what drives most Asian industry wages are effects that pass through the alternative wage. In the long term, openness might affect alternative wages through general equilibrium mechanisms but this is beyond the objective of our study here that is more trying to assess short or medium run effects of trade on wages. 6. Do the β's really inform on labor and product market imperfections? Most of the rent sharing studies are conducted on micro data levels, in general, where one can account for most of the determinants of the competitive wages (i.e. workers and/or firms' characteristics) and thus, be able to isolate rent sharing effects. However, as our study is based only on industry data, we run the risk of not isolating potential rent-sharing behavior through our β estimates. We tend to show in this section however, that these parameters are really capturing what the theory wants them to, that is: the double existence of rents and unions in case β N 0 or the absence of at least one of those for β = 0. We emphasize two sets of arguments related first, to the methodology undertaken and second, to the consistency of the β parameters with empirical results and indexes found in the literature. 6.1. The methodology argument 1/ First, although usually the relation is expected to be negative between market shares and real wages, due to the fact that higher wages harm competitiveness,14 we find a positive effect however in most of cases between these two variables. One of the few reasons that could explain this relation when productivity is already accounted for, is precisely the extraction and distribution of rents from selling to a particular market.

14

Recall that this could be due to the fact that contracts between unions and employers are not efficient, or the result of other wage setting policies (minimum wages, etc…).

546

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Table 10 Estimation results at the industry level for Latin American countries Industry

Av. wage Pty. dif

Beverages

0.444⁎⁎⁎ 0.069 0.686⁎⁎⁎ 0.026 0.794⁎⁎⁎ 0.032 0.419⁎⁎⁎ 0.025 0.57⁎⁎⁎ 0.025 0.901⁎⁎⁎ 0.068 1.129⁎⁎⁎ 0.029 0.708⁎⁎⁎ 0.107 0.645⁎⁎⁎ 0.026 1.062⁎⁎⁎ 0.024 0.98⁎⁎⁎ 0.023 0.707⁎⁎⁎ 0.082 0.739⁎⁎⁎ 0.045 1.029⁎⁎⁎ 0.052 0.703⁎⁎⁎ 0.019 0.631⁎⁎⁎ 0.042 0.888⁎⁎⁎ 0.057 0.188 0.139 0.764⁎⁎⁎ 0.047 0.77⁎⁎⁎ 0.046 0.803⁎⁎⁎ 0.043 0.965⁎⁎⁎ 0.04 0.861⁎⁎⁎ 0.052 0.785⁎⁎⁎ 0.008 0.025 0.081

Fab. metal prod. Food products Footwear Furniture Glass and products Industrial chemicals Iron and steel Leather products Machinery, electric Machinery Misc. petrol. prod. Non-ferrous metals Other chemicals Other manuf. Prod. Other nonmetallic prod. Paper and products Petroleum refineries Plastic products Pottery/china Printing and publishing Professional and scient. Rubber products Textiles Tobacco

R. Dom share βdom ′

R. Share Ind R. Share Dev DWH β′Ind β′Dev P-val.

0.0001 0.505⁎ 117.808 53.707 0.002 0.305 120.129 88.984 0.005⁎⁎⁎ 0.141⁎⁎⁎ 4.079 10.752 0.001 0.045 8.16 16.301 − 0.003⁎⁎ 0.095⁎⁎ − 17.925 70.876 0.001 0.043 49.255 48.703 0.006⁎⁎⁎ 0.28⁎⁎⁎ 9.899 − 65.93⁎⁎⁎ 0.001 0.047 6.363 11.727 0.006⁎⁎⁎ 0.019 1.699 − 2.645 0.001 0.023 2.812 27.574 0.017⁎⁎⁎ 0.076 11.84 20.508 0.004 0.05 11.292 31.947 0.02⁎⁎⁎ 0.004 10.181 − 12.448 0.005 0.003 11.521 17.491 0.006⁎ 0.245⁎⁎⁎ −1.15 − 35.045⁎ 0.003 0.094 27.369 20.326 0.002 − 0.104 −8.691 9.641 0.002 0.07 16.176 7.656 0.001 0.0001 0.62⁎⁎⁎ 11.377⁎⁎⁎ 0.002 0.0001 0.17 1.557 − 0.002 − 0.058⁎⁎ 2.958 23.208⁎⁎⁎ 0.002 0.024 2.94 4.406 0.001 0.047 −11.921 − 59.679 0.001 0.048 16.341 93.782 − 0.001 0.001 −1.328 − 1.296 0.001 0.002 16.221 2.854 0.009⁎⁎⁎ 0.258⁎⁎⁎ − 10.26 131.963⁎⁎⁎ 0.002 0.098 28.52 41.697 − 0.002⁎⁎ 0.0001 −1.388 0.718 0.001 0.003 1.396 1.948 0.005⁎⁎⁎ 0.016⁎ 3.34 − 81.372⁎⁎ 0.002 0.008 92.672 34.219 0.007⁎⁎ 0.235⁎⁎⁎ 35.591 16.835 0.003 0.074 38.034 18.447 − 0.001 0.024 5.989 37.693 0.001 0.017 15.738 63.879 0.008⁎⁎⁎ 0.168⁎⁎ 7.945 87.221 0.001 0.075 10.429 100.983 0.014⁎⁎⁎ − 0.013 −4.324 − 7.066 0.003 0.014 6.393 15.221 0.004⁎⁎ 0.202⁎⁎ 28.992 143.556⁎⁎⁎ 0.002 0.081 88.645 53.869 0.011⁎⁎⁎ 0.006 −2.632⁎⁎ 18.465⁎⁎ 0.004 0.004 1.336 7.23 0.013⁎⁎⁎ 0.39⁎⁎⁎ 11.591 45.914⁎ 0.002 0.05 19.579 27.655 0.004⁎⁎⁎ 0.101⁎ 12.315 − 16.403⁎⁎ 0.001 0.056 8.791 7.686 0.0001 0.351⁎⁎⁎ 160.438⁎⁎⁎ − 152.31⁎⁎⁎ 0.0001 0.135 49.167 55.494

Overid P-val.

Meth

Obs

0.27

0.38

WKD 154

0.79

0.39

WKD 166

0.65

0.52

WKD 175

0.24

0.12

WKD 139

0.65

0.13

WKD 145

0.77

0.42

WKD 125

0.86

0.5

WKD 148

0.35

0.53

WKD 104

0.23

0.65

WKD 128

0.92

0.64

WKD 141

0.86

0.56

WKD 142

0.86

0.64

WKD

0.2

0.72

WKD 120

0.37

0.71

WKD 151

0.7

0.51

WKD 153

0.46

0.39

WKD 150

0.6

0.91

WKD 142

0.78

0.95

WKD

0.25

0.33

WKD 140

0.58

0.57

WKD 129

0.95

0.18

WKD 140

0.45

0.47

WKD 108

0.61

0.54

WKD 134

0.44

0.77

WKD 167

0.92

0.46

WKD

62

75

56

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

547

Table 10 (continued ) Industry

Av. wage Pty. dif

Total manufacturing Transport equipment Wearing apparel

0.914⁎⁎⁎ 0.017 0.991⁎⁎⁎ 0.041 0.649⁎⁎⁎ 0.01 0.627⁎⁎⁎ 0.015

Wood products

R. Dom share βdom ′

0.01⁎⁎⁎ 0.124⁎⁎⁎ 0.001 0.028 0.014⁎⁎⁎ 0.192⁎⁎⁎ 0.003 0.04 0.011⁎⁎⁎ − 0.003 0.001 0.004 0.002 − 0.002 0.001 0.004

R. Share Ind R. Share Dev DWH β′Ind β′Dev P-val. 3.786⁎⁎ 1.903 4.751⁎ 2.495 − 0.344 0.383 13 15.171

10.429⁎⁎ 4.86 − 19.021⁎ 9.906 0.198 3.439 − 22.881⁎⁎ 11.015

Overid P-val.

Meth

Obs

0.24

0.38

WKD 150

0.32

0.79

WKD 124

0.97

0.43

WKD 150

0.12

0.18

WKD 162

Parameter estimates are in bold characters and standard errors in italics. ⁎⁎⁎, ⁎⁎ and ⁎ significant respectively at 1, 5 and 10%. Instruments used in GMM: Av. wage (t-2) and (t-3), App. Productivity (t-2) and (t-3), Market Shares in Domestic, Developing and Industrialized markets (t-2) and (t-3).

2/ Our theoretical framework reveals that the β parameters relative to each market should be correlated. To see this, let us re-express the parameter β for each market j (∀j ∈ {dom, Ind, Dev}) as: 3 2 power 6 unions′ 6 zffl}|ffl{ bj ¼ 6 6 kdom 4

jij rej |ffl{zffl} wj

7 7 7 ¼ kdom fj 7 5

frims′ market power

j rj

where fj ¼ wij ije . First, one can easily deduce from the above relation that the βDom should be positively related to the two others βInd and βDev, as they all contain λi and assuming market structures are similar across markets (ζdom = ζInd = ζDev). Indeed, Table 13, based on developed countries sample, shows a very high and positive correlation between βDom and βInd estimators across industries (r = 0.70).15 However, the correlation appears to be negative but low between the developing market parameter βDev and the rest. These results suggest that market structures are similar across developed countries (i.e.: ζdom similar to ζInd), while they seem to be different between developed and developing markets. 3/ Our methodology provides a third reason for which our parameters are indeed informing about the theoretical story we tell. Appendix A shows how we properly aggregate the wage relation from the firm level to the industry level in order to be able to test it. The corresponding theoretical parameters turn out to contain the same information on market imperfections and power of unions than that obtained from the theoretical firm relation. 6.2. Further investigating the beta parameters 1/ We had access to 1990 OECD data on unions density and union coverage as two alternative indicators of unions' power (respectively identified by λd and λc). Union density expresses the percentage of workers belonging to trade unions while union coverage indicator represents the percentage of workers who are covered by collective agreement. These data are country specific. 15 Similar levels of correlations are obtained for each other group of countries. Results are provided by the authors upon request.

548

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Table 11 Estimation results at the industry level for East Asian countries Industry

Av. wage Pty. dif

Beverages

0.678⁎⁎⁎ 0.002 0.195 − 141.095 33.662 0.66 0.102 0.003 0.18 108.689 21.256 1.005⁎⁎⁎ 0.004⁎⁎ − 0.112⁎⁎⁎ − 58.965 − 8.198⁎⁎ 0.54 0.029 0.002 0.04 38.085 3.57 0.861⁎⁎⁎ 0.006⁎⁎⁎ − 0.068 131.1⁎ − 15.514⁎⁎ 0.94 0.042 0.001 0.129 69.854 6.815 0.602⁎⁎⁎ 0.004 0.403⁎⁎⁎ 4.866 12.513 0.89 0.036 0.003 0.071 3.683 12.313 0.782⁎⁎⁎ 0.01⁎⁎⁎ 0.259⁎⁎⁎ 26.68⁎⁎ − 2.52 0.66 0.042 0.001 0.073 12.457 6.428 1.188⁎⁎⁎ 0.017⁎⁎⁎ 0.146⁎⁎ 20.406 −10.419 0.57 0.122 0.004 0.064 35.633 10.788 1.081⁎⁎⁎ 0.013⁎⁎⁎ 0.001 0.486 3.626⁎⁎ 0.78 0.022 0.002 0.003 6.84 1.622 1.478⁎⁎⁎ 0.006 − 0.205⁎ 68.017⁎ − 1.935 0.7 0.123 0.005 0.111 40.614 11.508 0.85⁎⁎⁎ 0.007⁎ − 0.138⁎⁎⁎ − 27.819 2.342⁎ 0.2 0.029 0.004 0.046 19.701 1.237 1.092⁎⁎⁎ 0.008⁎⁎ 0.0001 0.411 −0.14 0.9 0.029 0.004 0.001 1.713 0.243 1.011⁎⁎⁎ − 0.006⁎⁎ 0.027⁎⁎ − 9.901⁎⁎⁎ −0.8 0.37 0.028 0.002 0.011 2.007 1.109 0.515⁎⁎⁎ 0.01⁎⁎⁎ 0.038 − 34.153 − 188.754⁎ 0.87 0.134 0.003 0.041 33.377 102.209 1.049⁎⁎⁎ 0.005⁎⁎⁎ 0.004 7.723 0.753 0.47 0.092 0.002 0.003 17.158 0.611 1.038⁎⁎⁎ 0.004 0.032 15.532 10.986⁎⁎⁎ 0.89 0.089 0.004 0.053 31.34 3.34 0.78⁎⁎⁎ 0.017⁎⁎⁎ − 0.001 − 1.507 1.021⁎⁎ 0.83 0.035 0.003 0.005 5.067 0.401 0.92⁎⁎⁎ 0.025⁎⁎⁎ 0.002 269.981⁎⁎ − 8.819 0.63 0.077 0.002 0.001 137.293 5.748 0.925⁎⁎⁎ 0.011⁎⁎⁎ 0.177⁎⁎⁎ − 5.615 16.496⁎⁎⁎ 0.72 0.046 0.002 0.065 26.204 5.32 0.258⁎⁎ − 0.001⁎⁎⁎ − 0.004 − 20.865 − 51.421⁎⁎⁎ 0.45 0.123 0.0001 0.008 30.688 12.758 0.877⁎⁎⁎ 0.005⁎⁎ 0.154⁎⁎⁎ − 54.059⁎⁎ − 17.263 0.4 0.039 0.002 0.058 25.364 10.548 0.635⁎⁎⁎ − 0.001 0.0001 − 28.834⁎⁎⁎ 32.554⁎⁎ 0.91 0.082 0.003 0.001 10.245 15.904 0.843⁎⁎⁎ 0.002⁎⁎⁎ 0.167⁎ 205.075 −50.412 0.72 0.068 0.001 0.099 249.896 41.162 0.915⁎⁎⁎ 0.015⁎⁎ − 0.003 2.069 0.765 0.39 0.029 0.007 0.002 2.246 1.75 0.656⁎⁎⁎ 0.011⁎⁎⁎ 0.004 13.545 9.962⁎⁎⁎ 0.69 0.063 0.001 0.019 22.904 2.64 0.79⁎⁎⁎ 0.006⁎⁎ − 0.115⁎⁎⁎ − 53.454 2.365 0.43 0.011 0.003 0.025 43.843 2.438 0.083 0.014⁎⁎⁎ 0.016 − 193.236⁎⁎ − 8.196 0.93 0.115 0.003 0.027 77.491 11.06

Fab. metal prod. Food products Footwear Furniture Glass and products Industrial chemicals Iron and steel Leather products Machinery, electric Machinery Misc. petrol. prod. Non-ferrous metals Other chemicals Other manuf. prod. Other nonmetallic prod. Paper and products Petroleum refineries Plastic products Pottery/china Printing and publishing Professional and scient. Rubber products Textiles Tobacco

R. Dom share βdom ′

R. Share Ind β Ind ′

R. Share Dev DWH β Dev ′ P-val.

Overid P-val.

Meth

Obs

0.66

WKD

91

0.8

WKD

97

0.49

WKD

97

0.75

WKD

42

0.33

WKD

89

0.65

WKD

68

0.5

WKD

65

0.37

WKD

70

0.57

WKD

67

0.84

WKD

98

0.51

WKD

75

0.66

WKD

48

0.65

WKD

59

0.83

WKD

90

0.2

WKD

90

0.57

WKD

97

0.62

WKD 100

0.62

WKD

51

0.09

WKD

82

0.38

WKD

52

0.62

WKD

99

1

WKD

66

0.23

WKD

79

0.27

WKD 101

0.7

WKD

83

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

549

Table 11 (continued ) Industry

Av. wage Pty. dif

Total manufacturing Transport equipment Wearing apparel

1.013⁎⁎⁎ 0.028 1.19⁎⁎⁎ 0.067 0.563⁎⁎⁎ 0.01 0.796⁎⁎⁎ 0.01

Wood products

R. Dom share βdom ′

0.008⁎⁎⁎ − 0.071⁎⁎ 0.002 0.034 0.004 0.02 0.003 0.059 0.0001 0.003 0.002 0.02 0.006⁎⁎⁎ − 0.227⁎⁎⁎ 0.001 0.034

R. Share Ind β Ind ′ − 3.351 7.034 33.852⁎⁎⁎ 10.784 0.709 4.895 29.815⁎ 16.994

R. Share Dev DWH β Dev ′ P-val. − 0.095 1.369 − 5.514⁎ 3.068 − 0.874 3.615 − 6.831⁎⁎⁎ 1.516

Overid P-val.

Meth

Obs

0.52

1

WKD

99

0.76

0.37

WKD

94

0.82

0.5

WKD

27

0.72

0.95

WKD 102

Parameter estimates are in bold characters and standard errors in italics. ⁎⁎⁎, ⁎⁎ and ⁎ significant respectively at 1, 5 and 10%. Instruments used in GMM: Av. wage (t-2) and (t-3), App. Productivity (t-2) and (t-3), Market Shares in Domestic, Developing and Industrialized markets (t-2) and (t-3).

Then, we constructed new variables where now, market shares interact with one of these indicators that is supposed to roughly represent our λ parameter. This enables us to run an alternative regression for wages where the parameters ζ to be now estimated represent only the state of market power within firms. Accounting for the unbalanced panel data method we estimate two alternative equations where 1/ λ = λd and 2/ λ = λc. These regressions are based on the following equation: wi;t wi ; d t ¼ fdom ðkdom ⁎E⁎SÞdom þ fInd ðkdom ⁎E⁎SÞInd þ fDev ðkdom ⁎E⁎SÞDev þ b3;i ∼ ∼ pi;t pi;t P ½Ptyi;t − Ptyi;t  þ b4;i þ ui þ ut þ ui;t ∼ Pi;t

P

ð9Þ

The resulted ζ estimators are then compared to the β estimators. In fact, our theory suggests that when the β parameter is positive and statistically significant, ζ has to have the same sign and be statistically significant as well (see βj expression above). Table 13 provides another evidence of the relevance of our βj parameters as informing about market power. Indeed, the sign and significance of the β estimators appear to be very correlated with that of their direct correspondent ζj coefficients when using either of the unions' power indicators (between 0.73 and 0.85 when applying the λd indicator; between 0.49 and 0.67 when using λc).16 The cross-correlation is also positive and relatively high between the parameters relative to the Industrialized markets and those of domestic markets (0.41 using λd and 0.37 using λc). This is another evidence for similarity of market structures across developed countries' markets. 2/ Another piece of evidence is obtained from confronting mark-up estimates at the same level of ISIC aggregation from Oliveira Martins et al. (1996) (OMSP, hereafter) with our β estimators. These authors estimate average industrial mark-ups from total production and costs data following Hall's and Roeger's methods for 14 OECD countries. As they do not split mark-ups into profit rates obtained on different markets, their estimators are not perfectly comparable to ours from a theoretical point of view. Yet, the same table shows a correlation rate of 0.28 between 16 The correlation degree of ‘signs and significance’ is obtained by constructing a qualitative variable for each of the estimators ζj and βj, ∀j, that takes on 1 when the parameters are positive and statistically significant, 0 when they are insignificant and − 1 when they appear to be negative (few values of − 1).

550

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Table 12 Estimation results at the industry level for Asian countries Industry

Av. wage Pty. dif

Beverages

0.196 0.167 1.005⁎⁎⁎ 0.05 0.881⁎⁎⁎ 0.045 0.681⁎⁎⁎ 0.103 0.675⁎⁎⁎ 0.064 0.903⁎⁎⁎ 0.069 1.057⁎⁎⁎ 0.017 1.065⁎⁎⁎ 0.174 0.795⁎⁎⁎ 0.029 1.23⁎⁎⁎ 0.067 1.276⁎⁎⁎ 0.043 1.032⁎⁎⁎ 0.219 1.057⁎⁎⁎ 0.128 1.158⁎⁎⁎ 0.081 0.69⁎⁎⁎ 0.024 0.763⁎⁎⁎ 0.085 0.997⁎⁎⁎ 0.071 1.604⁎⁎⁎ 0.282 0.455⁎⁎⁎ 0.103 0.877⁎⁎⁎ 0.041 0.913⁎⁎⁎ 0.089 0.9⁎⁎⁎ 0.066 0.924⁎⁎⁎ 0.08 0.803⁎⁎⁎ 0.011 0.122⁎⁎ 0.052

Fab. metal prod. Food products Footwear Furniture Glass and products Industrial chemicals Iron and steel Leather products Machinery, electric Machinery Misc. petrol. prod. Non-ferrous metals Other chemicals Other manuf. Prod. Other nonmetallic prod. Paper and products Petroleum refineries Plastic products Pottery/china Printing and publishing Professional and scient. Rubber products Textiles Tobacco

0.014⁎⁎⁎ 0.004 0.003 0.003 0.005⁎⁎⁎ 0.002 0.003 0.004 0.005 0.004 0.021⁎⁎⁎ 0.006 0.009 0.006 0.003 0.004 0.012⁎⁎⁎ 0.002 0.009⁎⁎ 0.004 0.011⁎⁎⁎ 0.004 0.008⁎⁎ 0.003 0.01⁎⁎ 0.004 0.019⁎⁎⁎ 0.003 0.0001 0.002 0.018⁎⁎⁎ 0.004 0.02⁎⁎⁎ 0.004 0.0001 0.001 0.011 0.007 0.005⁎⁎⁎ 0.002 0.007 0.004 0.004 0.007 0.014⁎⁎⁎ 0.002 0.008⁎⁎⁎ 0.003 0.003 0.002

R. Dom share βdom ′

R. Share Ind β′Ind

R. Share Dev βDev ′

1.091⁎⁎ 97.705 44.242 0.424 361.215 274.932 0.021 − 45.588 24.397⁎ 0.033 49.204 13.35 − 0.101⁎ − 288.005⁎⁎⁎ 9.615 0.055 77.652 23.563 0.436⁎⁎⁎ − 2.803 − 36.25 0.091 4.154 57.505 − 0.045 8.137 − 32.084⁎⁎ 0.047 8.451 15.777 0.073⁎ − 92.752⁎⁎⁎ − 24.489⁎⁎⁎ 0.04 13.142 7.996 − 0.001 19.305⁎⁎⁎ − 5.246 0.002 6.256 11.401 0.13 23.658 3.205 0.116 40.401 14.598 − 0.105⁎⁎⁎ − 5.708 −13.111 0.038 4.405 8.798 0.0001 3.314 5.218⁎ 0.001 3.014 3.155 0.005 0.927 12.768⁎⁎⁎ 0.015 2.501 3.093 − 0.032 29.913 27.343 0.028 22.241 31.136 0.009 82.398⁎⁎⁎ 1.81 0.008 17.043 1.731 − 0.051 114.594⁎⁎⁎ − 77.356⁎⁎ 0.069 22.242 35.635 0.0001 4.014 − 6.567⁎⁎⁎ 0.0001 3.478 2.182 0.017⁎⁎ − 515.055⁎⁎⁎ −60.154⁎ 0.008 163.764 34.048 0.0001 1.59 9.738⁎⁎⁎ 0.034 19.926 2.477 − 0.012 114.199 −30.883 0.044 102.548 187.887 0.141 253.518 23.09 0.116 156.634 238.746 − 0.007 − 13.726 − 61.468⁎⁎⁎ 0.012 22.264 19.594 0.196⁎⁎ − 1709.032⁎⁎⁎ − 265.569 0.088 550.604 192.33 0.001 − 1.351 4.388 0.002 2.181 3.916 0.068 78.038⁎⁎⁎ − 49.692 0.076 19.594 34.832 − 0.07⁎⁎ 27.718 9.509⁎⁎ 0.034 17.62 4.694 0.621⁎ 210.041 − 211.146 0.333 139.757 138.725

DWH P-val.

Overid P-val.

Meth

Obs

0.59

0.72

WKD 45

0.65

0.78

WKD 71

0.86

0.55

WKD 76

0.68

0.37

WKD 45

0.98

0.56

WKD 48

0.85

0.41

WKD 49

0.4

0.09

WKD 57

0.82

0.61

WKD 47

0.11

0.43

WKD 57

0.54

0.45

WKD 71

0.92

0.15

WKD 66

0.97

0.81

WKD 22

0.85

0.6

WKD 30

0.62

0.72

WKD 65

0.98

0.91

WKD 69

0.83

0.77

WKD 59

0.62

0.27

WKD 56

0.95

0.83

WKD 24

0.92

0.79

WKD 52

0.48

0.67

WKD 53

0.64

0.76

WKD 66

0.87

0.69

WKD 40

0.93

0.84

WKD 55

0.31

0.78

WKD 71

0.79

0.72

WKD 53

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

551

Table 12 (continued ) Industry

Av. wage Pty. dif

Total manufacturing Transport equipment Wearing apparel

0.834⁎⁎⁎ 0.008⁎⁎⁎ − 0.012 0.034 0.002 0.025 1.448⁎⁎⁎ 0.004 − 0.099⁎⁎ 0.092 0.005 0.043 0.744⁎⁎⁎ − 0.003⁎⁎ − 0.086⁎ 0.074 0.002 0.044 0.578⁎⁎⁎ 0.003 0.126⁎⁎ 0.05 0.004 0.054

Wood products

R. Dom share βdom ′

R. Share Ind β′Ind − 20.756⁎⁎ 8.464 41.784⁎⁎⁎ 8.458 −6.484 6.353 − 37.581⁎⁎ 15.697

R. Share Dev βDev ′ 27.667⁎⁎⁎ 10.661 − 4.572 6.032 1.23 4.301 4.406 2.793

DWH P-val.

Overid P-val.

Meth

Obs

0.66

0.2

WKD 68

0.12

0.58

WKD 71

0.98

0.64

WKD 19

0.5

0.8

WKD 66

Parameter estimates are in bold characters and standard errors in italics. ⁎⁎⁎, ⁎⁎ and ⁎ significant respectively at 1, 5 and 10%. Instruments used in GMM: Av. wage (t-2) and (t-3), App. Productivity (t-2) and (t-3), Market Shares in Domestic, Developing and Industrialized markets (t-2) and (t-3).

industrial mark-ups from OMSP and βdom, while the correlation is around 0.10 with βInd and 0.06 with βDev. 3/ Finally, the value of the βdom parameters relative to domestic shares are mainly between 0 (non significant) and 0.5, which is consistent with our theory as well as other studies that try to evaluate properly the unions' market power parameter λdom. In fact, Abowd and Lemieux (1993) and Abowd and Allain (1996)) evaluated the ‘revenue shifter’ λdom to be around 0.25 and 0.40 on average. As it is not a strong assumption to consider a hypothesis of perfect integration for the domestic market, κdom is then expected to be near or a little above unity. Meanwhile, given that 0 b ψdom b 1 and for values of elasticity of demand σ around or above unity (see Goldstein and Khan, 1985 or Erkel-Rousse and Mirza, 2002), the ratio ψdom / σdom should be smaller than unity. This is perfectly consistent with our results on βdom. All these evidence seem to support that our estimators are revealing some information about market power of firms and unions. They are consistent with our theoretical story of rents acquired and shared from exporting as long as firms have some market power and unions have sufficient bargaining power to shift them to employees. Table 13 Correlation between β, ζ and OMSP estimators Variable

βDom

βInd

βDev

βDom βInd βDev

1 0.70 −0.03

1 −0.18

1

Using λd signζDomd signζdInd signζdDev Using λc signζcDom signζcInd signζcDev Mark ups (OMSP)

0.28

0.10

0.06

signβDom

signβInd

signβDev

0.76

0.41 0.85

0.11 − 0.19 0.73

0.67

0.37 0.63

0.01 − 0.18 0.49

552

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

7. Conclusion This paper has focused on rent sharing issues consecutive to openness. We asked whether openness, through exporting, is a source of rents for an industry that are shared between its workers and capital holders. We asked in particular to which extent rents are shifted between employers and employees from different countries. In that respect, we aimed at considering the short or medium run impact of openness instead of looking at general equilibrium effects from the Stolper–Samuelson type. We have derived then tested a theoretical equation, based on rent sharing theories, linking industry wages to openness variables. The real wage equation, net from the alternative wage, is shown to be a linear combination of the domestic market share and export market shares weighted by the rate of sales to each destination country. As the domestic market share variable is by construction inversely related to import penetration, the impact of openness has been tackled here through both import and export type variables. The equation has another feature. It stresses explicitly the interaction between union power on the labor market and market power of firms on the domestic and each of the export markets, when studying wages' response to openness. We then used industrial trade and activity data from two UNIDO databases on 65 developed and developing countries to test this equation. We found, for developed countries, that an increase in export market shares held on other developed markets (as well as domestic market shares) is associated with growth in wages in roughly half (third) of the industries. Then, rents to be captured in other developed countries seem to matter even more for wages than that captured when selling to the domestic markets. On the other hand, rents do not seem to be shifted systematically from developing in direction of developed economies as in more than two thirds of the industries the relationship between exporting to LDCs and increasing the wage premiums breaks down. In Latin America and the Mediterranean, things are slightly different as domestic market shares are more positively linked to wages than exports are. This suggests that unions do have a certain bargaining power to be able to redistribute rents from selling at home. However, selling abroad does not seem to be a principal source of rents to be shared with workers. In these countries, exporters are unable to shift rents away from other LDCs or rich economies to redistribute them back to home employees. The most striking results however are relative to Asia and East-Asia groups. Openness variables do not seem to be related in general to industry wages. Either firms do not have enough power on average on the domestic or export markets to extract rents, or unions in Asia are not strong enough to shift a part of them to workers. In sum, openness through exports and imports, is not systematically associated with gains and losses of rents respectively. The outcome depends on the characteristics of the industries, the power of unions and/or group of countries considered. A robust result persists however, across all groups of developing countries: when they open to trade with other developing and most importantly with developed economies they do not seem to be able to shift rents and redistribute them back to their employees through higher wage premiums. Appendix A. Aggregation strategy to industry level 1/ How can we obtain a wage relation at the industry level (Eq. (6) in the text) from the firm level Eq. (5) (in the text)? 2/ How can weighted market shares and wages be related at the industry level?

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

553

Before manipulating the wage equation, an aggregation of the mark-ups from the firm to the industry level is necessary to understand our story. Reconsider the mark-up expression on one market, say the foreign market (that of domestic market will be deduced in the same manner). Here we proceed as in most industrial organization textbooks to obtain average mark-ups at the industry level from firm level mark-ups (see Martin, 1993). Thus, weight the export mark-up Eq. (4) for a firm by its market share relative to all other firms from the same nationality (xfn / Xdf), and sum over all Nd firms exporting from d to market f. We have: X  xfn  pf −wu pf Xdf −wu Xdf ¼ Xdf pf pf Xdf Nd

with wf ¼

h i2 xfn Xdf

¼

1 X x2fn 1 rf Nd Xdf Xd f

¼

  1 X xf ;n 2 Xdf rf Nd Xdf Xd f

¼

1 w Sf rf f

.

As noticed, industry mark-up to exports (i.e. weighted average of firm mark ups) is now related to elasticity of demand (σf), a concentration index that informs about the degree of competition within all the firms of the same nationality selling to the foreign market f (i.e. ψf) and finally, total market share of these firms on the foreign market. How do the industrial wage rate per employee and industry mark-ups interact? To see this, multiply the wage rate (wn) expression by ln, and sum over the Nd firms we have: X

wn ln ¼ bd

n¼1 N Nd

X n

pn ln

X pf xfn xfn pd xdn xdn þ bf pn l n þ wu L pn yn Xd d pn yn Xd f n

P recall that if ln = yn then L ¼ n ln ¼ Y . Then, by dividing the wage bill at the industry level by L in order to obtain an expression for the industry wage per employee, we have: P w ¼¼

¼ pd k

¼ pd k

n¼1 N N

wn ln

L 1 X 2 1 1 1X 2 1 1 þ pf k þ wu xdn x rd n Xd d Y rf n fn Xd f Y   1 X xdn 2 Xdd rd n Xdd Xd d |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} Industry average markup on d

Xdd þ pf k Y

  1 X xfn 2 Xdf rf n Xdf Xd f |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} Industry average markup on f

Xdf þ wu Y

554

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

From this expression we can already see the relation between industry average wage rate and industry average mark-ups. Recall from the text Sd = Xdd / XUd and Sf = Xdf / XUf to be the market shares held on the domestic and foreign markets for a given industry. Recall Ed and Ef to be the corresponding industrial rates of domestic sales and exports. Finally we have defined wd ¼   P x 2 (resp. ψf) to be a concentration index that informs about the degree of competition n X d;n dd

within all the firms of the observed country selling to its market d (resp. to market f ). We can now derive the industry relation of real average wages from above: Xdf pf Xdf w Xdd pd Xdd ¼ bd þ bf þ wu p Xd d pY Xd f pY ¼ bd Sd Ed þ bf Sf Ef þ wu where bd ¼

kwd rd

bf ¼

kwf rf

and

Given that the range of values that could be taken by the parameters λ, ψf and σf, the coefficients βs should lie between 0 and 1. By detailing how the wage relation is obtained at the industrial level, one can now understand why weighted market shares are key regressors. In fact, an increase in a weighted market share held in one market will lead to an increase in the average markup of that industry as far as the market is imperfectly competitive (i.e. the value of σ is not so high and that of ψ not too small). Whether or not this affects positively wages in return depends then on the bargaining power of the unions through λ. This is precisely the relation that is estimated in the text (See the interpretation in more details in the text). Appendix B. Generalization of the theory We generalize our 2 country case to J countries' model. The corresponding markets are assumed to be internationally segmented. Note that j can be the domestic market ( j = i) or the foreign market ( j ≠ i), so that each time that the firm serves the domestic market we shall say that it ‘exports’ to this market. The objective function to maximize would be " ! #1−ki J X ki ðli;n ½wi;n −wu;i Þ pij;n xij;n −wi;n li;n ð10Þ j¼1

From the first order conditions we derive the following wage equation:
P  j pij;n xij;n −wu;i li;n wi;n ¼ ki þ wu;i li;n

ð11Þ

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

555

Besides, equating marginal revenue to marginal cost in each market, and considering Eq. (11) we derive the following quasi mark-up equation on each export market j: pij;n −wu;i 1 ¼ sij;n rj pij;n

ð12Þ

Eq. (12) is closely related to the Structure–Conduct–Performance type expressions in industrial economics, since ‘quasi’ mark-ups depend on conjectural variation α, price-elasticity of demand 17 σ and market share sij,n = xij,n / Xj, with Xj representing total sales in the market Pj. For ease ofP exposition, we assume that output equals labor demand y ¼ j¼1 N J xij;n ¼ li;n .   i;n Let pi;n yi;n ¼ j pij;n xij;n be the total revenue for firm n and bij ¼ ki r1 ; 8ja1 N J . Expressing by j′ a foreign market different from the domestic one i, then Eqs. (11) and (12) give the following real wage function that generalizes Eq. (2) for the case of J markets: j

X wi;n wu;i ¼ bii eii;n sii;n þ ½bij Veij V;n sij V;n  þ pi;n pi;n j Vpi


p x

ð13Þ



where, eij;n ¼ pi;nij yij;ni;n stands for the export rate of firm n in the market j. Then, the real wage equation, net from the real alternative wage, is a linear combination of the sum of export market shares weighted by the export rate to each country j. How does the firm level expression extends to industrial level. Let Sij = Xij / Xj be the country's
 pij Xij i market share in country j for an industry, Eij ¼ pi Yi being its industry's export rate and P ¼ n li;n representing total demand for labor at the industry level. Moreover, let wij ¼ L "i #

P
xij;n 2 Xij

n

be the export concentration on the bilateral market {ij}. This concentration index

18 informs us about the degree of competition within all the exporting firms from P i to the market j. Then considering Eq. (13) and computing the real average wage wi pi ¼ ½ n wi;n li;n =Li =pi at the industry level we can derive the following expression that generalizes Eq. (6) to the general case: X wi wu;i ¼ b1;ii Eii Sii þ b2;ij VEij VSij V þ ð14Þ pi pi j Vpi

where b1;ii and

17



1 ¼ ki wii ri





 1 b2;ij V ¼ ki wij V ; 8j Vpi; j Vaf1 N J −1g rj V

One could easily extend the Cournot case to a more general case of conjectural variations. However, such an extension leads to the same type of equations to test except that the coefficient βs to be inferred contain a conjectural variation parameter that reveals the form of the firms' behavior. 18 The export concentration on the bilateral market is the export concentration relative to a country i exporting to j. For the sake of clarity, assume the particular case where all the exporting firms have symmetric characteristics in costs then this bilateral concentration index reduces to the inverse of the number of firms that export to the market j. Then, it is easy to understand why this index of concentration reveals the degree of competition within these exporting firms on the market j.

556

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Appendix C. Robustness of the Specified Equation to differentiation We follow Geroski's (1983) specification in the presence of product differentiation. Goods are differentiated because each good is assumed to have its specific market. Put differently, every variety is unique as it is only partially comparable to the others.19 However, goods could also be geographically differentiated. For instance, in a big region j, where local markets are distant from one another, demand addressed to a firm in a market, say m1, could have little if any effect on the perceived demand of other firms selling in another marketplace, m2. On the opposite, if markets m1 and m2 are very close, and thus, tend to be integrated into one overall market j, then consumers' total demand perceived by each firm in this region P

to match total supply from these firms. P j tends e Hence, let Xj;ni ¼ xij;n þ hj ; x þ X n Vpn ij;n V i Vpi i Vj be the total ‘effective’ demand faced by firm n. The parameter θj can be considered either an indicator of product or spatial differentiation or a combination of both. The value of θj varies between 0 (perfectly differentiated good or geographically segmented markets) and 1 (perfectly homogeneous good or perfectly integrated markets within j). Then, the Lerner index for firm n is determined by the same function of that expressed for the homogeneous good and perfectly integrated market Eq. (11), except that pricee e elasticity σij,n , and the firm n's market share sij,n are defined in terms of ‘effective’ quantities. Recalling the mark up equation we then have: pij;n −wu;i ¼ ½1=reij;n ⁎seij;n pij;n with

seij;n ¼

xij;n Xj Xj ¼ ¼ sij;n e Xj Xj;ni Xj;ni

ð15Þ

representing the effective share of firm n on region j. Notice that

‘effective’ or ‘perceived’ market share is systematically higher than observed market share

xij;n Xj

which results in higher firms' rents at equilibrium. Following Martin's (1993) specification, let e e = αje, σij,n = σje. Note jij;n ¼ XXej . This parameter equals αij,n  1 when goods are perfectly j;n X homogeneous (resp. perfectly integrated region), and reaches xij;nj ; 8jafi; jVg when the variety that is produced by firm n is perfectly differentiated (resp. perfect market segmentation), (i.e.: θj = 0). Then Eq. (13) becomes: " ! #
 X jij;n wi;n jii;n wu;i ¼ ki ð16Þ eii sii;n þ ki eij sij;n þ e pi;n rei r pi;n j j i

add the assumption WeP

are sufficiently small in each market j. In that case, theevalue eof P that firms ≈ Xj,n′i, hj n Vpn ; xij;n V þ hj i Vpi Xi Vj is sufficiently large, which enables us to consider that Xj,n i ∀n, n′ ∈ i. Hence, κij,n ≈ κij,n ′ ≈ κij, ∀n, n′ ∈ i. Aggregating at the industry level leads to the following average real wage equation: X wi wu;i ¼ b1;ii V Eii Sii þ b2;ij V V Eij Sij V þ ð17Þ pi pi j Vpi where   jii V ¼ ki w i e b1;i ri see also Geroski (1998) who defines the market in ‘strategic’ terms. The main idea is that managers think about conceiving a product that creates its own market. 19

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

557

and ∀j′ ≠ i " b2;V j V¼ ki wij V

# jeij V rejV

Considering three markets j, with j ∈ {i, Ind, Dev}, replacing the real price by its estimate and the alternative wage by its function (Eq. (7)) in the wage relation (17), corresponds exactly to the Eq. (8) we have chosen to estimate. References Abowd, J., 1989. The effect of wage bargains on the stock market value of the firm. American Economic Review 79 (4). Abowd, J., Allain, L., 1996. Compensation Structure and Product Market Competition. NBER Working Paper March. Abowd, J., Lemieux, T., 1993. The effects of product market competition on collective bargaining agreements: the case of foreign competition in Canada. Quarterly Journal of Economics 108 (4), 983–1014 (November). Blom, A., Goldberg, P., Pavcnik, N., Schady, N., 2003. Trade policy and industry wage structure: evidence from Brazil. Dartmouth Working paper. Borjas, G., Ramey, V., 1995. Foreign competition, market power, and wage inequality. Quarterly Journal of Economics 110 (4), 1075–1110 (December). Brander, J., Krugman, P., 1983. A ‘reciprocal dumping’ model of international trade. Journal of International Economics 15, 313–321. Brown, J., Ashenfelter, O., 1986. Testing the efficiency of employment contracts. Journal of Political Economy 94 (3), S55–S78. Card, D., 1990. Unexpected inflation, real wages, and employment determination in union contracts. American Economic Review LXXX, 669–688. Christofides, L.N., Oswald, A., 1991. Efficient and inefficient employment outcomes: a study based on Canadian contract data. Research in Labor Economics XII, 173–190. Davis, P., 2002. Estimating multi way error components models with unbalanced data structures. Journal of Econometrics 106, 67–95. Erkel-Rousse, H., Mirza, D., 2002. Import price-elasticity: reconsidering the evidence. The Canadian Journal of Economics 2, 282–306. Francois, J., Nelson, D., 1998. Trade, technology and wages: general equilibrium mechanics. Economic Journal 108, 1483–1499. Geroski, P., 1983. Some reflexions on the theory and application of concentration indices. Journal of Industrial Economics 1, 79–94. Geroski, P., 1998. Thinking creatively about markets. International Journal of Industrial Organization 16 (2), 677–695. Goldberg, P., Pavcnik, N., 2001. Trade protection and wages: evidence from the Colombian trade reforms. NBER Working Paper, vol. 8575. Goldstein, M., Khan, M., 1985. Income and Price Effects in Foreign Trade. In: Handbook of International Economics (North Holland ed.), Volume 2. Ed. by R.W Jones and P.B. Kenen. Görg, H., Hijzen, A., Hine, R., 2001. International Fragmentation and Relative Wages in the UK. GEP working paper, vol. 01/33. Harrison, A., Hanson, G., 1999. Who gains from trade reform? Some remaining puzzles. Journal of Development Economics 59, 125–154. Hosken, D., Margolis, D., 1997. The Efficiency of Collective Bargaining in Public Schools. CREST-INSEE Working Paper, vol. 9755. Martin, S., 1993. Advanced Industrial Economics. Blackwell. Oliveira Martins, Y., 1994. Market structure, trade and industry wages. OECD Economic Studies 22, 131–154. Oliveira Martins, Y., Scarpetta, S., Pilat, D., 1996. Mark-up in manufacturing industries: estimates for 14 OECD countries. OECD working paper, vol. 162. Schmalensee, R. 1989. Inter Industry Studies of Structure and Performance (North Holland ed.), Volume 2, pp. 951–1009. Handbook of Industrial Economics, ed. by R. Shmalensee and R.D. Willig.

558

L. Fontagné, D. Mirza / Economic Modelling 24 (2007) 523–558

Sen, A., Dutt, A., 1995. Wage bargaining, imperfect competition and the markup: optimizing microfoundations. Economics Letters 48 (1), 15–20. Sutton, J., 1991. Sunk Costs and Market Structure: Price Competition, Advertising and the Evolution of Concentration. The MIT press. Wansbeek, A., Kapteyn, A., 1989. Estimation of the error-component model with incomplete panels. Journal of Economics 341–361.