Skill Content Tests of Endowment Models of Inter-a

research paper series Globalisation and Labour Markets Research Paper 2006/20 Skill Content Tests of Endowment Models of Inter-and Intra-Industry Trade: Evidence for Some High Income Countries by Manuel Cabral, Rod Falvey and Chris Milner The Centre acknowledges financial support from The Leverhulme Trust under Programme Grant F114/BF The Authors Manuel Cabral is a Professor Auxiliar at the University of Minho, Rod Falvey is Professor of International Economics and Research Fellow in GEP, University of Nottingham and Chris Milner is Professor of International Economics and Research Fellow in GEP, University of Nottingham. Acknowledgements Falvey and Milner gratefully acknowledge financial support from The Leverhulme Trust under Programme Grant F 114/BF. Skill Content Tests of Endowment Models of Inter-and Intra-Industry Trade: Evidence for Some High Income Countries by Manuel Cabral, Rod Falvey and Chris Milner Abstract The present study compares the results of factor (skill) content tests for different types of trade flows under alternative assumptions about the technologies used to produce imports and exports. Using data on trade, technologies (skill requirements) and national endowments for some high income countries, we show that the match between the actual factor content of trade and that predicted by endowments in an H-O-V framework improves substantially if technological heterogeneity across countries is allowed for and if the factor content of intra- industry trade (in particular in vertically differentiated goods) is included along with that in inter-industry or net trade. JEL classification: F11, F14 Keywords: Skill Content Tests, Intra-Industry Trade, Technology Differences … Outline 1. Introduction 2. Factor Content Measurement with Technology Differences 3. ‘Factor Content Tests’ with and without Technological Homogeneity 4. The Endowments-Factor Content Relationship for Different Types of Trade Flows 5. Summary and Conclusions Non-Technical Summary The hypothesis that trade patterns are at least partly determined by relative factor endowments has often been tested using the factor content of commodity trade. The traditional factor content methodology considered only net or inter-industry trade flows, and assumed that common technologies existed across trading partners so that the input matrix of one reference country was sufficient to measure the factor content of trade. This approach reveals trade to embody only relatively small net exchanges of factors when compared with the magnitude of the relative differences in endowments between countries, an outcome Trefler labelled the “mystery of the missing trade”. Subsequent empirical work using data on the input requirements of more than one country to calculate the actual factor content of total trade and/or measuring the factor content of intra-industry trade as well as net trade, has found stronger empirical support for an endowments explanation of trade. Here we extend this literature by allowing for product differentiation and distinguishing between the factor content of vertical and horizontal intra-industry trade when testing for the role of endowments in explaining trade flows. Vertical intra-industry involves two-way trade flows of vertically differentiated varieties, and can be explained in terms of differences in the relative factor input requirements of imported and exported varieties and of differences in the relative factor endowments of trading partners, much like inter-industry trade. Intra-industry trade in horizontally differentiated varieties is usually explained in terms of the scale economy motives for specialisation between trading partners with similar factor endowments. One would expect stronger support for a factor content test of an endowments explanation of vertical than of horizontal intra-industry trade flows. The tests that we employ have become standard in this literature. The sign test measures how often the pattern of factor exchange embodied in commodity trade can be predicted on the basis of the trading partners’ factor endowments. As its name implies, the rank order test compares the ranking of embodied factor trade with the ranking of relative factor abundance. The slope test regresses the measured factor content on that predicted by endowment differences. We apply these tests to the trade of the high income Western European countries with 27 middle income and developing countries. Decomposition of intra-industry trade into its vertical and horizontal components requires a high level of disaggregation, and our data sources allow us to use 201 different subsectors of manufacturing industry. The labour force is divided into four skill levels. Our results confirm that the actual factor content of all trade should be measured when exploring an endowments’ explanation of trade flows. The tests show that, once technology differences and product differentiation are allowed for, endowments can predict the factor content of both inter- and intra-industry trade flows with a fair degree of accuracy. The match of the measured skill content of vertical IIT flows is similar to that of net or inter-industry trade flows, while endowments have very little explanatory power in accounting for horizontal IIT trade flows. Our results also suggest that differences in the factor requirements of vertically differentiated varieties might play a more important role than differences in factor prices or technology, in explaining differences in factors used in the same sector in different countries. 1. Introduction There is now a sizeable literature using the “factor content” of commodity trade patterns to “test” the hypothesis that trade patterns are (at least partly) determined by countries’ relative factor endowments. The traditional factor content methodology considered only net or inter-industry trade flows, and assumed that common technologies existed across trading partners so that the input matrix of one reference country was sufficient to measure the factor content of trade. This was in part because the empirical methodology related directly to a well-established model - the Heckscher-Ohlin-Vanek (H- O-V) model – whose assumptions implied that outcome. It was in part also empirically convenient because it avoided the need to gather information about input requirements for more than one country. Most studies in fact used the input requirements of the US to measure the factor content of trade with both other developed countries and developing countries. This study, in line with other recent factor content studies, departs from the HOV approach by allowing for international differences in technology (Trefler, 1995). It also incorporates the implications of product differentiation in internationally traded goods in a way that significantly affects the measurement and interpretation of factor content evidence. In this we follow the recent studies of Hakura (1999; 2001), Davis and Weinstein (2001a,b), and Trefler and Zhu (2000) in measuring the factor content of trade using information on techniques of production from more than one country. We go further, however, by using more disaggregated data on industry and labour classifications and by taking into account the effects of product differentiation on the type of trade flow. This allows us to distinguish between the factor content of inter- and intra-industry trade and between horizontal and vertical intra-industry trade flows. When subjected to empirical verification by the traditional approach the H-O-V model tends to be rejected. 1 The traditional approach reveals trade to embody only 1 The studies that presented complete tests to the model using simultaneously data on factor endowments, trade and input requirements (Maskus, 1985; Bowen et al., 1987; Brecher and Choudhri, 1988; Staiger, 1988; Kohler, 1991) reveal contradictory results. A large number of exceptions to the prediction of the Vanek version of the H-O model were found – with the signs and ranks of the net export of factors not matching closely those predicted by the relative factor abundance. 1 relatively small net exchanges of factors when compared with the magnitude of the relative differences in endowments between countries. Trefler (1995) labels this the “mystery of the missing trade”. Generalising the model to allow Hicks-neutral differences in technology or differences in demand conditions between countries does contribute to bringing it closer to the data (see Trefler, 1995), but it would be difficult to conclude that allowing for these international differences alone empirically salvages the factor endowments explanation. Subsequent empirical work has found more empirical support for an endowments explanation of trade. Hakura (1996), for example, uses data on the input requirements of several European countries to calculate the actual factor content of total trade. Davis and Weinstein (2001b) measure the factor content of intra-industry trade as well as net trade and find that “in half of the rich OECD countries in our sample, intra-industry trade is more important than inter-industry trade in the net export or import of factor services” and that “intra-industry trade is in fact one of the principal conduits of net factor trade” (Davis and Weinstein 2001b, p 17). But while they allow for the existence of exchanges of factors in matched trade flows due to differences in technology or factor prices, they do not consider the role of product differentiation in generating different types of intra-industry trade2. Here we draw on the recent literature by allowing technological differences and for factor exchanges in other than inter-industry (net) trade, but also extend it by allowing for product differentiation and distinguishing between the factor content of vertical and horizontal intra-industry trade when testing for the role of endowments in explaining trade flows. Vertical intra-industry trade models (e.g. Falvey, 1981) explain two-way trade flows in vertically differentiated varieties in terms of differences in the relative factor input requirements of imported and exported varieties and of differences in the relative factor endowments of trading partners. Intra-industry trade in horizontally differentiated varieties is usually explained in terms of the scale economy motives for specialisation between trading partners with similar factor endowments (e.g. Krugman, 1979; Lancaster, 1980). One would expect stronger support for a factor content test of an endowments explanation of vertical than of horizontal intra-industry trade flows. One reason that other recent factor content studies may not have decomposed total intra-industry trade is that they have used relatively aggregated concepts of an industry - typically a disaggregation of trade data close to the 2 digit SIC classification (i.e. about 20 2 Admitting the latter explanation for differences in factors used in the exports and imports of each industry has important implications for trade- induced adjustment costs, as noted below. 2 industries). Decomposition of intra-industry trade into its vertical and horizontal components at such a high level of aggregation would be inappropriate. Further, as Feenstra and Hanson (2000) show, aggregation poses important bias problems for the calculation of measured factor content. In the present study we integrate their suggestions using the maximum disaggregation that was possible to match the different data sources. This in the end resulted in 201 different sectors of the manufacturing industry, a level of disaggregation more in line with what is considered appropriate in the intra-industry trade literature. The rest of the paper is organised as follows. In section 2 the modelling framework is set out for measuring the actual factor content of different types of trade in the presence of international technology differences. Section 3 outlines the various factor content tests (sign, rank and regression tests) employed and compares these test results for net trade flows only with and without technological differences. For the preferred case of allowance for technological differences, we then in section 4 report the factor content tests for different types of trade flows (inter- and intra-industry; vertical and horizontal intra- industry). Finally our summary conclusions are set out in section 5. 2 Factor Content Measurement with International Technology Differences Studies that use the factor content method traditionally consider the input matrix of factor requirements of only one country to measure the factor content of both imports and exports. This approach assumes that products are homogeneous, there are identical technologies in all countries and that trade leads to factor price equalisation so that identical techniques are employed. The input requirements of exports and imports would then be identical.3 In this context the net factor content of intra-industry trade (IIT) is zero because the factors embodied in symmetric trade flows are matched. We begin with a HOV equation4 explaining the factor content of bilateral trade between two countries on the basis of their factor endowment differences. We label the two countries P and U since we later 3 This is true both in the context of the Heckscher-Ohlin model, which does not predict the existence of IIT, and in monopolistic competition general equilibrium models (Helpman and Krugman, 1985) which assume that all intra-industry trade is horizontal, and production technologies in the differentiated goods sector are identical across countries. 4 Strictly speaking the bilateral factor content of trade cannot be predicted under the strict assumptions of the HOV setting, namely with factor price equalization. Equation (1) follows the “bilateral comparison” used by Davis and Weinstein (2001b), and is included simply to compare the results obtained in this way with those obtained when other specifications are chosen. 3 apply this method using factor requirements matrices for Portugal and the UK. The equation is: FjUP = EjU – sU(EjU + EjP) (1) with FjUP = AUNTUP (2) Here FjUP is the embodied trade in factor j measured from the direct requirements5 of the bilateral trade between countries U and P, jiE is the endowment of factor j in country i, is the bilateral income share of country i, and is the input requirements matrix used (which is implicitly assumed to be the same in both countries). The vector of trade flows considered in this case is only that of net trade flows in bilateral trade between the two countries (NT is UA UP). This approach can be modified readily to allow for uniform factor productivity differences between countries. Here the factor endowments are transformed by productivity differences so that they are equivalent in terms of production potential.6 Introducing this “correction” yields: FjUP = EjU – sU(EjU + δPEjP) (3) where δP is the productivity parameter for country P, and Uδ has been normalised at unity. Once we drop the assumptions that countries have identical technologies or that industries produce homogeneous goods (and allow imports and exports to represent different varieties produced using different technologies), the importer’s input matrix no longer captures the actual factor content of imports. We then need measures based on two different input requirements matrices. For bilateral trade between U and P the equation is: (AFjUP)/ψUP = [EjU – sU(EjU + EjP)] (4) With the actual factor content being measured by: AFjUP= AUXUP – APMUP (5) 5 Several authors (Staiger, 1986; Maskus et al, 1994; Bowen et al, 1987) discuss whether direct factor requirements or indirect factor requirements (including the factors used in the intermediates) should be considered in the calculations of the factor content. We accept the position of Maskus et al (1994), following Staiger (1986), who argue that direct requirements are more appropriate for the case of small open economies that trade intermediate goods freely at world prices, and measure only the direct factor content of UK trade. This is also convenient given data constraints on measuring indirect factor requirements at the level of trade disaggregation used here. See Trefler and Zhu (2000), however, for a discussion of how using direct inputs only may affect the measurement of factor content. 6 Here the adjustment consists of transforming the available man years of different countries into efficiency equivalent units of man years. A range of proxies were considered to capture productivity differences, including differences in output per worker and per capita GDP. The results reported use differences in average wages in manufacturing across the countries. Results for alternative means of correcting for productivity differences are available from the authors on request. 4 where XUP and MUP are, respectively, exports of U to P and imports of U from P represented in different types of trade flows (net trade; intra-industry trade; horizontal and vertical intra- industry trade). In (4) each type of trade flow is scaled by the parameter ψUP, which is the proportion of that type of trade flow in total bilateral trade.7 Equation 4 can also be subject to a factor productivity correction yielding: (AFjUP)/ψUP= [EjU – sU(EjU + δPEjP)] (6) A technique that can be applied only to a particular bilateral trade flow is of limited interest, and ideally we would like to consider the factor content of all bilateral trades for a range of countries. However, complete information on input requirements for all trading partners is unlikely be available, leaving little choice but to proxy the technologies of most countries. We suggest two ways in which this might be done. Each is based on the notion that countries at similar levels of development have similar technologies. Our first approach is to apply a “representative” matrix for all countries at a similar level of development. Under this approach we apply the UK matrix for all high income developed countries (HID) and the Portuguese matrix for all middle income countries (MID). Our alternative approach is to estimate the matrix for a third country (R) using a linear combination of PA and where the weights are based on the per capita GDPs of the three countries (i.e. , j = R, P and U). Thus UA jy [1 ] [ ]R R U R P R U PA A A A A PAθ θ θ= + − = − + (7) where is the estimated input requirements matrix for country R. Assuming that , the weighting parameter RA 0U Py y− > Rθ is given by: [ ] [R R P U P ]y y y yθ = − − , for all countries where 0R Py y− > ; and [ ] [R R P U R ]y y y yθ = − − , for all countries where