facilitating development - role of bgs

7/30/2019 Facilitating Development - Role of BGs

1/50

Harvard Business School

Competition & Strategy Working Paper Series

Working paper number: 98-076

Working paper date: February 1998

Facilitating Development: The Role of

Business Groups

Raymond J. Fisman (The World Bank)

and Tarun Khanna (Harvard Business School)

This paper can be downloaded without charge from the

Social Science Research Network electronic library at:http://papers.ssrn.com/paper.taf?abstract_id=98801
http://papers.ssrn.com/paper.taf?abstract_id=98801http://papers.ssrn.com/paper.taf?abstract_id=98801http://papers.ssrn.com/paper.taf?abstract_id=98801http://papers.ssrn.com/paper.taf?abstract_id=98801


2/50

FACILITATING DEVELOPMENT:THE ROLE OF BUSINESS GROUPS

Raymond Fisman & Tarun Khanna

Harvard Business School

July 2, 1998

Abstract

A defining characteristic of developing countries is the inadequacy of basic services

normally required to support organized economic activity. One way in which the private

sector acts to facilitate development is through investments orchestrated by

agglomerations of firms called business groups. Such groups dominate the landscape of

virtually all developing countries. Our study of plant location decisions in India shows

that group-affiliates are more likely to (profitably) locate in less-developed states than

unaffiliated firms. The magnitudes of the effects are large and significant, with group

affiliates being between 20% and 33% more likely to locate in less-developed states than

unaffiliated firms. We suggest that this is because the scale and scope of groups, and thede facto property rights enforcement within groups in environments where legal

enforcement is lacking, permit them to overcome of the difficulties that impair production

in underdeveloped regions.

___________

We thank Bharat Anand, Eli Berman, Richard Caves, Alex Dyck, Kei Hirano, Devesh Kapur,Jonathan Morduch, Sendhil Mullainathan, Aviv Nevo, Jan Rivkin, Nicolaj Siggelkow, Steve Spear,Peter Timmer, Catherine Wolfram, Yishay Yafeh, and seminar participants at HIID/MITDevelopment Seminar, Organizations Seminar at Harvard Economics, Harvard Business School andthe University of Western Ontario for many helpful comments and discussions. Many thanks toMargaret Schotte, for her diligent and tireless research assistance. The usual disclaimer applies.


3/50

2

One of the defining characteristics of developing countries is the inadequacy of

basic inputs normally required to support organized economic activity. Physical

infrastructure amenities are either inadequate or lacking entirely, while social services are

often severely underfunded, hampering the hiring and maintenance of a productive

workforce. Owing to such inadequacy of basic services, shortages in ancillary industries

that provide other essential inputs are also common.1 These difficulties are often so

extreme that it is a mystery how anything gets done at all. In this paper, we draw

attention to the role played by agglomerations of firms in the private sector, called

business groups, in ameliorating the costs of such shortages. Since business groups are a

prominent feature of the landscape of most, if not all, developing countries, their role in

facilitating development deserves attention.

Economic theory predicts that private enterprise should step in to provide these

basic services where it is economically efficient to do so. However, in most countries

where such failures take place, private firms are constrained from making this provision.

In many cases, government monopolies control industries such as telecommunications,

health care, transportation, and power; competition from the private sector is prohibited.

In recent years, these industries have been opened to competition in a host of countries.

Nonetheless, private enterprise has made little headway, as government bureaucracy and

regulation often make entry too expensive and time-consuming, and weak property rights

and contract enforcement prevent firms from capturing economic rents from the goods

and services they provide.2 Finally, government reputations may also play a role often,

regimes are unstable in developing countries, and the nationalization of infrastructure

assets may discourage firms from making investments in such projects.

Much time and effort has gone into imputing the costs of these inefficiencies and

1

The mere presence of physical infrastructure does not guarantee reliable service for example, in Nigeria,less than half of the power produced actually makes it to the end users, while in a host of developingcountries, the call failure rate for the public telecommunications system exceeds 50%. By comparison,industrialized countries have efficiencies of about 95%.2

For example, in India, a lot of power is taken illegally from the government power grid, resulting inleakages of up to 90%; it is not clear that a private company could do any better at controlling access to its


4/50

3

evaluating the productivity of various infrastructure goods.3 However, relatively little has

been done to understand how firms behave given the constraints of operating in

underserviced regions. Understanding how firms react to such shortages is essential to

determining the true costs and benefits of infrastructure provision, and is likely to be

useful in informing future development policy.

The reality is that firms do exercise choices in the manner in which they deal with

these shortages. Some firms simply make do with the services that are publicly provided,

accepting low productivity and poor product quality as the price of doing business in an

underdeveloped area. One obvious alternative is for companies to supply inputs for their

own private use. A host of companies in India have invested in V-Sat (very small

aperture terminal) connections to bypass ubiquitous communications bottlenecks. In

Nigeria, where public telecommunications are similarly unreliable, a 1988 World Bank

Study found that many firms had purchased radio equipment for communications. The

study also found that 92 percent of Nigerian manufacturers and 64 percent of Indonesian

manufacturers owned their own electricity generators. Firms in less developed regions

may similarly be required to ensure their own supply of physical inputs. For example, in

a study of Indian automobile parts manufacturers, it has been found that plants in less

developed regions keep higher inventories, since most inputs cannot be sourced locally4.

We contend that the business groups5 that dominate the landscape of virtually all

developing countries provide an organizational structure that is better suited to dealing

with the poor availability of basic inputs and services. Such groups are comprised of a

diverse set of businesses, often initiated by a single family, and bound together by equity

cross-ownership and common board membership. Group affiliates often share a common

brand identity (e.g. Salim Group in Indonesia, the Tata Group in India, Samsung in Korea,

investments.3

See Jimenez (1995) for a comprehensive literature survey; see Morrison and Schwartz (1996) for a recentcontribution to the literature on physical infrastructure; see Psacharopoulos (1994) for a review of theliterature on the productivity of educational expenditure.4

Personal communication, Sumila Gulyani, Department of Economics, MIT-5

These take different names in different countries, e.g. grupos in Latin America, chaebol in Korea,business houses in India.


5/50

4

etc.), draw on a common labor pool, and rely on each other for financing. The common

family bond that often runs through group affiliates acts as a social mechanism that

reduces the likelihood of reneging of contracts (see, for example, Granovetter, 1994).

Several of these features put groups at an advantage in dealing with the difficulties

facing producers in underdeveloped areas. The access to capital and the de facto property

rights protection that characterizes intra-group investments ensure that groups are

particularly well placed to make the large investments in public infrastructure substitutes

that small, unaffiliated firms might not find it rational to undertake.6 Greater size and

diversification also allow groups to better cope with the increased risk that often comes

with locating production facilities in less-developed areas. Because of shared labor pools,

groups are able to rotate managers and technicians through production facilities in areas

where skilled workers might be reluctant to reside in the long term. Finally, groups are

better able to compensate for shortages in physical inputs, either by providing these

inputs directly from group-affiliates elsewhere in the country, or by making arrangements

for their provision through already existing supplier networks.

If our hypothesis is true, then diversified groups may be an important means by

which industrial activity is brought to regions where governments have failed to provide

basic services. That is, groups may provide one mechanism for providing a partial

solution to one of the fundamental issues in economic development, with significant

implications for government policies in developing countries.

In this paper, we attempt to infer the extent to which group-affiliates are better

positioned to deal with the difficulties associated with underdeveloped areas by looking at

firms plant location decisions. If our hypothesis is true, we should find relatively more

group-affiliates locating in underdeveloped regions than non-group-affiliates; similarly,

large groups should be more likely to locate in less-developed regions than smaller groups.

Furthermore, in equilibrium, we should find that group firms are more profitable in low-

infrastructure regions, relative to unaffiliated firms.

6

Several of the investments in question are expensive for example, a V-Sat connection costs US$25,000,


6/50

5

Our empirical estimation focuses on a single country because both features that

we wish to explore level of development and business groups are difficult to measure,

making attention to detail important. Membership of business groups is typically quite

secretive. Often firms are members of more than one group, raising the potential for

specification error. There is no universal definition of a group that adequately captures

what it means to be a group affiliate across different countries. These factors argue against

a cross-country study.

India has a number of characteristics that make it particularly well suited for such

a study. A large proportion of Indian group-affiliates are publicly traded, and firms are a

member of only one group.7 Further, the large number of groups in India facilitates

statistical analysis. There is also much variation in the level of development across the

regions of the country, which allows us to look at firm decisions as a function of

development. While the variance in the level of development is high, the average is low.8

This is important, as we are primarily interested in studying firm reaction to difficulties

stemming from underdevelopment.9

We proceed as follows: Section 1 introduces and describes our data; in section 2,

we look at why firms might locate in low-infrastructure states; section 3 looks at the

specific advantages that groups have in locating in underdeveloped regions; we describe

our main results in section 4; alternative explanations and issues of robustness are

examined in sections 5; a discussion of our results and our conclusions are contained in

section 6.

while the cost of a power plant of an efficient scale is in the tens of millions of dollars.7

In most developing countries, group affiliates are, for the most part, privately held. For example,Indonesias Salim Group has an estimated 300-400 firms, but only five of these are publicly listed. Inaddition, firms in Indonesia, Central America and other countries are often partially held by several groups,making it difficult to determine the group membership of a particular firm.8

As a headline in an Indian news magazine proclaimed in February 1996, Indias infrastructure isstretched beyond capacity and will soon have domestic production in a hopeless bind. The article citeschronic shortages of electricity, communications, and transportation facilities as among the most seriousproblems facing Indian industry.9

Other countries satisfy the low level of development and high variance in level of development criteria, forexample, Brazil and China, but the lack of English language statements or of reliable measures of group


7/50

6

1. DataFirm-Level Data Sources

The data are obtained primarily from a publicly available database maintained by CMIE,

Centre for Monitoring the Indian Economy. CMIE is a privately run, twenty-year- old,

Bombay-based firm that maintains databases on private and public sector economic

activity in India. These databases have become the standard sources of information on

microeconomic and macroeconomic activity, and are commonly used by public and

private sector firms in India, international development banks, and the media.10 While

there are other possible data sources, CMIE data are the most reliable and comprehensive.

The database11 from which we draw our information contains data similar to those

contained in Compustat, CRSP and Mead Data Centrals Nexis databases, though our

data are not nearly as comprehensive or detailed. The database has computerized

information drawn from annual reports, other regulatory filings, and press releases of

several thousand firms operating in India, as well as daily stock price data for stockstraded on the Bombay Stock Exchange. The accounting information is available from the

mid-1980s, with coverage improving greatly in 1989. The year for which there is the best

coverage in the version of the database to which we have access is 1993. Coverage for

1994 and 1995 is sparser, due to delayed release of information by the firms, and delays

in updating the database. We therefore use 1993 data in the analyses that follow. The

majority of firms in the database are public firms (95%). Of these, approximately half are

traded on the Bombay Stock Exchange, with the remainder traded on several of the other

stock exchanges in the country.

affiliation, make these less appropriate sites for our analysis.10

As a recent example, Ahuja and Majumdar (1995) use these data to examine the performance of Indianstate-owned enterprises, and Khanna and Palepu (1997) use these data to show that the performanceconsequences of diversification in India are different from those in advanced economies.11

CMIE database: Computerized Information on Magnetic Medium.


8/50

7

Business Group Classification

The identification of business groups is of particular interest for this study. Between the

years 1989 and 1995, the database identifies a range of business groups, from the smallest

two-firm groups to the largest Tata Group. For the analysis in this paper, we adopt the

databases classification of firms into groups.12 Unlike in a variety of countries (Strachan

(1976), Goto (1982)), firms are members of only one group. Further, there is virtually no

movement of firms across groups.13 Khanna & Palepu (1997) have carefully examined

the quality of the CMIEgroup classification, and found that it passes a number of reality

checks.14

Our initial sample, consisting of the entire cross-section of firms in the CMIE

database, contained 4230 firms; many of these are multi-plant firms, yielding a total

sample of 5653 observations on location decisions. Since we were unable to obtain

adequate aggregate data for Union Territories (see below), plants located in these regions

were dropped (452 plant-level observations). We also removed government-run firms and

foreign-owned firms from the sample, because we felt that they were sufficiently different

12

While a group is not a legal construct, CMIE uses a variety of sources to classify firms into groups.Prior to the repeal of the MRTP (Monopolies and Restrictive Trade Practices) Act in 1991, acomprehensive list of firms belonging to Large Industrial Houses was published by the government.This forms a starting point for the CMIE classification for a number of the groups. Other significant sourcesof information include the following: (a) identifying the promoters of a firm when it was first started andtracing whether the original owners retained their affiliation with the firm (b) announcement by individualfirms of the groups of which they are a part, and by the groups of lists of affiliated firms. Suchannouncements appear periodically in annual reports, advertisements, or at the time of public offerings andnews releases about the groups and firms future plans (c) identifying the interest that a group has in aparticular firm through the membership of its board of directors. CMIE also regularly monitors changes ingroup structure. Shifts in group affiliation are extremely rare, but when they do occur, they are reflected inthe database. Note that the database does not contain a historical record of each firms affiliation with

different groups; rather the group membership variable reflects the most current affiliation. There is noambiguity between CMIEs classification of firms into groups and those attempted by other sources againstwhich we cross-checked the data.13

In the case of family-controlled groups, succession from one generation to another often results in thegroup being split into multiple parts. We identified several prominent groups that had gone through suchperiods of succession in the past twenty years, and checked to see that CMIE had indeed classified eachsub-group separately. Thus, the Birla group is classified in several different parts, as is the group originallyrun by the Goenkas.


9/50

8

from other Indian firms as to distort our results (217 government-run or joint-sector firms

and 90 foreign firms dropped)15. Firms were then removed that lacked observations on

the variables used in our basic regression, i.e., SALES (1027)16,17,18, age (23), and plant

location (890).

During interviews with both executives and government officials in Bombay and

Delhi, we discovered that, prior to 1975, plant location decisions were made by

government bureaucrats rather than business leaders. This was because a government

license was required for any new enterprise, and the license was generally given

conditional on the firm locating in the home state of the government official who made the

licensing decision (or in the state of another official to whom a political favor was owed).

Hence, our analyses are only relevant for locational decisions made after 1975. Our best

proxy for the dates of these decisions is firm incorporation date, so we drop all firms with

incorporation dates prior to 1975. This yields a final sample of 957 firms and 1193

plant-level observations. We outline in Appendix 1 the evolution of Indian industrial

reform that gradually led to more profit-based plant location decisions, and explain

precisely why we choose 1975 as a cutoff date19.

14

Note that there are a small number of groups for which information on only one firm is available for aparticular year. Such firms are nonetheless classified as group-affiliated. The classification into group ornon-group is not inferred solely from the number of entries appearing in the database.15

The case for excluding government-run firms is strong, as they differ markedly from other firms in oursample along many important observable dimensions, for example, earning negative profits on average.Moreover, it is very likely that the locational choice decisions of government-run firms would be madebased on entirely different criteria from private firms. Similarly, we expect that foreign firms would usedifferent criteria, though their inclusion/exclusion does not qualitatively affect our result (including foreignfirms as non-group firms actually increases the size of the group effect, since, as expected, foreign firms areless likely to locate in less-developed regions than domestic firms).16

Number of observations dropped are listed in brackets; the attrition is done sequentially17

Firms with zero sales were also omitted, since this often entered if data are unavailable.18 For these observations, no accounting data were available. We might be concerned that this would causethe following selection bias: suppose that data were unavailable for less well-established firms. Then wemight expect that the dropped observations would be excessively weighting non-group firms, as well asthose in less-developed areas. The combination of these two biases could mean that our results are biasedupward. While the former is definitely the case (76% of the firms dropped were non-group firms), the latterdoes not seem to be true (the average value of DEV for these firms is 0.62, somewhat higher than the fullsample average of 0.55). This suggests that the bias is more likely to run the other way. When we look atsome basic regressions without size controls, we do in fact find that the group effect for this subset of firmsis larger than the full sample value.19

Of course, our results are not substantively affected by the exact year chosen.


10/50

9

Measuring Development

In assessing a states level of development, we attempted to aggregate a number of

distinct elements that we felt were important for firm location decisions. Also, we hoped

that using an aggregate score would allow us to iron out some of the idiosyncratic

components that are present in each individual measure20. The four basic categories

included were power generation, telecommunications, transportation, and social services.

Where possible, we tried to obtain proxies that reflected the quality, rather than the

quantity, of services available21. Also, we avoided per unit area measures, using instead

per capita statistics, since vast undeveloped expanses, such as the Thar Desert in

Rajasthan, could seriously bias the former type22. Telecommunications and

transportation data were taken from CMIEsInfrastructure in India (1996), power

generation statistics were drawn from CMIEsIndias Energy Sector(1996), and

information on social services were taken from the Indian governments Statistical

Abstract(1992).

We used three basic measures of energy infrastructure: deficit in production, i.e.,

the percentage by which demand exceeds supply; percentage loss in transmission; and

generating capacity per capita. All of these factors affect the likelihood and frequency of

brownouts and blackouts. Moreover, percentage loss in transmission is correlated with

the consistency of voltage supplied (i.e., greater losses imply higher variance in voltage),

which is of serious concern for firms using advanced technologies23.

Our data on telecommunications and transportation are somewhat sparser. To

assess the accessibility of communications, we used direct phone lines per capita and

telephone exchanges per capita24. For transportation, we used the percentage of roads

20

We did look at each development measure individually. Unsurprisingly, the results were qualitativelysimilar to those using the composite measure described below, and were distributed around the overallresults we report below.21

Quantity measures were used in an earlier version of this paper, which obtained similar results.22

Per unit area measures were also used in earlier versions of this paper; our results were more or less thesame, though our standard errors were marginally bigger.23

Personal communication, Devesh Kapur, Government Department, Harvard University24

Unfortunately, we have not yet found more direct measures of telecommunications quality. However,it is of some consolation to note that, in cross-country data, telephone lines per capita is highly correlated


11/50

10

that are surfaced25.

Finally, for social services, we look at the public provision of education and

healthcare. For education, we use the percentage of teachers with formal training, as well

as total expenditure on education per capita. Our healthcare proxies are hospital beds per

capita and healthcare workers per capita.

We convert each of these values into a normalized score using the following:

TV V

Vsi

si i

i

=

where Vsi is the measure of amenity i in state s, and Vi is the mean over states for amenity

of type i. This transformation preserves the degree of dispersion associated with each

measure, while normalizing its mean to zero26. The score for each type of service is a

simple average of its components; our overall measure of development (DEV) is the sum

of these four scores, scaled to make the lowest score equal to zero and the highest equal to

one27. Table 2 summarizes our scores by state28.

Summary statistics of our variables from the CMIEdatabase are listed in Table 3;

we also include statistics broken down by group affiliation in Table 4. Note that the size

with more quality oriented measures, including telecom investment (0.76), overall quality as assessed bythe World Competitiveness Report(0.65), and faults per 100 calls (-0.52).25

It has been pointed out that the quality of surfaced roads varies drastically; this is only a problem if webelieve that surfaced road quality is negatively related to the percentage of roads that are surfaced.Conventional wisdom, however, would argue for the opposite.26

With this measure, the magnitudes of the differences in each measure matters. We also ran our analysesusing a rank order statistic to measure the various components of development. This resulted in a measurethat was highly correlated with the one used in our analyses below. Not surprisingly, using this rankorder-based measure yielded almost identical results to those we report here.

Another concern might be the weighting that the various measures are accorded. To check of thedegree to which this might be driving how states are classified, we repeated our classification algorithm,adopting the (non-agricultural) weights used by Shah (1970) in his assessment of infrastructure in India.This had no effect on our results.27

Similarly, our results hold if we use Indias premier business newsmagazineBusiness Today's recentranking of the 'Best States to Invest In' as our measure of development (see our section on robustnesschecks below).28

There are 25 states in India, although we list only 16; the rest were omitted because there are no firms inthe CMIEdata with plants located in these states. Moreover, there are actually two kinds of administrativeentities in India states and union territories. The fundamental difference between the two is that territoriesare centrally administered, while the states have a greater degree of political autonomy. Unfortunately, thedata available for territories is extremely incomplete; for this reason, we were unable to include plantlocated in union territories in our sample.


12/50

11

distribution of firms is very different for the group and non-group samples. It will

therefore be particularly important to allow for a lot of flexibility in the size-location

relationship, which we look at in Section 6 below. Note also that mean DEV is lower for

group-affiliates than for unaffiliated firms (with the difference in means significant at the

1% level), consistent with the hypothesis that groups should locate more in low-

infrastructure regions.

2. Why Locate in a Low-Infrastructure Region?

A. A (Very) Simple Illustrative Model

Consider a continuum of identical firms with measure 1 in an economy with two regions,

A and B, and two factors of production, I ('infrastructure') and L (labor). Firms decide

which region to locate in, and how much labor to hire. Infrastructure varies across the

regions, is given exogenously, and enters into production as: f(L,IX), where X = A or B.

All firms produce an identical product, whose price is defined by Q=D(p), and hire labor

from (immobile) labor pools according to LX=S(wX), where wX is the wage in each region.

Assume the f2, f12>0, and that IA>IB, i.e., region A is high-infrastructure. If firms areprice-takers in product and labor markets, each firm faces:

LwILpfMax XXL

= ),(

Since firms may locate in either region, we get the arbitrage condition:

BBBBAAAA LwILpfLwILpf = ),(),(

where wi is the market-clearing wage. Sincepf(L,IA)>pf(L,IB) for allL, we must have

wA>wB, i.e., in order for markets to clear, labor must be less expensive in the low-


13/50

12

infrastructure region to compensate companies locating there for the higher costs of

production resulting from low-infrastructure.

Now, imagine that there are some firms of type g (with production function g)

that are better able to deal with infrastructure shortages, i.e., g(L,IB)>f(L,IB);

g1(L,IB)>fl(L,IB)29, while g(L,IA)=f(L,IA). Then, in equilibrium, we must have one of the

following conditions hold:

)(),(),( iLwILpfLwILpf BBBBAAAA =

)(),(),( iiLwILpgLwILpg BBBBAAAA =

If a firm of type f is making the marginal location decision, then (i) will hold; otherwise,

(ii) holds. In the first case, all type g firms will locate in region B, with type f firms being

split between the two regions such that markets clear; type g firms earn strictly higher

profits in this scenario. In the latter case, type f firms all locate in region A, while type g

firms are distributed so as to make markets clear; sincef=g in region A, all firms earn the

same profits. Thus, in both cases, there is a higher ratio of type g firms in region B thanin region A, and these firms weakly earn higher profits than type f firms. In all cases, we

have wA


14/50

13

30% less, on average, than wage rates in more developed states) and tax rates (average of

8.2% in developed, vs. 7.0% in undeveloped states), as well as a higher average rate of

government 'fiscal benefits' (0.8% vs. 0.6%)30. All of these differences are significant at

10% or higher31

.

3. The Role of Groups in Less-Developed Regions

Unfortunately, it is difficult to infer precisely why group-affiliated firms are more likely

to locate in less-developed regions from the data alone. We therefore conducted a series

of interviews of Indian managers and executives from both group and non-group firms to

determine the factors that cause group firms to locate in less-developed regions. The

dominant reasons are described below. All but the last two are probably welfare-

enhancing reasons for why groups might locate in underdeveloped regions.

A. Direct Infrastructure Substitution

Because of the high degree of fixedness in the costs of the infrastructure investments

described in the introduction, it would be efficient to spread their expenses over a large

volume of assets32. Capital sharing across unaffiliated firms may be difficult because of

government regulations and restrictions, and because the absence of property rights

reduces the incentive to invest33. Agglomeration within a single firm is also a possibility,

but this may not always be desirable or practical.

29

Though g12is still positive30

CMIEdescribes fiscal benefits as benefits given by the government to companies operating in certainindustries and/or to promote specific objectives. Fertilizer companies receive subsidies under the retentionprice support system. Tea companies receive replanting subsidies. Similarly, exporting companies receivesubsidies under the duty-drawback scheme or the cash compensatory scheme or the international price re-imbusrsement scheme.31

The differences remain if we block on assets or industry.32

As a result, smaller firms may bear a greater burden where public infrastructure is lacking the WorldBank study described above found that, while larger Nigerian firms spent around 10 percent of theirmachinery and equipment budget on infrastructure expenditures, smaller firms were spending closer to 25%.In the case of Indonesia, smaller firms were spending up to 25 times more for private power than largercompanies.33

For example, Nigerian firms are prohibited from selling any excess power capacity.


15/50

14

An alternative solution to the problem of scale is the sharing of capital within

closely knit sets of companies affiliated with a single group. This sharing of resources is

observed in varying degrees throughout India, but is most striking at the self-contained

industrial 'cities' constructed by a number of India's largest groups. These facilities

include Jamshedpur (or Tatanagar, affiliated with the Tata Group), Pirojshanagar (Godrej

Group), Birlagram (Birla Group), and Modinagar (Modi Group). In each case, the group

provides such basic services as power generation, roads, schools, and employee housing

for the joint use of its co-located companies.

The locational decisions of the Godrej Group provide a compelling illustration of

the diversified groups provision of a partial solution to infrastructure shortages. The

Group was founded in 1897, primarily as a producer of locks and safes. The lack of

reliable tool suppliers in India at the turn of the century led Godrej to integrate backward

into the production of machine tools, fork lifts, castings, and finally steel. Meanwhile,

Godrej had diversified horizontally into such unrelated product lines as soaps, pesticides,

typewriters, and refrigerators. During its first fifty years, almost all Godrej-affiliated

companies had located their production facilities in Bombay. However, increasing labor

and property costs in the city, as well as the 'octroi' surtax on bringing raw materials into

Bombay, were steadily eroding the Group's profitability. In what was considered a bold

and risky decision, Mr. Pirojsha Godrej, the group's chairman, decided to move the bulk

of Godrej's operations to a new facility where these costs would be minimized. Thus, in

the early 1950s, the group set up a multi-purpose industrial garden township called

Pirojshanagar ('City of Pirojsha') at Vikroli, about fifty miles from the center of Bombay.

Cheaper land and labor came at a cost, however -- while the ever-expanding

suburbs of Bombay have now absorbed Vikroli, fifty years ago the region was mostly

jungle. Before construction of the new facility could even begin, new roads had to be built

into the area, and because of its isolation, Pirojshanagar was required to be self-sufficient.

All basic infrastructure services were provided for, and shared by, group-affiliated

companies. A captive power plant was built which satisfied the needs of the entire


16/50

15

facility (and did so far more reliably than power supplied from the government grid). In

contrast to the few virtually impassable roads that surrounded Pirojshanagar, those within

the compound were all paved and well maintained. A tool shop and furniture workshop

provided non-specialized physical inputs for all the factories; security, waste disposal,

stationary supplies, external capital goods purchasing, gardening, sewage treatment, dining

services, and communications were also jointly provided.

At the time, travelling from Bombay to Vikroli was an all-day affair, which meant

that workers had to be convinced to reside locally. Since no housing or social services

existed, these too had to be provided by Godrej. Apartment complexes were built to

accommodate lower-level employees, and a number of large and comfortable houses were

constructed to help convince the factories' managers to leave their homes in Bombay. A

major consideration for managers considering relocating to less-developed areas has

always been the quality of education for their children. To address these concerns,

Godrej set up a modern school that was (and still is) considered to be as good as those

found in Bombay. Other shared facilities included a hospital, a store selling basic

household goods, a family planning clinic, and even places of worship34.

Very few of these services were paid for directly by individual employees. Along

with the joint factory services provided by the facility, these expenses were documented

by a central accounting office, then allocated by a complicated but well-defined system to

individual factories35. Common ownership ensured that the guidelines were followed and

rarely disputed.

The primary motivation for colocation was notvertical integration. From the

beginning, Pirojshanagar supported a diversity of manufacturing activities, with few 'intra-

township' supply relationships. The advantages of colocation derived from the sharing of

general purpose services that would be externally supplied in a more developed region.

34

In recognition of the diversity of beliefs among his employees, Godrej ordered that a mosque, church,and Hindu temple be built at Pirojshanagar.35

A few small non-Godrej companies have recently been given leases at Pirojshanagar to utilize whatwould otherwise be vacant facilities. They are required to share the costs of shared facilities according tothe same formula as Godrej companies. However, they may access to only a small proportion of the joint


17/50

16

The management at Pirojshanagar today recognizes that the overall project was feasible

only because these costs could be spread over a large volume of industrial activity.

B. Human Resource Considerations

The lower quality of schools and healthcare, as well as the lack of cultural amenities in

less-developed areas, makes it a serious challenge to recruit competent managers and

engineers to work in these regions. Skilled workers are often much sought after in the

labor market, and have ample opportunities available to remain in more cosmopolitan

areas. Because of the shared infrastructure facilities listed above, groups are often able to

make life in backward areas more palatable for skilled employees. Another advantage was

described by a manager from the Tata Group, who explained that groups often recruit

young managers to spend a few years at a facility in a less-developed region, with the

promise that a job would later be available with the Group in a major center36. Also,

since many groups have large pools of shared labor resources, they are able to rotate

skilled workers through facilities in less developed areas on shorter-term assignments.

C. Supply of Inputs

Since the concentration of industries is relatively sparse in less-developed areas, there is

usually insufficient demand to stimulate the development of supporting industries. Basic

inputs and repair services must therefore be sourced from relatively distant suppliers. A

potential producer with only a local operation may have trouble establishing and

coordinating these relationships, particularly given the poor quality of communications in

less-developed areas. Group firms, on the other hand, already have well-established

supplier networks, and are able to coordinate the delivery of materials through afacilities available to group-affiliated companies.36

In a world of perfect labor markets, this would not be an important consideration -- an unaffiliatedcompany in a less-developed region could recruit skilled workers who would be able to simply find a newjob in a more developed area after a few years. However, given the imperfect information flows that exist inreality, finding a job in Bombay if one is working in rural Orissa would be a tremendous challenge. Thereare also signaling considerations -- if a Tata employee in Bihar wishes to look for a job in another state, hehas the advantage of the Tata reputation, which is transferable across regions. The same cannot be said of a


18/50

17

centralized bureaucracy. Often, the supplies will come from a company which is itself a

group-affiliate. For example, when the Godrej Group opened a new factory in Goa, it had

materials shipped to it from the Group's principal facility near Bombay for the first few

years of its operation.

D. Financing

One particularly important input that may be difficult to access from less-developed

regions is bank financing. India's financial institutions are concentrated in the country's

major urban centers, particularly Bombay; because of deficient contract enforcement,

reputations established via word-of-mouth contact are very important in banking

relationships. The establishment and maintenance of the necessary reputation requires a

strong presence in a major center, which makes it particularly difficult for independent

firms in less-developed areas to obtain financing. In contrast, a group is likely to have a

head office in a major city that may handle financing on behalf of its rural subsidiaries.

Also, many groups reportedly utilize internal capital markets, even though there

are some legal restrictions on borrowing and lending among the publicly traded group

companies. This may be yet another avenue for financing that is not available to

unaffiliated firms.

E. Risk and Diversification

Locating a production facility in a region without supporting services is more risky,

simply because there are more things that can go seriously wrong. Groups are more

capable of bearing this risk because of their greater size and diversification.

F. Land-Intensive Projects Groups and the Government Bureaucracy

While many aspects of the Indian economy have been liberalized since 1975, land

acquisition and land use are still highly regulated. For smaller plots of land, there are few

local firm's reputation.


19/50

18

bureaucractic barriers to be overcome. However, if a project requires a lot of land, it must

be rezoned for industrial use. This process may take years to work its way through the

bureaucracy, resulting in potentially costly project delays. There exist considerable

economies of scale in dealing with the Indian bureaucracy many large groups reportedly

maintain 'industrial embassies' in New Delhi whose sole purpose is to handle group

relations with the government (Encarnation (1989)). Large groups are thus often able to

expedite the rezoning process due to superior political contacts.

Now, it is usually preferable to locate projects with significant land requirements

in less-developed regions, where property costs are lower37. Hence, land-intensive

projects, for which group-affiliated firms are at an advantage, will be more common in

less-developed (i.e., low property cost) regions.

G. Accounting Profits

Firms locating in backward areas receive, among other benefits, a five year tax holiday.

Since these areas are overwhelmingly concentrated in less developed states38, firms in

these states will, on average, pay lower taxes on their profits (see section 3 for aconfirmation of this fact). As an accountant with a large Indian group pointed out, this

provides an incentive for firms to locate their most profitable operations in less-

developed states. This effect will be particularly strong for vertically integrated groups

for the following reason: a group may build an upstream plant in a backward area, and

sell its output to a group affiliate that does not enjoy the same tax advantages at an

inflated transfer price. This will artificially increase the profits of the untaxed operation

at the expense of its customer, thereby lowering the total taxes paid by the group as a

whole.

37

Unfortunately, we have not been able to obtain any data to confirm these assertions; this section is basedpurely on interviews with Indian executives, and discussions with D. Kapur.38

For firms located in states with DEV0.5


20/50

19

4. Results

A. The Basic Result: Differences in Locational Choice

If it is true that group affiliated firms are better able to minimize the costs of locating in

less-developed areas we would expect that, all else equal, these firms would be more

likely to locate their plants in states with lower values of DEV. The basic summary

statistics suggest that this theory is promising: the mean value of DEV for group firms is

0.531, while it is .582 for non-group firms 39,40. Figure 1 shows the relationship between

a state's development score, and the percent of plants in that state that are group-

affiliated -- the pattern is quite striking.

This relationship may derive from systematic differences in firm characteristics,

however. To account for possible systematic differences in industries chosen by group

versus non-group firms, we look at the difference in industry-weighted means, thus

comparing only firms within the same industry. Our weighted estimate of this difference:

is given by:

( )( ) ( )( )][ ,,*#

#IinFirmsNonGroupDEVMean

I

IinFirmsprouGDEVMean

FirmsofTotal

IIndustryinFirmsof

The resulting difference is 0.051, almost identical to the unadjusted estimate (see Table

6)41.

39

Statistically significant at 1%.40

While locational choices are observed that the plant level, these decisions are probably made at the firmlevel. There is therefore likely to be some correlation among a firms locational choices. To deal withmulti-plant firms, plant location observations are weighted by 1/n, where n is the number of plants ownedby the firm. This effectively assumes a high degree of correlation among the firms decisions, collapsing nobservations into one; this weighting is used for all non-regression calculations below. An alternativeapproach would be to assume complete independence among locational choices for each firm. Thiscorresponds to using the unweighted data in the calculations; this approach yielded almost identical results.41

Another approach to the problem is to look at regressions of state-level aggregates. We take the rate ofgroup affiliation as the dependent variable (AVMEMB), and use similarly averaged state-level variables asregressors. This allows us to look at the individual effects of different components of development. Thecoefficients are mostly negative, as expected, suggesting that group affiliation is greater in underdevelopedstates. However, given that we have so few observations and so many (highly correlated) regressors, it isprobably not appropriate to attach much meaning to individual coefficients.


21/50

20

Our basic regression model is given by

DEVi= + *MEMBi + *Xi + ij + (1)

whereXi is a vector of covariates, i is a firm index,j indexes the firm's plants, and is a

vector of industry dummies. DEVij captures the level of development of the state in

which plant j belonging to firm i is located42. We allow the error structure to be correlated

across plant-level observations within the same firm (i.e., cov(ij,ik)0)43. The results

from this model are shown in Table 7; we add SALES and AGE as controls in

specifications (2) and (3)44. To allow the coefficients on these controls to differ between

group and non-group firms, we allow for interaction effects in specification (4). We

repeated these analyses using log(SALES) in specifications (5)-(7). In all specifications,

the coefficient on MEMB is significant at the 10% level or better, and takes on values

between -0.038 and -0.054.

It is interesting to note that the coefficient on the MEMB*AGE interaction termis positive and (often) significant, implying that our reported effect is stronger for newer

firms. We interpret this observation as resulting from the steady liberalization in

industrial policy that has taken place in India since 1975 as firms face fewer constraints

in locational choice, location decisions should increasingly reflect direct profit maximizing

behavior, rather than political objectives. If the group effect we observe reflects a cost

advantage enjoyed by firms in less-developed regions, then this effect should be increasing

over time. Hence, the positive coefficient on MEMB*AGE.

42

Note thatj differs across firms, since each firm has a different number of plants.43

We also performed these analyses, allowing for correlation among observations within the same group;this increased our standard errors marginally, but did not substantively affect the results that follow.44

See the next paragraph for an explanation of why we believe the age control to be important.


22/50

21

B. Size Effects

So far, all we have shown is that there is evidence of a group effect of some kind. We

now turn to evidence of a size effect, by adding total group sales (TSA) to the basic

regression given in (1)45

. If the advantages of group affiliation are increasing with size

then, all else equal, a firm associated with a large group should be more likely to locate in a

less-developed area than a firm associated with a smaller group. This would imply a

negative coefficient on TSA. Moreover, since there may be threshold effects in the cost

advantages described in Section 3, we should observe decreasing returns to group size. To

capture this effect, we ran a specification using log(1+TSA)46. Our results (see Table 9)

weakly support the existence of a group size effect in the linear model. When we allow

for decreasing returns by using a log-linear specification, the coefficient on log(1+TSA) is

negative and significant.

C. Group Leadership Effects

The literature on groups has previously noted the leading role that one particular firm

within the group usually takes in coordinating the group's activities. In the case of the

Japanese keiretsu, the group's primary lending institution coordinates the distribution of

financial capital, and also provides an external source of governance for group-affiliates

(see Hoshi et al, 1991). Recent work on foreign investment in Indian firms (Fisman and

Khanna, 1998) has suggested that in India, group financing is also coordinated by a

particular firm within the group, generally the largest. We conjecture that this 'flagship

firm', in the face of the more extensive market failures present in the Indian economy,

plays a much broader coordinating role. In particular, we expect that this firm will be

active in coordinating the labor market, capital market, and supply relationships of other

group-affiliates. As outlined in Section xx, it is much easier to coordinate these types of

45

We set TSA equal to zero for all non-group firms.46

Since log(1+TSA) and MEMB are highly correlation (=0.95), we include regressions showing the effectof log(1+TSA), with the sample limited to group-affiliated firms. Since there might also be issues ofidentification for the linear TSA specification as well, we also include the results of a groups-onlyregression for the linear specification.


23/50

22

activities from more developed areas, where there is likely to be a relatively high

concentration of suppliers, banking institutions, and skilled workers. We posit, therefore,

that this 'flagship firm' is more likely to be located in a more developed region47.

We define a dummy variable LEAD, which takes on a value of one if the firm is

the largest in its group48. If such firms are coordinating group activities, we would expect

that they would have to be situated in more developed regions. When LEAD is added to

our 'standard' regressions, we find that its coefficient is indeed positive (see Table 8); the

inclusion of this leadership dummy also substantially increases the size of the

membership effect. This is as expected: many of the advantages that group-affiliates

enjoy in less developed regions relate to the coordinating roles played by other group

entities in more developed regions. Adding LEAD to the regressions nets out this

leadership effect from MEMB.

D. Interpreting the Results

The problem with the model as outlined above is that it does not lend itself to

straightforward interpretation. In particular, it is difficult to assess the economic

significance of the coefficients in the model. A more natural way to think about the

location decision is as a discrete choice problem, with location (i.e., state) as the choice

variable, and firm and state characteristics as the explanatory variables (see McFadden,

1974). In this setup, the probability that a firm makes a particular locational choice is

inferred based on a vector of firm characteristics. Unfortunately, models of this type tend

to fall apart when the number of choices is much greater than 3 or 449. Certainly our

model, with 16 potential locations, far exceeds this threshold. What is usually done in

such cases is to collapse the choice variable into one with fewer options. As our

condensed measure, we convert DEV into a dichotomous variable, DDEV, with DDEV=0

47

The presence of a central coordinating entity of this type is, in fact, implicit in the discussion in sectionxx that outlines the advantages that group-affiliates enjoy in less-developed regions.48

Note that, for many groups, LEAD=0 for all firms in our sample, since the largest (coordinating) firmhad an incorporation date prior to 1975, and was therefore dropped from our sample.49

Personal communication, Aviv Nevo, Department of Economics, University of California, Berkeley.


24/50

23

if DEV0.5(undeveloped states). This

allows us to work with a relatively simple binary choice model50.

All of our previous analyses were repeated, using DDEV as the dependent

variable; the analyses were also done using logit and probit binary choice models. Table

10 shows the results of the first set of specifications for our linear regressions, with

DDEV as the dependent variable51. The signs and significance levels of our coefficients

are similar to those we obtained with our continuous infrastructure variable, but the

interpretation of the coefficients is more straightforward. In particular, the coefficient on

MEMB reflects the difference in probability that a group firm will locate in a high-

infrastructure region, relative to a non-group firm. This coefficient ranges from -0.077 to

-0.108. The average probability of locating in a low-infrastructure area is about 40%,

implying that group affiliated firms are 20-25% more likely to locate in these areas than

non-group firms. The results from the probit and logit models were similar but even

stronger, with effects for a mean firm ranging from -0.115 to -0.130.

It is worth noting that there are many other aspects to a firms plant location

decision. The Indian executives we interviewed listed proximity to market, proximity to

raw materials, state investment incentives, quality of infrastructure, quality of workforce

and labor relations as the primary considerations (generally in that order). Given that the

tradeoffs that we discussed in the previous section are but one part of the overall

decision-making process, we consider the observed effect to be extremely large.

Our calculations for total group assets were similarly repeated using DDEV as the

dependent variable (see Table 11). The results were qualitatively similar to those

obtained previously the point estimates on log(1+TSA) imply that a firm associated

50 It seems somewhat arbitrary to choose 0.5 as the high infrastructure cutoff point -- a state with a score of0.47 is presumably much closer to a state with a score of 0.52 than to one with a score of 0.15. In anattempt to mitigate this problem, we repeated all of our analyses, discarding all observations for firmslocated in states with scores between 0.33 and 0.67, thereby limiting our sample to truly high and lowinfrastructure states. As expected, this yields similar (though slightly stronger) results. This effect is notsensitive to the choice of cutoff values -- for example, {0.3,0.7} and {0.4,0.6} both yield very similarresults (though admittedly affects the classifications of only a few observations).51

We show the results from the linear regression, rather than either of the discrete choice models, for ease ofexposition. Since our observations are not near the boundary values of zero or one, and given our relatively


25/50

24

with an average group (log(1+TSA)=5.7) is eight percent more likely to locate in an

undeveloped region than a non-group firm with similar characteristics. Among group-

affiliates, the results indicate that a firm affiliated with a group in the ninetieth percentile

in terms of total group assets (log(1+TSA)=7.6) is six percent more likely to locate in an

underdeveloped state than a group in the tenth percentile (log(1+TSA)=4.0). We note,

however, that there is a substantial decline in the significance of these results, which we

attribute to the loss of information resulting from the transformation to a dichotomous

variable52. Again, the results from the binary logit and probit models were very similar to

those from the linear model.

D. Using Performance as a Dependent Variable

The above models assume a high degree of economic rationality in the location decisions

of firms. However, these decisions likely involve a lot of unobservable, non-economic,

considerations. For example, many groups are family, rather than professionally,

managed; in such cases, close proximity to familial homes tends to dominate locational

choices in such cases. Moreover, issues of endogeneity arise with any firm-level

characteristics that we might use as regressors. Using locational choice as the dependent

variable may thus introduce a lot of unnecessary noise and potential biases. While any

observational study encounters such problems, they would be reduced here by looking at

a purer economic outcome as the dependent variable. As our model in section 2 (as well

as common sense) suggests, if group-affiliates are better able to minimize the costs of

locating in less-developed regions, then they should be more profitable in these regions

relative to non-group firms.

It may also be the case that, while groups facilitate development by locating in

underdeveloped regions, they do not do so efficiently; for example, it may be that, in the

extreme, the company towns described in the earlier section (Pirojshanagar, Tatanagar,

large sample size, the linear model is a very close approximation to the true models.52

We do not report here our results for a linear TSA model, since it was not supported by the previousmodels. This specification performs even more poorly with DDEV as the dependent variable.


26/50

25

Birlagram etc.) are not manifestations of any particular efficiency, but rather monuments

to the family that constructed them. By looking at economic outcomes, we seek to verify,

at a minimum, that groups are at least no worse than non-groups in bringing development

to underdeveloped regions.

We use an approximation of Tobins Q, as well as accounting profits (ROA) as

our measures of performance53. There are disadvantages associated with each measure of

profitability, which is why we report both sets of results. By using ROA, we are relying

on accounting data, which may not be reliable and are certainly very noisy. There are also

difficulties with using Tobins Q: due to data limitations, we may only use a coarse

approximation of the true value of Q; also, the lack of efficient equity markets adds a lot

of noise to any measure based on stock price.

In spite of these problems, the raw data seem to bear out our hypothesis -- mean

ROA is higher for firms in undeveloped regions (0.91 vs. 0.087 in developed regions), and

mean Q differs only slightly (1.56 in developed vs. 1.57 in undeveloped). However, for

non-group firms, both mean ROA and Q are higher in developed states (0.096 vs. 0.093

for ROA; 1.48 vs. 1.35 for Q). This effect remains when these profit differentials are

taken as weighted averages by industry54.

The regressions in Table 12 also support the hypothesis that group-affiliated

firms are better able to profitably locate in low-infrastructure regions than non-group

firms. The coefficient on the MEMB*DEV interaction term is negative in both the ROA

and Q regressions55, as is the MEMB*DDEV coefficient in the discrete variable

regressions. In the ROA regression, the coefficient on MEMB*DDEV equals -.011,

53 We approximate Q byAssetsTotalofValueBk

DebtofValueBkStockprefofValueBkEquityofValueMkt ++, where market value

is based on the closing price on the last trading day of 1993. Data availability considerations preclude usfrom computing a better approximation to Q (See Lindenberg and Ross, 1981). The relevant measure ofreturns is operating income (see Khanna and Palepu, 1997). We therefore use tax-adjusted ROA, defined asROA = {Net Profit + (1 Tax Rate)*Interest}/Assets. The results are similar (though somewhat stronger)if we use non-adjusted ROA (i.e., (Net Profit)/Assets).54

Admittedly, these differences are not significant at conventional levels.55

We also ran a number of other specifications, including log(SALES) as well as interaction terms; theresults were the same for these models, which are not included in the interests of conserving space.


27/50

26

while the mean of ROA equals approximately 0.095, implying that a shift to locating in a

low infrastructure region will reduce the ROA of a non-group firm by about 12% relative

to the effect of such a shift on a group-affiliated firm. The comparable value when Q is

the dependent variable is also 12%. While our results are not significant at conventional

levels, we interpret these as indicating that groups are no less efficient than non-groups in

bringing industrial activity to underdeveloped regions.

5. Robustness Checks

A. Non-Linearities in the Sales-Location Relationship

In theory, our fixed-effects regressions control for size effects by including SALES as a

regressor. Note, however, that the distribution of SALES is substantially different for

group versus non-group firms (see Table 4). Our specifications are reasonable as long as

the relationship between SALES and DEV does not deviate too drastically from the linear

and/or log linear specifications that we use. If this is not so, however, our results may be

driven simply by our assumptions of functional form. To examine this further, we allow

more flexibility in our specification, letting the slope on SALES change every 20 units,

i.e., we use the following:

i

m

iMSALESMiii SALESIMEMBXDEV i ++++= =

?

5

1

*20 *

where

?

0, while 3 < 0, though neither coefficient is statistically significant.

There is therefore weak support for the hypothesis that group firms earn higher subsidies

56

Note that theBusiness Today score is not ideal for our purposes, since it takes into account such factorsas financial investment incentives that are relevant for a company's location decision, but do not reflect thestate's level of economic development.


29/50

28

in underdeveloped regions.

If this were driving our earlier results, however, then the subset of firms receiving

no fiscal benefits should not display our previous group effect. Of our sample of 1193,

there are 848 firms that did not receive any fiscal benefits. When our previous analyses

were performed on this subsample of firms, the results were virtually unchanged, in terms

of both magnitudes and statistical significance. Therefore, while there may be a group

effect in obtaining fiscal benefits, it is unlikely to be driving our results here.

Another proxy for the degree of bureaucratic corruption comes from the results of

a 1997 survey on corruption conducted byIndia Today. In this survey, respondents

were asked to rate the three most corrupt states in India in order. They were then asked

to do likewise for the three least corrupt states. Based on the survey results, each state

was assigned a scalar corruption score (CORR). This measure was only moderately

correlated with DEV (=-0.32). Moreover, CORR and MEMB were almost completely

(though slightly negatively) correlated. There was similarly almost no relationship

between CORR and MEMB in multivariate analyses. Thus, the bureaucratic corruption

hypothesis does not seem to explain our observed pattern in locational choice.

D. Intrastate Heterogeneity

Many of Indias states are vast, both in size and population, and there is good reason to

believe that there would be considerable variation in development within states. While it

is not clear that this should be the source of any systematic bias, replicating our findings

with some district level data would certainly bolster our confidence in the results. While

there are obvious benefits from moving to more disaggregated data, there is a cost as well.

Our only district-level data come from the Census of India, 1981, which provides muchweaker infrastructure proxies than are available at the state level.

We obtained district-level data for Gujarat, Karnataka, Maharashtra, Tamil Nadu,

57

Using a log(SALES) in place of SALES did not affect the results.


30/50

29

and West Bengal58. As our measure of telecommunications development, we use the

percent of villages with access to telephones. We similarly define measures for other

development amenities by looking at the percent of villages with the following:

availability of power (Power); access to the village via permanent road (Transportation);

access to primary schools and medical services (Social Services). We aggregate these

items in the same manner as with the state-level data, and ran our analyses with this new

district-level development measure. Our results were very similar in terms of the

magnitude of the implied effect, though admittedly somewhat noisier (p-values ranging

from 0.01 to 0.15). We attribute this to the extremely coarse nature of our district-level

development proxies.

E. Firm Diffusion

Suppose that the first few plants of any organization are more likely to be located in a

more developed region due to the proximity of product markets and input suppliers.

Suppose further that the construction of future plants diffuses outward to take advantage

of new opportunities in other areas. Under this process of plant location, a larger

organization will have a larger proportion of plants in less-developed regions simply

because of this diffusion process, rather than any productive advantages59. However, this

story yields a separate prediction: the plants that an organization builds early on should

be more likely to be located in a more developed region than those that it builds later. To

test for this possibility, we repeated the basic analyses of this paper, adding a counter

designating the order in which a groups firms were incorporated. So, the groups first

firm was assigned a value of one, the second firm to be incorporated was assigned a two,

and so on. When this variable was added to our regressions, its coefficient was extremely

close to zero. Hence, we do not find evidence to support the firm diffusion hypothesis.

58

These states were chosen because they had the largest number of plant observations; we did not use UttarPradesh simply because it is divided into so many districts that entering data on all of its districts seemedimpractical.59

Note that the basic argument made in this paper is consistent with (though does not necessarily imply)the presence of a diffusion effect. That is, group-affiliates are able to diffuse outward because of the


31/50

30

6. Discussion

In this paper, we have shown that firms associated with business groups are more

likely to locate in less-developed regions. We interpret this as evidence that business

groups are locating in these areas because they are better able to cope with the shortages

associated with less-developed regions.

This result has important implications for evaluating both the development process

and the value of groups. If we are to fully evaluate the costs of underdevelopment, we

need to first understand how firms are dealing with current shortages, and how they might

react to changes in the provision of basic services. Until now, there has been very little

work done on the behavior of firms (and industrial organization in general) in developing

countries.

Private versus Public Sector Provision of Infrastructure

Of course, an alternative mechanism for stimulating development in underdeveloped

regions is for government-owned firms to deliberately locate their plants in less-developed

areas. When we revisited the sample of government-owned companies that we discardedfor our main analyses (a total of 235 plants, 217 firms), we found a mean level of DEV for

these firms of 0.47, considerably lower than the average for any class of private firms.

This basic finding held after we controlled for industry and size effects.

This begs the question of who is more effective at bringing industry to

underdeveloped areas: government firms or group firms. If we take profitability as a

measure of performance, then the basic statistics suggest that government firms are not

doing a very good job of anything at all -- operating ROA is only 0.4%, while net ROA is

well below zero. While profits may not be a fair benchmark for analyzing the

performance of government-run entities, which probably have other primary objectives,

Indian government run firms are notorious enough for their operating inefficiency (Kelkar

advantages they have in producing in less-developed regions.


32/50

31

and Rao (1996)) that it would be a stretch to imagine that they could facilitate

development more efficiently than private sector group-affiliates.

The Effect of Group Plant Location on Development

Godrej's experiences at Pirojshanagar highlight another important role of groups in the

development process. Soon after the township was built, small-scale industries sprang

up in the surrounding areas to serve the needs of the factories and their employees.

Eventually, basic infrastructure was built to serve the needs of this new population.

Thus, with basic services and ancillary industries present, other enterprises were able to

locate in the area. Therefore, Godrej's original decision to locate at Vikroli subsequently

spurred development in the surrounding area60.

The Tata Group's steel facility at Jamshedpur (sometimes called Tatanagar) is

another particularly striking illustration of this phenomenon. Jamshedpur began as a self-

contained facility in the jungles of Bihar, but has grown into a major city of nearly 700

thousand inhabitants.

Discussions with Indian managers suggest that this is part of a more general

phenomenon being the first to locate in a less-developed region is difficult because of

the absence of infrastructure and ancillary industries61. However, an enterprise of

sufficient scale will spur the development of these services, which in turn makes the area

more attractive as a site for other industries. Our discussions suggest that group firms

often play this leading role. Unfortunately, our current data do not allow us to test this

hypothesis directly, but we hope that it will be a topic for future research.

Recognition of the role of business groups in facilitating development is important60

Ironically, this has resulted in an escalation in the costs that Godrej had hoped to avoid by buildingPirojshanagar. The octroi surtax now applies to raw materials coming into Vikroli, while property pricesand labor costs have skyrocketed. Because of its investments in infrastructure, as well as Indian regulationsprohibiting the dismissal of employees, Godrej has faced many obstacles in moving once again to a low-cost region.61

Today, larger enterprises are usually able to convince local governments to install the necessaryinfrastructure as a condition for locating in a particular area. The absence of ancillary industries remains a


33/50

32

in evaluating the effects of groups, particularly given that they are so often portrayed as

rent-seeking monopolists. We are certainly not claiming that this is the reason for the

existence of business groups certainly, there are many (see, for example, Leff (1976),

Amsden & Hikino (1994), Granovetter (1994), Ghemawat and Khanna (1997)).

However, given the presence of groups in developing countries, understanding their role

in the development process is important in evaluating the implications of their existence.

significant problem, however.


34/50

33

Bibliography

Ahuja, Gautam, and Majumdar, Sumit (1995): An Assessment of the Performance of

Indian State-Owned Enterprises, University of Michigan, School of Business

Administration, Working Paper #9550-10 Revision 1

Amsden , Alice (1986):Asias Next Giant: South Korea and Late Industrialization, New

York: Oxford University Press.

Amsden, Alice and Hikino, Takashi (1994): Project Execution Capability, Organizational

Know-How and Conglomerate Corporate Growth in Late Industrialization,Industrial

and Corporate Change, pp.111-148

Amsden, Alice (1996): Korea: The Dynamics of Business-Government Relations,

forthcoming in Alfred Chandler, Jr., Franco Amatori and Takashi Hikino (eds.)Big

Business and the Wealth of Nations, Cambridge: Cambridge University Press.

Blasi, Joseph, Kroumova, Maya, & Kruse, Douglas, 1997, Kremlin Capitalism:

Privatizing the Russian Economy, Ithaca: Cornell University Press.

Business India, Park Spark, March 24-April 6, 1997

BusinessWorld, Cracking Under the Strain, 7-20 February 1996.

BusinessWorld, India Incs Win-Win Wish List, 15-28 May 1996.

Camp, Roderic A. (1989):Entrepreneurs and Politics in Twentieth-Century Mexico, New

York: Oxford University Press.

Caves, Richard, and Uekusa, M. (1976):Industrial Organization in Japan, Washington,

D.C.: The Brookings Institution.

Chang, Sea Jin, and Choi, Unghwan (1988): Strategy, Structure, and Performance of

Korean Business Groups: A Transactions Cost Approach, Journal of Industrial

Economics, December

Daems, Herman (1977): The Holding Company and Corporate Control, Nijenrode Studies

in Economics, Volume 3, The Netherlands.

Economist Intelligence Service (1995):India's Energy Sector, Bombay: Centre for

Monitoring Indian Economy.

Economist Intelligence Service (1996):Infrastructure in India, v. 2, Bombay: Centre for


35/50

34

Monitoring Indian Economy.

Ellison, Glenn, and Glaeser, Edward (1994): Geographic Concentration in U.S.

Manufacturing Industries: A Dartboard Approach, Harvard Institute of Economic

Research Working Paper, no. 1691.

Encaoua, David and Jacquemin, Alexis (1982): Organizational Efficiency and Monopoly

Power: The Case of French Industrial Groups,European Economic Review, 19:25-51

Ghemawat, Pankaj, and Khanna, Tarun (1996): The Nature of Diversified Business

Groups: A Research Design and two Case Studies, Harvard Business School Working

Paper, no. 97-043, forthcomingJournal of Industrial Economics, Dec. 1997

Goto, Akira (1982): Business Groups in a Market Economy, European Economic

Review, 19:53-70.

Granovetter, Mark (1994): Business Groups Chapter 18 inHandbook of Economic

Sociology, Princeton University Press

Herdeck, Margaret and Piramal, Gita (1985):Indias Industrialists, Volume One

Washington, D.C.: Three Continents Press.

Hoshi, Takeo, Kashyap, Anil, and Scharfstein, David (1991): Corporate Structure,

Liquidity, and Investment: Evidence from Japanese Industrial Groups, Quarterly

Journal of Economics, p. 33-60.

Jacquemin, Alexis, and de Ghellinck, Elisabeth (1980): Familial Control, Size and

Performance in the largest French Firms,European Economic Review, 13, 81-91.

Joshi, Vijay and Little, I.M.D. (1997): Indias Economic Reforms: 1991-2001, Oxford

University Press, Delhi

Jimenez, Emmanuel (1995): Human and Physical Infrastructure: Investment and Pricing

Policies in Developing Countries, inHandbook of Development Economics, v. IIIB,

ed. J. Behrman and T. Srinivasan, Amsterdam: Elsevier Science B.V.

Karanjia, B.K. (1997): Godrej, A Hundred Years: 1897-1997, New Delhi:Penguin Bks

Kelkar, Vijay and Rao, V.V.B. (1996):India: Development Policy Imperatives, Tata

McGraw-Hill Publishing Company Limited, New Delhi

Khanna, Tarun, and Palepu, Krishna (1997): Corporate Scope and Institutional Context:

An Empirical Analysis of Diversified Indian Business Groups, Harvard Business


36/50

35

School Working Paper #96-051.

Khanna, Tarun, Yoshino, Michael, and Melito, Danielle (1996): Sime Darby 1995,

Harvard Business School Case Study #9-797-017

Knoop, C. and Yoshino, M. (1995): First Pacific Company Limited: From Letters ofCredit to Personal Communications Networks, Harvard Business School Case 9-396-

139

Kunio, Yoshihara (1988): The Rise of Ersatz Capitalism in South-East Asia , Oxford

University Press: Singapore.

Leff, Nathaniel (1976): Capital Markets in the Less Developed Countries: The Group

Principle, in Ronald McKinnon (ed.),Money and Finance in Economic Growth and

DevelopmentNew York: Marcel Dekker.

Ling, Sieh Lee Mei (1992): The Transformation of Malaysian Business Groups, in Ruth

McVey (ed.) Southeast Asian Capitalists, Southeast Asia Program, Cornell University,

NY

Majluf, N., Abarca, N., Rodriguez, D., and Fuentes, L., 1996, The Ownership Structure

of Economic Groups in Chile, Pontificia Universidad Catolica de Chile, mimeograph

McFadden, D. (1974): Conditional Logit Analysis of Qualitative Choice Behavior, in

Frontiers in Econometrics, ed. P. Zarembka, New York: Academic Press.

McVey, Ruth (1992):(ed.) Southeast Asian Capitalists, Southeast Asia Program, Cornell

University, NY.

Piramal, Gita (1996):Business Maharajahs (Viking Press, Bombay).

Robison, Richard (1986):Indonesia: The Rise of Capital, Sydney: Allen and Unwin.

Sato, Yuri (1993): The Salim Group of Indonesia: The Development and Behavior of the

Largest Conglomerate in Southeast Asia, 31(4): 408-41.

Shah, Narottam (1970): Infrastructure for the Indian Economy, inInfrastructure for the

Indian Economy, ed. V. Dagli, Bombay: Vora & Co.

Statistical Abstract, India (1992), India Central Statistical Organization (1992), New

Delhi: Manager Of Publications.

Strachan, Harry (1976): Family and Other Business Groups in Economic Development:


37/50

36

The Case of Nicaragua, New York: Praeger.

Suehiro, Akira, (1992): Capitalist Development in Postwar Thailand: Commercial

Bankers, Industrial Elite, and Agribusiness Groups, in Ruth McVey (ed.) Southeast

Asian Capitalists, Southeast Asia Program, Cornell University, NY.

Wang, N. T., ed., (1992): Taiwans Enterprises in Global Perspective, Armonk, N.Y.:

M.E. Sharpe.

Weinstein, David and Yafeh, Yishay, (1995): Japans Corporate Groups: Collusive or

Competitive? An Empirical Investigation ofKeiretsu Behavior, Journal of Industrial

Economics, Vol. XLIII, No. 4, December, 359-376.

White, Lawrence J. (1974):Industrial Concentration and Economic Power in Pakistan,

Princeton: .Princeton University Press

World Bank (1994): World Development Report 1994: Infrastructure for Development,

New York: Oxford University Press.

Zeitlin, Maurice, Ewen, Lynda Ann, & Ratcliff, Richard Earl, (1974): New Princes "

for Old? The Large Corporation and the Capitalist Class in Chile,American Journal

of Sociology, 80, 1, Jul, 87-123.


38/50

37

Appendix A: Evolution of Indian Industrial Policy

The history of Indian industrial regulation is sufficiently convoluted and bizarre that it

would be impossible to capture its many nuances and subtleties in a short description.We try here to give only a brief overview to illustrate the point that firms were given very

little discretion in plant location decisions during 1950 - 1975, and that beginning in 1975,

there has been a fairly steady process of liberalization which continues to the present.

There are several sources of information from which this section draws, including Kelkar

and Rao (1996) and Joshi and Little (1997).

Industries Act, 1951

In 1948, the Industrial Policy Resolution was drafted as a programme of planned

development in order to secure the most effective use of scarce resources, both domestic

and external, for promoting balanced economic growth (xxx, xxx). This resolution was

given legislative backing with the Industries (Development and Regulation) Act of 1951,

which essentially gave the government complete control over Indias industrial output.

The Act created a licensing system under which government permission was required for

any new enterprise, as well as any increase in the output of an existing enterprise62. Since

imports were also severely limited, any firm that was able to procure a license enjoyed a

considerable degree of market power. This lack of competition meant that a company

could invest heavily in political rent-seeking, run a relatively inefficient operation, and still

be highly profitable. One component of the bargain struck between politicians and

prospective enterprises was usually that the location of any new production facilities be

chosen at the discretion of the license-granting government official. Since this type of

rent-seeking was less overt (and therefore easier to engineer) than outright bribery, it was

almost always a part of the 'informal' licensing agreement. Given Indias democratic

political structure, the incentives for license-granting officials are obvious: enterprises

were often required to locate in the officials home states, or in the states of other officials

to whom the decision maker owed political favors. Thus, the home states of top

ministers received what was perceived to be a disproportionately large volume of new

investment throughout the period 1951-1975.

1975 Notifications

Beginning in 1973, the Indian government began to recognize that growth in certain

industries was necessary in order to provide for a steadily growing national population(and economy). In 1975, the government announced that an annual capacity increase of

5% would automatically be allowed in 15 industries that were deemed to be important

from the point of view of the national economy or are engaged in the production of

articles of mass consumption.

62

The government was actually given direct control over far more than just total output. For example, thedegree of product differentiation was, strangely enough, dictated by government bureaucrats.


39/50

38

1977 Elections

From independence until the mid-1970s, the Congress party controlled both the federal

and state governments, and did not face any serious threats to its rule. This stability

facilitated the kinds of political deal making that led to the politicization of plant location

decisions outlined above. The situation changed drastically when the J

facilitating development - role of bgs

Documents