TRADE, TECHNOLOGY AND WAGE INEQUALITY IN DEVELOPING COUNTRIES: EVIDENCE FROM INDIAN MANUFACTURING. Chris Milner 1 Dev Vencappa Peter Wright.

TRADE, TECHNOLOGY AND WAGE INEQUALITY IN DEVELOPING COUNTRIES: EVIDENCE FROM INDIAN MANUFACTURING Chris Milner 1 Dev Vencappa Peter Wright Abstract This paper explores the roles of trade and technological change behind the rising wage inequality observed in Indian manufacturing following the 1991 trade policy reforms. We use the mandated wage methodology and explore the effects of technological change on wage inequality within a unified general equilibrium framework that accounts for the wage effects of both price and technological change. Two versions of the model are estimated, depending on whether the maintained assumption is that price and TFP growth are exogenous or endogenous. Under the assumption of exogeneity, we find that post-reforms, price changes and technological change pull in opposite and expected directions on wage inequality, but the forces of technological change are of a greater magnitude. Assuming endogeneity of price and technological change, we find that the rise in inequality post-reform is due only to technological change, and not price changes, although the magnitude of wage changes mandated by the model are quite high compared to actual wage changes. Our results confirm the findings of Berman, Somanathan and Tan (2005), who argue that part of the increase in the relative demand for skilled workers is due to skill-biased technological change. While our technology variable captures all types of technological change, our results seem to point out that a large component of this technical change may actually be skill-biased in nature. Work-in-progress: comments welcome April 2005 Preliminary draft: please do not quote. 1 The authors are all at the School of Economics, University of Nottingham. 1

1.0 Introduction Virtually all analysts of labour markets now accept as fact the deterioration in employment prospects and relative returns to labour market participation for unskilled workers in developed countries in the 1980s. Many studies have documented these trends and changes (Levy and Murnane, 1992; Bloom & Brender, 1993; OECD, 1993; Goldin and Katz, 1995; Mincer, 1996; OECD, 1997; Gottschalk & Smeeding, 1997; Murphy & Topel, 1997). What is controversial, however, is the relationship between these trends and patterns and any (or all) of the elements of globalisation. Just about a decade ago, these trends were generally ignored by trade economists, and were mostly the concern of labour economists; for instance, analyses of causes of growing earnings inequality, like Levy and Murnane (1992), gave little weight to global factors, such as trade. However, the last two decades witnessed an unprecedented number of research papers on the possible links between globalisation and the labour market changes of the types witnessed in the 1980s. The specific focus of such research was on the possible impact of globalisation on wages and employment and more specifically, on increasing wage inequality between skilled and unskilled workers in developed countries. Many have tried to research the causes behind such rising inequalities, and have come up with several competing explanations. Of the many candidates explored, trade and technological change have come out as the most plausible explanations. Trade economists have turned to the Hecksher-Ohlin-Samuelson (HOS) model as a potential explanation to the worsening income positions of unskilled workers, because of its simple postulate relating greater trade to factor rewards. According to this model, international trade will induce a decline in the reward to the production factor(s) a country is relatively badly endowed with. Viewed from a global perspective, the implications of the model are that developed countries will witness an increase in wage differentials between skilled and unskilled workers, while the converse is true for developing countries due to their endowments of these two categories of labour. The main competing alternative explanation to the observed skill wage differentials is skill-biased technological change, going back to the observation of Lawrence and Slaughter (1993) that despite the increase in the skill premium in the U.S since the 1970s, the skilled workers share in employment has been increasing as well. The extensive research into the outcomes of the HOS model led to the 2

conclusion that the evidence has weighed more in favour of skill-biased technological change. While initially such research was focused on developed countries, over time it has moved slowly towards developing countries (mostly East Asian and Latin American countries). An interesting facet of these developing countries labour markets has been the coincidental increase or decrease in skill-wage differentials with their trade liberalisation episodes. The East-Asian evidence, has witnessed reduced wage inequality and this has been attributed to greater trade: in a critical review of these East-Asian studies, Wood (1994) point out gaps and deficiencies in the relative wages data used in these studies, but nevertheless claims that: most of the evidence supports the conventional view that the adoption of more outward-oriented polices increases the demand for workers with only a basic general education relative to the demand for workers with more education and skills. The evidence is also consistent with the theoretical prediction that a switch in trade regime causes a step (or once-and-for-all) change in the composition of demand, whose effects on wage differentials appear to be spread over a period of about ten years. Wood (1997, p.42) While this statement is close to concluding that trade is responsible for the narrowing wage differentials in these countries, Wood (1997) however emphasises that more research into other developing countries, African and South Asian, is needed as the available empirical evidence is limited to East Asia and Latin America only. The Latin American experience has been different, with increasing skill premiums coinciding with their trade liberalisation episodes (e.g. Cragg and Epelbaum, 1996; Hanson and Harrison, 1999 and Feliciano, 2001 for Mexico; Beyer et al, 1999 for Chile). An in-depth exploration of the causes of such rising wage inequality has however produced mixed explanations, of which trade and technological change have been isolated as mutually exclusive forces. 2 Evidence for technological change is 2 As Cragg and Epelbaum (1994) point out, there are at least two other competing explanations to this shift in relative demand in favour of skilled workers: (a) the reallocation or dissipation of rents which affect unskilled workers disproportionately; (b) entrepreneurs or managers who use the new opportunities created by institutional changes to catalyse change, and who are paid a premium since 3

produced by Cragg and Epelbaum (1996), Feenstra and Hanson (1997b), Lopez- Acevedo (2002a, 2002b) for Mexico, Tan and Batra (1997) for Colombia, Mexico and Taiwan and Pavcnik (2000) for Chile. Pro-trade explanations, on the other hand, include Hanson and Harrison (1999), 3 Robertson (2000a, 2000b) for Mexico and Beyer et al (1999) for Chile. While the above studies view trade and technological change as two mutually exclusive forces, other studies have explored the possibility that trade brings about technological change and interact to affect the wage structure. Robertson (2000c) uses logit and multinomial logit models for Mexico and finds that exposure to foreign markets increases the likelihood of introducing new and more advanced technologies. Alvarez and Robertson (2002) find a positive relationship between exposure to foreign market and firm-level innovation for Chile and Mexico, with the role of exporting being more pronounced. These studies suggest that such innovative technology, via the trade liberalisation channel, if skill-biased in nature, will drive up wage differentials. On balance, the evidence for the Latin American studies quoted above seems to be inclined towards skill-biased technological change. Over time, a few studies from other parts of the developing world have emerged although not all of them have focused explicitly on wage inequality. In the African context, Teal (2000) reports a rise in the relative wages of skilled workers in Ghana, but observes that this has not resulted in a rise in the share of skilled labour in total wages, as this rise in relative wages has led to the substitution to unskilled workers. Other studies have attempted to test the impact of trade openness on labour demand elasticities. For instance, for the Middle East, Krishna, Mitra and Chinoy (2001) do not find any evidence of any relationship between trade openness and labour demand elasticities using Turkish plant-level data. India s experience with trade reforms in 1991 has not been very different. Wage differentials in manufacturing have been increasing at exactly the same time as the reforms were implemented (see figure 1), suggesting a possible link between the two. While the rising wage inequality has not been as drastic as in the developed economies, there is nonetheless cause to worry, as the clear pattern emerging from their supply is relatively limited. Empirical studies have however discarded these two explanations as insignificant. 3 While Hanson and Harrison find that tariff reductions were larger for less skill-intensive industries, they do not find any significant evidence between changes in output prices and relative wages. 4

figure 1 is that wage inequality looks set to have adopted an increasing path. 4 There have been very few studies that have attempted to analyse the link between globalisation and wage inequality for India. A few studies have however attempted to link the Indian trade reforms to changes in it s labour market. For instance, in an attempt to provide quantitative estimates of the effects of trade on manufacturing output and employment, Nambiar and Tadas (1994) find that trade had squeezed manufacturing output and employment except in resource-intensive industries. They conclude that this observation was pointing to a gradual increased demand for unskilled labour. Nambiar, Mangekar and Tadas (1999), using an extended data set found similar results: trade had actually shrunk manufacturing base both in terms of value added and employment and was shifting manufacturing from high skilled to low-skilled and labour-intensive production. They also find that the existing wage disparity between skilled and unskilled labour was worsening but merely concluded that these patterns were at odds with the simple Hecksher-Ohlin-Samuelson model, without attempting to explain this controversy. 5 Using survey level data, Kambhampati and Howell (1998) find that trade liberalisation had led to a decline in formal sector employment in the cotton mills in India through closure, downsizing and a shift towards more capital-intensive technologies. Wage levels have however been maintained for those still employed, due to protective legislations and formal agreements between unions and employers. Where firms have begun to modernise, demand for skilled labour had increased, leading to higher wages. More rigorous methods of testing for the impact of trade reforms on labour markets have also been used in the context of Indian studies. For instance, Khambhampati, Krishna and Mitra (1997) utilise firm level data in five different import competing industries to assess the impact of trade reforms on the Indian labour market. First, they find only a small and insignificant effect of the reforms overall and in each of the five import competing industries. By investigating the relationship between labour demand and mark-ups in imperfectly competitive settings, they find that there is a significant negative relationship between the two, 6 overall and in four of the 4 Our data covers the period 1984-97 only, and this observation may not necessarily be true after 1997. 5 They went further to argue that this finding was at odds with their finding that trade had increased the weight of value added in low skill-intensive segments of manufacturing. 6 This reduction in industry markups following trade reforms is documented by Krishna and Mitra (1995) for India. Levinsohn (1991) and Harrison (1994) provide similar evidence for other countries. 5

industries. Their results provide evidentiary support for the theoretically suggested pro-competitive effects of trade reforms: trade liberalisation increases the demand elasticity perceived by firms and induces them to reduce mark-ups and increase their output. 7 This induces an increase in the demand for labour, which may at least partially offset the reductions in the demand for labour caused by other factors. Hasan, Mitra and Ramaswamy (2003) find that lower protection raises labour demand elasticities, but these responses are conditioned by the nature of labour institutions: states with more flexible labour markets experience larger increases in labour demand elasticities as protection levels are reduced. Dutt (2005) does not find any significant relationship between net exports and real wages, but finds that across sectors, the post-reform period witnessed a significant improvement in wages and employment, but argues that this should not be attributed to trade liberalisation, but rather to the rigidities in labour markets that hamper the labour mobility to expanding sectors. In a study of Indian manufacturing, Berman, Somanathan and Tan (2005) attribute the rising demand for skilled workers to three factors: increased output, an increase in capital-skill complimentarity, and skill-biased technological change. Beyond the above papers, we are not aware of any studies that directly attempt to link the wage inequality patterns to trade and technological change within the commonly invoked Hecksher-Ohlin-Samuelson framework. This study contributes to the developing country literature by exploring the causes behind the rising wage inequality in Indian manufacturing using the HOS framework. In particular, we look at trade and technological change as the two most commonly invoked causes behind changing patterns in wage inequality. We use a mandated wage approach along the lines of the model by Haskel and Slaughter (2001). Rather than looking at the factor bias of technological change, they argue that in a multi-sector framework, it is the sector bias of technological change that matters most. Using this model, we find evidence that the sector bias of price changes mandated a fall in wage inequality prereforms while technological change mandated a rise in wage inequality post-reforms. The plan of the paper is as follows. In section 2, we present some stylised facts on wage inequality in Indian manufacturing. Section 3 describes the theoretical 7 This is in direct contradiction to the predictions of competitive models of trade. 6

framework guiding our empirical analysis while section 4 describes the data set used and discusses the results. A concluding section follows. 2.0 Stylised facts on wage inequality in India Like many other countries, the phenomenon of increasing wage differentials in India coincided with a period of extensive trade reforms. Table 1 shows the picture for overall Indian registered manufacturing. We use the production/non-production dichotomy to reflect the unskilled/skilled labour distinction. 8 The annual average nominal wages are presented overall and for the two categories of labour. This is based on the average of wages across all manufacturing industries. We compute wage inequality as the ratio of skilled wages to unskilled wages. In 1984, skilled workers were receiving about twice the salary of unskilled workers, and by 1991, they were receiving only about 70 % more. By 1997 however, wage differentials had increased again to almost reach the level in 1984. These trends are clearer from figure 1, which tracks the movement in wage inequality for each of the years 1984-97. Table 1: Annual nominal earnings and wage inequality in Indian manufacturing Year Avg. wages (Rs) Avg. unskilled wages (Rs) Avg. skilled wages (Rs) Wage Inequality 1984 12408 10121 19641 1.94 1987 16438 13561 25120 1.85 1991 25955 22055 37261 1.69 1994 35043 29415 51125 1.74 1997 48404 39928 73235 1.83 Source: Author s computations from Annual Survey of Industries data. Note: All wages are annual nominal earnings per worker, averaged across the industries for each year. India s drastic reforms were implemented in 1991, and from figure 1, the coincidental increase in wage inequality from around 1991 onwards makes it tempting to attribute the rising wage inequality to these reforms. In the following sub-section, we analyse the relationship between skilled workers employment and their wages. 8 We recognise that there is a huge literature on the skilled/unskilled distinction that may question this classification, but data unavailability, meant we could not produce more suitable proxies for skill over a long period of time. Bhalotra (1996), Nambiar and Tadas (1994) use a similar classification. 7

1.95 Relative wages of skilled workers in Indian manufacturing 1.90 1.85 1.80 1.75 1.70 1984 1985 1986 1987 1988 1989 1990 Year 1991 1992 1993 1994 1995 1996 1997 Figure 1 2.1 Trends in employment and wages Figure 2 plots the share of skilled labour in total employment and their share of wages in total wage bill for overall manufacturing. 9 The share of skilled labour in total.4 Skilled workers share of employment and wage bill.35.3.25.2 1984 1985 1986 1987 1988 1989 1990 year 1991 1992 1993 1994 1995 1996 1997 Wage bill share Employment share Figure 2 9 These are averaged across all industries for each year. 8

manufacturing employment has been increasing slowly from around the late 1980s onwards while their share of wage bills has been increasing more rapidly. Since wage inequality has been increasing from around 1991 onwards, one explanation as to why skilled workers share of employment has increased (albeit slowly) over this period could be skill-biased technological change, going back to the observation of Lawrence and Slaughter (1993) that despite the increase in the skill premium in the U.S since the 1970s, the skilled workers share in employment has been increasing as well. It may be more insightful to see whether a similar pattern emerges across broadly defined skill-intensive and unskill-intensive industries. We define any industry above the median of the share of skilled workers in total employment as skill-intensive and the rest as unskill-intensive. This classification is intended to be illustrative only, and shows that the bulk of the industries are unskill-intensive industries (altogether, they represent about 68 % of total employment). Nambiar et al (1994) also provide a similar classification for Indian industries, but using different cut-off points. In their study, industries that are defined as skill- and medium-intensive fall under our list of skill-intensive industries. Figure 3 graphs the evolution of the share of skilled employment and relative wages in skill- and unskill-intensive industries. Skilled labour share of employment and wages.25.3.35.4.45 Skill-intensive Industries.15.2.25.3.35 Unskill-intensive Industries 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 year 1994 1995 1996 1997 1984 1985 1986 1987 1988 1989 1990 1991 1992 year 1993 1994 1995 1996 1997 Employment share Wage share Figure 3 9

On average, there seems to be a clear increase in skilled labour s share of employment and total wage bill in both skill- and unskill-intensive industries, suggesting that skillbiased technical change must have taken place across all industries. 2.2 Trade policy changes and relative prices. The underlying notion behind the HO model is that trade policy changes influence relative prices, which in turn affects relative wages by affecting intersectoral profitability. Therefore, the explicit assumption is that trade liberalisation changes relative prices. One of the main aspects of India s trade policy reforms was drastic cuts in tariff rates in 1991 followed by reductions in NTBs afterwards. Tables 2 & 3 detail these changes for two-digit manufacturing industries. Between 1989 and 1997, nominal tariff protection has gone down drastically in all manufacturing industries, although these rates are still considered high by developed countries standards. 10 India has also made some effort on the non-tariff barriers front, although these barriers are still high in many industries. The changes in these trade policy variables reflect the trade liberalisation process that India is going through and we use these to assess their impact on wage inequality via price and technological changes in our empirical application. 11 Figure 4 graphs nominal protection against the relative employment of skilled to unskilled workers. Note that the tariffs data are available from 1991 only on a consistent year-to-year basis. 12 In the left panel, we graph tariff levels in 1991 against relative employment levels (factor intensity) in that year. There seems to be no indication that trade policy was biased towards a particular category of labour prior to the reforms. In the right panel of figure 4, we graph changes in the tariffs between 1991 and 1997 against average relative employment over that period. Again, there seems to be no indication that India was protecting a particular category of labour at 10 Note however that the reforms in India are ongoing, and in the latest Trade Policy Review (2002) India pledged to reduce trade restrictions even further. 11 We have trade policy data at three-digits industry level. We wish to thank Rajesh Chadha, from the National Council of Applied Economic Research for kindly providing us with these data via Hasan Rana. 12 Earlier data is available for 1988, but there are no data for 1989 and 1990, and hence we preferred to construct the graph using data from 1991 onwards. There was no change in tariffs between 1988 and 1991 anyway. 10

Table 2- Structure of nominal rates of protection in India 1988-89 1991-92 1997-98 1998-99 Average all sectors N/A 105 34.4 40.2 Activity based Primary 92.25 91.86 20.61 23.85 Secondary 152.46 153.40 37.67 42.46 Industry based 1. Food, beverages and tobacco 144.70 147.45 40.81 46.27 2. Textiles and leather 145.36 142.38 42.57 46.06 3. Wood, cork and products 127.76 127.76 45 46.93 4. Paper and printing 149.56 131.57 35.11 40.40 5. Chemicals, petrol and coal 177.71 175.79 37.01 41.64 6. Non-metallic minerals 145.94 145.66 42.62 47.94 7. Basic metal industries 197.51 218.93 32.82 39.10 8. Metal products and machinery 138.17 139.44 35.06 39.28 9. Other manufacturing 151.48 151.25 37.76 42.04 10. Agriculture 91.09 90.05 22.08 23.98 11. Mining 106.46 114.04 18.80 22.26 Use based 1. Consumer non-durables 99.41 98.04 24.72 42.52 2.Consumer durables 153.21 153.97 38.12 44.77 3. Intermediate goods 128.87 131.96 29.08 38.88 4. Basic goods 188.77 201.32 33.73 40.19 5. Capital goods 124.70 125.99 32.32 38.79 Source: National Council of Applied Economic Research (NCAER), Delhi, India. Table 3- Import Coverage ratio for NTBs on Indian imports 1988-89 1997-98 1998-99 1999-00 Average all sectors 95.21 64.03 62.16 24.24 Activity based Primary 99.96 76.22 75.01 57.41 Secondary 87.43 39.42 36.3 27.71 Industry based 1. Food, beverages and tobacco 100 66.92 63.98 47.95 2. Textiles and leather 100 54.88 53.37 45.07 3. Wood, cork and products 100 34.48 26.41 5.74 4. Paper and printing 100 30.9 26.93 22.54 5. Chemicals, petrol and coal 97.54 30.74 26.24 15.45 6. Non-metallic minerals 98.25 50.52 47.04 36.28 7. Basic metal industries 53.37 15.85 11.41 8. Metal products and machinery 80.11 34.55 31.57 25.03 9. Other manufacturing 78.48 37.28 29.76 21.53 10. Agriculture 100 80.23 78.87 59.88 Use based 1. Consumer non-durables 100 75.65 74.07 56.19 2.Consumer durables 84.34 46.77 41.56 32.80 3. Intermediate goods 98.45 42.02 39.71 33.53 4. Basic goods 70.34 22.72 23.23 16.09 5. Capital goods 74.12 20.29 18.26 13.81 Source: Pandey (1999): National Council of Applied Economic Research (NCAER), Delhi, India. Note: all forms of NTBs observed assigned equal weight of 100 percent to arrive at these figures. 11

Tariffs and factor intensity Tariff levels, 1991 100 150 200 250 300 350 0.1.2.3.4.5.6.7.8.9 1 1.11.21.3 Relative employment Tariff change, 1991-97 -300-250 -200-150 -100-50 0.1.2.3.4.5.6.7.8.9 1 1.11.21.3 Relative employment Figure 4 the expense of the other, post-reforms. Unfortunately, there are no objective and detailed analyses of the changes in trade policy in India (e.g. the political economy behind the tariff changes) to inform us. In figure 5, we replicate the above exercise using NTB data (available from 1994 only) and although it is hard to establish the scenario pre-reforms, since import coverage of NTBs was 100 % in almost all industries during that period, post-reforms, there is some evidence that India was protecting its unskilled workers. NTB changes were lower in industries with low skillintensity. Changes in NTB and factor intensity Change in NTB, 1994-97 -100-50 0 50 0.1.2.3.4.5.6.7.8.9 1 1.1 1.2 1.3 Relative employment Figure 5 12

These tariff changes would have affected wages if they affected relative prices first. First, if these changes affected relative prices across industries, and if industry output responded to these relative price changes, then resources would have shifted towards the unskill-intensive industries. This shift would raise the demand for unskilled workers, and thus their wages. In figure 6, we plot relative wages along with relative prices. Both series are normalised so that 1984=1. The relative price series represents the ratio of prices in skill-intensive to unskill-intensive industries, using our earlier skill-intensity classification. Figure 6 shows that relative prices increased until around 1992, and starting falling afterwards. Relative wages on the other hand, were moving in an exactly opposite path to relative prices..85.9.95 1 1.05 Relative Prices and Relative wages 1984 1985 1986 1987 1988 1989 1990 year 1991 1992 1993 1994 1995 1996 1997 Relative prices Relative wages Figure 6 What does the above picture tell us in terms of the Stolper-Samuelson mechanics? Post-reforms, relative prices of skill-intensive industries seem to have gone down. Since these skill-intensive industries have become less profitable, one would expect an overall increased demand for unskilled labour, a rise in their wages and hence a fall in overall wage inequality. Nambiar and Tadas (1994) and Nambiar, Mangekar and Tadas (1999) note a gradual increase in demand towards unskilled labour. However, as we have seen from figure 4, exactly the opposite happened. Industries (whether skill-intensive or unskill-intensive) have been employing more skilled labour and this has given rise to correspondingly higher wages for this category of labour. These 13

findings rebel against the neat geometry of the simple HOS model, which predicts that wage inequality in developing countries should fall with reductions in protection rates. This suggests that other factors are at play, such as technological change. These findings are preliminary, and are only indicative of the overall picture in total manufacturing. In the following sections, we analyse the causes of wage inequality in Indian manufacturing using the more rigorous methods of assessing the impact of trade reforms and technological change on relative wages that have been used in the literature. First, we describe the theoretical and empirical framework guiding our empirical analysis. 3.0 The mandated wage methodology: a theoretical framework While the HOS model has clear predictions, they are not met in real world. However, testing the implications of the model is not in itself a straightforward task. Empirical studies have attempted to provide checks on whether changes in labour markets are consistent with the predictions of trade theory instead of looking for proof that the changes in labour markets are the results of trade liberalisation. Yet, one cannot discard the model as irrelevant, because there are a number of qualifications that apply to any empirical application of the HOS model. First, Stolper-Samuelson effects are explicitly long run effects. Given that real world processes rarely reflect pure equilibrium states, it is important that any empirical application of the HOS model takes account of both the long run nature of the equilibrium relationship predicted by the theory, as well as the fact that adjustment to equilibrium and dynamics may be an important feature of the modelling of the impact of globalisation on labour markets. A second consideration is that the theory explicitly presumes a uniform impact of trade liberalisation across all sectors in an economy. This assumption of homogeneity across sectors is clearly questionable. One needs to understand the political economy behind trade liberalisation, which often explains why sectors differ substantially, from the degree or even the direction of liberalisation. The presence of non-tariff barriers and changes in them, developments within labour market institutions, composition of trade between developed and developing trading partners, and many other factors can materially affect the extent to which the HOS predictions of the impacts of 14

globalisation are met in real world. 13 In the Indian context, for instance, the trade policy data shows varying degrees of liberalisation across sectors, whether in terms of tariffs reductions, or in terms of eliminating NTBs on imports. A third consideration is that trade liberalisation is not the only possible reason for price and demand changes in labour markets. Changes in the pattern of wages and employment may also reflect the impact of skill-biased technological change (see Leamer, 1997 and Feenstra and Hanson, 1999). As we mentioned earlier, evidence to date has weighed in favour of such technological change as more plausible explanations of labour market adjustments in terms of wage differentials. In a survey of nine well-known product-price studies, Slaughter (2000) reviews the emergence of three distinct approaches to testing the HOS theorem. The first relies on a simple factor proportions regression, controlling for technology in an ad-hoc manner. The second takes care to disentangle the impact of globalisation effects and the impact of technology on factor usage. 14 This approach assumes that price changes and technological change are exogenous. A third approach, started by Feenstra and Hanson (1997, 1999), accounts for the endogeneity of price and technological change. 15 The second and third approaches are more commonly known as the mandated wage methodology, which incorporates the one- and the two-stage methodologies, corresponding to the assumptions of exogeneity and endogeneity of price and technological change respectively. Besides the above difficulties, a further difficulty of the HO trade theory lies in the dimensionality of the model. The predictions of the HOS model rest on the simplistic assumption of a two-factor-two product world that does not extend unambiguously to multi-factor and multi-product contexts. For this reason, the most prominent test of the impact of trade on labour markets has adopted what Deardoff calls the correlation version of the Stolper-Samuelson theorem. This relates any vector of relative product price changes to relative factor price changes, and predicts that on average, factors 13 In an empirical application to South African manufacturing, Federke, Shin and Vaze (2002) address these two concerns using dynamic heterogenous panel estimation techniques. 14 See Leamer (1997) and Baldwin and Cain (1997) for this approach. 15 See Feenstra and Hanson (1999) and Haskel and Slaughter (2001) for an application to the US and the UK respectively. 15

used intensively in industries with rising (falling) prices will experience relative price increases (decreases). Yet, the correlation version of the Stolper-Samuelson theorem is nothing more than a consistency check of the trade theory: it is difficult to unambiguously associate product price changes with trade effects. Furthermore, empirical application has frequently linked product price changes to factor proportions rather than relative factor price changes. Thus, for developed countries, a common check is whether observed price changes of unskilled labour-intensive goods after liberalisation are consistent with factor scarcity, i.e. whether unskilled labourintensive product prices fell. A typical regression specification estimated across i industries (see, e.g., Lawrence and Slaughter, 1993; Sachs and Satz, 1994; Desjonques et al, 1997) is given by (1). N s ln pit it (1) Nu where, ln p it represents price changes and N, with the relevant subscript, is employment of skilled (s) and unskilled (u) workers respectively and it is a random error term. The ratio of the two represents relative employment. If the coefficient is positive, then this is taken as supportive of the predictions of the Stolper-Samuelson theorem for industrialised countries. 16 This simple consistency check is restrictive. In order to be interpreted as a test of the impact of international trade, it must assume that all domestic prices are exogenously set internationally. To argue that all domestic price changes are caused by tradeinduced changes would have to rely on the assumption that for a small economy, domestic industries are international price takers. This assumption is valid as long as tariff changes are not changing the differential between domestic prices and international prices and only as long as we ignore the industry and factor specific impacts of technological progress. Yet, there is no reason to suppose that technological progress will be factor-neutral, in which case one should expect that the relative factor productivity and hence factor prices would reflect its shifts. One solution to the ambiguity introduced by technological change has been to control directly for some identifiable technological changes (see Sachs and Shatz, 1994). The 16 Lawrence and Slaughter (1993) find a non-positive value for and interpret this as evidence against the Stolper-Samuelson theorem. 16

most widely used measure of technological change is TFP growth. Applying this argument to (1), the TFP-augmented consistency check can be specified as follows: ln p N TFP sit, it it it N uit, (2) Even then, equations (1) and (2) still suffer from many shortcomings. First, it considers only the intensity of two factors. Also, the Stolper-Samuelson price effects arise from the assumption that each sector in the economy makes zero profits, so that, when price changes, relative wages have to change to restore zero-profit equilibrium. The zero-profit relation links the level of prices and levels of factor inputs. Yet equation (1) regresses the change in prices on the level of factor inputs. Since the HO model is based on zero profit conditions, a more informative approach would be to allow for the impact of technological change on sounder theoretical foundations. Leamer (1997) proposes to estimate the zero-profit conditions. In what follows, we describe the model as initially proposed by Leamer (1997), and later extended by Feenstra and Hanson (1997a) and used by Haskel and Slaughter (2001) amongst others. Suppose an economy produces I different tradable goods, each of which requires some combination of J primary factors and I intermediate inputs. Sectors in the economy differ in their factor intensity at the same relative factor prices. Then for each sector i, we can write the zero profit condition as: G i jj p a w b p i=1.i (3) ji j ii ii G i where input; G pi is the domestic gross-output price in sector i; w j is the unit cost of the jth aij is the employment of input j per unit of output in sector i; and b ii is the amount of intermediate input required to produce a unit of good i. There are I equations in (3), one for each sector where production occurs. Totally differentiating (1) with respect to time gives: 17 17 See appendix for full derivation. 17

log pit logtfpit vjitlog wjt (4) jj g g where log pit ( log pit vjitlog pit ) is the change in value added prices, jj log TFP is the v jit is the share of factor j in total costs in sector i at time t; and it growth in total factor productivity for sector i. The final term in (2), log w jt, is the change in the wage of factor j, which again is economy-wide since all factors are mobile across sectors. Equation (4) shows how changes in product prices log p ) or technology ( it ( logtfpit ) cause shifts in relative demand and lead to adjustments in economywide factor prices log w ) to restore zero profits in all sectors. Two interesting ( jt points from equation (4) are worth noting. First, we note that logtfpit captures technical change of any form, including skill-biased technological change (SBTC). 18 A second point is that the sector bias of price or technology shocks will determine their qualitative wage effects. Thus, a larger price or TFP increase in the skillintensive (unskill-intensive) sectors will cause skill premiums to rise (fall). Equation (4) can be used in two ways depending on the assumption made about technology and prices changes. If they are assumed to be exogenous, for example if India were a small open economy, and technical change were exogenous, 19 then the effects of technology and trade (via prices) on wages are respectively estimated via the following equations: 20 (5) logtfpit v j J jit jt it log p it v j J jit jt it (6) 18 Note that as Jonhson and Stafford (1998) point out, different types of SBTC have different impacts on TFP. Thus it is not necessarily the case that SBTC raises TFP in a sector (quoted in Haskel and Slaughter, 2001) 19 The assumption of open economy may not be particularly appropriate for India before 1991, but there were modest attempts by India to open up in the 1980s. (see, e.g. Srivastava, 1996) 20 Note that TFP growth does not affect prices with a small economy assumption, i.e. there is zero passthrough from TFP growth to price changes. 18

where it is an additive error term in each equation. We interpret each estimate and jt as each factor j s wage change mandated to restore zero profits in all sectors in response to the sector bias of TFP and trade-induced product prices, respectively, ceteris paribus. Adding together the s and s from (3) and (4) for each factor j gives the net wage change mandated by trade plus technology. jt If on the other hand, changes in technology and prices are assumed to be endogenous, then estimation warrants a two-stage approach. In the first stage, TFP and price changes are each regressed on a set of underlying regressors, Z, which are assumed to drive TFP and price changes, respectively over some period t, as shown by equations (5) and (6): The log TFP log p it it Z tec,it tecs tec Z pri,it pris pri pri,t tec,t it Ztec vector includes many factors. Studies on the determinants of TFP growth have identified many forces, such as trade policy, industrial structures and domestic competition. For instance, for the UK, Haskel and Slaughter (2001) identify these forces as computerisation, innovations, industry concentration, union density and foreign prices based on the large literature explaining TFP in this country. In the context of developing countries, many similar forces have been identified although quantifying them in econometric studies depends very much on data availability (see Das, 2002 for a critical review of studies on the impact of trade liberalisation on productivity growth in developing countries). In the Indian context, we identify several forces. Trade policy measures represent trade-related influences. Industrial policy reforms will be represented by capital intensity, 21 while a measure of industrial concentration will highlight the role of domestic competition in enhancing productivity improvements. Trade union density will also be used as a force that might impede improvements in productivity growth. The change in Indian prices relative to foreign prices (proxied by US prices) will test whether international it (7) (8) 21 Das (2002) uses capital intensity as one of the three non-trade determinants in his study of industrial productivity in India. 19

competition induces technical change. As in Haskel and Slaughter, our Z pri vector includes a measure of foreign prices (proxied by U.S. prices), tariffs, and industry concentration. The first two forces represent the effects of trade-related influences on prices, while concentration represents a domestic influence on prices. In the second stage, each underlying variable in the Z vector that contributes to price and technology changes are regressed on cost shares: Z v (9) ˆtec tec, it jj jit jt, tec it Z v (10) ˆpri pri, it jj jit jt, pri it The share coefficients ( jt, tec and jt, pri ) give the wage changes mandated by the sector bias of each structural force working through either log TFPit or log pit. For instance, a measure of trade policy in (6), such as tariffs will explain the quantity of observed product-price variations accounted for by tariff changes. Using this quantity in turn as the regressor in (8) then determines the wage changes mandated by the sector bias of tariff changes working through product prices. 4.0 Empirical estimation and results. In this section, we discuss the estimation results for the model described in section 3.0. We first describe the data set used in estimation, explain how we computed TFP and price changes, and then move on to estimation. 4.1 Data description The main data set used in this study is the Annual Survey of Industries (ASI), which we arranged into a balanced panel of 158 three-digits manufacturing industries at the Indian level of classification. The data set is augmented with other crucial variables that are not present in ASI, and that are not reported under the same classification. We 20

therefore matched the different classifications to create a combined data set. 22 However, because these additional variables were not reported for all the industries under ASI, our sample size varied according to each of the additional variables used in this analysis. Table 1 shows the number of three digits industries that have data for each of the relevant non-asi variables, as well as the share of these industries in total ASI employment, value-added and output. Because these shares are computed on a year-to-year basis, we report the range of values for the shares over the 14 year period. Both value-added and output proxy industry size very closely. To a lesser extent, employment also proxies industry size reasonably closely and it can be seen that in general and as expected, industries with higher output share also tend to be industries with higher employment share. Clearly, the sample size used in each part of the analysis will largely depend on the number of industries that have data for the relevant variable. While this is not a problem for the one-stage mandated wage regressions, it represents a potential problem for the two-stage analysis. For instance, in the one-stage mandated wage methodology, we will be using price data as the only non-asi variable. Therefore, the results will be based on a sample of 98 industries for which the price data was originally computed. In the two-stage analysis, we are likely to combine the simultaneous use of all these variables, which do not have common industries in terms of data availability, and therefore represents an additional source of sample size constraint. 23 Table 4: Data availability of non-asi variables 24 Non-ASI variables Matched ASI Range of industries share (%) in total ASI s: industries Employ. Val. Add. Output 1. Wholesale Price Index 98 77-79 % 82-85% 83-87% 2. Trade policy 59 50-54% 63-67% 62-67% 3. Herfindahl index 53 42-46% 50-54% 51-55% 5. U.S Wholesale Price Index 25 158 27-33 % 33-43% 33-43 % available only from year 1991 onwards. 22 Concordance codes used are available from authors on request. 23 Where possible, we have attempted to make use of as many observations as analysis permits. This was not always possible though, as is clear from the sample sizes of our results. 24 We are grateful to Rajesh Chadha, Rana Hasan and Uma Kambhampati for providing us with the price data, trade policy data, and Herfindalh indices, respectively. 25 The U.S. Wholesale Price Index data is given at the ISIC revision 2 level. The matching with the ASI data was carried out via a matching of ASI industries with ISIC revision 2 industries. 21

4.2 TFP growth, price growth and factor shares. Our measure of TFP growth is the more commonly used primal Tornqvist index, which equals the log change of output minus the share-weighted growth of inputs: lntfp ln Y ( ln L v ) ( ln L v ) ( ln M v ) ( ln K v ) it it s, it s, it u, it u, it it k, it it m, it Where Y, M and K are output, intermediate inputs and capital stock respectively. The L s with the relevant subscript denote skilled (s) and unskilled (u) labour respectively, while the v s are the shares of each factor in gross output and the subscripts s and u as before, denote the two categories of labour, skilled and unskilled. The shares of skilled labour, unskilled labour and materials are the shares of skilled wages, unskilled wages and expenditure on materials as a proportion of output. This is computed for each industry for each year. Capital share is one less the sum of all other shares. We compute the growth in value-added prices in a similar way: lnvadpri ln PRI ( ln PRI v ) ( ln PRI v ) ( ln PRI v ) it it it s, it it u, it it k, it where ln PRI and lnvadpri are output and value-added prices respectively, with the it v s representing the factor shares as before. it The main approach of interest is the mandated wage methodology. Since almost all studies of the mandated wage methodology are cross-sectional in nature, we will adopt a similar approach using standard cross-sectional regression analyses. We define two periods: the pre-reform periods (1984-1991) and the post-reforms period (1991-97). We compute price changes and TFP changes as changes over each of these periods. For factor shares, we use each industry s average of beginning of period and end of period factor shares 26 in the regressions, in view of the fact that factor shares in 26 Our data is split into two distinct periods, 1984-91 and 1991-97, corresponding to the pre- and postreforms period respectively. If we use long period changes, such as seven-year changes, then 1984 and 1991 are the beginning points of each period while 1991 and 1997 are the endpoints of each period. 22

Indian manufacturing seem to have evolved over these 14 years under study. 27 This is clear from figure 8, where there appears to have been a gradual fall in both categories of labour in favour of an increasing share of capital in production. capital share.26.28.3.32.34.36 Average factor shares.03.04.05.06.07 skilled & unskilled labour share 1984 1985 1986 1987 1988 1989 1990 year 1991 1992 1993 1994 1995 1996 1997 capital share unskilled labour share skilled labour share Figure 8 4.3 A simple consistency check: only trade determines product prices. Although mandated wages is the main approach here, we perform some simple consistency checks, as described in section 3.0. This is done by estimating equation (1). While Lawrence and Slaughter (1993) use relative employment as the skillintensity variable, Sachs and Shatz (1994) use the share of unskilled workers in total employment. We adopt both definitions and report the results in table 5. Note that since we have domestic price data only, we could only go as far as replicating the exercise on such data only, as compared to these two studies that also use U.S. import and export prices. N st ln pit it (1) Nut Lawrence and Slaughter (1993) estimate as zero or negative, and thus interpret this as evidence against the hypothesis that international trade contributed to rising U.S. 27 We also note that some of the most influential mandated wages studies have also used average of beginning and end of period factor shares, e.g. Feenstra and Hanson (1997a, 1999) and Haskel and Slaughter (2001). 23

wage inequality by raising the relative price of skilled-labor-intensive products. Sachs and Shatz (1994) find to be negative: industries employing a larger share of production workers had lower relative price increases over the 1980s. Sachs and Shatz conclude that this supports the hypothesis that international trade contributed to rising U.S. wage inequality via the rise in relative price of skilled-labour-intensive products. Table 5: Simple consistency checks Relative employment of skilled workers Production workers share of employment Prereforms Postreforms Prereforms Postreforms Factor proportions -0.02-0.31** 0.07 0.64** (0.21) (0.10) (0.37) (0.20) Constant 0.68*** 0.65*** 0.62** 0.06 (0.08) (0.04) (0.27) (0.15) No. of observations 98 99 99 99 Adjusted R-squared 0.00 0.06 0.00 0.07 Note: robust standard errors in parentheses What is of interest in table 5 is the coefficient for the post-reforms period, because this is the period when rising wage inequality is witnessed, and which we seek to explain. Interpreted along the lines of the findings of Lawrence and Slaughter (1993), since is non-positive, it suggests that India s greater engagement in the process of international trade that would have accompanied the trade reforms did not contribute to the rising wage inequality: industries with higher relative employment of skilled workers did not have larger price increases post-reforms. Along the lines of Sachs and Shatz (1994), the positive coefficient for also suggests that trade was not responsible for the rising wage inequality in Indian manufacturing, since the results imply that industries with a higher share of unskilled workers witnessed large price increases post-reforms. These confirm our earlier findings from figure 6, that relative prices fell post-reforms. 4.4 Extending the simple consistency check: allowing for technological progress. The above simple consistency check is restrictive. It does not account for the possible influence of technological progress on factor prices, especially if such technological progress is not factor-neutral. Therefore, there is the need to control for technological 24