Market Imperfections, Communities and the Organization of Production : An Empirical Analysis of Tirupur’s Garment-Export Network ¤ Abhijit Banerjeey

Kaivan Munshiz

October 1999

¤ PRELIMINARY- DO NOT CIRCULATE. This project could not have been completed without the support and assistance that we received from the Export Credit Guarantee Corporation of India (ECGC) and the PSG Institute of Management. Professor A. Govindan and T.J. Sivan organized the survey and supervised the data-collection. We thank Esther Du‡o and Petra Todd for helpful discussions. We are responsible for any errors that may remain. y Massachusetts Institute of Technology z University of Pennsylvania

1

Introduction

The past few years have witnessed a growing awareness among economists that informal communitybased institutions and networks more generally may have strong e¤ects on the way production is organized in a world where markets function imperfectly. There are however no studies that we are aware of which attempt to rigorously quantify the e¤ects of networks or identify the precise channels through which they a¤ect industrial production.1 Our objective in this paper is to start to …ll this gap in the speci…c context of a community-based network. The data we use to study the e¤ect of community-based networks comes from the town of Tirupur. Tirupur is a small town in South India, which, rather remarkably, is now estimated to produce 50 % of India’s knitted-garment exports.2 The garment industry in Tiruppur is characterized by a low level of vertical and horizontal integration. Of the seven principal stages in the production process, the most important of which are knitting, dyeing and stitching, most exporters own machinery for one or two. The rest of the work is farmed out to job-workers who are specialized producers, owning machinery in a single stage of the production process. Furthermore, exporters typically do not have the capacity to meet their biggest export orders. Therefore they also make use of indirect exporters, who, on order, supply them with …nished goods (which they in turn produce using a combination of the machinery they own and job-workers). Indeed most exporters in the industry are indirect exporters who do not deal directly with the foreign buyer - getting the export orders is the privilege of a relatively small number of direct exporters. What makes Tirupur especially suitable for our study is however a sociological characteristic of the area. The production network in Tirupur was established by the Kongu Vellala Gounders in the mid-1960s. The Gounders are the dominant cultivating caste in Kongunad, the region surrounding Tirupur. They are a known as a hard working, industrious, community who made their fortune in agriculture, through the cultivation of cash-cops such as cotton, sugarcane, tobacco and groundnut. Through the 1960s and 1970s they invested a part of their agricultural surplus into the knitted-garment industry in Tirupur and the industry grew slowly but steadily, catering to the domestic market for 1

See Banerjee(1996, 1998), Cranton (1996), Cranton and Minehart (1997), Greif (1993), McMillan (1997) for largely theoretical discussions of the role of networks and some anecdotal evidence. Townsend (1996) provides some interesting evidence on the role of production teams in agricultural production. 2 While direct exports from Tirupur account for roughly 50 % of the country’s exports, a substantial volume of production also ‡ows through export-houses in Bombay. Taking into account these indirect exports, Tirupur is estimated to account for close to 70 % of India’s total knitted-garment exports.

1

under-garments. The mid-1980s saw a dramatic change with the arrival of the …rst export orders from abroad. Thereafter exports grew at the rate of 50-100 % per year, through the 1990s. The dramatic growth in exports was accompanied by a corresponding change in the sociological composition of the industry, with entrepreneurs arriving from di¤erent parts of the country to do business in Tirupur. These new entrants for the most part belong to traditional trading communities, with generations of experience in the textile business. The major outside communities in Tirupur today are the Gujaratis, Punjabis, Marwaris and Sindhis from North India and the Chettiars from South India. We will see in our sample that the direct exporters are evenly divided between Gounders and Outsiders. In contrast, the indirect exporters and job-workers are mostly Gounders. It is this substantial presence among direct exporters of both Gounders and Outsiders that allows us to try to identify the e¤ect of participating in a community network. Both Gounder and Outsider direct exporters produce and export the same kind of goods using exactly the same technology. The one apparent di¤erence between them, given that the Gounders in Tirupur are surrounded by their own people while the outsiders are not, is in the access to a community-based network. Access to a community-based network a¤ects production decisions for three reasons. First, it can be a substitute for vertical and horizontal integration which is clearly valuable since capital is costly and capital markets are highly imperfect. Non-integrated network-based production has the obvious disadvantage that a non-integrated producer is at the mercy of the indirect exporters and job-workers who he needs to complete the job.3 Being socially connected to indirect exporters and job-workers must reduce the extent of the hold-up problems that this would otherwise inevitably raise and therefore makes it less costly to be unintegrated.4 Second, it gives members of the community privileged access to the capital market internal to the network. The Gounders are a wealthy community. Investing this wealth pro…tably outside the knitted-garment industry in Tirupur is a potential problem for them for a number of reasons: …rst, the land ceiling acts prevent further investment in land. Second, they have virtually no connections with industry elsewhere in India and therefore direct investment into any other industry is not really an option. Finally, given the poor functioning of most …nancial sector institutions in India, putting the 3

Vertical integration reduces hold-up problems and improves managerial control. This reduces production delays and improves product-quality. For instance Cawthorne (1995) quotes one of the Tirupur exporters as saying “I want to be like the spinning mills here. That is my ambition. Then I will have all the stages as one operation ... This way I have much better managerial control than otherwise. ... With a large factory you know exactly what is going on.” While Cawthorne does not share his view, a large number of exporters that we spoke with expressed the same sentiment. 4 For the argument that lack of vertical integration leads to hold up problems see Hart and Moore (1990).

2

money into …nancial assets is unlikely to be very rewarding. For all these reasons, the Gounders prefer to invest their surplus in Tirupur, in …rms owned by members of their extended family (and perhaps friends) even when there may be more pro…table investment opportunities elsewhere in India.5 The Outsiders, by contrast, have no particular reason to invest in Tirupur. They have no natural ties to the area and they are much more likely to have relatives who are in business somewhere else in India. They will therefore only invest in Tirupur as long as the rate of return on investing in Tirupur is comparable to the best opportunities available elsewhere in India. The opportunity cost of capital faced by the Outsiders is therefore likely to be higher than that faced by Gounders. Finally, the very fact that Gounders do not expect to move from Tirupur reduces the option value of being able to pack up their business and move. To the extent that …xed investments also tend to be irreversible at least in part, this further widens the gap between the Gounders and the Outsiders in the cost of making …xed investments. The three arguments made above are potentially in con‡ict with each other in terms of behavioral predictions. The …rst argument says that Gounders perhaps bene…t less from owning more machinery than the Outsiders. The second and the third arguments say that Gounders …nd it cheaper to invest in machinery. The net predicted e¤ect on investment is ambiguous. Therefore, if we found that the investment patterns of Gounders and Outsiders were indistinguishable, we would not know what to make of it - does the community network have no e¤ect or do the e¤ects neutralize each other? Fortunately, as we will see, the data is quite unambiguous. The data that we use to test for the role of the community comes from a survey of six hundred direct exporters, indirect exporters and job-workers that we conducted in 1995. Detailed information on exports (and production), jobworking, investment and access to …nance was collected from each …rm over a four-year period, 1991-94. A careful reading of this data tells us that despite the fact that they are both in the same industry and produce essentially the exact same goods, Gounders sink substantially more capital into production compared to the Outsiders: the di¤erence in the starting (period-0) capital is striking, with the Gounders investing roughly three times more in machinery. The Gounders are also more 5

We are implicitly arguing here that Gounders may be trading o¤ the higher returns available elsewhere with the fact that investments in the businesses owned by friends and extended family are less subject to agency problems. In part this could be because of common objectives (the father wants his son to succeed). In part also this could be because of the repeated nature of the interaction with the extended family, as well as the superior information and social control mechanisms that are available when dealing with members of the extended family. Banerjee (1996) provides a more detailed exposition of the theory underlying this view, drawing on the formal model in Banerjee and Newman (1997).

3

vertically integrated and make less use of job-workers, re‡ecting a much higher capital to export ratio Nevertheless the Outsiders are the ones who have faster export growth and who end up with higher exports. These facts, we argue in this paper, allow only one straightforward interpretation: the cost of investing in Tirupur must be lower for the Gounders, to the point where it overwhelms the bene…ts if any of being socially connected to the job-workers and indirect exporters. At the same time, ability, interpreted as something that makes exporting easier, must be lower for the Gounders. The latter conclusion is unremarkable and probably re‡ects a greater degree of self-selection among outsiders, combined with the fact that outsiders are better educated and have more of a history of involvement with garment production. The …rst conclusion is, however, rather striking. It provides the …rst evidence, to the extent that we are aware, of the fact that segmented capital markets distort investment decisions to the point where the less able own more of the capital6 While we cannot directly identify the presence of a network in this paper, the segmentation of the capital market is a natural outcome when wealth-‡ows do not cross community boundaries. This paper is organized in six sections. Section 2 provides an overview of the institutional setting with a historical description of the region, the shift from agriculture to industry and Tirupur’s rise as an industrial center. Section 3 describes the production network using data from the 1995 survey. Section 4 presents a simple model of investment and production in the industry. Section 5 provides the estimation results, which have been discussed above, using the results from Section 4 to interpret the patterns in the data. Section 6 concludes the paper.

2

The Institutional Setting

This Section attempts to place Tirupur’s production network in historical perspective. We begin with a discussion on agriculture in Kongunad, the region surrounding Tirupur. Here we pay special attention to the changing fortunes of the Gounders who constitute the traditional cultivating class in the region. Thereafter, we discuss the Gounder’s transition from agriculture to industry which is also associated with Tirupur’s rise as an industrial center. 6 The fact that credit markets in developing countries are often segmented is well-known and as been documented by Aleem (1990), Lanjouw and Stern (1999) and others. None of these look at the e¤ect on the pattern of investment.. The fact that it might lead to ine¢cient use of capital, is in Banerjee (1996).

4

2.1

Agriculture in Kongunad

Kongunad is a large upland plain, one of the …ve big sub-divisions of Tamilnad (the Tamil-speaking country) prior to the British conquest of the region. These sub-divisions were controlled by di¤erent castes and Kongunad is believed to have been colonized by the Vellala Gounder caste in about the twelfth century (Beck, 1979). Vellala is the term used to describe the cultivator class, which is considered to occupy the …rst place in the social scale among non-Brahmins in the region. The Gounders thus occupy an elite place in their society. Kongunad includes most of Coimbatore district as well as small pieces of Salem, Tiruchi and Madurai districts in the modern Indian state of Tamil Nadu. However most historical sources use data drawn from Coimbatore district to describe change in Kongunad. This practice is actually quite convenient in our case since Tirupur is located in the heart of Coimbatore district. Coimbatore district is exceptionally dry and its rainfall is scanty and uncertain. It has the lowest rainfall among all the districts in Tamil Nadu, with an annual average of twenty-three inches (Coimbatore District Gazetteer, 1951). Most of the soil in Kongunad consists of poor quality red sand and gravel. However, the area is also endowed with tracts of black cotton soil as well as very fertile red soil. Moreover it has substantial reserves of subsoil water. Where well irrigation is available, extremely high yields are obtained. The main source of irrigation in the region is wells, which require a substantial personal investment.7 The black cotton soil also requires substantial investment in draught power since it is heavy to work. The cultivators in the region thus had to commit to a capital-intensive style of farming, on the limited high-quality land that was available, from early times (Baker, 1984). This contrasts with the subsistence agriculture that predominated elsewhere in the region.8 Cash crops had traditionally been grown in the heavily irrigated thottam gardens, surrounding the wells, to …nance the capital-investment in irrigation and bullocks. However it was only in the last quarter of the nineteenth century, with the advent of the railways, that cultivation began to shift predominantly into cash crops and Kongunad emerged as the most commercialized zone in the 7

Irrigation by tanks, used extensively elsewhere in Tamilnad, is infeasible in Kongunad. Rainfall is low, the soil is absorbent and tank water quickly evaporates in the heat. Wells were thus the dominant mode of irrigation from medieval times. 8 The British, who took control of Kongunad at the beginning of the nineteenth century, were most impressed by the farming skills of the Kongu Vellala Gounders. Nicholson (1871, pp. 184) writes that “ the art (of farming) is both excellently known and practiced by the average and substantial ryots (cultivators) ... it is doubtful that he (the ryot) will su¤er by comparison with any farmer anywhere."

5

region.9 The most important cash crop was cotton. Roughly 150,000 acres were allocated to cotton in the early nineteenth century. The acreage increased to 200,000 towards the end of that century with the advent of the railways. The introduction of the long-staple Cambodia variety, which grows well in Kongunad, increased the cotton-acreage to 300,000 by the …rst quarter of the twentieth century. Finally the emergence of Coimbatore’s cotton-mill industry in the 1930s pushed cotton cultivation up towards 400,000 acres. Increased access to markets, with the advent of the railways, increased cultivation of other cash crops as well. Tobacco and sugarcane, which require heavy irrigation, were particularly well- suited to cultivation on the thottam gardens. Acreage under tobacco doubled over the 1880s, from 15,000 to 30,000 acres, remaining stable thereafter. Sugarcane cultivation also increased steadily, from 4,000 acres in the 1880s to 20,000 acres in the 1950s. Finally groundnut emerged as an important cash crop in the twentieth century. By the 1950s Coimbatore had 20% of its land allocated to cash crops, which is the largest share of any district in Tamil Nadu. While the availability of good land was limited, a substantial cultivator (Vellala) class nevertheless managed to establish itself in Kongunad. In the 1891 Census of India the Vellala caste accounted for 31 % of the population of Coimbatore district. In no other district did it account for more than 12 % of the population.

2.2

From Agriculture to Industry

While the Kongu Vellala Gounders had always been wealthy, the cultivation of crash crops transformed this community into one of the wealthiest in Tamil Nadu. In the nineteenth century, land in Kongunad was roughly the same value as agricultural land in other parts of Tamil Nadu. By the middle of this century, once opportunities for cash-crop cultivation were opened up, this land was among the most valuable in the State (Baker, 1984). The Gounder community was now faced with the question of what to do with the agricultural surplus that it had accumulated. While the Gounder cultivating households traditionally marketed their own produce at the periodic local markets, or shandies, they have never ventured into trade. When Tirupur emerged as the major regional textile market early in the twentieth century, the Gounder cultivators did initially bring their produce directly to the market. However by the late 1920s, 95% of the cotton kapas was sold in the market by commission agents. 9

The subsequent discussion draws on Baker, 1984, and the Gazetteer of the Coimbatore District, 1951.

6

Turning to industrial activity, many villages in the area did acquire their own cotton ginning factories, which convert the kapas to lint. However this activity was not at su¢cient scale to absorb the Gounders’ surplus capital. A more promising economic opportunity presented itself in the 1920s with the establishment of the …rst spinning and weaving mills in Coimbatore city. Here it was the Kammavar Naidu community that edged out the Gounders for control of this industry. The Naidus, who are originally from the neighboring state of Andhra Pradesh, constitute the second largest agricultural community in Coimbatore district. Their fortunes grew, together with the Gounders, over the last quarter of the nineteenth century and the beginning of this century. Most of the thirty-odd mills started in Coimbatore between 1932 and 1939 were controlled by Naidus. While Coimbatore soon emerged as the third largest textile center in the country, behind Bombay and Ahmedabad, the Coimbatore mills tended to be substantially smaller than those elsewhere. Most of the capital for these mills came from friends and relatives of the promoter, in contrast with the Bombay mills where the stocks were widely distributed, cutting across communal lines (Baker, 1984, Kulke, 1975). Thus while the Gounder community did help …nance the Coimbatore mill-industry to a limited extent, it would have to wait three more decades before it could …nd an industrial opportunity of its own. As described earlier, the advent of the railways and the introduction of the long-staple Cambodia variety provided a huge boost to cotton cultivation in Kongunad. At the beginning of this century, a group of merchants from Bombay established the …rst cotton market in Kongunad, choosing Tirupur a small ginning town as the location for the market. The Noyyal river runs past the town and its soft salty water was traditionally used for bleaching cloth (Harris, 1996). Nicholson (1887) describes Tirupur as a busy railway station from where large shipments of cotton were despatched. However it was only with the establishment of the cotton market that Tirupur began to emerge as an important regional center. Its population grew from 6,000 at the beginning of the century to 18,000 in 1931 and …nally to 52,000 by 1951. Note however that Tirupur remained a trading-town at this point. It was not until the 1960s that it would begin its rise as an industrial center.10 The …rst hosiery factory, with hand-operated machines, was set up in 1935. For the next thirty years the industry in Tirupur mainly produced grey and bleached banians (vests). Up to this time the Chettiars, who are traditionally traders, dominated the industry. Most of the production was centralized, with little subcontracting. There was a prolonged period of labor unrest in the mid-1960s which 10 The following discussion is based on personal correspondence with G. Karthikeyan, a local businessman, and Swaminathan and Jeyaranjan, (1994).

7

culminated in the Chettiars withdrawing from Tirupur, to be replaced by the Gounders. The Gounders reduced the scale of operations of individual units, with a corresponding expansion in subcontracting. The Tirupur production network, with the Gounders at its core, was …nally established.

2.3

Tirupur’s Rise as an Industrial Center

After catering to the domestic market for four decades, Tirupur received its …rst export orders in the 1970s. After an initial transition period, exports began to increase rapidly from 1985 onwards. Turning to Figure 1 we see that direct exports from Tirupur increased at 50-100 % per year over the 1990s. Taking into account indirect exports through industrial houses in Bombay, Tirupur is estimated to produce close to 70 % of India’s knitted-garment exports today. The town’s population kept pace with the growth in the industry, increasing from 113,303 in 1971 to 187,700 in 1981 and 235,000 in 1991 (Cawthorne, 1995). The population of Tirupur most likely exceeds 300,000 today, a …fty-fold increase over the past two centuries. Most of the new entrants into the industry are Gounders, making the transition from agriculture. However a large number of entrepreneurs from other parts of the country also moved into Tirupur, mostly as direct exporters, to exploit the new business opportunities that were available. These were seasoned businessmen from traditional business-communities, with generations of family-experience in trade and industry. Where did the Gounders get their capital from? As discussed earlier, the Tirupur industry provided a channel through which the community could invest its considerable agricultural surplus. Apart from the direct pro…ts from farming, the Gounders were also by this point wealthy landowners. The price of land in Kongunad, particularly the black cotton soil and the thottam gardens, was exceptionally high by the beginning of the century. This re‡ected the demand for land, stemming from the high pro…ts from cash-cropping, as well as the capital- investments in irrigation that had been made over time. The land market in Kongunad was also more active than elsewhere in Tamilnad (Baker, 1984). Thus most of the Gounder businessmen in Tirupur sold or pledged land to invest in machinery. As we will see in the data, the level of investment was very high. As one Gounder entrepreneur put it, “the Gounders do business like they do agriculture. Their objective is to accumulate more machinery.” Of course, this is precisely the investment-strategy that they had used to farm so successfully on the heavily- irrigated thottam gardens of Kongunad for generations. In contrast, the Outsiders in the industry such as the Chettiars, Gujaratis and Marwaris all 8

belong to traditional trading communities. A bank manager described their investment behavior in the following manner: “their objective is to turn the capital over as quickly as possible. They are not interested in tying the capital up by investing in machinery.” In fact, these communities have historically moved in and out of Kongunad as economic opportunities changed over time. Baker (1984) describes how the Bombay merchants, who established Tirupur as a cotton-market, left when the market collapsed following the boom years of the 1920s. He uses Census data to uncover similar movements among the Marwari businessmen in the area. Finally he describes how the Chettiars, who are local Tamil traders, wound up their business and moved to South East Asia when the market collapsed in the 1930s. All of these communities returned to Tirupur as its fortunes grew and now play an important role in the industry. They however continue to search for new business opportunities. Given the relatively thin second-hand market for machinery in Tirupur, they maintain their mobility by restricting investment in …xed-machinery.11 The industry today contains a mix of Insiders and Outsiders, pursuing very di¤erent investment strategies. The Gounders channel their surplus capital into the network by investing in machinery. The Outsiders prefer to exploit the capital-resources that the network provides, without investing in machinery of their own. We will keep these distinctions in mind when we look next at the data. The di¤erence between the two communities will also be an important feature of the more formal empirical analysis that follows.

3

A Description of the Industry in Tirupur

The easiest way to understand the organization of the industry is to follow an order through the various stages of the production process. The major stages are knitting, dyeing and stitching. Among the broad classes of agents in the production network, exporters may own machinery in multiple stages of the production process whereas job-workers are restricted to a single stage. Job-work allows for vertical decentralization in the production process. When an order is received by the direct exporter he passes a fraction through to one or more indirect exporters. Indirect exporters are entirely responsible for their share of an order, sheparding it through the various stages of the production process before delivering the …nished product to the direct exporter. For the remainder of the order, for which he takes responsibility himself, the direct 11 Second-hand machinery accounted for less than 5 % of the capital stock (total value of machinery) held by direct exporters, from both communities, in a short follow-up survey of direct exporters that we conducted in 1997.

9

exporter will supplement his own machinery with job-work from the production network. We use two sources of data in the paper. Six hundred direct exporters, indirect exporters and jobworkers were surveyed in 1995. Details of the entrepreneur’s background, his access to …nance, as well as export (production) and investment information over a four-year period, 1991- 94, was collected from each …rm. Subsequently we re-surveyed the one hundred and forty-seven direct exporters in the original sample in 1997. Additional information on the exporter’s background, which we had neglected to ask in the original survey, as well as partnership details, were collected at this time from the surviving …rms. Before turning to a description of the industry, we brie‡y describe the sampling procedure employed in the 1995 Survey, which is non-standard. The Tirupur production network is a complex institution comprising at least a couple-thousand production units. Most of the units are unregistered, so there is no “list” of …rms in the town. Moreover, accurate maps of the town are unavailable. Tirupur, like most small Indian towns, is a maze of lanes and by-lanes. In these circumstances we were unable to conduct a census of all production units, which would have allowed us to randomly sample …rms for the survey. Instead, we divided the town into ten zones and then proceeded from one zone to the next until the entire network was covered. Production units in Tirupur are located for the most part in converted residential structures. There is no segregation of homes and production units today, so in principle we would have to cover every street in the town if we wished to collect information from all the …rms. Our approach was to search through a zone, covering the streets within it as best as we could. To maintain control of the survey, the number of interviewers never exceeded eight at a time and much of the survey was conducted with four interviewers. The 1995 Survey ultimately took three months to complete, exceeding the estimated duration by a month. In our …rst pass through the town we focused on exporters only, the job-workers were surveyed later. Direct exporters and indirect exporters are ex ante indistinguishable. We ended up with one hundred and twenty …ve …rms in the former category and three hundred …rms in the latter category. To increase the number of direct exporters in the sample we took advantage of our connections with the Export Credit Guarantee Corporation of India (ECGC), a government agency that insures exporters, to interview some of their clients whom we had missed earlier, bringing the number of direct exporters close to one hundred and …fty. This over-sampling is in principle not a cause for concern since we do not compare direct and indirect exporters in the analysis that follows. Furthermore, the name 10

of a …rm or the entrepreneur’s appearance does not reveal his community. So it is unlikely that our search for production units was biased towards either the Gounders or the Outsiders. Sampling bias could still arise, for instance, if the larger Outsiders were more hostile to the interviewers and tended to turn them away. We made every attempt to avoid such a problem, returning repeatedly to …rms to complete un…nished interviews. While our sampling procedure is not standard, we nevertheless remain fairly con…dent about the integrity of the data that was collected. The discussion that follows focuses for the most part on the one hundred and forty-seven direct exporters in the 1995 Survey. Since we are particularly interested in comparing the investment behavior and the export performance of the Gounders and the Outsiders, the sample is partitioned by community. We also divide …rms into Young and Old units, where the cut-o¤ separating these …rms is speci…ed to be …ve years of export-experience. Very few …rms in our sample have more than ten years of experience. Note that we have data over a four- year period, 1991-94, for most of the variables that we discuss below. We begin with the individual characteristics of the direct exporters in Columns 1-4 of Table 1. Each exporter provides information on when he …rst arrived in Tirupur as well as when he received his …rst direct export order. The entrepreneur’s age and experience can then be computed at each point in time over the sample-period where age is the number of years elapsed since he …rst arrived in Tirupur while experience is the number of years since he became a direct exporter. The age minus the experience tells us how long it took for the entrepreneur to receive his …rst direct order. The samplestatistics for the two communities are fairly similar, with the exception of the Old direct exporters who have more than …ve years of direct-export experience. In this category we see that the age variable is signi…cantly larger for the Gounders. This tells us that the older Gounders waited a relatively long time before they received their …rst export orders. Turning to the sources of …nance for machinery, self-…nance is clearly the dominant source although more experienced exporters appear to have greater access to bank-credit. Gounders rely more on their own capital than the Outsiders in all categories, although the di¤erence in means is not signi…cant at the 5 % level. When listing the sources of …nance for buying machinery, the proprietor was allowed to choose from self-…nance, friends, relatives, the bank and informal …nancial institutions. As noted above, self-…nance and bank …nance completely dominated the other sources of …nance, for both communities and for Young as well as Old exporters. After the 1995 Survey was completed, we returned to some of 11

the respondents to probe this question further. It turns out that self-…nance was interpreted by many of the respondents to include outside partnership in the …rm. Since partnership is one channel through which family wealth, or community wealth more generally, could be passed through to the …rm, we re-surveyed the direct exporters in our sample in 1997 in order to collect more information about the partnership arrangement. It turns out that as many as 25% of the Outsiders and 31% of the Gounders did have outside partners in their …rms. Among these partnership …rms, 29% reported an immediate family member (father or brother), 51% reported an extended relative (uncle, cousin or in-law), and 32% reported a friend or business associate, as one of the partners. There were on average three outside partners in these …rms, and the proprietor’s share of the ownership was 44% for the Outsiders and 39% for the Gounders. Formal partnership will certainly under-estimate the role of the family and the community in the industry, since much of the capital would ‡ow without direct involvement in the business. Nevertheless, we observe a strong role for the family in a fairly large proportion of the …rms, in both communities. Further, while the 1995 Survey collected extensive information on exports, production and …nance from each …rm, we neglected to ask details about the exporter’s background. In particular, we did not ask the exporters about their education. It turns out that the Outsiders received more schooling than the Gounders; the average years of education for the two communities, with standard deviations in parentheses, are 13.41(2.62) versus 11.90(3.96). We can reject the equality of means for the two communities with greater than 95% con…dence, which is not surprising given their histories. Finally, 74% of the Outsiders versus 57% of the Gounders belong to families with previous experience in the textile industry. There is therefore some prima facie evidence to suggest that the Outsiders may have higher ”ability” in the sense of being better prepared to become successful producers. With this background in hand, we return to the 1995 Survey to study the investment and export variables that lie at the heart of the empirical analysis. Average exports for the two communities are very similar for Young direct exporters, and among the Old exporters average exports are higher among the Gounders. Gounders own signi…cantly more machinery at each stage of the direct exporter’s life-cycle. Looking at the capital-export ratio, the di¤erence between communities, particularly for the Young exporters, is striking. The Gounders appear to stock up on capital early in their careers. Consistent with this view, a signi…cantly greater fraction of production is allocated to indirect exporting by the Gounders. Remember that a direct exporter will accept indirect orders from other exporters when his machinery is idle. All exporters accept less indirect exports as they grow more experienced, 12

presumably because demand stabilizes over time. Yet Old Gounders continue to maintain a substantial level of indirect exporting which suggests that their own orders are never su¢cient to keep their machinery running at full capacity. In contrast, the Old Outsiders in our sample focus exclusively on direct exporting. There are also clear di¤erences in the extent of vertical integration. While we focus on the level of capital stock held by the …rm throughout the analysis, does this translate into a greater degree of vertical integration? De…ning vertical integration as ownership of machinery in all three stages of production we see that 15 % of the Young Gounders are vertically integrated, as opposed to 7.5 % for the Young Outsiders. These numbers increase when we study partial vertical integration, which is de…ned as ownership in two or more stages of production, but the distinction between the two communities remains. These are however average di¤erences which do not control for di¤erences in the distribution of experience within the two communities. In other words, we do not compare Gounders and Outsiders who have the exactly the same level of experience. As we will see later, controlling in that way does change some of the results substantially. Here we report one set of numbers where the comparability is not an issue: the capital stock in the year prior to the …rst export order is available for all direct exporters who commenced during the sample-period. The distinction between the communities, noted above, holds for the starting capital stock as well. Gounders start with nearly three times as much capital as the Outsiders. To conclude we brie‡y describe the characteristics of the indirect exporters, who are very di¤erent from the direct exporters in our sample. Looking at Columns 5-8 in Table 1 the following di¤erences are readily discernable. First, they are younger. This is not surprising since many hope to move up and receive direct orders of their own. Very few indirect exporters, particularly among the Outsiders, remain in the network once they have crossed …ve years of age. Pro…t-margins are small for these producers and most will leave the network if they do not receive direct export orders within a few years. Second, the indirect exporters are much more reliant on private sources of …nance than the direct exporters. Capital stock and production are also far lower than the corresponding levels for the direct exporters. Notice that there is little di¤erence between Gounders and Outsiders among the indirect exporters. Furthermore, …rm characteristics hardly change with the exporter’s age, although this pattern in the data may be due to selected-exit among the older indirect exporters. 13

Third, most of the indirect exporters are Gounders. In contrast, the direct exporters are evenly divided between Gounders and Outsiders. Migrants into the network appear to be primarily interested in ultimately obtaining direct orders of their own.

4

A Simple Model of Investment and Exports

In the previous Section we described the pattern of investment and exports, across communities and over the …rm’s life-cycle. Before turning to a more detailed analysis of the data, we present a model of investment and exports below. This model will help us interpret the regression results that we obtain in Section 5. In this section we propose a very simple model that captures the essential features of the Tirupur industry. We imagine that there is a large population of …rms producing knitted garments for exports. In each period …rms lose a fraction of their regular buyers but acquire a random fraction of new buyers as well. This is captured by writing the following equation for Xt+1 , which should be thought of the amount of exports that the …rm is sure to get in period t + 1 : Xt+1 = ¦tXt (1 + ²t ): We think of ²t as a random shock distributed on [0; ²] and ¦t as a number between 0 and 1. f(²t ) is the distribution of ²t .12 This equation is easy to interpret if we think of Xt (1 + ²t) as the realized exports in period t. Then this equation says that the minimum possible exports in the next period is a fraction ¦t of the current period’s actual exports. Firms produce exports using one input, capital. Capital stock for period t is chosen in period t ¡ 1

(when Xt is known but not Xt (1 + ²t )). Denote it by Kt¡1 . We now make the central assumption of the model: we assume that ¦t is given by the equation ¦t = ¦(

Kt¡1 ; ®) Xt (1 + ²t )

with ¦1 > 0; ¦2 > 0 and ¦11 < 0: This equation represents the idea that a …rm that uses more capital per unit of exports retains more of its buyers. This is reasonable since a …rm that owns less capital and yet produces the same amount of exports, presumably relies more on indirect exporters and jobworkers, both of which make it harder for it to control its quality. ® is parameter that measures the 12 We can assume that f((¢) is the same for everyone because any …xed di¤erence in the f (¢) is completely equivalent to a di¤erence in the ability ®, which is introduced in the discussion below.

14

quality of the …rm: a higher ® …rm retains more of its clients. Finally we assume that there is a diminishing return to capital. The exporter maximizes his expected discounted pro…ts, which can be written as:

V (Xt) = maxE K

"

1 X t=1

t

± [Xt (1 + ²t ) ¡ rKt¡1 ]

#

under the assumption that the owner of the …rm is risk-neutral and discounts the future at rate ± and r is the interest rate that applies to the …rm. The pro…t maximization is subject to the constraint

Xt+1 = ¦(

Kt¡1 ; ®)Xt (1 + ²t ) Xt(1 + ²t )

The assumption implicit in this way of writing the problem is that the …rm can borrow as much as it wants at an interest rate of r which however may be speci…c to the …rm. This is a speci…c (though standard) way of modeling a capital market imperfection: we could have alternatively assumed that the …rm is credit rationed or faces an increasing interest rate schedule. However, this speci…cation will allow us to conveniently capture the notion that the two communities operate in di¤erent capital markets. Observe from the structure of the maximization problem, that the value of V will double if we simply double Xt and Kt¡1 for all t. It follows that

V (Xt ) = Xt V (1): Using this fact, we can write the maximand for this problem as

Ef[Xt (1 + ²t ) ¡ rKt¡1 ] + ±Xt+1 V (1)g We maximize this subject to:

Xt+1 = ¦(

Kt¡1 ; ®)Xt (1 + ²t ): X + (1 + ²t ) 15

Substituting this in, we can write the maximand as

EfXt [(1 + ²t) ¡ r Writing

Kt¡1 Xt

Kt¡1 Kt¡1 + ±¦( ; ®)(1 + ²t )V (1)]g Xt Xt (1 + ²t )

= zt , we write this as

Xt Ef[(1 + ²t ) ¡ rzt + ±¦(

zt ; ®)(1 + ²t )V (1)]g 1 + ²t

Maximizing this w.r.t. zt gives us

r = ±Ef¦1 (

zt ; ®)gV (1) 1 + ²t

This determines z ¤ (®; r): It follows from the second-order condition for this maximization that an increase in r leads to fall in z ¤ : The e¤ect of an increase in ® on z, when z and ® are complements (¦12 > 0) is unambiguously positive, since an increase in ® both raises ¦1 and V (1) (higher ® means higher lifetime income). The e¤ect may be either positive or negative when z and ® are substitutes (¦12 < 0); since the positive e¤ect on V (1) may or may not dominate the negative e¤ect on ¦1 : Consider next the accumulation process of someone who starts out with an expectation X1 . Then

X2 = X1 (1 + ²1 )¦(

X3 = X2 (1 + ²2 )¦(

z¤ ; ®) 1 + ²1

z¤ z¤ z¤ ; ®) = X1 (1 + ²1 )(1 + ²2 )¦( ; ®)¦( ; ®) 1 + ²2 1 + ²1 1 + ²2

Xt = X1 (1 + ²1 )::::::::(1 + ²t¡1 )¦(

) log Xt = log X1 +

t¡1 X

z¤ z¤ ; ®)::::::::::¦( ; ®) 1 + ²1 1 + ²t¡1

log(1 + ²s ) +

S=1

t¡1 X

S=1

log ¦(

z¤ , ®) 1 + ²s

Taking expectations we get the dynamic path of evolution of exports: 16

E log Xt = E log X1 + (t ¡ 1)E log(1 + ²s ) + (t ¡ 1) E log ¦(

z ¤ (®; r) , ®): 1 + ²s

To get the corresponding equation for capital stock, we observe that Kt¡1 = Xt ¢ z ¤ (®; r) which

gives us

E log Kt¡1 = E log X1 + E log z ¤ (®; r) + (t ¡ 1)E log(1 + ²s ) + (t ¡ 1) E log ¦(

z ¤ (®; r) , ®): 1 + ²s

Since E log(1 + ²s ) is the same for everyone,13 both capital stock and exports grow faster if and only if ¦ is higher. Con…ning ourselves …rst to the case where ¦12 > 0; it is easy to see that both lower r and higher ® lead to higher values of both ¦ and z ¤ . Thus if one …rm has steeper export and capital-stock trajectories, as well as a higher z ¤ , than another …rm, this could be either because it faces a lower interest rate or because it has higher ability. However, in our case we will see that the Gounders have a higher z ¤ but a lower ¦(z ¤ ; ®) than the Outsiders. This tells us that the Outsiders must have a higher ®. So the Gounders must have a lower r to sustain a higher z ¤ . This is the basic result that we exploit in the paper to show that Gounders face lower interest rates than the Outsiders. Up to this point we have assumed that ¦12 > 0. The results derived above extend to the case where ¦12 < 0 but small, since the e¤ect of a higher ® is still to increase z ¤ . The more di¢cult case is where the substitution e¤ect is large enough that a decrease in ® increases z ¤ . In this case a …rm with a higher z ¤ could still have a lower ¦, even if r were the same for everyone. This would explain the higher z ¤ and the shallower trajectories for the Gounders, as a consequence of their lower ability. There are however two reasons to be skeptical of this explanation of the data. First, a priori, if all participants in the industry face the same interest rate, then the fact that the Gounders seem to be able to survive in the industry despite their lower ability and despite the fact that they spend more on capital suggests that the Outsiders must earn huge rents. It seems puzzling that this would not have attracted a much larger in‡ow of Outsiders over the rather long period that the industry has existed. Second, if substitution e¤ects are the entire story, the same kind of negative relation between z ¤ and ¦ should also be found within each community. Note that by contrast, in our story with di¤erent interest rates across communities, the relation between z ¤ and ¦ does not have to be the same within and across communities. In the case where ¦12 is positive (or negative and not too large) the relation 13

This is almost by construction: di¤erences among people have been swept into the ¦ function.

17

between z ¤ and ¦ will be positive within the community (where interest rates ought to be similar across …rms), and yet could be negative across communities. The problem in estimating this relationship within the community is that z ¤ is not observed by the econometrician: what we observe instead is z ¤ =(1 + ²t) in period t, and the corresponding level of realized exports Xt (1 + ²t ). These variables are negatively correlated by construction. To avoid this problem, we look at the correlation between capital stock Kt¡1 and exports instead. Since capital stock would also be correlated with lagged export-shocks, which could be serially correlated, we work backward to a point in time just before the exporter received his …rst direct order. Taking the substitution view, this initial capital stock K0 should be higher for low-ability …rms and therefore negatively correlated with subsequent exports. The initial capital stock is available for …rms that received their …rst order during the sampleperiod. We estimate the K0 ¡ Xt(1 + ²t) correlation separately for the …rst three years of experi-

ence, by community. A positive and signi…cant correlation is obtained, without exception, for both communities. Higher initial capital is associated with higher future exports, consistent with the more standard view that ability and capital are complements (see, for instance, Griliches, 1957, and Olley and Pakes, 1996). We consequently maintain the assumption that ¦12 > 0 in the discussion that follows.

5

Estimation

We saw in Section 3 that the Gounders held higher levels of capital on average than the Outsiders. This di¤erence translated as expected into a lower level of network-utilization by the Gounders. We also saw that the Older Gounders had higher exports than comparable Outsiders, whereas di¤erences between the two communities were insigni…cant among the Young exporters. We now subject these patterns in the data to more careful scrutiny by comparing investment behavior and export outcomes for the two communities at each point in the exporter’s life-cycle. Exports, capital-stock, the capitalexport ratio and the extent of indirect exporting are regressed separately on the exporter’s experience in Table 2. Recall that experience refers to the number of years of direct exporting. To allow for variation in the investment and export trajectory over the life-cycle, separate export-coe¢cients are estimated for Young and Old exporters in Table 2. The cut-o¤ separating these categories is speci…ed as …ve years of direct export experience. We will also present the corresponding nonparametric kernel

18

estimates of these trajectories in Figures 2-7. We begin with the export regression in Column 1. Exports, in logs, are increasing over time for both communities, although the trajectory is ‡atter for the Old exporters. We cannot reject the null hypothesis, at the 5 percent signi…cance level, that the coe¢cients for the two communities are the same, in both stages of the life-cycle. While the Gounder community-dummy is positive in Column 1, it is also not signi…cant. In general, the trajectories for the two communities are statistically indistinguishable. This feature of the data is more clearly demonstrated in the corresponding nonparametric regression, presented in Figure 2. While the Older Gounders appear to grow faster than the Old Outsiders, consistent with the patterns in Table 1, the 95% con…dence-bands for the two communities overlap throughout the exporter’s life-cycle. Part of the variation that we are picking up in Column 1 and Figure 2 is due to changes in each exporter’s experience over the sample-period. The rest of the variation is due to cross-sectional heterogeneity across …rms with di¤erent levels of experience in the industry. This is a potential source of bias, since the characteristics of …rms entering or exiting the industry could have changed over time as it evolved. To control for such heterogeneity, we include …rm …xed-e¤ects in Column 2 of Table 2.14 Notice now that we cannot reject the null that the export-coe¢cient is constant over time, for both communities. The trajectory is also steeper for the Outsiders, both among Young and Old exporters. However, the Gounder community- dummy remains positive, and is now statistically signi…cant. These patterns in the data are once more easy to visualize with the corresponding nonparametric regression, presented in Figure 3.15 The Gounders begin with higher exports, but the Outsiders surge ahead after about …ve years of direct exporting. Contrast the linearity in Figure 3, once the …xed-e¤ects are di¤erenced out, with the concave trajectory that we observed in Figure 2. We interpret this change to be due to inter-…rm heterogeneity, which is purged in Figure 3. Once the community dummy is included in the export regression, the 14

Using within estimation to purge the …xed-e¤ects we are e¤ectively estimating a set of year dummies in this case. With a balanced panel, EXPit ¡ EXP it is the same for all …rms in a given year, where EXPit is …rm i’s experience in year t and EXP it is its average experience over the sample period. The experience-coe¢cient thus measures the linear time- trend in these dummies. An additional year dummy must therefore be dropped in the …xed- e¤ects regression. 15 To obtain the nonparametric kernel estimates in Figure 3 we …rst di¤erence-out the …xed-e¤ects from the nonparametric series approximation, presented in Column 2 (following an approach suggested by Porter, 1996). We assume here that the …rst stage is ‡exible enough to capture the basic features of the export trajectory, and indeed the linearity in the kernel estimates is consistent with the patterns that we obtain with the series estimator. All the nonparametric regressions in this paper utilize the Epanechnikov kernel function. Pointwise con…dence intervals are computed using a method suggested by HÄ ardle (1990). We assume that the estimated …xed-e¤ects are “…xed" when computing the nonparametric con…dence intervals since the kernel estimates converge much more slowly than the …xed-e¤ects.

19

…xed-e¤ect measures the deviation in the …rm’s characteristics from the community-average. Figure 2 and Figure 3 suggest that these deviations are systematically lower for more experienced …rms. The industry in Tirupur evolved very rapidly over the past two decades, so the …rms that entered more recently are very likely to be di¤erent from those that were already operating when the export-boom began. Once we have accounted for this compositional change in the industry, a clear distinction between the two communities begins to emerge. Turning to the capital-stock regression, we will see that much of the preceding discussion applies here as well. Starting with the regression without …xed-e¤ects, we see in Column 3 that the capitalstock trajectory, in logs, is positive and concave for both communities. The export-coe¢cients for the two communities are statistically indistinguishable and the Gounder community-dummy is positive but insigni…cant. Looking at the corresponding nonparametric estimates in Figure 4, the capital stock for the Gounders is signi…cantly higher than the corresponding level for the Outsiders, consistent with what we saw in Table 1. This distinction between the two communities is observed at each point in the exporter’s life-cycle, except for a brief period around the …ve-year experience mark. Introducing …xed-e¤ects in the capital stock regression in Column 4, the di¤erence between the communities widens. The trajectory is now linear for both communities and steeper for the Outsiders. Finally, the Gounder community-dummy is positive and signi…cant. All of these patterns are observed in the corresponding nonparametric regression presented in Figure 5. Including …xed-e¤ects in both the export and the capital-stock regressions appears to purge unobserved …rm heterogeneity, which would otherwise bias the estimated trajectories for the two communities. In the regressions that follow, we consequently report results with …xed-e¤ects only. Since the Gounders invest in signi…cantly more capital-stock than the Outsiders, yet are surpassed by them in export performance, we would expect the capital-export ratio to be higher for the Gounder community. The community-dummy is indeed signi…cantly larger in Column 5 of Table 2, which regresses the capital-export ratio (in logs) on the exporter’s experience. We noted in Section 3 that the exporters, particularly among the Gounders, seemed to load their investments up front. The capital-export ratio is as expected declining for both communities in Column 5, more steeply for the Outsiders. The decline over the life-cycle appears to be linear once more. These features of the data are all discernable in the parametric regression presented in Figure 6. What is important for our purpose is that the capital-export ratio is signi…cantly higher for Gounders at every point in the exporter’s life-cycle. 20

We conclude the reduced-form regressions by studying the extent of indirect exporting. The production network in Tirupur is very ‡exible, allowing direct exporters to function as indirect exporters when they do not receive orders of their own. We would expect that …rms have excess capacity more frequently when the capital-export ratio is high. Indeed, we see in Column 6 that the Gounder community-dummy is positive (but insigni…cant), and that direct exports as a share of total production is declining in experience for both communities. The nonparametric estimates in Figure 7 reveal that the Gounders start with about 25 % of their production as indirect exports, with the corresponding statistic for Outsiders around 20 %. Thereafter, the importance of indirect exporting declines for both communities, but more steeply for the Outsiders. By the end of their life-cycle, the Outsiders focus entirely on direct orders, whereas the Gounders remain with about 15 % of their production in indirect exports. The patterns that we observe in the data can be easily interpreted, using the simple model presented in Section 4. The export and capital-stock trajectories are both steeper for the Outsiders, which implies that ¦(z ¤ ; ®) must be greater for them. In contrast, the capital-export ratio z ¤ is higher for the Gounders at every stage of the exporter’s life- cycle. Looking at the export trajectory in more detail, notice in Figure 3 that the Gounders start with higher exports than the Outsiders. This tells us that the ElogX1 term in the dynamic export equation in Section 4 is higher for the Gounders. The corresponding term in the capital-stock equation is ElogX1 + Elogz ¤ (®; r). So we would expect the initial gap between the communities to be even wider in the capital regression. Indeed this is the case in Figure 5. In fact, the Outsiders never overcome the Gounders’ lead in the level of capital stock. In contrast, their exports surge ahead of the Gounders within the space of …ve years. The simple model that we have presented seems to capture many features of the data very well. However it fails to explain the decline in z ¤ (®; r) that is observed for both communities. This is because we assume that ¦ is constant over time. It is quite plausible that ¦ may be less responsive to z for …rms with higher levels of experience - because, for example, their reputation may be much t¡1 more secure. If we introduce the assumption that ¦ varies with time (i.e. ¦ = ¦( XtK(1+² ; ®; t)) and t)

in particular, that ¦1 falls as the …rm ages, into the model in section 4, it is easily checked that the basic analysis still goes through. The value function is still linear in expected exports. and as long as ¦12 > 0 continues to hold for all values of t, we can still explain why the Gounders have a higher z ¤ but both capital and exports grow faster for the outsiders. z ¤ is however no longer constant and 21

will in fact decline over time - which …ts the facts well. The one complication it introduces is the possibility that the time paths of log-capital and log-exports may not be linear, though it does not rule out their being linear.

6

Conclusion

This paper studies the role of the community in organizing production, in a world where markets function imperfectly. Our data comes from a survey of six hundred …rms that we conducted in Tirupur’s garment-export industry. This industry was set up by the Gounders, the dominant cultivating caste in the region surrounding Tirupur. The dramatic growth in exports from Tirupur over the past two decades was accompanied by a corresponding change in the sociological composition of the industry, with entrepreneurs arriving from di¤erent parts of the country to do business there. It is the substantial presence of both Gounders and Outsiders in Tirupur that allows us to identify the e¤ect of participating in a community network. The basic idea is that the Gounders will invest most of their wealth in Tirupur, since they have limited opportunities elsewhere. In contrast, the Outsiders will only invest in Tirupur if the rate of return is comparable to the best opportunities available elsewhere in India. The opportunity cost of capital faced by the Outsiders is therefore likely to be higher than that faced by the Gounders. We wish to empirically establish to basic propositions in this paper. First, that the organizational structure of the Gounder …rms and the Outsider …rms is di¤erent, in the sense that the Gounders own more machinery and so are less reliant on the production-network. Second, that these organizational di¤erences arise because the two communities operate in di¤erent capital markets. The data show very clearly that the Gounders sink substantially more capital into production than the Outsiders, at every stage of the exporter’s life-cycle. Nevertheless, exports grow faster for the Outsiders than for the Gounders. The Outsiders begin with signi…cantly lower exports than the Gounders, but surpass them after about …ve years of export experience. We show in the paper that these patterns in the data, taken together, must imply that the Outsiders have higher ability while the Gounders face a lower cost of capital. We are thus in a position to establish that the capital market in Tirupur is segmented along communal lines, distorting investment to the point where the less able own more capital. While we cannot directly identify the presence of a network in this paper, the segmentation of the capital market is a natural outcome when wealth-‡ows fail to

22

cross community boundaries.

References [1] Baker, Christopher (1984). An Indian Rural Economy, 1880-1955: The Tamilnad Countryside, Oxford University Press. [2] Banerjee, Abhijit and Andrew F. Newman (1998). "Information, the Dual Economy and Development," The Review of Economic Studies, Vol. 65, No. 4. [3] Beck, Brenda E.F. (1079). Peasant Society in Konku: Subcastes in South India, Vikas: New Delhi. [4] Cawthorne, Pamela M. (1995). "Of Networks and Markets: The Rise and Rise of a South Indian Town, the Example of Tiruppur’s Cotton Knitwear Industry," World Development, Vol. 23, No. 1. [5] Greif, Avner (1993). "Contract Enforceability and Economic Institutions in Early Trade: the Maghribi Traders’ Coalition," American Economic Review, Vol. 83, No. 3. [6] Griliches, Zvi (1957). "Speci…cation Bias in Estimates of Production Functions," Journal of Farm Economics, Vol. 39. [7] Harris-White, Barbara (1996). A Political Economy of Agricultural Markets in South India: Masters of the Countryside, Sage Publications: New Delhi. [8] Kulke, Eckehard (1974). The Parsees in India: A Minority as Agent of Social Change, Welforu Verlag: Munchen. [9] Olley, Steven G. and Ariel Pakes (1996). "The Dynamics of Productivity in the Telecommunications Equipment Industry," Econometrica, Vol. 64, No. 6.

23