Factor Endowments and the Location of Industry in India

Factor Endowments and the Location of Industry in India¤ Kwok Tong Sooy London School of Economics October 29, 2002

Abstract This pap er explores the relationship between factor endowments and the location of industrial production, using a panel dataset on Indian industries. The method developed allows us to consider di¤erences in technology across states and time, in addition to the usual identical technology assumption. Use of within-country data avoids many of the problems associated with cross-country data that is usually used in such studies. In addition, the factor endowments approach is tested against several alternative hyp otheses on the determinants of the location of production. We …nd that factor endowments and technology play important roles in explaining the share of an industry. The factor endowments model is found to be superior to alternative explanations for share of industry, including economic development, government development expenditure, and market access. The lib eralisation of the economy beginning in 1991 represents a clear structural break in the industrial sector.

JEL Classi…cation: F11, F14, R11 Keywords: India; Factor endowments; technological di¤erences; location of production. ¤ Incomple te. I am very grateful to Steve Redding for his constant encouragement and help throughout this project, to Alejandro Cunat, Kala Krishna and Tony Venables for valuable suggestions and comments, Robin Burgess for access to data, and Marco Schonborn and Martin Stewart for help with the data. Financial support from the ORS Award Scheme and the LSE are gratefully acknowledged. All remaining errors are mine. y Correspondence: Centre for Economic Performance, London School of Economics, Houghton Street, London WC2A 2AE, UK. Tel: 0207 955 7080; Email: [email protected].

1

1

Introduction

The factor endowments model is one of the key theories in the international economist’s toolkit. This is despite the fact that many empirical studies have found that the model does not seem to hold empirically (see, for example, Leontief (1953), Bowen, Leamer and Sveikauskas (1987), Tre‡er (1993, 1995)). However, it is clear that many of the strict assumptions associated with the basic factor-endowments model - identical technologies and preferences across countries, idential qualities of inputs and outputs, no trade barriers or transport costs - do not hold in the real world. Consequently, tests of the model that allow for di¤erences in the above factors (e.g. Tre‡er (1993, 1995), Harrigan (1995, 1997), Davis and Weinstein (2001a), Redding and Vera-Martin (2001)) have been able to …t the data more closely than strict versions of the model.1 The main motivation of this paper is to test the relationship between factor endowments, technology, and the location of production using state-level data on industries in India. This relationship, which is postulated by the model, should hold both across and within countries, provided that the assumptions of the model hold. In this regard, there are clear bene…ts of using withincountry rather than cross-country data, since within-country data are more likely to be comparable, any measurement error biases may be expected to work in a similar direction, and more generally the assumptions of the basic model listed above are more likely to hold across states within a country than across di¤erent countries. Also, it relates to the original work of Ohlin, who notes in the preface to The Theory of Trade: “... international trade is only a special case of what could be called interlocal trade, that is, exchange between locations which are characterised by incomplete mobility of factors and commodities between them.”2 There are several reasons why India makes an interesting case study for the relationship between technology, factor endowments, and the location of production. First, India is a developing country. This contrasts with previous studies which have mainly concentrated on developed countries (e.g. Davis, Weinstein, Bradford and Shimpo (1997) and Bernstein and Weinstein (2002) who look at the case of Japan). Second, India is also developing rapidly. While the development in India has not been quite as spectacular as China, for example, the average GDP growth rate of 5.8 percent per year between 1980 and 2001 certainly places it as one of the top performers in terms of economic growth in the last two decades.3 Third, as a large country, both in terms of land 1 For some excellent recent surveys of the empirical literature on factor endowments and trade-related issues, see e.g. Helpman (1999), Davis and Weinstein (2001b), or Harrigan (2001). 2 Ohlin (1924). 3 There is also some concern about the reliability of data on China’s growth performance . See e.g. Young (2000). This of course does not make any presumption on the reliability of Indian data.

2

area and population, the performance of the Indian economy has been very diverse across states (see for example Besley and Burgess (2000, 2002), Ba jpai and Sachs (1999)). This makes the analysis in this paper closer to Redding and Vera-Martin’s (2001) study on European regions than say Bernstein and Weinstein’s (2002) study on Japanese regions. A …nal and perhaps most interesting reason for choosing India is that the Indian economy has been undergoing a process of liberalization beginning in 1991. While the process is still under way, early signs have shown that the liberalization has improved overall economic performance; despite the general slowdown of the world economy at the end of the 1990s, GDP growth has improved from an average of 5.6 percent per year in the 1980s to an average of 6.0 percent between 1992 and 2001.4 This paper is closely related to three strands of work on the empirical testing of factor-endowment models, the …rst in terms of the framework used, the second in terms of the use of within-country data, and the third in terms of the existence of multiple cones of diversi…cation. The …rst strand comprises the papers by Harrigan (1997), Harrigan and Zaka jsek (2000), Redding and Vera-Martin (2001) (Redding (2001) uses a similar framework, but his focus is on the evolutionary dynamics of sector shares in manufacturing). While earlier work (see above) tended to use restrictive models, Harrigan (1997) used a more ‡exible model that did not rely of factor prize equalization, and allowed for non-neutral technology di¤erences. Using a panel of OECD countries, he found that both factor endowments and relative technology levels are important determinants of specialization. Harrigan and Zakrajsek (2000) further generalize the model, by dropping the assumption that cross-country technology di¤erences are exclusively Hicks-neutral at the industry level. In addition to …nding that factor endowments play an important role in determining the location of production in a cross-country sample, they also …nd that an alternative ladder-of-development model dominates the factor proportions model on purely statistical grounds. Redding and Vera-Martin (2001) use European regional data (which is likely to be cleaner than cross-country data), and …nd again that factor endowments are important in explaining production patterns. In addition, they perform two speci…cation tests to determine the reasonableness of the assumptions of the model. The second strand of related literature includes tests of the factor proportions model using within-country data. Davis, Weinstein, Bradford and Shimpo (1997) test the Heckscher-Ohlin-Vanek (HOV) model using both Japanese regional data and cross-country data, and …nd that the HOV prediction of trade in factor services …ts Japanese regional data better than cross-country data. Bernstein and Weinstein (2002), testing the relationship between factor endowments 4 The average GDP growth rate betwee n 1991 and 1996 was 6.7 percent per year. See Basu and Pattanaik (1997) for background on the reforms, and Ahluwalia (2002) on the e¤ects of the reforms after 10 years.

3

and production, …nd indeterminacy of production patterns across Japanese regions, which disappears in the international dataset. Their interpretation is that trade costs help to render international production patterns determinate. The third strand of related literature is that on multiple cones of diversi…cation in a Heckscher-Ohlin framework. This is the idea that factor price equalization (FPE) will not be an outcome of free trade, and that therefore di¤erent relative factor prices will lead to di¤erent factor intensities across countries. While the framework that we use does not assume FPE, we can nevertheless test whether there is evidence of the existence of multiple Heckscher-Ohlin cones of diversi…cation in the data. This is a much simpli…ed version of such a test; more careful analysis of the idea of multiple cones can be found in Cunat (2000), who develops a criterion for checking whether the condition for FPE holds. He …nds that FPE fails for a cross-section of countries, and that there is evidence of multiple cones of diversi…cation. Bernard, Jensen and Schott (2001), using a di¤erent approach from Cunat (2000), …nd that the US can be divided into three distinct cones of diversi…cation. The contributions of the present paper are as follows. First, we use a new panel of cross-state industry data on India to test for the relationship between factor endowments, technological di¤erences, and the location of production. Second, following Redding and Vera-Martin (2001), we perform several speci…cation tests on the model’s assumptions. Third, furthering the idea of testing the factor endowments model against alternatives as suggested by Harrigan and Zakrajsek (2000), we propose and test several alternative models against the factor endowments model. In addition to the ladder-of-development model suggested by Harrigan and Zakra jsek, we also consider a model of government determination of the location of production, and a model of access to markets as a simple test of the importance of the international aspect of location. The latter two tests are important in the context of India, since our dataset includes data from 1980 to 1997, which falls on both sides of the massive economic liberalization programme begun in India in 1991. Prior to liberalization, the Indian economy was relatively closed and heavily regulated, with many industries reserved for the government sector. The liberalization programme, which reduced trade barriers and regulation of industry, would be expected to have di¤erential impacts on the relative roles of government and market access in the location of production. Finally, as a further development of an idea that originates in Harrigan and Zakrajsek (2000), we consider whether there exists multiple cones of diversi…cation in India using a simple extension to the basic equation, as suggested by the theory of multiple cones. To brie‡y preview our results: our estimate of Total Factor Productivity indicates that technological progress has not been the engine of growth in manufacturing in India in the sample period. Nevertheless, we …nd that both factor endowments and technology play important roles in determining the share of an industry in the Net Domestic Product of a state. Only in industries which are 4

natural-resource intensive do we …nd that factor endowments play little or no role, as we do not include endowments of natural resources in our independent variables. When we compare the factor endowments model with the alternatives suggested above, we …nd that the factor endowments model statistically dominates the other models in determining the share of industry in 8 of our 19 industries. The conclusion to be drawn from this is that factor endowments clearly play an important role in determining the location of industries in India. Further, there is strong evidence that the liberalisation of the economy beginning in 1991 represents a structural break in the industrial sector. The rest of this paper is structured as follows. The next section outlines the empirical factor endowments model to be estimated. This is followed by a detailed description of the data. Section 4 discusses estimation issues. Section 5 details the results, and section 6 concludes.

2

An Empirical Model

This section …rst outlines the theoretical background of the model, then presents the econometric speci…cation, and …nally discusses some alternative hypotheses to the factor-endowments model.

2.1

Factor endowments and technology

The model is derived from neoclassical trade theory. We consider …rst the most basic speci…cation. This makes the usual assumptions of constant returns to scale and perfect competition, and identical technologies and preferences across states. States are indexed by z 2 f1; :::; Zg, goods by j 2 f1; :::; N g, factors of production by i 2 f1; :::; M g, and time by t. Each state is endowed with an exogenous vector of factors of production, v zt . The revenue function r (p zt ; vz t) characterises general equilibrium in production. As long as the revenue function is twice continuously di¤erentiable, the vector of net output supplies y (p zt ; vz t) is given by the gradient of r (pz t; v zt ) with respect to p. We follow Harrigan (1997) and Redding and Vera-Martin (2001) in assuming a translog revenue function:

ln r (pz t ; vzt )

1P P ¯ ln (pz jt) ln (p zkt ) 2 j k jk P 1P P + i ± 0i ln (v zit) + ±ih ln (v zit ) ln (vzh t) (1) 2 i h P P + j i ° ji ln (pz jt ) ln (v zit )

= ¯ 00 +

P

j

¯ 0j ln (p z jt) +

where j; k 2 f1; :::; N g index goods and i; h 2 f1; :::; M g index factors. Symmetry of cross e¤ects requires that, for all j; k; i and h: 5

and

¯ jk = ¯ kj

(2)

± ih = ±hi

Linear homogeneity in v and p requires: P

j

¯ 0j = 1

P

i ± 0i

P

=1

j

P

¯ jk = 0

i

P

±ih = 0

i

° ji = 0

(3)

Di¤erentiating ln r (p zt ; vz t) with respect to each pj gives the share of good j in GDP as a function of prices and factor supplies:

Szjt =

P P p zjt yz jt (p zt ; vz t) = ¯ 0j + j ¯ jk ln (p zjt ) + i ° ji ln (vz it ) r (p z t; v zt )

(4)

Relaxing the assumption of identical technologies across states, we assume Hicks-neutral technology di¤erences such that the production function takes the form (see Dixit and Norman p. 138) yz jt = µz jt Fj (vz jt ). Then the revenue function may be written as r (µz t pz t; v zt ), where µz t is an n £ n diagonal matrix of the technology parameters µz jt. Now, di¤erentiating the translog form of this revenue function with respect to pj gives: Sz jt = ¯ 0j +

2.2

P

j

¯ jk ln (pz jt) +

P

j

¯ jk ln (µz jt ) +

Econometric speci…cation

P

i

°ji ln (v zit )

(5)

In our econometric speci…cation, we follow Redding and Vera-Martin (2001) in their methodical progression from the simplest setup to the most general speci…cation. We …rst write equation (4) as: Sz jt = ¯ 0j +

P

j

¯ jk ln (pz jt) +

P

i ° ji ln

(vz it) + "zjt

(6)

where " zjt is a random error. Equation (6) is estimated for each industry, pooling observations across states z and time t. However, one problem with estimating (6) is that prices of individual industries are not observable across states. If prices and factor endowments are uncorrelated, then we can still obtain consistent estimates of °ji by estimating the following equation: Szjt = ¯ 0j +

P

i

°ji ln (v zit ) + "z jt

(7)

To assume that prices and endowments are uncorrelated, we need to assume that all goods are traded. However, states in India are large, hence there will be some nontraded goods. Since the prices of non-traded goods are likely to be correlated with endowments, equation (7) is likely to yield inconsistent estimates of the parameters ° ji. 6

If instead we assume that all goods are perfectly tradeable, then prices are equalized across states, and we can get around the problem of unobservable prices by adding a set of time dummies djt to (7): Szjt = ¯ 0j + Áj djt +

P

i

°ji ln (v zit ) + "z jt

(8)

where the time dummies also control for common shocks across states. To allow for non-tradeables, ¡ …rst ¢ we partition the vector ¡ ¢ of goods prices p zt into the vectors of tradeable pTzt and non-tradeable p NT goods prices: zt ¡ ¢ T 0 p 0zt = pTzt : pN zt

T NT where pTzt is an 1£nT vector of traded goods prices, pN vector z t is an 1 £n T NT of non-traded go ods prices, and n + n = n. Since free trade still allows equalized goods prices among tradeables, while in general we have di¤erent non-traded goods prices, we can write equation (6) as:

Szjt = ¯ 0j + Áj djt +

Pn

j= nT + 1

¡ ¢ P ¯ jk ln p NT zjt + i ° ji ln (v zit) + "z jt

(9)

Since we once again do not have data on the prices of non-tradeables, we follow Harrigan (1997) and Redding and Vera-Martin (2001) in treating the price of non-traded goods as a random variable with some estimable probability distribution. Thus, let Pn

j=nT +1

¡ ¢ ¯ jk ln p NT zjt = ´ zj + ¹jt + uz jt

¡ ¢ uz jt » N 0; ¾ 2j

(10)

such that the price of non-traded goods comprises state …xed e¤ects ´ zj , time dummies ¹jt, and a random component uz jt with constant variance ¾ 2j . Combining equations (9) and (10), we get: Szjt = ¯ 0j + ´ zj + ³ j djt +

P

i

° ji ln (vz it ) + ! z jt

(11)

where ! zjt = "z jt + uz jt . The state …xed e¤ect ´ z j will also control for unobserved time-invariant di¤erences across states. Equation (11), which is SP5 in Redding and Vera-Martin (2001), is the main estimated equation. However, the framework used by Harrigan (1997) allows us to do even better than (11). If, instead of starting from (4), we begin developing our empirical framework from (5) exploiting the Hicks-neutral technology di¤erences across states, then the analogue to equation (11) would be:

7

Sz jt = ¯ 0j + ´ zj + ³ j djt +

P

j

¯ jk ln (µ zjt ) +

P

i

° ji ln (vzit ) + ! z jt

(12)

This is Harrigan’s (1997) equation (5). The advantage of using equation (12) rather than equation (11) is that, with the former, we can evaluate not only the impact of factor endowments on the production structure, but also the impact of technological di¤erences. This is the second equation which we estimate.

2.3

Alternative hypotheses

In agreement with the suggestion by Davis and Weinstein (2001b)5 , one goal of this paper is to test the factor-endowments mo del against several alternative hypotheses on the factors which in‡uence the location of industrial activity. The bene…t of such testing is that we can then determine which factors play a more important role in the location of a particular industry. Three alternative hypotheses are considered: …rst, that a state’s output mix is determined by its level of development; second, that it is determined by the government; and third, that it is determined by access to markets. The …rst hypothesis we consider derives from Harrigan and Zakra jsek (2000). Under this hypothesis, states develop through capital accumulation and technological progress. As states develop, their output mix changes, as they move from specialization in agriculture and labour-intensive manufactures, to capitalintensive manufactures, and …nally to high-tech goods and services. A simple version of this ladder-of-development model is that, controlling for time and state …xed e¤ects, output shares depend only on the aggregate productivity level, or: Sz jt = ¯ 0j + ´ zj + ³ j djt + ±j µ zt + ! zjt

(13)

where µz t is aggregate output per worker. The second alternative hypothesis on the determinants of the location of production is that it is the government which determines (or at least heavily in‡uences) the location of industry. Thus for example the Industrial Policy Resolution of 1956 lists 17 industries that would be the exclusive responsibility of the state. This includes defence equipment, atomic energy, iron and steel, coal and lignite, aircraft, railways, shipbuilding, telecommunications, and generation of electricity. A further 12 industries were to be progressively state-owned, whereby the government will generally take the initiative in establishing new 5 And, incidentally, in direct contrast to Leamer’s (1994) injunction to “estimate , don’t test”. Note, howeve r, that while we take the testing procedure seriously, our interpretation will be couched in terms of which factors operate more strongly in which industries, rather than a dichotomous accept/reject approach.

8

undertakings, supplemented by private enterprise. In this category are most of the remaining heavy industries: all other minerals, machine tools, basic products required by chemical industries, road and sea transport. The remaining industries were left to private enterprise. In addition, the statement is explicit on the government’s intention of reducing disparities in the level of development between regions. Subsequent industrial policy statements in 1973, 1977 and 1980 took the …rst steps towards liberalizing industry, but it was the Statement on Industrial Policy 1991 that had the most impact on liberalization, especially since it was combined with a package of …nancial and trade liberalization in the wake of the 1991 public sector …nancial crisis. In this statement, while eight industries remain the exclusive province of the government, the remaining industries either require industrial licensing (18 industries), or grant automatic approval of foreign technology agreements and for 51% foreign equity approvals (34 industries). Therefore, the alternative speci…cation that we estimate takes the following form: Sz jt = ¯ 0j + ´ z j + ³ j djt + ! j G zt + ! zjt

(14)

where G zt is state government development expenditure. That is, controlling for state and time e¤ects, we seek to investigate the impact of government development expenditure on the share of industry in the state. The third alternative hypothesis seeks to relate the location of industries to a simple measure of market access. As is well known in the literature on economic geography (see e.g. Fujita, Krugman and Venables (1999)), proximity to suppliers and …nal demand has positive e¤ects on the location of production. This idea has been implemented in Midelfart-Knarvik, Overman and Venables (2001), who …nd that both factor endowments and forward and backward linkages are important determinants of industrial structure. Internal market access, that is, the proximity of states to other centres of economic activity within India, is likely to be important. Further, access to foreign markets has become increasingly important since India’s exports as a share of GDP has increased from 15% in 1980 to 27% in 1999. Thus, while total GDP in current US dollars has increased 2.5 times in that time, exports have increased by four times, representing an average growth rate of 7%.6 Thus, we can write this speci…cation as: S zjt = ¯ 0j + ´ zj + ³ j djt + ½1j DM Azt + ½2j F M Az t + ! zjt

(15)

6 This obscures the fact that the larger proportion of this growth occured after the 1991 reform. Between 1992 and 1999, exports increased 2.5 times, representing an average growth rate of over 10%.

9

where DM Az t is domestic market access, measured following Harris (1954) as distance-weighted SDP in all locations: DM Az t =

PZ

z =1

SDPz t dyz

(16)

where dyz is the bilateral distance between the capitals of states y and z. Foreign market access is similarly de…ned to be distance-weighted size of ports: F M Azt =

PP

p=1

T RADE pt dpz

(17)

where the summation is over all major ports in the country, and T RADEpt is the total trade volume of the port. As an aside, the distance between a state and itself is arbitrarily set at 100 km; the same is true when the state capital is itself a ma jor port.7 In calculating values for DM Azt , we include Delhi in the summation, since it is a Union Territory that has a large SDP.

2.4

Multiple diversi…cation cones

A cone of diversi…cation is a region in the endowment space where a unique set of factor prices, and hence a unique set of factor intensities, exists. In the simple two goods, two factor mo del, this region is simply the area in the Lerner diagram between the optimal capital-labour ratios for a given factor price ratio. If relative factor prices are not equalized across states, then more than one diversi…cation cone will exist. To see how the above framework can be extended to explore the possibility of the existence of multiple cones of diversi…cation, consider the Lerner diagram Figure 1. In this diagram, are two possible cones of diversi…cation: cone ¡ there ¢ ¡ ¢0 1 with factor price ratio wr , and cone 2 at factor price ratio wr . There are 2 goods, good Y is the capital-intensive good, good X is the labour-intensive good. The equilibrium capital-labour ratios for at¢ each factor price ¡ each ¢ good¡ K ratio is given by the vectors from the origin: K and at factor price L Y ¡w ¢ ¡ K ¢0 ¡ K ¢0 ¡ w ¢0L X ratio r , and L Y and L X at factor price ratio r . As the diagram is ¡ ¢ drawn, good X has the same capital-labour ratio at factor price ratio wr as ¡ ¢0 good Y does at factor price ratio wr . This is a simpli…cation for our thought experiment. Now, suppose that the endowment of an ¡economy is at point E0 . Then the ¢ w economy is in cone 1, faces factor price ratio , and hence uses capital-labour r ¡ ¢ ¡K ¢ ratios K and to produce the two goods. But more importantly, since L Y ¡ ¢L X ¡K¢ E 0 is closer to K than , the economy produces more of good Y than L Y L X 7 See Overman, Redding and Venables (2001) for a discussion of di¤erent measures of market access.

10

good X (assuming of course full employment). Suppose now that the economy’s endowment moves downward from E 0 . As the endowment of capital falls relative to labour in the economy, the share of good X in production rises while the¡share ¢0 of good Y falls. This is so until the economy’s endowment hits the vector K L Y ¡ ¢ = K . As the economy crosses this vector, factor prices change, the economy L X is now in cone 2, produces using a di¤erent mix of factors, and is now producing more of good Y than good X. After is crosses this threshold, the process of switching towards the more labour-intensive good X resumes as the economy’s endowment moves towards increasing labour-abundance. This analysis carries over to the more general many-factor, many-goods case, although as usual the result will be weaker, in the form of correlations rather than direct relationships. Also, it works equally well if we assume that di¤erences in per capita GDP, or government involvement in the economy, or market access, a¤ects relative factor prices. The key result of the above is that, in the presence of multiple cones of diversi…cation, there need not be a stable linear relationship between factor endowments and output shares of the two goods, even if there are no factor intensity reversals. Therefore, one simple test for the existence of more than one diversi…cation cone is whether the coe¢cient on a quadratic term in augmented versions of equations (11), (12), (13), (14) and (15) are signi…cant, and to use a simple visual inspection of the relationship between factor endowments, technology, per capita GDP, government, or market access and the output share to determine whether there exists non-linearities in the relationship between these factors and shares of output. This is the method suggested by Harrigan and Zakra jsek (2000).

3

Data

The main dataset that we use comes from the Annual Survey of Industries (ASI), produced by the Central Statistical Organisation of India. This annual publication consists of data at the 3-digit level at the state level. This gives us a total of 138 industries. However, to reduce the mass of tables and …gures, we aggregate up to the two digit level where we obtain 19 industries. There is a total of 25 states and 7 union territories in the sample period (in 2000, the borders of several states have been redrawn and three new states created). The analysis is performed on the 16 largest states in terms of industrial output due to data limitations. For each industry-state pair, data on a wide range of variables is available, from number of factories, to capital employed, workers employed, total inputs and output, value added, and capital formation. We have data for the period 1980 to 1997, which is an especially interesting period to investigate for the reason mentioned above: the liberalization of the economy, which began somewhat hesitantly in the 1980s, and was rapidly pushed forward in 1991 as a

11

direct result of the …nancial crisis faced by the Indian government. The sample period is thus one of rapid change and growth of the Indian economy. Data on endowments comes from a variety of sources. A key source of data is the dataset compiled by Ozler, Datt and Ravallion (1996), augmented by Besley and Burgess (2000, 2002), and further extended by the author using data from the Statistical Abstract of India (various years).8 Key data sources include the Reserve Bank of India Report on Currency and Finance for public …nance variables, the Census of India for population …gures, and Butler, Lahiri and Roy (1991) for political variables. A map of India is provided as Figure A1 for convenience.

3.1

Factor endowments

Tables 1 and 2 provide summary statistics of factor endowments. States can be divided into those with high per capita State Domestic Product (SDP) (>15000 rupees in 1997): Gujarat, Haryana, Maharashtra and Punjab; those with intermediate levels of SDP (8000