The role of skilled labor in developing countries Globalization and Wage Inequality

Globalization and Wage Inequality

The role of skilled labor in developing countries

Janneke Pieters July 25, 2007

Thesis for the Research Master International Economics & Business

University of Groningen

Faculty of Economics, SOM Research School

Supervisor: dr. Marcel P. Timmer Co-assessor: Prof. dr. Steven Brakman

Abstract

2 CONTENTS Foreword 1 – Introduction 4 2- Theory 7 2.1 Factor Endowments 7

2.2 Technology and Capital 12

2.3 Trade and Technology 16

3 FOREWORD

This thesis is the final result of two years of challenging courses and research, completing my Research Master in International Economics and Business. For many students, their thesis marks the end of their academic endeavors, but for me it is just the beginning. I look forward to starting my Ph.D. project at this faculty in September, to continue my research and find answers to some of the many questions surrounding development and income inequality.

Of course, I owe gratitude to a number of people. Most of all, I want to thank my supervisor Marcel Timmer for his enthusiasm about my research and his many helpful suggestions during the writing of this thesis and during my research for the course Learning and Practicing Research last year. For the rest, I would like to thank my co-assessor Steven Brakman for valuable feedback, Bart Los for his help as coordinator of the Research Master, and my fellow students for a lot of fun and inspiration. For the ‘moral support’ throughout my studies, I want to thank my parents, my grandmother, and my friends, especially Lotte, Jolien, Ruth, and Esther.

4 1 - INTRODUCTION

Wage inequality has become an important subject in light of fast growing global economic integration, because the extent to which countries are involved in international trade and ongoing technological development has an impact on the relative demand for different production factors, and thus on their relative rewards. Early research on the relationship between trade, technology, and wage inequality focused mainly on the United States and other advanced countries (e.g., Katz and Murphy, 1992; Berman et al., 1994), where growing supply of skilled workers has been accompanied by rising relative wages of skilled workers (the skill premium). The two most widely accepted explanations are increased trade with developing countries - drawing on the predictions of neoclassical Heckscher-Ohlin (H-O) theory - and skill-biased technical change. Most discussion now concerns the relative importance of trade versus technology in explaining relative wage changes, their interrelatedness, and methods of empirical analysis (see Krugman, 2000; Slaughter, 2000).

However, the fact that the skill premium is rising in many less developed countries as well poses a challenge to researchers, as H-O theory predicts a negative effect of trade on wage inequality in developing countries, due to their comparative advantage in unskilled-labor intensive production. This has spurred new theoretical explanations for changing wage inequality and a range of developing country studies since the second half of the 1990s (e.g., Feenstra and Hanson, 1996; Robbins and Gindling, 1999). Unfortunately, the empirical evidence on the causes of wage inequality in developing countries is scattered, as most studies have a national or regional focus and are hardly comparable, and the results are far from mutually consistent. An important question is what country characteristics can account for these contradicting results. This study focuses on the relative supply of skilled labor as a possible answer to this question, using a large panel-data set of 28 developing countries over the period 1970-2000, and provides a step forward towards understanding the causes of skill premium changes in the developing world.

5 with low education are dealing with increasing global competition. Particularly the opening up of China and India caused an enormous growth of the global unskilled labor supply, while at the same time, a number of developing countries (e.g. Mexico, Korea) have experienced a rapid increase in the working-age population share of highly educated workers. Consequently, among developing countries, important shifts in comparative advantage have taken place. Given these shifts, the limited attention for the role of relative skill supply in wage inequality research may well be an explanation for the lack of consistent evidence.

Figure 1: World labor supply – total (top panel) and by education level (bottom panel)

1

National labor forces scaled by export/GDP ratio 3

More educated labor force is defined by persons with university-level education

Source: IMF World Economic Outlook, 2007

6 supply-price relationship. The main novelty of this study is, therefore, the explicit modeling of skill supply in demand-side determinants of skill premium changes. Our results confirm the importance of relative skill supply for trade’s effect on the skill premium, in line with H-O predictions. In addition to this compositional trade effect, trade has a strictly inequality-enhancing effect as well, which is not influenced by relative skill supply. Regarding the relationship between technology (including capital) and the skill premium, we find that the lack of cross-country consistent evidence cannot be explained by the supply of skilled labor. Finally, our results show significant trend growth in the skill premium in the 1980s and 1990s, indicating global skill-biased technical change, which was not present in the 1970s.

7 2 – THEORY

This chapter describes the theories linking factor endowments, trade and/or technological progress to the skill premium. First, the role of trade and technology is discussed in sections 2.1 and 2.2, including trade theory and models of skill-biased technical change. We then review the theories that combine trade and technology in explaining skill premium changes in section 2.3, and conclude with an overview in section 2.4

2.1 – Factor Endowments

Within a simple supply and demand framework, an increase in the relative supply of skilled labor results in a decline of the skill premium, ceteris paribus. Apart from this, factor endowments play a central role in the direction and effects of international trade. Because trade changes the composition of domestic production, it affects the relative demand for production factors within a country. Therefore, trade is a channel through with relative supply can affect relative demand.

Neoclassical Trade Theory

A widely used model to predict patterns of international trade is the neoclassical Heckscher-Ohlin (H-O) model, which was developed in the 1920s and has gained great popularity among economists (see, for example, Husted and Melvin, 2004: Chapter 4). Relative factor endowments are essential in H-O theory, because they determine a country’s comparative advantage. The basic version of the H-O model consists of two countries facing the same available technology, two goods, and two factors of production, which will be skilled and unskilled labor in the current context. One good is skilled-labor intensive in production while the other is unskilled-labor intensive, and workers are perfectly mobile between industries but immobile between countries. The country with the larger relative supply of skilled workers has a comparative advantage in production of the skilled-labor intensive good, and the unskilled-labor abundant country has a comparative advantage in the unskilled-labor intensive good.

8 follows from the famous Stolper-Samuelson theorem: an increase in the relative price of a good will increase the return to the factor used intensively in that good, and reduce the return to the other factor. All in all, this means that the abundant factor gains from trade while the scarce factor loses.

Compared to advanced countries, developing countries are relatively unskilled-labor abundant. According to the H-O predictions, then, developing countries will export unskilled-labor intensive products and import skilled-labor intensive products under free trade with advanced countries. Therefore, trade increases the relative demand for unskilled labor in developing countries, decreasing the skill premium. In advanced countries the opposite should happen: increased international trade increases demand for skilled-labor intensive products and thus increases relative demand for skilled labor, raising the skill premium. This is as far as reasoning goes in most of the developing-country studies on skill premium changes.

Importantly, the basic H-O predictions are based on the assumption that both countries will produce both goods (i.e., there is no specialization), as a result of which factor prices are insensitive to factor endowments and solely determined by goods prices. If, however, countries differ enough in their relative endowments, such that they are outside the so-called cone of diversification, it is possible that at least one country will specialize, and no FPE will occur. Giving up FPE is a first step towards a more realistic link between trade and wage inequality.

9 Figure 2 – Unit Cost Curves in the Dornbusch-Fischer-Samuelson Model

Source: Figure 3.10 in Feenstra (2004)

The real world consists of more than two countries, and these countries have more than one trading partners, of which some are more skilled-labor abundant, and some less. To predict the effect of more trade on relative wages, it would thus be wrong to simply assume all developing countries are skill-scarce. As pointed out in the introduction, some developing countries have become relatively skilled-labor abundant in comparison with newer international trade partners, like China and India (see Wood, 1997). Such a loss of comparative advantage in unskilled-labor intensive goods can be expected to affect the composition of trade, and thereby the impact of trade on relative wages.

To illustrate, in Figure 3 we add a third country (LDC2) to the DFS model with a lower relative supply of skilled labor than LDC1. Its unit cost curve is steeper and intersects with LDC1’s unit cost curve somewhere to the right of z*(otherwise, LDC1 would not produce anything), for example at z’. LDC2 is the cheapest producer of the range [0, z’), so that LDC1 now only produces (z’, z*), which is more skill-intensive on average. Thus, the opening up of LDC2 shifts LDC1’s comparative advantage away from the least skill-intensive goods, which raises the relative demand for skilled workers in LDC1.

10 Figure 3 – Dornbusch-Fischer-Samuelson Model with Three Countries

Source: Based on Figure 3.10 in Feenstra (2004)

As shown in Figure 1, the global relative skill supply has declined over the past decades. Since in most countries and periods, the share of skilled workers has grown since the 1970s, the global decline is largely due to skill-scarce countries opening up to trade or increasing their trade participation. We have shown here how the change in a country’s composition of trade depends on the change of relative skill supply compared to other trading countries. In general, either the opening up of more unskilled-labor abundant countries, or relatively fast growth of domestic skill supply will shift comparative advantage to more skill-intensive goods. The other way around, if relative skilled-labor endowment in an LDC decreases compared to other trading countries (because more skilled-labor intensive countries open up to trade or because other trading countries have faster growth of relative skill supply) it will gain comparative advantage in goods with lower skill-intensity than before, and the skill premium will decline with more trade.

Clearly, this line of argumentation differs from the Stolper-Samuelson prediction (the abundant factor will gain from trade), as we drop FPE and allow for more goods and countries. Wage inequality studies thus far have not accounted for the above predictions on changing comparative advantage in the developing world, but it appears important in determining the effect of trade on wage inequality, and should thus be considered when testing H-O theory.

11 Foreign Direct Investment

Another important consequence of disappearing trade barriers is that the production location of firms has become less tied to their sales market, such that firms can fragment the production process and move each activity to its lowest-cost location (outsourcing). This means that firms can substitute foreign for domestic labor, so workers in countries all over the world increasingly compete with one another. Foreign direct investment (FDI) is an important channel through which this outsourcing takes place, and the supply of skilled workers is likely to play a role in attracting this FDI. That is, the type of activities that are moved to a country, and specifically the level of skill required in those activities, will depend on the availability of sufficiently trained workers. The more unskilled-labor abundant a country is, the more it will attract FDI in unskilled-labor intensive production, stimulating demand for unskilled workers, and vice versa. The relationship between FDI and relative wages is thus likely to be positive in case of a relatively skilled labor force, and negative when the host country is unskilled-labor abundant. FDI will be further discussed later on in section 2.2 in relation to international technology diffusion and in section 2.3 in the model of Feenstra and Hanson (1996) on trade in intermediate inputs.

Structure of Protection

A different perspective on the trade-wages discussion relates more specifically to tariffs and other trade barriers, and particularly to the structure of protection. Through trade barriers, particular industries are protected from foreign competition. If protection is initially high in unskilled-labor intensive industries, trade liberalization will increase competition most in these industries and thereby disproportionately harm the unskilled workers. Especially if reform targets are set exogenously (by the World Trade Organization, for example) at some uniform low level of tariff rates, the tariff reduction will be largest in the most protected industries.

12 competitive disadvantage. Hanson and Harrison (1999) find the same phenomenon in Mexico and argue that this is the case in more developing countries: “The evidence suggests that developing countries often protect sectors in which they are likely to have a comparative advantage, such as the sectors with a high share of unskilled workers. In this light, it is not surprising that increasing wage inequality is observed in developing countries undergoing trade reforms.” Given that developing countries’ tariff reductions are greatest for unskilled-labor intensive industries, trade liberalization increases competition for these industries and thereby has a positive effect on the skill premium.

Regarding the impact of trade on wage inequality, we limit our empirical analysis to H-O theory, which allows us to focus on the role of relative endowments and comparative advantage. FDI will be considered as a mechanism for international technology diffusion, while we do not investigate the structure of protection in this study. In the next section we discuss how technology and capital as such can affect skill premia, and in section 2.3 we describe more comprehensive theories that combine factor endowments, trade, and technology.

2.2 –Technology

Besides international trade, technological developments play a major role in global economic integration. Information and communication technologies, for example, have greatly reduced virtual distances between people and firms. Think of the ICT sector in India, sometimes called the ‘back-office’ of the Western world: technology can induce major structural changes, and do so in very short periods of time. The link between technology and skill premia derives mainly from the skill-biased nature of technological change, and the idea that capital (which embodies technology) and skilled labor are complements in production.

Skill-Biased Technical Change and Capital-Skill Complementarity

13 complementarity is a related concept. The capital-skill complementarity hypothesis was originally formalized by Griliches (1969), and describes the idea that capital and skilled labor are complementary inputs in the production process, more so than capital and unskilled labor. That is, the elasticity of substitution between capital and unskilled labor is higher than that between capital and skilled labor. Consequently, as well formulated by Krusell et al. (2000: 1047), ‘…low-wage foreign labor is not the only factor competing with domestic labor. Unskilled labor is also competing with persistently cheaper and better capital equipment.’

In the nineteenth century, new machinery and mechanization of production processes in the US and Europe were typically skill replacing. As Goldin and Katz (1998) describe it (supported by empirical evidence), production by highly skilled craftsmen was replaced by factory production and assembly lines. The main change was in the division of labor: craftsmen had produced goods almost entirely by themselves, whereas factory workers specialized in a specific part of the production process. Capital-skill complementarity then emerged in the early 20th century, when continuous-process methods spread, which caused a greater relative demand for skilled machine operators and managerial employees, rather than unskilled manual labor.

14 There are some panel-data and cross-section studies of capital-skill complementarity that include developing countries as well. Duffy et al. (2004) use nonlinear least squares and instrumental variables estimation of a CES production function, to test for capital-skill complementarity in a large panel data set, including 73 countries over the period 1965-1990. They find only weak evidence to support capital-skill complementarity, which could be due to the fact that the extent of capital-skill complementarity varies over time or across countries, depending on a country’s stage of development. Such change in the impact of technological change and capital intensity was addressed already by Goldin and Katz (1998), who found that the effect of the capital-output ratio on the skilled labor wage share was much higher in the 1910s and 80s than in the 1960s and 70s in the US. In addition, Papageorgiou and Chmelarova (2005) find capital-skill complementarity to be nonlinear in development levels, based on cross-section analysis of 46 countries. When their sample is split into OECD and non-OECD countries, capital-skill complementarity is present only in the latter subsample. They also find that countries with low initial literacy rates but moderate initial per capita income have much stronger capital-skill complementarity than countries with high initial literacy rates or countries with low literacy and low per capita income. The authors suggest that middle-income countries with some skilled labor have more physical capital than other forms of technology. The scarce skilled labor is then most productively allocated to complement physical capital. Poor countries, on the other hand, would allocate physical capital to complement their abundant unskilled labor, and advanced countries allocate their skilled labor to technology-intensive sectors.

The nonlinear nature of SBTC may be explained by the view that the factor-bias of technical change is determined endogenously by the relative supply of skilled labor. Acemoglu (2002) shows how endogenous SBTC can induce a positive relationship between skill supply and the skill premium. As profit incentives determine the amount of research and development directed at different factors, these can result in a bias towards the relatively scarce or the relatively abundant production factor, depending on the elasticity of substitution between the factors. These profit incentives result from two effects: the ‘price effect’ is an incentive to develop technologies for production of more expensive goods, which are the goods produced with relatively more input of the scarce factor. The ‘market size effect’ is an incentive to develop technologies for the more abundant factor, as there is a larger market for the technology.

15 technology, so the market size effect will stimulate development of more skill-complementary technology. Acemoglu shows that, when the elasticity of substitution between skilled and unskilled labor is greater than one, the market size effect dominates. In that case, a relative supply increase of skilled labor leads to more innovation for this skilled labor, which raises its productivity. At the same time, demand for unskilled labor decreases because elasticity of substitution is high, so the relative reward for skilled labor increases. The skill-bias of technical change thus depends positively on the relative supply of skilled labor, and the relative demand curve will be upward sloping. It is doubtful, however, whether Acemoglu’s theory is relevant for developing countries, which do not develop most technologies domestically.

International Technology Diffusion

In general, developing countries are technology followers: they operate inside the technology frontier and lack the capabilities to innovate on their own. Instead, new technologies are invented and developed in advanced countries, or leader countries, and later imitated and adopted by developing countries. If the skill-bias of technological change is endogenous on the relative supply of skilled labor (Acemoglu, 2002), this refers mostly to the relative supply in advanced countries, where the new technologies are developed. Regardless, in developing countries, the level of technology and the pace of technical change may be endogenous on the domestic relative supply of skilled labor. That is, the adoption of foreign technology or capital may be affected by the absorptive capacity of the developing country. Educated workers are probably better able to adapt to changing technology (Nelson and Phelps, 1966), and therefore education of the labor force will enhance the process of technology diffusion. Given some degree of skill-bias or skill-complementarity, this enhanced technology diffusion will translate into higher relative wages.

16 (2001) use a direct measure of technology adoption, namely computer equipment imports, and find that these imports rose significantly with human capital in the 40 to 90 countries studied over the period 1970-1990. Savvides and Zachariadis (2005) measure the effect of foreign R&D, capital goods imports, and FDI on TFP growth, for 32 developing countries over the period 1965-1992. They find that foreign R&D intensity has the biggest and most robust effect on domestic manufacturing productivity, positively interacting with human capital. The same holds for FDI, but only if capital goods imports are decomposed into machinery and transportation equipments, of which only the latter increases TFP.

All in all, SBTC and capital-skill complementarity are phenomena that make sense intuitively and are supported by empirical evidence, although there is no consensus on their exact form. There is, for example, some indication that differences in the degree of skill-bias or capital-skill complementarity exist across time periods or a country’s level of development. Acemoglu’s upward sloping relative demand curve reflects a growing degree of skill-bias with relative skill supply, and predicts growth of the skill premium in periods of growing relative skill supply. It is, however, questionable whether his model as such is relevant for countries that do not produce new technology domestically. (Later on, we will discuss an extension of his model that distinguishes between the US as technological innovator, and developing countries as technology followers.) It can, however, be expected that the adoption and diffusion of technology in developing countries rise with their domestic supply of skilled labor. As developing economies are not the innovators in this world, the pace of diffusion and its endogeneity on skilled labor supply seem to be most relevant for skill premium changes in these countries. That is, though the degree of skill-bias of technology matters for skill premium changes, the developing countries may have relatively little impact on this nature of technology.

The third part of this chapter will be dedicated to the relationship between trade and technology. While a growing supply of skilled labor in a country may improve that country’s adoption and diffusion of technology, trade may be an important channel through which new technologies reach the developing world. Besides, there are theories linking trade to the degree of skill-bias in new technologies.

2.3 - Trade and Technology

17 the skill premium. While Feenstra and Hanson (1996) argue that technical change induces international trade, Acemoglu (2003) and Thoenig and Verdier (2003) show how trade can induce SBTC. These theories will be discussed below, after a more general overview of the channels through which trade enhances the international diffusion of technology.

Trade and the Pace of Technology Diffusion

The link between the pace of technology diffusion and trade may appear rather straightforward: part of the technology adoption in developing countries takes place in the form of capital goods imports, and more imports implies more capital goods imports. Caselli and Coleman (2001) remark that, since computer imports are only a tiny fraction of total manufacturing imports, this ‘accounting’ endogeneity is not the issue. Though their measure of technology imports is quite narrow, the message is valid. There are more important channels through which international trade raises technological development. Caselli and Coleman mention knowledge spillovers that result from manufacturing imports, as many kinds of manufactures produced in advanced countries reflect the use and knowledge of modern technologies. Connolly (2003) finds that imports of high technology goods from OECD countries increase domestic innovation and imitation activity (measured by patent applications), which also indicates the presence of knowledge spillovers. Keller (2004) reviews theoretical and empirical evidence on international technology diffusion. He argues that trade in intermediate goods involves a technology transfer, but since only the manufactured outcome of that technology is being transferred, it is a relatively weak form of technology diffusion. Rather, knowledge spillovers would arise mainly from the personal contacts associated with trade and FDI. His claim ignores the scope for learning-by-imitating, for example through reverse engineering, and the empirical evidence that does point to a significant role for imports in international technology diffusion (e.g. Coe and Helpman, 1995).

18 process requirements. Keller (2004) emphasizes, however, that the econometric evidence thus far rejects such learning-by-exporting effects, whereas case studies do confirm their importance. All in all, it is clear that trade openness does open up new channels of technology diffusion. Knowledge spillovers seem to be the accepted mechanism in case of imports from advanced countries, and they may also arise in case of exporting. If this international technology is skill-biased, which it most likely is, at least more skill-biased than domestic technology, it will stimulate demand for skilled workers. Therefore, for LDCs trade will always raise wage inequality to some degree.

Trade and the Degree of Skill-Bias

Regarding the effect of trade openness on the degree of skill bias in technology, Acemoglu (2003) provides an extension of his earlier model (Acemoglu, 2002) that includes less developed countries (LDCs) as purchasers of US-developed technology, or machines. Acemoglu (2002) predicted that the skill-bias of technical change increases with relative skill supply in the innovating country, due to the so-called market size effect. In the extended model, Acemoglu’s main prediction is that trade causes a price effect on SBTC: with free international trade, relative goods prices will equalize across countries. For the skill-abundant US, this means the relative price of skill-intensive goods rises, inducing more SBTC via the price effect. The skill premium will rise due to this SBTC, and will equalize across countries due to trade. In turn, the induced SBTC increases productivity of skilled workers, raising the world relative supply of skill-intensive goods, such that its relative price will decline again to the pre-trade US level. Acemoglu (2003) thus explains how relative wages in the US can rise without relative goods prices changing.

19 property rights in LDCs has, if anything, increased, there must have been a considerable market size effect in favor of unskilled labor, reducing SBTC.

Thoenig and Verdier (2003) argue that the process of trade opening in LDCs leads to more skill-biased technological change due to a lack of property rights protection in LDCs. Their key assumption is that, with a skilled- and an unskilled-labor intensive technology, only the latter technology is subject to information spillovers such that imitation is possible. When LDCs open up they gain access to foreign technology, so that effective global IPR enforcement decreases. This induces firms in advanced countries to engage in defensive innovation: firms will develop technologies with a higher degree of tacit knowledge, which are more skill-intensive and harder to copy. This skill-biased technical change will raise the skill premium in both advanced countries and LDCs. Thoenig and Verdier provide examples of empirical evidence that confirm the actual presence of defensive innovation by firms, in response to increased international competition. At the macro level, however, it will be hard to detect defensive innovation. Given that the opening up of a single country will not have a substantial effect on global IPR enforcement, with the exception of very large countries, the predictions of this model imply an increasing degree of skill-bias with global trade integration, which has been a more gradual process.

The main differences between the above two models are 1) that Thoenig and Verdier make a distinction between types of technology regarding their proneness to imitation, and 2) that Acemoglu does not incorporate a market size effect due to growing international trade, while Thoenig and Verdier do. As already argued, a trade-induced market size effect in Acemoglu’s model would negatively affect SBTC, which may or may not be offset by the trade-induced price effect. The fact that two opposing forces result from trade complicates empirical testing, especially because SBTC is not directly observable. Since Thoenig and Verdier assume only unskilled-labor-intensive technologies are subject to predation, they predict increasing trade will always raise SBTC. We discuss their model to provide an alternative view on trade-induced SBTC, but an empirical application will be most interesting when studying firm behavior, rather than countries.

Technology Catch-Up and Trade

20 distinctive feature of this model is that it focuses on trade in intermediate inputs, based on the notion that more and more firms from advanced countries outsource part of their production activities to less developed countries, for example through FDI (as briefly discussed in section 2.1). Capital growth or technological progress in developing countries increases outsourcing by firms in advanced countries, which raises the skill premium in both developing and advanced countries.

The model is based on Dornbusch et al. (1980), distinguishing a continuum of inputs z∈

[ ]

0,1 produced within an industry, ranked in increasing order of skill-intensity. All intermediate inputs are produced with both types of labor and capital, where capital has an equal cost share in all inputs. The assembly of intermediate inputs is a separate activity, which does not require any factor inputs. Figure 4 shows the unit cost curves similar to Figure 2, but now for intermediate products: the LDC produces [0, z*) and the US produces (z*, 1]. A US firm will import the more unskilled-labor intensive inputs from the LDC. If the price of the imported inputs falls, the firm can realize a higher output level by offshoring even more production, consequently shifting home production more towards inputs at the high end of the skill-intensity range. This is shown by a downward shift of the LDC unit cost curve in Figure 4, which causes a rightward shift of the intersection, from z* to z’. The range of inputs that is relocated is less skill-intensive than the range remaining in the US, but more skill-intensive than the range thus far produced in the developing country. As a consequence of this shift, therefore, the relative demand for skilled workers, and their relative wage, will rise in both countries.

Figure 4 – Unit cost curves of intermediate inputs

21 What causes the price of the imported inputs to fall? In Feenstra (2004) capital moves from the advanced to the developing country, because the return on capital is initially higher in the developing country. But in fact, the underlying mechanism can be any structural variable that affects production costs, as long as the cost reduction (due to, e.g., growth in capital stock or technological progress) is greater in the developing country than in the advanced country. In other words, it is technology catch-up that raises the skill premium in both countries, through its effect on trade in intermediate inputs. Notice that this technological progress does not need to be factor-biased: any relative cost reduction does the trick.

A novelty of this model is that trade is affected by technology catch-up. If we introduce a fixed cost share of capital in the production of final goods in the DFS model in Figure 2, technology catch-up by the LDC will have the same effect: the cost curve of the LDC shifts downwards, moving the intersection point z* to the right, increasing relative wages in the LDC and the US. This prediction does not depend on the type of goods traded (final or intermediate). However, it does depend on the cost share of capital in production, which is more likely to be equal across activities within an industry than across industries. To discriminate between final goods and intermediate goods trade, one must distinguish these goods in the data, which we will not do in the present study. It is not our goal to find which type of trade is more important, and we can easily show that it does not matter for the role of relative skill supply in trade’s effect on relative wages: adding a third country to the model of Feenstra and Hanson in Figure 4, in the same way as in Figure 2, will lead to the same predictions as we made in section 2.1: when a country’s relative skill supply increases compared to other trading countries, its production will shift towards more skill-intensive products (or away from less skill-intensive products), whereby trade will have a positive effect on the skill premium.

22 2.4 - Conclusions

What can we conclude from the large body of theory on skill premium changes? Clearly, trade, technology, and the supply of skilled labor are interrelated in the process of skill premium changes. As emphasized throughout this chapter, it is important to take into account the relative supply of skilled labor when analyzing the effects of trade and technology, for two reasons. Firstly, the predictions of the Heckscher-Ohlin model depend crucially on relative factor endowments. New countries open up to trade, and the relative supply of skilled labor grows at different rates across countries, causing shifts in comparative advantage. Therefore, one should not just classify any developing country as skill-scarce. Secondly, skilled labor is an important factor for the so-called absorptive capacity of a country, which is likely to determine the pace of technology diffusion.

23

Table 1 – Theories on skill premium changes

Theory Mechanism Variable

Effect on skill premium

Present Study

Heckscher-Ohlin theory Comparative advantage determines composition of trade Trade x Increase in comparative skill-endowment + *

Trade x Decline in comparative skill-endowment - *

Structure of protection Unskilled industries more protected pre-liberalization Trade barrier reductions +

Skill-biased technical change Linear SBTC Time + *

SBTC nonlinear over time Time x Period (dummy) +/- * Capital-skill

complementarity Linear CSC Capital intensity + *

CSC nonlinear over development Capital intensity x Development level +/- * International technology

diffusion Knowledge spillovers (Capital goods) Imports, Exports, FDI + *

Absorptive capacity Imports, Exports, FDI x Skill supply + * Induced innovation

(Acemoglu, 2003) Price effect of trade: LDC trade integration raises SBTC LDC Trade x Time + Market size effect of increased IPR enforcement: LDCs

join the technology market, but are relatively skill-scarce. Trade weighted world IPR x Time - Market size effect of increased trade: LDCs join the

technology market, but are relatively skill-scarce. LDC Trade x Time - Defensive innovation

(Thoenig and Verdier, 2003) LDC trade induces more SBTC: threat of predation LDC Trade x Time + Trade in intermediate goods

(Feenstra and Hanson, 2001) Technological catch-up in LDC stimulates offshoring

Technology catch-up
range of intermediate

24

3 – EMPIRICAL EVIDENCE

Though most research attention has been paid to the rising skill premium in advanced industrial countries, a number of empirical studies deal with less developed countries, mainly in Latin America and East Asia. The following review of part of these studies is meant to give an impression of the available evidence and to illustrate the lack of consensus in this area: most studies are single-country analyses, using a wide range of different models and methods, which have led to incomplete and sometimes conflicting conclusions.

3.1 – Empirical Evidence

To discriminate between trade and technology, an approach that has been used in several studies is to decompose shifts in the relative demand for skilled labor (measured as employment share or wage bill share) into within-industry and between-industry shifts (see Berman et al., 1994). The underlying idea is that demand shifts caused by trade would only take place between industries, because trade changes the relative demand for final goods, and thereby the relative size of industries (the domestic production structure). As industries differ in terms of skill-intensity, growth of skilled-labor-intensive industries relative to unskilled-labor-intensive industries would constitute a between-industry shift in employment towards skilled labor. If most of the employment shifts take place within industries, however, this would reflect skill-biased technical change within industries. Most studies find that within-industry shifts account for the largest part of relative employment or wage share shifts. This was the case in Mexico during the 1980s (Hanson and Harrison, 1995), where 80 to 90 percent of the growth in the wage share of skilled workers was due to employment shifts within industries. Mazumdar and Quispe-Agnoli (2004) employ the decomposition analysis at the four-digit (ISIC Rev.3) industry level for Peru and find that within-industry changes explain all of the increased demand for skilled labor in the second half of the 1990s, as the between-industry shifts were decreasing the wage-bill share of skilled workers. In India in the 1990s as well, most of the shifts in employment share and wage-bill share took place within industries (Kijima, 2005). As Kijima only distinguished 14 industries, which is a rather aggregate level, some of his within-industry shifts may actually be qualified as between-industry shifts, but it is not possible to tell how this affects his results.

25 Feenstra and Hanson (1997) for 9 Mexican industries over the period 1975-1988, indicates that within-industry shifts in relative demand for skilled labor are consistent with increased outsourcing, or increased trade in general. At the same time, increased FDI is positively correlated with relative demand for skilled labor and explains more than half the increase in skilled labor wage share in the late 1980s. These results illustrate the difficulty of separating effects of trade from those of technology.

To more formally study the impact of trade liberalization on skill premia, some studies use relative price changes to test directly for Stolper-Samuelson effects. One possible method for this is the so-called mandated wage approach. This method provides a test of the Stolper-Samuelson theorem, based on an interpretation by Slaughter (2000): “For any vector of goods price changes, the accompanying vector of factor price changes will be positively correlated with the factor intensity-weighted averages of the goods price changes”. Esquivel and Rodriguez-Lopez (2003) use the mandated wage approach with data on 49 manufacturing branches in Mexico before and after NAFTA, which came into effect in 1994. Though most of Mexico’s trade protection was dramatically reduced between 1985 and 1988, NAFTA was expected to lead to a deepening of the trade effects on wage inequality. The study shows that between 1988 and 1994 trade liberalization reduced the wage gap in Mexico, but at the same time technological development (as proxied by labor productivity growth) greatly increased the wage gap. In the 1994-2000 period, trade liberalization had hardly any effect, while technological progress caused a slight increase in wage inequality. These findings are in line with the skill-biased technical change hypothesis, as well as with H-O-type trade effects, given that Mexico is abundant in unskilled labor relative to Canada and the US, its NAFTA trade partners. Gonzaga et al. (2006) use the mandated wage approach as well, and find evidence for trade-induced relative demand changes in Brazil, where trade liberalization took place between 1988 and 1995. Unlike most other countries, the wage inequality in Brazil declined during this period. Tariff changes were not related to skill intensity, but due to differences in import penetration across sectors, relative prices of skill-intensive products decreased. This accounted for more than the observed decline in wage differentials, which indicates there were some counteracting forces at work, such as skill-biased technical change. Unfortunately, the presence of SBTC is not further examined.

26 changes across industries, from which they conclude increased international trade was not very important for the growth of the skill premium in Mexico in the 1980s. However, they do find a positive association of export orientation and foreign investment shares with wage inequality, at both the industry and the plant level. This suggests that trade does affect the skill premium, but not through tariff or quota reductions, which were implemented at the industry level. Further, capital intensity was negatively related to skill premia at the industry level, but positively at the plant level. All in all, neither the effect of trade, nor that of technology, is strongly confirmed by the evidence. In a later study on Mexico, Hanson and Harrison (1999) find that the 1985 tariff reductions were greater for low-skilled industries, from which they conclude that competition from countries with larger reserves of cheap unskilled labor harmed unskilled workers in Mexico. There is, however, only a weak relation between tariffs and product price changes, so the Stolper-Samuelson theorem is not really supported.

In Kenya, the skill premium declined over the period 1964-2000, according to Bigsten and Durevall (2006). Cointegration analysis shows that the degree of openness, as proxied by the ratio of Kenyan to UK manufacturing prices, was the main cause of this decline. The authors state that foreign technology inflows are very limited in Kenya, so exposure to foreign competition would not lead to productivity improvements. No long-term relationship was found between the skill premium and the capital-labor ratio, educational attainment, relative labor productivity, or domestic relative goods prices.

27 International technology diffusion is addressed by Berman and Machin (2000b), who find that skill-upgrading in the 1980s in 18 middle- and lower-income countries took place in the same industries as in the US. Furthermore, US computer usage and OECD R&D intensity in the 1970s and 1980s are positively correlated with the wage-bill share of skilled labor in middle-income countries, at the industry level. Some confirmation of the importance of international technology diffusion may also be found in Gopinath and Chen (2003). With panel data for 1970-1992 on 26 countries, of which 11 developing countries (Colombia, Ecuador, Egypt, El Salvador, Honduras, India, Korea, Peru, Philippines, Sri Lanka and Venezuela), they study the effects of capital inflows on the wage gap. They derive the labor share of Gross National Product (GNP) from a translog GNP function with net FDI stock as one of the cost components. Inward FDI raises the labor share of GNP in developing countries, while results for a sub-sample of six developing countries indicate that inward FDI widened the wage gap between skilled and unskilled workers between 1980 and 1992.

Direct evidence for capital-skill complementarity was found for Peru 1995-2000, by Mazumdar and Quispe-Agnoli (2004), who (besides the demand shift decomposition described above) estimate the wage-bill share of skilled workers in a translog cost function. They find capital intensity has a strong positive impact on the relative demand for skilled labor. Still, the authors cautiously remark that their findings do not necessarily rule out the relevance of trade: “if trade stimulates investment in developing countries then it could also increase the relative demand for skilled labor in the presence of strong capital-skill complementarities” (p. 23). This mechanism, however, is not further investigated.

28 Finally, a direct test of the model of Feenstra and Hanson (1996) is provided by Zhu and Trefler (2005). The analysis includes twenty countries from all developing regions in the world, mainly from Asia and Latin America, over four periods between 1983 and 1997. Evidence confirms the prediction that technological catch-up in developing countries, as measured by the growth of manufacturing labor productivity, causes a shift of their exports towards more skill-intensive goods. This shift, in turn, increases relative wages of skilled labor. Importantly, technology catch-up does not increase the skill premium directly, but only through changes in export composition. Two weaknesses in this study are the measure of relative wages, which will be discussed in the next chapter, and the measure of technology-catch up: this is measured as the growth of manufacturing labor productivity, without controlling for changes in the share of skilled workers in manufacturing. Therefore, what is interpreted as a technology catch-up may well reflect increasing skill-intensity of production, which would be in line with our prediction that shifts in comparative advantage change the composition of trade and thereby determine the impact of trade on relative wages.

3.2 - Conclusions

The main results of these empirical studies are summarized in Table 2. All in all, the conclusions concerning trade effects on skill premia are mixed. With respect to technology, there seems to be more agreement, as the majority of studies find that capital intensity, FDI or some other measure of technological progress raises the skill premium. Te Velde and Morrissey (2004), however, find that FDI increased the skill premium only in Thailand, while there was even a negative impact of FDI in Hong Kong and the Philippines. These results together mainly emphasize the lack of cross-country consistency in most of the mechanisms studied so far. We believe this illustrates the importance of panel data studies, at least if we are interested in drawing general conclusions for the developing world. That is not to say that country studies are not valuable, but a challenge now is to find which country characteristics can account for the dissimilarities found. We expect that the supply of skilled labor is such a country characteristic, so the focus of the empirical part of this study will be on the role of skilled labor supply in shaping the effects of trade and technology on skill premia, as described in the previous chapter.

30

Table 2 – Overview empirical literature

Focus Study Analysis Country

Effect on skill premium Trade Hanson and Harrison (1995) Decomposition of demand shifts Mexico small

Mazumdar and Quispe-Agnoli (2004) “ Peru -

Kijima (2005) “ India small

Esquivel and Rodriquez-Lopez (2003) Mandated Wage Approach Mexico -/0

Gonzaga et al. (2006) “ Brazil -

Hanson and Harrison (1995) Correlation price changes-skill intensity Mexico 0 Hanson and Harrison (1999) Correlation tariff changes-skill intensity Mexico - Bigsten and Durevall (2006) Cointegration openness-rel. wages Kenya - Te Velde and Morrissey (2004) Regression openness-rel. wages Korea +

“ Thailand +

“ Philippines -

Technology diffusion Robbins and Gindling (1999) Imported capital stock Costa Rica + Gopinath and Chen (2003) Regression FDI-skilled income share 11 LDCs + Te Velde and Morrissey (2004) Regression FDI-rel. wages Thailand +

“ Philippines -

“ Hong Kong -

Hanson and Harrison (1995) Correlation foreign investment-rel. wages Mexico + Capital-Skill

Complementarity Mazumdar and Quispe-Agnoli (2004)

Regression capital intensity- skilled labor

wage bill share Peru +

Hanson and Harrison (1995) Correlation capital intensity-rel wages Mexico +/-

SBTC Te Velde and Morrissey (2004) Time trend (linear) East Asia 0 Trade in intermediate

31

4 – METHODOLOGY

4.1 - Model

As discussed above, our analysis will focus on the role of relative skill supply in the relationship between trade and technology and the skill premium. Given the fact that this study will employ macroeconomic data for a broad cross-section of countries, a suitable approach is to use a supply and demand framework. An empirical model can be derived from a constant elasticities of substitution (CES) production function. The CES production function is attractive for the possibility to include demand- and supply-side effects, and through the specification of labor efficiency indices various trade and technology variables can easily be included.

The simultaneous rise of wages and supply of skilled workers relative to unskilled workers in many countries can be graphically illustrated in a simple supply and demand framework. In Figure 5, growth in the relative supply of skilled workers shifts the relative supply curve from S0 to S1, which would cause a drop in the relative wage along the demand curve D0. If this growth in relative supply is accompanied by a large enough demand shift, however, the relative wage rises, which is the case with a rightward shift of the relative demand curve from, say, D0 to D1.

Figure 5 – Relative Supply and Demand

32 The size of such a demand shift can be identified within a two-factor CES production function, assuming full employment and exogenous relative supply (Katz and Murphy, 1992: 68):

(1)             −       =       ) ( ) ( log ) ( 1 ) ( ) ( log t x t x t D t w t w u s u s

σ

,

where the subscripts s and u refer to skilled and unskilled labor, respectively, and w and x refer to wages and supplies, respectively. D(t) is the unobservable time series of relative demand shifts in log quantity units, and σ is the unknown elasticity of substitution between skilled and unskilled labor. Equation (1) illustrates that with a greater elasticity of substitution, relative supply changes have a smaller effect on relative wages, so to explain a given change in relative wages for some change in relative supplies, we need greater demand shifts D(t). Assuming a value σ0 for the elasticity of substitution, the demand shift time series can be obtained from

(2)

_











+













=

)

(

)

(

log

)

(

)

(

log

)

(

₀

t

x

t

x

t

w

t

w

t

D

u s u s

σ

.

The approach that Katz and Murphy (1992) take, is to obtain an estimate of σ from equation (1) assuming D(t) is a linear time trend. This value for σ, and some values around it, is then used to impute D(t) from equation (2).

Te Velde and Morrissey (2004) also start from the two-factor CES technology, and add functions of labor efficiency units to the model. This is an elegant way to incorporate factors that may affect the productivity of skilled labor differently than that of unskilled labor. We use a similar specification as Te Velde and Morrissey, which is a production function with unskilled labor (U) and skilled labor (S):

(3) f(U_t,S_t)=

{

λ

(

ψ

_UtU_t)ρ +(1−

λ

)(

ψ

_StS_t)ρ

}

1/ρ ,

33

technical change, and country specific technological change (tech) affecting labor efficiency, these indices are

(4)

ϕ

_Ut ≡

ln

ψ

_Ut ;

ϕ

_Ut =

γ

_1Ut+

γ

_2U

ln(

tech

)

_t St

St

ψ

ϕ

≡

ln

;

ϕ

_St =

γ

_1St+

γ

_2S

ln(

tech

)

_t.

Under perfect competition, real wages equal the marginal products per unit of labor input, which gives the relative wage equation

(5) Ut St t Ut t St Ut St U S w w

ψ

ψ

ψ

ψ

λ

λ

ρ ⋅       ⋅ − = −1 1

Substituting the labor efficiency indices given in equation (4) and taking the log, the estimation equation looks as follows (with subscript i for country and t for time)

(6) _{it it} it it U S tech t U S w w

ε

γ

σ

σ

γ

σ

σ

σ

λ

λ

+ − + − +       −       − =      

)

ln(

1

1

ln

1

1

ln

ln

_{1 2} ,

where the first term on the right-hand side is a constant, γ1= γ1S – γ1U , and γ2 = γ2S – γ2U. A positive γ1 would reflect skill-biased global technical change, and a positive γ2 indicates that country-specific technology (in the form of FDI, capital, high-technology imports, etc.) is skill-biased. Following this approach, the elasticity of substitution is estimated along with the other coefficients, with relative wages as the dependent variable. This differs from the procedure Katz and Murphy (1992) follow, who first net out relative supply shifts from relative wage changes and then examine the relative demand shifts. A priori there is no reason not to estimate (6) directly. The estimated elasticity of substitution can be compared to available evidence, and supply and demand variables can interact in the determination of relative wages.

34

(including differences in data gathering and reporting standards) is probably larger for levels than for changes. (7) _{it it it} it it U S trade tech U S w w

ε

β

β

β

β

 + ∆ + ∆ +      ∆ + =       ∆

ln

_{0 1}

ln

₂

ln(

)

₃

ln(

)

,

where β0 replaces the time trend coefficient in equation (6), which reflects global skill-biased technical change. The second coefficient, β1, is the negative and the inverse of the elasticity of substitution between skilled and unskilled labor, and β2 measures the effect of country-specific technology on the skill premium. The last coefficient, β3, measures the impact of trade. To analyze the role of relative skill supply beyond its main effect (the second right hand side variable), we will later add interaction terms.

4.2 – Data

The period of study is 1970-2000, which is determined mainly by data availability. For all data, average annual growth for six periods is calculated: 1970-1975, 1975-1980, 1980-1985, 1985-1990, 1990-1995, and 1995-2000. The reason for taking these period averages is that the available skill supply data are quinquennial, which limits the sample size considerably.

Skill premium

The main challenge in obtaining data for this study is in finding a good proxy for the skill premium. As discussed in Chapter 3, many different calculation methods have been used in research so far, based on different definitions of skilled and unskilled workers. The International Labor Organization (ILO) is a source of data with broad geographical coverage and systematic reporting over the past decades, and the ILO online source LABORSTA includes two useful datasets. One is the dataset Wages and Hours of Work in 159 Occupations for the period 1983-2005, based on the ILO October Inquiry. The other is the dataset Wages in Manufacturing for the period 1969-2005 (table 5B in Yearly Data). We calculate different skill premium series from these two sources, to be able to assess and compare two common methods of calculation.

35

From this dataset we remove all European countries, the US, Canada, Australia, New Zealand, former Soviet Union countries, countries in the Middle-East and a number of small countries with a population below one million (e.g. Comoros, French Polynesia, Guam, Samoa). The final sample consists of 62 less developed countries in Asia, Africa and Latin America.

To obtain a measure of the skill premium from these data, two methods can be used. The first one is to take the ratio of the 90th to the 10th percentile in the wage distribution, as do Freeman and Oostendorp (2001). The second method is to classify all occupations as either skilled or unskilled, and then calculate the average wage per group and take their ratio. Te Velde and Morrissey (2004) use this method, and define as skilled workers all occupations in groups 0/1, 2 and 3 of the 1968 International Standard Classification on Occupations (ISCO-68). These are “Professional, technical and related workers”, “Administrative and managerial workers”, and “Clerical and related workers”. In the ISCO-88, the skilled workers are groups 1, 2, 3 and 4: “Legislators, senior officials and managers”, “Professionals”, “Technicians and associate professionals”, and “Clerks”. All other occupations are classified as unskilled. Freeman and Oostendorp have stressed the fact that cross-country and cross-time comparability depends on the number and type of occupations reported. Since there is neither overlap across countries, nor complete overlap within countries across years, the skill premia are not strictly comparable.

We calculate the skill premium both ways, using all available data. In addition, to improve the cross-time comparability, we make the same calculations using only overlapping occupations within countries (the ‘overlap’ set). A problem that remains is the cross-country comparability: the number of observations per occupation ranges from 85 to 448 in the full set and from 26 to 397 in the overlap set. With 542 country-year points in the data, this means there is not a single occupation reported for every country-year. To somewhat improve the quality of the final series we remove all skill premia based on less than 25 occupations. In the overlap set, this is the case for 162 country-years, which constitutes 18 out of 62 complete countries.

36

classified. This will result in a less precise skill premium, for all industries actually employ both types of workers. Still, this type of classification is not uncommon in the relative wage literature, and the advantages pointed out above are important. We identify the manufacturing industries with lowest and highest skill-intensity within the International Standard Industrial Classification (ISIC), based on studies by Keesing (1971), Cheon (1999), Peneder (1999), and Wörz (2003). The unskilled industries are Textiles, Wearing apparel, Leather, and Footwear (321-324 in ISIC-Rev.2 and 17-19 in ISIC-Rev.3). The skilled industries are chemicals and chemical products, petroleum and petroleum products, and a number of machinery and equipment industries (351-354 and 382-383 in ISIC-Rev.2 and 23, 24, 29-32 in ISIC-Rev.3). To summarize, Table 3 describes the five different measures of the skill premium and Table 4 shows descriptive statistics for all series.

Table 3 –Skill Premium Series

W1 Skilled/Unskilled, total occupational wages dataset W2 Skilled/Unskilled, overlapping occupational wages dataset W3 90th/10th percentile, total occupational wages dataset W4 90th/10th percentile, overlapping occupational wages dataset W5 Skilled/Unskilled, manufacturing wages dataset

Table 4 – Skill Premium, Descriptive Statistics

W1 W2 W3 W4 W5 Mean 1.721 1.658 3.994 3.663 1.876 Median 1.665 1.605 3.505 3.334 1.750 Maximum 3.857 3.501 13.70 13.18 3.997 Minimum 0.336 0.327 1.071 1.067 0.738 Std. Dev. 0.548 0.572 2.050 1.866 0.664 N 459 356 461 356 556 N countries 60 46 60 46 31

Sources: W1-W4 from Freeman and Oostendorp (2001) and own calculations,

W5 from ILO Wages in Manufacturing Industries and own calculations.

37

Table 5 – Skill Premium, Pairwise Sample Correlation Coefficients

W1 W2 W3 W4 W5 W1 1 .781* .529* .286* -.087 W2 1 .365* .379* .084 W3 1 .827* -.107 W4 1 .254* W5 1

* Significant at the 1 percent level

The correlation coefficients in Table 5 and descriptive statistics further indicate that the data are comparable between the full and the overlap dataset (compare W1 to W2 and W3 to W4), but not across different methods of calculation. These statistics conceal, however, that for particular countries, the difference between W1 and W2 or between W3 and W4 is substantial. For Mexico, for example, wages are reported for 39 overlapping occupations. In the original data, however, the number of occupations ranges from 39 in 1992 and 1993 to 134 in 2000. As Figure 6 below shows, this variation in the number of occupations causes sizeable variation in the skill premium, which almost only reflects measurement error. This one example is enough to illustrate the great effect that the coverage of the data can have on the skill premium calculations. As there is no set of occupations for which all countries report wages for one or more years, it is impossible to rule out this type of variation across countries. That is, the cross-country variance in the first four skill premium series will always partly reflect the cross-country differences in the number and type of occupations covered by the data. Therefore, the occupational wage data should not be used for panel data analyses.

38

Figure 6 – Skill Premia Mexico

0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 W1 W2 W3 W4 W5

Source: Freeman and Oostendorp (2001) and own calculations

All in all, the relative wages based on manufacturing industry wages may be a less accurate measure of the skill premium, but the cross-country comparability is much better, since all countries report on nearly all the same industries. Based on the foregoing discussion, we believe it makes little sense to perform a panel estimation using the occupational wages, so we will only use the W5 series in our analysis. This means the final sample will consist of 31 countries at most, which are listed in the appendix.

39

Figure 7 – Skill premium W5, cross-country average +/- 1 standard deviation

0.8 1.2 1.6 2.0 2.4 2.8 3.2 1970 1975 1980 1985 1990 1995 2000 Mean +/- 1 S.D.

Source: ILO Wages in Manufacturing Industries and own calculations

Skilled labor supply

The supply of skilled workers is obtained from the Barro and Lee (2001) dataset on educational attainment, which includes quinquennial enrollment rates and average years of education over the period 1960-2000. We use the data for the population of age 15 and over, rather than age 25 and over, as this “corresponds better to the labor force for many developing countries” (Barro and Lee, 2001:2). Skilled workers are those who have completed secondary education or are enrolled in/have completed postsecondary education, and the remaining population is unskilled. Ideally, the definition of skilled labor perfectly matches the definition of skilled industries, from which the skill premium is calculated. Such a perfect match is impossible to realize on a macroeconomic level. Recognizing this, we construct a second measure of the relative skill supply with a broader definition of skilled labor, which also includes the working age population enrolled in secondary education. The educational attainment data are not available for Puerto Rico, which therefore drops from the sample.

40

Table 6 – Cross-country average relative skill supply

SU1 SU2 Year Mean Std. Dev. Coefficient of Variation N Mean Std. Dev. Coefficient of Variation N 1970 0.080 0.064 0.800 29 0.249 0.167 0.671 28 1975 0.100 0.080 0.800 30 0.304 0.205 0.674 30 1980 0.147 0.132 0.898 30 0.433 0.302 0.697 30 1985 0.162 0.128 0.790 30 0.523 0.417 0.797 30 1990 0.241 0.275 1.141 30 0.672 0.573 0.853 30 1995 0.290 0.353 1.217 30 0.777 0.688 0.885 30 2000 0.323 0.392 1.214 30 0.870 0.780 0.897 30 All 0.192 0.251 1.307 209 0.550 0.540 0.982 208

SU1: skilled workers are those with at least completed secondary education, SU2: skilled workers are those at least enrolled in secondary education. Source: Barro and Lee (2001) and own calculations.