Contractual Imperfections, Credit Markets and Vertical Integration: Theory and Cross-Country Evidence

1 Contractual Imperfections, Credit Markets and Vertical Integration: Theory and Cross-Country Evidence Rocco Macchiavello First Version: April 2005 T...

Author:  Charity York
0 downloads 0 Views 782KB Size

Contractual Imperfections, Credit Markets and Vertical Integration: Theory and Cross-Country Evidence Rocco Macchiavello First Version: April 2005 This Verion: November 2005 JOB MARKET PAPER

Abstract Anecdotal evidence, as well as theoretical considerations, suggest the possibility of important cross-country di¤erences in the organization of production in general, and in the degree of vertical integration in particular. This paper examines the institutional determinants of vertical integration, asking, in particular, whether contractual frictions in input and …nancial markets have a di¤erential impact on the degree of vertical integration across industries and countries. I discover new patterns in cross country di¤erences in vertical integration. First, contrary to conventional wisdom, I …nd some evidence of higher vertical integration in developed countries. Second, I show that industries that are more dependent on external …nance tend to be relatively more vertically integrated in developed countries. These facts are not consistent with existing theories of vertical integration and suggest that contractual frictions in input and in …nancial markets have a radically di¤erent impact on vertical integration across industries. I develop an industry equilibrium model of vertical integration and imperfect contracting in input and …nancial markets. Fewer contractual imperfections in input markets unambiguously lead to less vertical integration while fewer contractual imperfections in …nancial markets are associated with lower degrees of vertical integration in industries that are dominated by small …rms. I …nd empirical support for these predictions in an analysis of cross-country-industry data for the manufacturing sector. Keywords: Vertical Integration, Credit Constraints, Contracts Enforcement, Developing Countries, Industry Equilibrium. JEL Codes: D23, L11, L22, O14. London School of Economics, Sticerd and PSE (Paris School of Economics). I am especially indebted to Tim Besley and Maitreesh Ghatak for their support, and Abijhit Banerjee for encouraging me at the beginning of this project. I also thank Daron Acemoglu, Pol Antras, Oriana Bandiera, Marianne Bertrand, Robin Burgess, Irma Clots-Figueras, Dave Donaldson, Leonardo Felli, Patrick Francois, Nathan Nunn, Steve Pisckhe, Imran Rasul, Mark Rosenzweig, Chad Syverson, Philipp Schmidt-Dengler, Thierry Verdier, Fabian Waldinger and participants at NEUDC 2005 conference, and seminars at LSE and Munich for useful comments. All errors are mine. E-Mail: [email protected]

1

1

Introduction

Anecdotal evidence, as well as theoretical considerations, suggests the possibility of important cross-country di¤erences in the organization of production in general, and in the degree of vertical integration in particular. Despite the importance of understanding the determinants of these di¤erences and their implications for the observed cross-country di¤erences in productivity, there has been no systematic analysis of cross-country determinants of vertical integration. This paper studies the institutional determinants of vertical integration, asking, in particular, whether contractual frictions in input and …nancial markets have a di¤erential impact on the degree of vertical integration across industries and countries. Novel patterns in cross-country di¤erences in the extent of vertical integration are revealed, and an industry equilibrium model of vertical integration and imperfect contracting in input and …nancial markets is developed. Finally, the predictions of the model are tested using cross-country-industry data. Existing theories of the …rm suggest that the more prevalent the imperfections in input markets, the more …rms tend to be vertically integrated (see e.g. Williamson (1971, 1975, 1985), Carlton (1979)). Since imperfections such as poor contractual enforcement or low quality and productivity of suppliers characterize the economies of less developed countries, …rms in those countries are often thought to be larger and more vertically integrated (see e.g. Khanna and Palepu (1997, 2000)). Figure 1 casts some doubts on these conventional views. Figure 1 plots (the log of) an average measure of vertical integration against the (log of the) GDP per capita of each country. For each country in my sample I compute the unweighted average of the ratios of value added over output (a commonly used proxy of vertical integration in the industrial organization literature) across 25 industries in the manufacturing sector. Figure 1 does not support the view that there is a higher propensity for …rms to vertically integrate in poorer countries.1 The absence of a negative relationship is instead consistent with anecdotal evidence suggesting that subcontracting arrangements are fairly extensive in the developing world, and have played an important role in the industrialization of late industrializing countries.2 1

As the index is an unweighted average, Figure 1 conveys the same information of a regression of vertical integration in industry i and country c on GDP per capita in country c and industry …xed e¤ects. In the Appendix I discuss results from regressions of this form, in which the e¤ect of country level variable is assumed to be the same across all the industries. While Figure 1 provides interesting motivating evidence for this work, this paper is mostly concerned with the identi…cation of the di¤erential e¤ects of institutions across industries. 2 See, for example, the experience of Italian industrial districts or the clusters in the early computer industry in

2

-2

Average Vertical Integration (Log) -1.5 -1

-.5

Figure 1: Vertical Integration and GDP Per Capita

6

7

8 GDP per Capita (Log)

9

10

Vertical Integration=Value Added / Output

The transaction cost approach to vertical integration (see e.g. Williamson (1975, 1985)) argues that when it is di¢ cult to write detailed contracts, trading at arm’s length results in excessive transaction costs. Vertical integration reduces the costs of arm’s length transactions by replacing the bargaining process with authority. In the presence of weak contract enforcement institutions, the transaction costs associated with the use of markets increase, and vertical integration becomes (relatively) more appealing. On the other hand, contractual imperfections severely a¤ect the e¢ ciency of transactions in …nancial markets.3 When applied to …nancial markets, transaction costs theories argue that, in Taiwan (Levy (1990)). Further examples from developing countries are given by the cotton industry in Tiruppur in southern India, the Guadalajara shoe cluster in Mexico (Woodru¤ (2002)), or the Sinos Valley in Brazil (Schmitz (1996)). Andrabi et al. (2004) provide an insightful analysis of the subcontracting arrangements of a large tractor producer in Pakistan. 3 An extensive cross-country literature documents how the degree of …nancial development is strongly associated with its legal origin (see e.g. LaPorta et al. (1998)). There is also a substantial amount of evidence that …rms are …nancially constrained in developed and developing countries as well (see e.g. Banerjee and Munshi (2004), Banerjee,

3

.1

Figure 2: Di¤erential Vertical Integration and External Financial Dependency transportation equipement beverageswood products food products

Differential Vertical Integration -.2 -.1 0

nonferrous metal

textile iron paper and steel and products furniture machinery footwear leather printing and publishing non metal products other industrieselectric machinery apparel professional goods pottery

glass rubber products metal products

plastic products

-.3

other chemicals

-.5

0

.5 External Financial Dependency

1

1.5

Differential Vertical Integration = Vertical Integration(non-OECD) - Vertical Integration(OECD)

the presence of …nancial market imperfections, …rms vertically integrate and create internal capital markets to substitute for the lack of (e¢ cient) external capital markets. Figure 2 reveals a second puzzling pattern in the relationship between …nancial market imperfections and vertical integration. For each of the 26 industries in the sample, I construct an index of "Di¤erential Vertical Integration" as the di¤erence between the average index of vertical integration across non-OECD countries and the average index of vertical integration in OECD countries. I then plot this index against the measure of external …nancial dependency at the industry level in the U. S., a proxy for technological needs for external …nance.4 Du‡o and Munshi (2003) for …rms in India). Moreover, …nancial development has been shown to be a key determinant of cross country di¤erences in industry growth and productivity (see e.g. Rajan and Zingales (1998), and the survey in Levine (2005)). 4 It is important to stress that Figure 2 does not tell us anything on the levels of vertical integration across industries. The measure of external …nancial dependency is from Rajan and Zingales (1998). As in their seminal paper, the key assumption underlying Figure 2 is that industry characteristics in the United States proxy for technological characteristics at the industry level. Results are not driven by the comparison between OECD and non-OECD

4

Figure 2 shows that industries that rely more on external …nance are not relatively more vertically integrated in less developed countries. This is not consistent with the transaction costs view applied to imperfections in …nancial markets. Figure 2 suggests that poor contract enforcement may have a very di¤erent impact on vertical integration through imperfections in speci…c (intermediate) versus non-speci…c (capital) input markets.5 To shed some light on the stylized facts reported above, I develop a theoretical framework to study the relationship between vertical integration and contractual imperfections in input and …nancial markets. Building on Grossman and Helpman (2002), I assume that vertical integration centralizes the control over non-contractible investments in the production of intermediate inputs, and thus mitigates hold-up problems arising from contractual imperfections. However, since additional investments are required to e¤ectively acquire control over non-contractible investments, vertical integration essentially comes at the cost of higher …nancial requirements. I embed this model of vertical integration in an industry equilibrium model in which …rms are heterogeneous with respect to their productivity levels, as in Melitz (2003). Under these assumptions, the model generates a natural sorting of …rms into organizational forms. More productive …rms become larger both in terms of revenue and size. For this reason more productive …rms gain relatively more from solving contractual imperfections, and thus become vertically integrated. I introduce contractual frictions in the intermediate input market and in the …nancial market and show that these imperfections have radically di¤erent implications for the extent of vertical integration in the industry. Fewer contractual frictions in the intermediate input market mitigate the hold-up problem associated with arm’s length transactions. Better contractual institutions in intermediate input markets reduce the relative returns of vertical integration and hence unambigously increase the degree of non-integration in the industry. In other words, the bene…ts of vertical integration are particularly strong in industries that rely more heavily on contracts, and in countries with poor contract enforcement institutions. countries. Identical results are obtained if the average for non-OECD countries is compared with the average for the U.S. or if the index of "Di¤erential Vertical integration" for industry i is computed as the coe¢ cient i of the regression IN Tic = + i + GDP pcc + i GDP pc + "ic : 5 This intuition is con…rmed by the contrast with Figure 9, in which the index of di¤erential vertical integration is plotted against a measure of contractual intensity (described in the text). At this point, a natural question is whether contract enforcement and external …nancial dependency are truly the crucial factors driving the correlations in Figures 2 and 9. Section 3 provides further econometric evidence on this. For reasons of space, this paper does not focus on other institutional variables, such as openness to trade, labor market regulation, antitrust institutions and social capital.

5

Frictions in …nancial markets are introduced through a simple moral hazard problem between the entrepreneur and the external investors. Contractual imperfections in …nancial markets impact the degree of vertical integration in the industry through two conceptually distinct channels. On one hand there is an "investment", or partial equilibrium, e¤ect. Since vertical integration implies higher …nancial needs, more developed …nancial markets allow more …rms to become vertically integrated. On the other hand, there is an "entry", or industry equilibrium, e¤ect. Better …nancial markets favor entry, and thus destroy the pro…t base that allows …rms to vertically integrate. Since the "investment" and "entry" e¤ects have opposite sign the net impact is shown to be ambiguous, and to depend crucially on the shape of the distribution of …rm productivity (and hence size) in the industry. In industries in which technology is such that the equilibrium industry structure displays a large number of relatively small …rms, the entry e¤ect dominates and better …nancial markets lead to less vertical integration. In industries in which technology is such that the equilibrium industry structure displays a large number of relatively large …rms, the opposite is true. Thus, the model shows that contractual frictions in the input markets and in the capital markets have very di¤erent impact on the incentives to vertically integrate, and provides a candidate explanation for the puzzling pattern in Figure 2.6 The model delivers clear-cut predictions on the di¤erential impact of better credit markets across industries. In section 3, I use cross-country-industry data to provide econometric evidence supporting those predictions, showing that the patterns in Figures 1 and 2 are robust to the inclusion of additional controls. I …nd strong evidence supporting the prediction of the model with respect to the di¤erential impact of better …nancial markets on vertical integration across industries. I also …nd some evidence of better contract enforcement being associated with lower vertical integration in industries that extensively rely on contracts. This work is closely related to several strands in the literature. On the theoretical side there is a large literature on vertical integration and …rm boundaries. The two dominant theories of …rm boundaries are the transaction costs theory (TC) developed by Williamson (1971, 1975, 1985) and the property rights theory (PR), developed by Grossman and Hart (1986) and Hart and Moore 6 A further extension shows how credit markets development may further reduce the incentives for vertical integration, by fostering the development of input markets. I allow the baseline model to accommodate for some search for potential upstream suppliers. The possibility of bottlenecks in the backward industries, due to worse …nancial markets or low quality of potential supplier implies, ceteris paribus, higher vertical integration in the industry.

6

(1990). The model in the theoretical section builds on these contributions.7 While most of the theoretical work on …rm boundaries presents partial equilibrium models, a new and rapidly growing literature analyzes models of …rm boundaries and organization in industry equilibrium. The model in this paper is more closely related to this literature, and in particular to Grossman and Helpman (2002), as mentioned before. With respect to their framework, I introduce …rm heterogeneity and credit market imperfections. The heterogeneity of …rms is essential in generating the predictions on the di¤erential e¤ects of credit market imperfections on vertical integration across industries. Other recent related contributions are Antras (2003) and Antras and Helpman (2004). The latter in particular, analyzes an industry equilibrium model in which heterogeneous …rms choose between FDI and outsourcing in a property rights framework.8 My focus on …nancial market imperfections brings this work close to an extensive literature on the relationship between credit markets, development and occupational choice pioneered by Banerjee and Newman (1993).9 On the empirical side, Acemoglu et al. (2005b) is a parallel and independent study that is most closely related to this paper. It also provides evidence on the cross countries determinants of vertical integration exploiting a large dataset of …rms around the world. In contrast to their work, the theoretical model I develop in section 2 allows me to separate, and identify, di¤erent opposing channels through which …nancial development di¤erentially a¤ects the degree of vertical integration across industries and countries. On the other hand, they estimate regressions without industry …xed e¤ects, and show that less developed countries are concentrated in industries that have higher propensity for vertical integration. Dispite di¤erences in the measure and data sources to proxy for 7

As it is now clear (see e.g. Whinston (2003)) the two theories are conceptually very di¤erent in the analysis of the costs and bene…ts of vertical integration, and have di¤erent empirical content. The PR theory has been re…ned and developed in several papers. Aghion and Tirole (1994), Legros and Newman (2004), and Acemoglu et al. (2005a) consider settings in which ex-ante transfers are banned, and thus the organizational form may not be chosen optimally. For an excellent survey and discussion of existing theories of the …rm, see Gibbons (2004). There is also an extensive industrial organization literature on vertical integration, that, however, does not focus on the institutional determinants of vertical integration (for a survey see e.g. Perry (1989)). Finally, at the end of section 2 I discuss how this work relates to recent contributions in the corporate …nance literature. 8 Other important recent contributions analyzing the internal organization of …rms in industry equilibrum are Grossman and Helpman (2004, 2005) and Marin and Verdier (2002). Acemoglu et al. (2005a) analyze an industry equilibrium model of the division of labor. 9 Gershenkron (1962) provides arguments and some anecdotal evidence on why industries in late industrializing countries seemed to be more vertically integrated. Acemoglu et al. (2003) formalizes similar arguments in a Shumpeterian endogenous growth model. Stigler (1951) noted that developing industries tend to be more vertically integrated since they are not able to rely on established and well developed supplier networks. These contributions share with this work an interest in the relationship between the stage of development and vertical integration.

7

vertical integration, we obtain very similar results in the subset of overlapping speci…cations.10 Antras (2003) and Acemoglu et al (2004) are recent examples of cross-industry studies of vertical integration. The …rst links …rm boundaries to trade ‡ows. Interestingly, it …nds that the share of intra…rm imports in total US imports is signi…cantly higher the higher the capital-labor ratio of the exporting country, a …nding that echoes our Figure 1. The second exploits a very disaggregated dataset on British manufacturing establishments to explore within-country determinants of vertical integration. Instead, I emphasize country level determinants of vertical integration.11 The rest of the paper is organized as follows. Section 2 analyzes the model and a simple extension to accommodate a third e¤ect of credit market development on vertical integration working through the development of upstream industries. Section 3 presents empirical evidence on the cross-country determinants of vertical integration using country-industry data, and discusses further evidence from case studies and business history. Section 4 provides some concluding remarks. All the proofs are in the Appendix.

2

Model

In this section I develop an industry equilibrium model of the relationship between vertical integration and contractual frictions in input and credit markets. I model vertical integration in a similar way to Grossman and Helpman (2002) and introduce …rm heterogeneity as in Melitz (2003). This section is divided into four subsection. I …rst set up the model and introduce the contractual imperfections in input and …nancial markets as well as the distinction between vertical integration and non-integration. I then derive the industry equilibrium. In the third part I discuss the e¤ect of contractual imperfections on the degree of vertical integration in the industry. Finally, I present a simple extension of the model that disentangles a further e¤ect of credit market imperfections on the degree of vertical integration. 10 This work is also related to institutional determinants of cross country-industry analysis of trade ‡ows (e.g. Nuun (2005)), and industry growth (e.g. Rajan and Zingales (1998), and the survey in Levine (2005)). The present study shares with these studies data and methods, however the focus on vertical integration is new. 11 There is a large literature on the determinants of vertical integration in speci…c industries in the United States (see Whinston (2003)). A large part of this literature focussed on testing the TC theory of the …rm. I am not aware of empirical papers that systematically examine the relationship between vertical integration, and credit markets. In the empirical section I discuss some evidence on the links between …nancial markets and vertical integration, available from case studies and business history. Fee et al (2005) analyze corporate equity ownership in vertical relationships in the United States.

8

2.1

Set Up

Environment I consider an economy with population L that produces goods using only labor. There are J + 1 sectors. One sector provides a single homogeneous good. This good is used as the numeraire, and its price is set equal to 1. It is produced under constant return to scale, with a technology employing 1 unit of labor to produce 1 unit of the homogeneous good. Provided that the economy produces the homogeneous good, the wage will be w = 1: In the remaining of the paper, I will assume that this is true. The other J sectors supply a continuum of di¤erentiated goods. In each of these sectors there is a …xed set of potential entrepreneurs described later. Each …rm is a monopolist over the variety it produces. The workers are the only consumers, each endowed with 1 unit of labor. They all have the same CES preferences over the di¤erentiated goods. A consumer that receives q0 units of the homogeneous good, and q( ) of each variety

2

j

(to be determined in equilibrium) of the

di¤erentiated goods produced by industry j 2 f1; ::j; ::Jg; gets a utility U U

q01

J'

J Z Y (

j=1

where "j =

1 1

j

q( ) j d ) 2

' j

(1)

j

> 1 is the elasticity of substitution between two varieties of the di¤erentiated

goods in industry j: If all varieties in the set

j

are available at a particular price p( ) these preferences yield

aggregate demand functions q( ) = Aj p( ) where p( ) is the price of a particular variety

The monopolist of variety

Aj = R (

"j

and

2

'L p( ) j

j "j

d )

in industry j treats Aj as a constant, and so perceives a constant hR i 1 j "j j "j d elasticity of demand "j : I denote Pj = p ( ) as the price index in industry j: j 2 j

The price index is inversely related to the "competitiveness" in the industry. Competitiveness is,

9

ceteris paribus, increasing in the number of varieties produced in the industry, and decreasing in the (average) price charged by competitors. Production and Firm Organization I now turn to the description of …rms’ technology and modes of organization in the industry. Since the set of potential entrepreneurs in each industry is exogenously given, and production of the homogeneous good in the economy implies w = 1; industries can be treated independently. Therefore, I suppress the subscript j from industry variables. With a slight abuse of notation, I assume that each di¤erentiated …nal product y( ) is produced under a constant marginal cost technology according to y( ) = I

(2)

where I is a specialized component described below.12 I also assume that the specialized component must be exact in its speci…cations, and that di¤erent …nal goods require distinct components. An input must also be of suitably high quality in order to be useful in the production of the …nal output. Furthermore, I assume that there are …xed costs associated with entering the market. Final goods may be produced by vertically integrated …rms, or by specialized producers that purchase their inputs at arm’s length from external suppliers (outsourcing). In either case I assume that an intermediate input of low quality can be produced at no cost. The intermediate speci…c input is produced undertaking speci…c investments in a unit measure of (symmetric) tasks, each entailing a constant marginal cost c: The (quality of) the intermediate input is then a Cobb-Douglas aggregate I = exp

Z

1

ln x(i)di

(3)

0

where x(i) denotes the level of investment in task i: I consider a setting with incomplete contracting where investments in some tasks i can be observed by the collaborating parties, but cannot be veri…ed by a court. The lack of veri…ability precludes contracts between input suppliers and customers stipulating a given price for an agreed quantity. If such a contract were signed, the supplier could raise its net pro…ts by reducing the 12

It is straightforward to extend the model to allow for labor as an additional factor in the production of y; by having a production function of the form y = ( 1 L )1 ( I ) : See the appendix for details.

10

investment in some task i: The buyer would be obliged to buy the lower quality input with no contractual protection.13 Legal institutions vary greatly across countries. In order to capture the e¤ects of di¤erent contractual institutions on vertical integration, I follow Acemoglu et al. (2005) in parametrizing the quality of the contract enforcement institutions in the following intuitive way. I assume that a measure

of the tasks necessary to complete the intermediate input can be perfectly contracted

upon, while a measure (1

) can not be contracted upon. While product and industry character-

istics certainly a¤ect the degree of contractual incompleteness in intermediate inputs transactions, contract enforcement institutions also a¤ect the degree of incompleteness of contracts. In countries with better contracting institutions

tends to be higher, i.e. it is relatively easy to enforce con-

tracts that give appropriate incentives to undertake ex-ante investments. Since industries can be treated in isolation, higher

has to be interpreted as a decrease in the industry-country speci…c

degree of contractual frictions in input markets.14 There are two alternative ways of organizing the …rm. Under vertical integration the entrepreneur retains control over all non contractible investments. Vertical integration entails centralized control, and thus the entrepreneur (e¢ ciently) decides all the relevant investments x(i). Alternatively, the entrepreneur may decide to outsource to an independent supplier the production of the intermediate input. Under outsourcing the independent supplier retains control over the non-contractible investments. The absence of ex-ante enforceable contracts exposes parties to a hold-up problem. Once a supplier specializes its inputs to a particular …nal product, these inputs have higher value within the relationship than in any alternative uses. Assuming for simplicity that the value in alternative uses is zero, the downstream producer can then threaten to refuse the transaction with the upstream supplier, unless the price, negotiated once the investments are sunk, is low enough. This leaves the upstream supplier in a relatively weak position. Anticipating this situation, the upstream supplier has insu¢ cient incentives to invest in the non-contractible tasks 13

While a rich literature has developed discussing alternative solutions to this form of contractual incompleteness (see e.g. Maskin and Tirole (1999) and Aghion et al. (1994)), I do not dwell on the micro-foundations of the assumption, and simply take incomplete contracts as a fact of commercial life. 14 The Cobb-Douglas aggregate formulation for the production of the intermediate input I implies that the elasticity of substitution between elementary investments is equal to 1. It is possible to use a general CES production function for the production of the intermediate input. It would then be possible to parametrize the contractual needs of the industry with the elasticity of substitution across investments, with higher elasticity parametrizing lower contract needs. Should labor be used to produce the …nal good, contractual needs could also be parametrized by ; as in footnote 12.

11

i:15 I assume that the marginal cost of investment x(i) is equal to c = 1 in both vertically integrated …rms and independent suppliers.16 The price for the intermediate input among two independent …rms is negotiated ex-post. Expost, once the uncontractible investments have been undertaken and the speci…c input produced, the two parties bargain over appropriable quasi rents which are given by the amount of pro…ts that can be obtained using the speci…c input to produce the …nal good. To simplify, I assume that in the ex-post bargaining process the downstream …rm retains a share (1

) of the revenues derived

from the transaction, while the upstream …rm retains the remaining share : I assume that in each industry there is a …xed pool of potential entrepreneurs that are heterogenous with respect to their productivity : Each entrepreneur draws her productivity level

from

a distribution with associated continuous cumulative function G( ) and observe her productivity before deciding whether to start production. To simplify, I also assume that the mass of potential entrepreneurs is equal to L in each industry. I take the distribution of the productivity parameter as exogenous. However the marginal cost of producing one unit of the …nal good depends on the organizational form. Since …rms with di¤erent

choose di¤erent organizational forms, the distrib-

ution of measurable productivity is endogenous and is determined by the same forces shaping the organizational form choice. I assume for now that there is a large supply of homogenous external suppliers. This assumption implies that a …rm deciding to "buy" the intermediate input always …nd a partner.17 Fixed costs and …nancial constraints In order to start production, …rms incur …xed costs, such as the costs of entering the market 15

As in the property rights theory of the …rm (Grossman and Hart (1986) and Hart and Moore (1990)) I assume that (some) tasks x(i) are not contractible. However, in contrast to the property rights framework, I assume that control over tasks is contractible and transferrable, as in Baker et al. (2004). This brings the theory of the …rm in this model closer to Transaction Costs theories of the …rm (see e.g. Williamson (1975, 1985) and Grossman and Helpman (2002)). 16 When labor is an additional factor of production the results in the model can be obtained regardless of the contractibility of L: It would also be possible to allow di¤erences in marginal costs of investment in task x(i) across organizational form, and in particular c < 1 for specialized suppliers, without a¤ecting the results. Specialized suppliers may be more e¢ cient than vertically integrated …rms due to diseconomies of scope, or the excessive bureaucratic costs associated with a more complex, vertically integrated …rm. 17 The assumption that suppliers are homogeneous implies that there are no matching / sorting issues between heterogeneous (in terms of ) downstream and homogeneous upstream units, and is made for the sake of analytical tractability. In subsection 2.4 I consider a setting in which specialized assemblers do not always …nd a supplier.

12

and setting up the organization and of designing the di¤erentiated product, as well as those …xed costs associated with the equipment necessary to perform assembly operations. These costs have to be paid by all …rms, regardless of the organizational form, and are denoted by f . In addition, …rms that decide to become vertically integrated have to acquire control over the extra equipment needed to produce the intermediate inputs. I assume that the cost of this additional machinery is equal to k; and that both f and k are strictly positive. For our purposes, it is irrelevant whether a …rm deciding to become vertically integrated incurs the extra cost k in order to acquire one of the existing suppliers, or instead builds at cost k the necessary equipment from scratch.18 I assume that …xed costs have to be paid up-front, i.e. before production takes place and that …rms have no liquidity and need to borrow from external investors in order to …nance the …xed cost investment. I assume for simplicity that the risk free interest rate in the economy is equal to zero, and that a large supply of risk neutral investors lend capital at this interest rate. However, because of contractual imperfections, credit markets are not perfect. I model credit constraints in a rather crude, but simple, way. Speci…cally, I assume that the …xed costs f and k need to be …nanced in advance, and are in fact composed of a continuum of small investments. I assume that a fraction 1

of this investments is contractible: external investors can easily make

sure that the capital is e¤ectively invested in the project (for instance renting corporate buildings, acquiring speci…c machines, etc.). In contrast, the remaining fraction

is not contractible, in the

sense that the external provider of …nance can not make sure that the capital is e¤ectively invested in production (e.g. hiring the appropriate product designer, purchase of some speci…c services, etc.). While

certainly captures characteristics of the industry, it also depends on the availability

of legal instruments protecting external investors, such as borrowers’ public register, courts, etc. Since industries can be treated in isolation, lower

has to be interpreted as a decrease in the

industry-country speci…c degree of contractual frictions between …rms and external investors. After borrowing from external investors, the entrepreneur can choose among two di¤erent behaviors. She can invest the borrowed cash to pay the …xed costs and start production. Alternatively 18

The assumption that vertical integration entails higher …xed costs is common in the literature (see e.g. Williamson (1971), Grossman and Helpman (2002), Antras and Helpman (2004)). The assumption can be justi…ed on di¤erent grounds such as, for example, "managerial" diseconomies of scales. Here I emphasize that in order to e¤ectively acquire control over the production of the intermediate input a …rm has to build its own plant, or must acquire an upstream supplier. In either case, these operations are assumed to be relatively more costly than dealing with an already established external supplier whose …xed costs are at least partially sunk.

13

she can divert the cash corresponding to the fraction

of non-contractible investments. For sim-

plicity, I assume that such a diversion of cash occurs at no cost.19 This describes the borrowing process of a vertically integrated …rm. The case of a specialized assembler is slightly more complicated, since the relationship between the external investors and the entrepreneur is also a¤ected by the presence of a third party, the independent supplier. For simplicity, I focus on bilateral contracts. The restriction on bilateral contracts can be justi…ed by the fact that the …xed costs f have to be …nanced before the speci…c match with the supplier is realized, and contracts involving investors, suppliers and assemblers cannot be signed simultaneously. The sequence of events is as follows. First the …nal assembler …nances the …xed costs f borrowing from the external investors. Once these …xed costs have been paid she is matched with an upstream supplier. Since the assembler has ex-ante bargaining power she can ask for an ex-ante transfer from the supplier. Ex-ante competition, drives suppliers pro…ts to zero. I assume that external investors hold claims on the assembler’s ex-post pro…ts and on the ex-ante transfer from the supplier. I denote with

v(

) and

o(

) the variable (i.e. net of …xed costs) pro…ts of vertically integrated

and specialized assembler …rms respectively, when the entrepreneur has productivity . Similarly, I denote with Fv = f +k and Fo = f the …nancial requirements of vertical integration and outsourcing respectively. We have Lemma 1 An entrepreneur with productivity i 2 fv; og if and only if

i(

)

obtains funding to set up a …rm with organizational form

(1 + )Fi :

Lemma 1 describes the e¤ects of the contractual frictions in the …nancial market associated with the two di¤erent organizational forms. If

is equal to zero, credit markets are perfect, and

all projects that generate (variable) pro…ts in excess of the …xed costs (i.e. with positive net present value), are …nanced. However, if

is positive, some projects that would generate positive

net present value cannot be …nanced because of the form of moral hazard introduced by contract incompleteness in the capital markets20 . 19 I assume for simplicity that all the …xed costs have to be paid in order for production to take place, so that the entrepreneur never …nds it pro…table to divert only a fraction of the borrowed cash. 20 The formulation relies on a form of ex-ante moral hazard. However one could imagine that, once …rms realize

14

It is crucial that k > 0: This assumption introduces a trade-o¤ between the two organizational forms which is central for the results. Vertical integration reduces the distortions associated with imperfect contracting in input markets, but comes at the cost of higher …nancial requirements for the …rm. Note that, under the assumptions speci…ed above, for a given level of …nancial requirements and variable pro…ts, the form of the credit constraints inequality is the same for vertically integrated and non-integrated …rms. In other words, I present a setting in which the organizational form of the …rm only a¤ects the …nancial position of the …rm through the associated …nancial needs.21 Since vertical integration requires higher …nancial needs, higher contractual frictions in the …nancial market, i.e. a higher ; make vertical integration more di¢ cult. To summarize, the timing of events is as follows. Entrepreneurs with heterogenous productivity decide whether to enter the market, as a vertically integrated …rm or as a specialized manufacturer of …nal goods. Those that choose vertical integration as the organizational form of their …rm and …nd external investors willing to …nance them, borrow from external investors and pay the corresponding …xed costs, f and k and undertake production decisions. Those that decide to enter as assemblers of …nal good and …nd external investors willing to …nance them, are matched to a supplier. Suppliers can make ex-ante transfers to attract assemblers. The assembler and the supplier write an ex-ante contract specifying an initial transfer from the supplier to the assembler, and investments on the fraction

of contractible tasks. After this contract is signed, the supplier

undertakes non-contractible investments i: Finally, bargaining over the surplus takes place, the …nal goods are produced and sold, and external investors are repaid.

2.2

Industry Equilibrium

I now turn to the determination of the industry equilibrium. In order to solve the industry equilibrium, I …rst compute the pro…t functions for a vertically integrated and for a non-integrated …rm respectively. I then analyze the organizational form and entry decision of entrepreneurs with revenues, the owner of the …rm can hide (a share of) pro…ts at some per unit cost 1; avoiding to repay the external investors. This form of ex-post moral hazard would generate a form of credit constraints equivalent to the one introduced in the text. 21 The assumptions are made for the sake of simplicity, and do not alter the logic of the results. The Appendix discusses a setting in which suppliers are also …nancially constrained. Under certain conditions, it is also possible to assume that is di¤erent across organizational forms, without a¤ecting any of the results. From a theoretical point of view, however, it is not obvious whether vertical integration increases or reduces :

15

productivity ; and de…ne the industry equilibrium. The derivation of pro…t functions and proofs of all the results are reported in the Appendix. I …rst consider a vertically integrated …rm. Since the elementary tasks in the production of the intermediate input are symmetric, I can restrict attention to the case in which a vertical integrated …rm invests xv in each task. The production function of a vertically integrated …rm thus becomes qv ( ) = xv Moreover, from the demand equation it is clear that …rms charge a constant mark up determined by the elasticity of substitution of consumer demand. For a vertically integrated …rm we have

pv ( ) =

1

By substitution in the demand equation, we obtain the variable pro…ts of a vertically integrated …rm as v(

)=

"

A

"

(1

)

(4)

I now turn to the pro…ts of a specialized assembler. In the proof of Lemma 1 in the Appendix I show that ex-ante competition among suppliers that are not …nancially constrained implies that we can restrict attention to the total surplus generated by a match between an assembler and a supplier. Since elementary tasks in the production of the intermediate input are symmetric, we can restrict attention to the case in which an ex-ante contractually speci…ed xc is invested in the fraction of contractible tasks, and xn is invested in the fraction (1

) of non-contractible tasks, and is

independently decided ex-post by the supplier. For a non-integrated …rm, the production function becomes qo ( ) = xc x1n The sequence of events is as follows. First, the assembler and the supplier sign a contract specifying investments xc : Then, for a given xc ; the supplier undertakes investments xn ; anticipating that, because of ex-post bargaining, she will only retain a fraction

16

of the revenues. In the

Appendix, I show that under this con…guration the optimal pricing policy for a non-integrated …rm is given by po ( ) = and !( ; ) =

1 1 (1 )( ) 1 (1 )

: If

1 !( ; )

= 1 contractual institutions are perfect. Since all the tasks

are contractible, there is no hold-up problem and a non-integrated …rm behaves as a vertically integrated …rm, i.e. !( ; 1) = 1: Alternatively, when severe, and !( ; 0) = : In the Appendix, I show that

= 0; the hold up problem is extremely

@!( ; ) @

0; implying that better contractual

institutions allow non-integrated …rms to reduce their marginal costs, and hence charge lower prices. This pricing rule implies that the pro…ts of a non-integrated …rm can be written as o(

)=

"

( ; )A

From the properties of !( ; ); it is intuitive that @ ( ; ) @

"

(1

)

(5)

( ; 1) = 1; while

( ; 0) =

1 1

: Finally,

0: Under both organizational forms, pro…ts are increasing in the productivity index ;

and increasing in the index A; i.e. decreasing in the number of …rms active in the industry. We have Lemma 2 There exists a unique threshold bv such that a …rm with productivity

bv earns higher pro…ts

choosing vertical integration rather than non integration. The opposite is true for In particular the unique threshold bv is determined by the equality b v( v)

(f + k) =

Note that the threshold markets

v

v

b o( v )

f () bv =

A

k " (1

) (1

1 ( ; ))

< bv :

1 " 1

(6)

is decreasing in A: This implies that, ceteris paribus, in more competitive

is higher, i.e. fewer …rms …nd it pro…table to integrate vertically. This is simply due to

the fact that in order for vertical integration to be pro…table, a …rm must generate enough pro…ts to repay the additional …xed costs. The sorting of …rms with heterogeneous productivity into di¤erent organizational forms is illustrated in Figure 3. On the x-axis I report the underlying source of heterogeneity, …rm unobserved productivity, while on the y-axis I report the pro…ts of …rms with

17

Firm Heterogeneity and Vertical Integration

EÝSÞ Non Integration

Integration

Se

Sv

S P?1

Fd

Fv

Figure 3: Sorting of Firms into Organizational Forms productivity

under the two organizational forms.

Vertical integration reduces ine¢ ciencies caused by contract incompleteness, and thus becomes relatively more attractive in the presence of high quasi rents. Since the specialized component is useless outside the relationship, rents are entirely determined by the scale of operation of the …rm. Entrepreneurs with higher

have lower marginal costs and choose a bigger scale of operations, and

thus have higher quasi-rents: for this reason they become vertically integrated.

Firms however, are not unconstrained in their choice of organizational form. When

> 0 …nan-

cial constraints prevent some …rms from adopting the optimal organizational form. Substituting the pro…t function of a vertically integrated …rm into the respective …nancial constraint, we obtain that a …rm with productivity

v(

)

can enter the industry as a vertically integrated …rm if and only if

(1 + )(f + k) ()

The relative position of the two thresholds

v

v

a¤ect the vertical integration decision of …rms. If

=

(k + f ) (1 + ) A " (1 )

1 " 1

(7)

and bv determines whether …nancial constraints

18

v

bv ; …nancial constraints are irrelevant when

making a vertical integration. However, if

v

> bv instead some entrepreneurs who would like

to enter the industry as vertically integrated …rms are prevented from doing so by the existence of …nancial constraints. In the remaining part of this paper I will focus on the case in which (1 + )(1

( ; )) >

k (k+f ) ;

and hence

v

> bv ; so that some entrepreneurs are constrained in their

organizational form decision. A …rst implication of lower contractual frictions in the …nancial mar-

kets is that more entrepreneurs will be able to vertically integrate. Financial market imperfections thus impact vertical integration through a direct "investment" e¤ect. Finally, in order to solve for the industry equilibrium of the model I have to derive the thresholds determining whether an entrepreneur can enter the industry as a non-integrated …rm. Combining the …nancial constraint inequality for a non-integrated …rm with its respective pro…t function, it is obvious that an entrepreneur with productivity

can enter the industry with a non-integrated

…rm if and only if

o(

)

(1 + )f ()

e

=

A

f (1 + ) " (1 ) ( ; )

1 " 1

(8)

The following proposition characterizes the equilibrium in the industry. Proposition 3 If

f (k+f )

( ; ) there exists a unique equilibrium de…ned by two thresholds

that entrepreneurs with

<

e

do not enter the industry, entrepreneurs with

industry as specialized assemblers, and entrepreneurs with

v

e

2 [ e;

and v)

v;

such

enter the

enter the industry as vertically

integrated …rms: If

f (k+f )

>

( ; ) there exists a unique equilibrium de…ned by the threshold

entrepreneurs with

<

v

do not enter the industry, and entrepreneurs with

v;

such that

v

enter the

industry as vertically integrated …rms: As noted above, the model generates an endogenous sorting of …rms with heterogeneous productivity into organizational forms: only relatively more productive …rms generate enough variable pro…ts to cover the extra …nancial requirements necessary to vertically integrate. When f (k+f )

( ; ); the distortions associated with outsourcing are relatively mild, and both orga-

nizational forms coexist in equilibrium. On the other hand, when

f (k+f )

> ( ; ); the ine¢ ciencies

caused by incomplete contracting are so strong that the industry only displays vertically integrated 19

…rms. A …rm trying to enter the industry as a non-integrated …rm would not generate enough pro…ts to credibly commit to repay external investors. The particular form of the endogenous sorting of …rms into organizational forms in the model is particularly appealing since it implies a positive correlation between …rm’s size and vertical integration. This is consistent with anecdotal as well as more formal evidence (see e.g. Acemoglu et al. (2005)). Consistently with the model, in the cement industry in the U.S., Hortacsu and Syevrson (2005) …nd that vertically integrated …rms are larger, more productive and charge lower prices22 . In the model the specialized component is useless outside the relationship, and hence quasi-rents are entirely determined by the scale of operation of the …rm. The model thus predicts a positive correlation between quasi-rents and vertical integration. It is a common …nding in empirical work in the transaction costs literature that higher quasi-rents are associated with higher vertical integration (see e.g. Whinston (2003) for a survey).23 Large, and perhaps vertically integrated …rms, are often believed to have easier access to …nancial markets.24 These considerations transposed into the model would suggest the possibility that

may vary with the organizational form of the …rm, and in particular

predictions of the model would be robust to this extension, as far as

1+ 1+

o v

f f +k

v

<

o:

The main

< ( ; ): The model,

however, suggests an alternative explanation for the presumption that vertically integrated …rms, being larger, …nd easier access to …nance. The underlying heterogeneity in productivity implies that vertically integrated …rms are larger, and, while vertical integration requires more external …nance, in a cross-section of …rms vertically integrated …rms are less likely to be …nancially constrained. This would also be true in a model in which vertical integration reduces the pleadgeable income, and worsen the access of …rms to external …nance, i.e. when

v

>

o:

Interestingly, this possibility

has been suggested in Williamson (1971), where it is suggested that it is more di¢ cult to monitor large and complex organizations, and for this reason investors may require higher expected returns 22

The implication that more productive …rms are vertically integrated however does not need to hold for the logic of the main results to be valid, and could be easily mitigated by assuming that …rms are also heterogeneous in their access to capital markets. 23 The theory of vertical integration in this paper is di¤erent from the property rights theory of the …rm and is close in spirit to transaction costs theories of …rm boundaries (see footnote 15). Under certain assumptions, it is possible to extend the property rights framework and derive the same sorting of heterogeneus …rms into organizational forms as in the current model without a¤ecting the results (see e.g. Antras and Helpman (2004)). 24 If suppliers are also …nancially constrained, vertical disintegration tends to reduce pleadgeable income, since the assembler is not a full residual claimant of revenues. I refer the reader to discussion after the proof of Lemma 1 in the Appendix for details.

20

as …nance requirements become progressively greater. In this case, even in the absence of the higher …nancial requirements of vertical integration, the sorting pattern of the model would be obtained, since only more productive …rms generate enough pro…ts to a¤ord vertical integration.25

2.3

Main Predictions of the Model

I now turn to the analysis of the role that the industry speci…c institutional variables

and

play

in determining the extent of vertical integration in the industry.26 Following the seminal contribution in Adelman (1955), the empirical literature has often measured vertical integration as the ratio of valued added over sales. Intuitively, the ratio tells us the percentage of the value of production that is carried on within …rm boundaries. In our model, the ratio of valued added over sales is equal to 1 for vertically integrated …rms, and is instead equal to (1

) for non-integrated …rms. At the industry level, a convenient index of vertical integration is given by the average index of

vertical integration of …rms active in the industry. Denoting by Nv and No the number (measure) of vertically integrated and non-integrated …rms respectively, the industry level index of vertical integration is given by IN T =

No (1

I focus on the more interesting case

) + Nv 1 =1 No + Nv

f (k+f )

No No + Nv

(9)

( ; ), in which both organizational forms coexist in

the industry. When this is the case, in the industry equilibrium there is a mass of non-integrated …rms equal to G( v )

G( e ); and a mass of vertically integrated …rms equal to 1

G( v ): The

average index of vertical integration in the industry is then given by IN T = 1

G( v ) G( e ) 1 G( e )

25

(10)

Inderst and Mueller (2004), Faure-Grimaud and Inderst (2005) analyze models in which the organizational form itself a¤ects the …nancial constraint of the …rm. Recent work on internal capital markets also sheds some light on the relationship between …rm boundaries and …nancial constraints (see the excellent review in Stein (2004)). In Macchiavello (2004, 2005) I consider settings in which the assembler and the supplier jointly contract with the external investor. The two papers provide a microfoundation for v > o : The key intuition is that vertical integration, by bringing the bargaining process inside the …rm, reduces the amount of information that can be used by external investors to monitor the …rm. For a related point, in the literature on trade credit, see Bukart and Ellingsen (2004). 26 Since @!(@ ; ) 0; an increase in reduces vertical integration. Arti…cial barriers to entry could be conceptualized as an exogenously given threshold be ; i.e. as a restriction on the number of …rms in the industry. In this case, arti…cial barriers to entry reduce competition in the industry, and imply lower vertical integration.

21

y

3

2.5

2

1.5

1

0.5

0 1

2

3

4

5 x

Figure 4: Generalized Pareto Distribution, for

= 2; and

2 f 13 ; 12 ; 23 g

As is clear from this expression, the industry level index of vertical integration critically depends on the shape of the underlying distribution of productivity, G( ): In order to derive clear cut comparative statics, it is useful to suppose that

is distributed in the population according to a

generalized Pareto distribution, i.e. g( ) = with by

1; =

R1 1

2 R2+ ; and dG( ) = 1 +

1

(

(1 +

2 (0; 1): The average 1

1)

)

1

1

in the pool of potential entrepreneurs is given

: In order to perform comparative statics on the shape parameter

without changing the average productivity in the pool of entrepreneurs ; I substitute function of

as a

and work with the corresponding cumulative function G( ) = 1

(1 +

( (1

)

1) ) 1

1

(11)

The average productivity of the potential pool of entrepreneurs is thus easily parametrized by ; while the shape of the distribution is conveniently parametrized by : For a given ; the e¤ect of di¤erent

is depicted in Figure 4.

22

When

1

= 1

; the distribution is a standard Pareto distribution. For

< 1

1

; the

distribution has relatively lower density at low levels of ; and higher density for large : The opposite is true for

> 1

1

: Proxying the size of the …rm with revenues, equations 4 and 5

show that more productive …rms (higher ) are larger. Industries with low (relatively) large …rms, while industries with high

are dominated by

are dominated by small …rms. We can thus

state Proposition 4 i) Higher average productivity in the pool of potential entrepreneurs (higher ) unambiguously increases vertical integration in the industry, ii) Better contractual institutions in input markets (higher

) unambiguously reduces vertical

integration in the industry, iii) Better contractual institutions in …nancial markets (lower

)

- increase vertical integration in industry dominated by large …rms ( - decrease vertical integration in industry dominated by small …rms ( - When

=1

1

vertical integration is independent of

<1 >1

1

) and

1

):

:

I …rst consider the e¤ects of changes in contractual institutions

on the degree of vertical

integration in the industry. There are two e¤ects: a partial equilibrium e¤ect and an industry equilibrium e¤ect. The partial equilibrium e¤ect is that the pro…ts of a non-integrated …rm increase, thus making non-integration relatively more pro…table. This e¤ect is illustrated in Figure 5.27 Ceteris paribus, better contract enforcement leads to a decrease in vertical integration in the industry, in the sense that fewer …rms are vertically integrated. The industry equilibrium e¤ect is due to the fact that, since non-integrated …rms are relatively more e¢ cient because of better contractual institutions, vertically integrated …rms face higher competition. This implies that the pro…ts of each …rm with productivity

bv decrease and this further shifts towards the right the

threshold bv : The industry equilibrium e¤ect thus pushes further away from vertical integration28 Next, I consider the case of better capital markets (i.e. lower

). As in the case of better

27 Since the industry equilibrium e¤ect goes in the same direction as the partial equilibrium e¤ect, it is not illustrated in Figure 5 in order to keep the analysis simpler. 28 This can be easily shown for the case in which follows a Pareto distribution in the industry. Under this circumstance the index of vertical integration only depends on the ratio be =bv ; it is easy to show that an increase in ; raising ( ; ); reduces be and thus implies lower vertical integration.

23

EÝSÞ

Firm Heterogeneity and Vertical Integration - Better Contractual Institutions Non Integration

Se Se

Integration

Sv

Sv

S P?1

Fd

Fv

Figure 5: Comparative Statics and Industry Equilibrium (1)

contractual enforcement there are two di¤erent e¤ects: a partial equilibrium e¤ect, and an industry equilibrium e¤ect. The e¤ects of imperfect …nancial markets are illustrated in Figure 6. First, whenever

> 0

some …rms are credit constrained and cannot integrate vertically. Because of the sorting e¤ect, only …rms with productivity above a certain threshold are vertically integrated. When capital markets improve, the threshold moves towards the left, as it becomes easier to raise external funds. Since lack of …nancial resources is the only constraint on vertical integration, better …nancial markets have a positive (partial equilibrium) "investment" e¤ect on the degree of vertical integration. However, better …nancial markets also favor the entry of new competitors in the industry. First of all, the marginal entrant is a non-integrated …rm, an e¤ect that counterbalances the previous e¤ect. More importantly, new competitors in the industry implies that each …rm earns fewer pro…ts. This e¤ect implies that fewer …rms can become vertically integrated.29 This second e¤ect di¤erentiates the mechanics of the response of vertical integration to bet29

In the Appendix I show that for a general distribution G( ) the net e¤ect of better …nancial markets (lower ) G( )) depends on whether d ln(1 7 0: See equation 24 for details. d ln( )

24

EÝSÞ

Firm Heterogeneity, Vertical Integration and Credit Markets Non Integration

Integration

VF v VF d Se

Sv

S P?1

Fd

Fv

Figure 6: Comparative Statics and Industry Equilibrium (2)

ter contractual institutions in the capital markets compared with the e¤ect of better contractual institutions in the speci…c input market. In the speci…c input market the industry equilibrium e¤ect works in the same direction as the partial equilibrium e¤ect, and the total e¤ect is thus unambiguous. The net impact of better …nancial markets on the degree of vertical integration is thus a priori ambiguous, and depends on additional industry characteristics. Better …nancial markets increase vertical integration in industries that have relatively high densities at high ; since the "investment" e¤ect is relatively stronger in such industries. These industries are, for technological reasons, dominated by large …rms, in the sense that a large share of output is produced in large …rms. On the other hand, better …nancial markets reduce vertical integration in industries that have relatively high densities at low ; since in those industries the "entry" e¤ect is relatively stronger. These industries are instead dominated by relatively small …rms, in the sense that a large share of output is produced by such …rms.30 30

I have solved the model without considering labor as additional factor of production. Introducing labor would establish a closer link between the prediction of the model and the empirical test in section 3. In the Appendix I

25

2.4

Entry in the Upstream Industry

It is often argued that vertical integration may be a response to the di¢ culties of …nding suitable and reliable suppliers. To the extent that institutional failures in developing countries hinder the development of upstream industries, it is expected that …rms in the developing world are relatively more vertically integrated. The model developed so far assumed that a downstream …rm always …nds a potential supplier in the market. I now relax this assumption in order to consider the e¤ects of contractual imperfections in …nancial markets on the degree of vertical integration by examining the e¤ect on the development of upstream industries. The analysis for a vertically integrated …rm is as described in the previous subsection. The case for non-integration needs to be modi…ed. In particular I assume that after the …xed costs f have been paid, the downstream entrepreneur searches for an upstream supplier. Let of downstream …rms in the market, and

denote the mass

the mass of upstream …rms in the market. To keep the

analysis simple I assume that the probability that a downstream …rm …nds an upstream supplier is given by Pd = minf

; 1g

while the probability that an upstream supplier …nds a downstream customer is given by Pu = minf

; 1g

Firms on the short side of the market always …nd a partner. If a …rm does not …nd a partner it is forced to exit the market, and the …xed costs are lost. Since we are interested in analyzing the e¤ects that the development of backward industries has on vertical integration, I restrict attention to equilibria in which

>

; so that upstream …rms are now on the short side of the market.

Upstream …rms enter the industry paying the …xed costs k: In order to keep the model simple, I assume that upstream …rms are homogenous in terms of their productivity, and are randomly matched with a downstream unit. At the time of entry, the upstream supplier faces uncertainty with respect to the productivity of the …nal assembler she matches. This uncertainty translates into uncertainty regarding the pro…ts of entering the industry. I introduce contractual imperfections in the …nancing of the entry of upstream suppliers. Since, show that the amount of labor employed by the …rm is proportional to

26

"

.

as in the previous subsection, the degree of …nancial market frictions is industry speci…c, I parametrize the degree of contractual imperfections with the share of the …xed investment of upstream suppliers k that is not contractible, e:31

Before entry, in order to satisfy the (expected) zero pro…t condition for upstream …rms, the

following condition has to be satis…ed

u

=E

2

[

o(

k = ek

)]

(12)

where the ek term on the right hand side follows from the fact that upstream …rms face some

borrowing constraints originating from contractual imperfections in …nancial markets. In equilibrium

is proportional to G( v )

G( e ); where the two thresholds

v

and

e

are determined in

an analogous manner as in the previous subsection. In an equilibrium in which upstream …rms are on the short side of the market, the thresholds

e

and

v

adjust in order to satisfy the zero

pro…ts constraint of upstream …rms. In particular, note that higher contractual imperfections in the …nancial market (higher e) imply that the expected productivity of specialized assemblers must

be higher. A similar argument applies to downstream …rms. The expected pro…ts of the marginal downstream …rm entering the market with productivity

o

= (1

)

o( e)

e

must satisfy

f= f

(13)

It is clear from this last equation that tighter intermediate input markets, i.e. smaller that the productivity level of the marginal …rm entering the industry,

e

; imply

must be higher: Finally,

restricting attention to equilibria in which downstream …rms are constrained in their organizational form decision, the marginal (least productive) …rm that can a¤ord vertical integration as organizational mode is de…ned as before by the productivity threshold

v

Since the threshold through which 31

v

=

v

given by

(k + f ) (1 + ) A " (1 )

does not depend on the thickness of the input market, the only way

can appropriately adjust to a higher e is through an increase in the threshold

I also assume that the match is speci…c and …rms do not negotiate ex-ante transfers. This assumption simpli…es the analysis without a¤ecting the results.

27

e

that implies a higher expected productivity of downstream assemblers. This movement in

e

in

turns implies a higher degree of vertical integration in the industry. I summarize this discussion with the following Proposition 5 Better contractual institutions in the …nancing of upstream industries (lower e) reduce vertical

integration in the downstream industry.

In the presence of credit market imperfections, the e¤ective …xed costs that have to be covered by the upstream …rm are given by k(1 + e): A direct implication of higher e is that the input market become tighter, and it is thus more di¢ cult to …nd a supplier for a non-integrated downstream assembler (lower

). This e¤ect reduces the relative returns of non-integration in the industry, and

is thus equivalent to a reduction in

3

( ; ) in the model in the previous subsection.

Empirical Evidence

In this section, I use data on value added and output for 26 industries in the manufacturing sector across a sample of 89 countries to test the empirical validity of the theoretical arguments relating institutions and vertical integration developed in the previous session. This can be thought of as a more robust illustration of the patterns in Figures 1, 2 and 9. I measure vertical integration in industry i in country c as the ratio of value added over output, i.e. IN Tic =

V Aic Yic

(14)

This measure was …rst introduced in a classical study on vertical integration in Adelman (1955). At the …rm level, the measure captures the proportion of the production process that is carried out within …rm boundaries. A higher value of the index is therefore associated with a higher degree of vertical integration. In section 3.4 I discuss in more detail some concerns arising from how the index aggregates up at the industry level. This section is divided into three subsections. In the …rst subsection I describe the data used. The second subsection investigates whether country level institutions have di¤erential impact across industries and whether those di¤erences are captured by the speci…c channels emphasized in section 28

2. The third subsection discusses some further issues arising from the aggregation of data at the industry level. Finally, the last subsection discusses some further evidence from case studies and American business history on the link between credit markets and vertical integration. The Appendix contains a discussion on whether measures of institutions at the country level are correlated with the average degree of vertical integration and a description of the main variables used in the analysis.

3.1

Data

The data used to compute the index of vertical integration are from the 2001 edition of the UNIDO Industrial Statistics Database. Data are available for the manufacturing sector, and are aggregated at the 3-Digit level of the second revision of the ISIC Code classi…cation system. This gives a total of 29 manufacturing industries. The data were supplied by national statistical o¢ ces and supplemented with estimates generated by UNIDO whenever necessary. The 2001 edition of the database covers 175 countries for the period 1963-1999. However, since period coverage as well as item coverage di¤er from country to country, I focus on a sample of 95 countries using the years from 1990 to 1999 inclusive. In the analysis I compute the index of vertical integration for every industry, country and year. Since most of the analysis does not exploit time variation, I use only the average of the variable of interest for the period between 1990 and 1999. Two sets of variables are added. I use data from the Census Bureau and from the input-output table of the United States to construct data on industry characteristics in the United States. In addition to these data, I use several institutional variables at the country level. Most of these variables are from the Doing Business database at WorldBank and have been developed by Djankov et al. (see e.g. (2002) and (2003)). The preferred measures of …nancial development come from the dataset discussed in Levine (2005). Table 1 shows descriptive statistics of the main variables used in the analysis. The sample includes 89 countries.

29

3.2

Di¤erential Cross-Country Propensities to Vertical Integration: Speci…c Channels

3.2.1

Basic Results

This subsection tests the empirical validity of the main predictions of the model by exploring whether country level variables have a di¤erential impact across industries. In doing so, I look for speci…c channels through which institutions a¤ect vertical integration. Linking country institutions to organizational form through speci…c channels also provides better grounds for establishing a causal relationship between institutions and organizational form. I run regressions of the following form IN Tic = where

i

and

c

+ Ii

Zc +

i

+

c

+ "ic

(15)

are a set of industry and country dummies, Ii are industry characteristics in the

United States and Zc are country level variables. Since the regression includes country …xed e¤ects, the estimate of the coe¢ cient

identi…es relative propensity towards vertical integration. For

example, the regression interacting external …nancial dependency with …nancial development tells us whether countries with more developed …nancial systems are relatively more or less integrated in industries that require more external …nance. It is important to stress that results should be interpreted in terms of relative degrees of vertical integration across industry within countries, and not in terms of di¤erential level of vertical integration across countries. The identi…cation strategy is closely related to the in‡uential work of Rajan and Zingales (1998).32 The key assumption is that characteristics of industries in the United States are representative of the same technological characteristics of industries in other countries. To clarify the assumption, consider a measure of external …nancial dependency. I assume that the measure of external …nancial dependency in the United States re‡ects technological features of the industries. These technological features do not need to be equal in other countries, although the ranking across industries does. For example, I do not require the …nancial needs of the textile industry in India to be the same as in the United States, but I require that if the textile industry is less dependent on external …nance than the glass industry in the United States, this relationship is also valid (on 32 The same identi…cation strategy is extensively used in the rapidly growing literature on industry performance and credit markets development as summarized in Levine (2005), and in more recent papers on the institutional determinants of trade (e.g. Nunn (2005)).

30

average) in India. For this reason I run regressions in this subsection substituting the industry level variable with its ranking in the United States. Since the model treats industries in isolation the contractual imperfections parameters

and

can be thought as industry-country speci…c. I proxy for contractual imperfections in industry i and country c using an interaction between an industry variable computed in the U.S. (capturing the need for contracts in either input or …nancial markets) and a country variable (capturing the development of speci…c contractual institutions). With respect to the e¤ects that contractual imperfections in the credit market have on the degree of vertical integration, the model predicts that Better credit markets allow more …rms to undertake the necessary investments to integrate vertically; However, better credit markets allow more …rms to enter the industry, and hence increase competition, and make vertical integration more di¢ cult; Finally, better credit markets allow more …rms to enter upstream industries. Since input markets become thicker, this reduces the incentives to integrate vertically. While the …rst e¤ect is positive, the second and third e¤ects are negative. However, the model predicts that the …rst e¤ect dominates the second in industries that for technological reasons employ a large share of workers in relatively large establishments, while the second e¤ect dominates the …rst one in industries that, for technological reasons, employ a large share of workers in relatively small establishments. Finally, for a generic industry i; the third e¤ect should work through the (average degree of) external …nancial dependency of all industries j 6= i supplying inputs to industry i: With the appropriate interactions between …nancial development and suitable industry characteristics in the United States, it is possible to disentangle the three di¤erent e¤ects. Table 2 focuses on credit markets, and disentangles the three e¤ects emphasized by the model. I proxy …nancial market development with the ratio of credit to the private sector over GDP. Column 1 shows that …nancial development has no di¤erential impact. This is consistent with the model, since the …rst and second e¤ects go in opposite directions. Column 2 and Column 3 capture the three e¤ects, and further con…rm the prediction of the model. The second e¤ect is captured in column 2 by further interacting external …nancial develop31

-.2

Better Credit Markets: Total Effect -.1 0 .1

.2

Figure 7: Di¤erential E¤ect of Better Credit Markets

0

5 10 15 20 25 Share of Employees in Establishments with < 500 Employees (Ranking)

ment with a measure of the relative importance of small …rms in the industry (the share of workers in establishments with less than 500 employees). I …nd a positive, direct e¤ect of …nancial development on vertical integration (the …rst coe¢ cient increases and becomes statistically signi…cant). However, the net e¤ect is weaker (and for some industries negative), for industries that employ a large share of workers in relatively small establishments. The coe¢ cient of the second interaction is negative, and statistically signi…cant. Figure 7 explains the di¤erential impact of better credit markets on vertical integration. For any given industry, the total e¤ect of the interaction between …nancial development and external …nancial dependency is given by the sum of the coe¢ cient in the …rst line of Table 2, plus the coe¢ cient in the second line multiplied by the ranking of the industry with respect to the share of employees working in establishments with less than 500 employees. The y-axis reports the appropriate linear combination of the estimated parameters, while the x-axis reports the corresponding ranking. The Figures also reports the appropriate interval of con…dence on the estimated total e¤ect of better …nancial markets. The total e¤ect is positive and statistically di¤erent from zero for the ten industries with the highest share of employees working in establishments with more than 500 employees. These industries include, for example, tobacco, iron and steel, transportation equipment and glass. The e¤ect is instead negative only for the 4 industries for which the share of 32

Figure 8: Quantitative E¤ect apparel footwear furniture professional goods electric machinery non metal products nonferrous metal glass machinery printing and publishing other chemicals wood products other industries rubber products paper and products leather metal products pottery food products plastic products iron and steel beverages transportation equipement textile tobacco -.1

-.05 0 .05 Percentage Change in Vertical Integration

Percentage Change=Effect of 1 st. dev. change in Financial Development

employees working in establishments with less than 500 employes is highest (wood manufacturing, leather, metal and non-metal products). Column 3 investigates whether …nancial institutions a¤ect vertical integration through a di¤erential impact on backward industries. Industries are di¤erent with respect to the input mix, i.e. use the products of other industries in di¤erential proportion. Using data from the Input-Output table in the United States, I construct for each industry in the sample an index of the average external …nancial dependency in industries from which the industry under consideration purchases inputs. If bad …nancial institutions are particularly detrimental to the development of industries that rely heavily on external …nance, I may expect higher degrees of vertical integration in industries that use a relatively high proportion of inputs from …nancially dependent industries.33 The third e¤ect emphasized by the model, is thus captured in column 3: industries that purchase inputs from industries that require more external …nance are relatively less vertically integrated in countries with higher …nancial development. The coe¢ cient is negative, and statistically signi…cant. 33

The use of Unites States Input-Output table is justi…ed by concerns that, because of various sources of input markets imperfections, industries in country c may substitute inputs in ways which are correlated with other determinants of vertical integration speci…c to country c.

33

What is the magnitude of the e¤ects we are identifying? Figure 8 provides an answer to this question. In Column 3 the country level measure of …nancial development a¤ects vertical integration through four di¤erent, and opposing, channels (three reported, plus the unreported interaction of the variable "small …rms" with …nancial development). Figure 8 reports, for each industry, the average percentage change induced in the index of vertical integration by an increase in the index of …nancial development of one standard deviation. In our sample, a di¤erence of one standard deviation in the index of …nancial development is equivalent to the di¤erence between the index of …nancial development in Algeria and South Korea (two standard deviation is somewhat smaller than the di¤erence between Algeria and France). Figure 8 identi…es that for some industries the net e¤ect is positive (e.g. textile and transport equipment) while for other industries is negative (e.g. footwear), and is in the order of 3-5% points.34 Columns 4 and 5 perform some preliminary robustness checks. Since …nancial development is strongly correlated with the country development stage, a …rst concern is whether the patterns in columns 2 and 3 are really due to …nancial institutions, or are instead capturing the e¤ects of other institutional characteristics correlated with development. To control for this possibility, Column 4 includes interactions of the industry variables with GDP per capita, a proxy of broader institutions. The relationship is robust, and GDP has no impact at all. It thus seems that …nancial development is not capturing the e¤ects of broader institutional context, as proxied by GDP per capita. Another concern is that, within industries, richer countries produce goods which are more similar to those produced in the US. In order to control for this possibility, Column 5 adds an interaction of vertical integration in the US and GDP per capita. This further interaction should partially account for the fact that, because of more similar technologies, richer countries tend to be relatively more vertically integrated in industries that are more vertically integrated in the United States. Column 5 shows that this further interaction is not signi…cant and does not a¤ect the results35 . 34

The e¤ect identi…ed in Figure 8 is the net e¤ect of four coe¢ cients working in di¤erent directions. This implies that each single e¤ect has a larger magnitude. Moreover, the inclusion of country …xed e¤ects prevents the identi…cation of the average e¤ect of …nancial development on vertical integration, which, as we have shown in table 1, is negative and quite large. The total e¤ect of a change of one standard deviation in the index of …nancial development implies a change in the ranking of industries in terms of vertical integration (for the average country) for 6 industries (footwear, machinery, professional goods, textile, iron and steel and metal products). 35 I have repeated the exercise using an index of investor’s rights as alternative measure of …nancial development. Statistical signi…cance is reduced, but the economic interpretation (the ratio of the coe¢ cient in line 2 and 1), and the magnitudes of the coe¢ cient, are not changed. Results available upon request.

34

3.2.2

Further Results

The results in Table 2 are not driven by omitted country characteristics that a¤ect vertical integration equally in all industries, or by industry characteristics that a¤ect vertical integration equally in all countries, since country and industry …xed e¤ects are included in the speci…cation. However, there may be omitted country characteristics that a¤ect vertical integration in some industries more than in others, and the interactions in the regressions are simply picking up these e¤ects. As noted above, one important concern regards cross-country di¤erences in product mixes within industries. Better …nancial markets may favor a switch towards products that require higher vertical integration, and this compositional e¤ect may be stronger in industries that rely more heavily on external …nance. Alternatively, …nancial development could simply be picking up the e¤ects of economic development, or other broader characteristics that happen to be correlated with it. Consider as an example how bad enforcement of contracts at the country level may push …rms towards vertical integration relatively more in industries that would otherwise rely on subcontracting. To the extent that our credit measures are correlated with contract enforcement, and our measure of technology at the industry level are correlated with reliance on subcontracting, our results could be biased. Moreover, the model emphasizes the di¤erential impact of better contractual institutions in speci…c versus generic (capital) input markets. For all these reasons, it is important to consider the role of contract enforcement institutions. This is done in table 3, in which I explore the robustness of the evidence in favor of the credit market story to the inclusion of additional controls considering contractual institutions. In particular, I ask whether the potential correlation between external …nancial development with contractual needs is responsible for the patterns in table 2. Table 3 shows that this is de…nitively not the case. I proxy contract enforcement with the number of procedures mandated by law or court regulation demanding interactions between the parties or between them and the judge, from Djankov et al. (2003). I interact this measure of contractual enforcement with measures of contractual intensity at the industry level in the U.S. I construct two measures to proxy for the contractual intensity of the industry (contractual needs and Her…ndahl index of input use) and use three di¤erent speci…cations. The …rst measure is a weighted average of the degree of speci…city of inputs used by the industry. I have combined information from the input-output table in the United States with the commodity classi…cation in Rauch (1999). Rauch (1999) classi…es goods according to the thickness of markets to procure 35

goods. At one extreme there are goods for which referenced markets exist. At the other extreme there are goods that are normally procured through speci…c contractual arrangements. I assume that industries that use higher shares of inputs classi…ed by Rauch (1999) as "di¤erentiated" rely more heavily on contracts, since these inputs are not readily available on markets. A higher value for the index represents higher contractual needs. I compute an index of speci…city at the 4-digit NAICS classi…cation industry level. I then use this speci…city index to compute a weighted average of input speci…city for each industry, at the same level of aggregation. I then take the median level of input speci…city within each industry classi…cation in the sample to obtain an index of contractual needs for the 26 industries in our sample. The second measure of contractual needs is the (negative of the) Her…ndahl index of input use. Starting from the input-output table in the United States, I construct for each 4-digit NAICS industry the Her…ndahl index of input use, in order to capture how heavily an industry depends on a (small) set of suppliers. Letting sij be the share of input use of industry i from industry j; the index is given by HIi =

2 j sij :

The rationale for using the Her…ndahl index instead of the number

of inputs used is that the number of inputs used would overestimate the importance of inputs that contribute only marginally to the production process. Instead I assume that industries that rely on a less concentrated set of suppliers are more exposed to hold-up problems, and thus require more contractual provisions to mitigate hold-up problems.36 Columns 1 and 4 show that the interactions with …nancial development are not simply picking up the e¤ects of contractual institutions. The results of table 2 are robust to the introduction of the interaction between contractual intensity and contract enforcement institutions. Moreover, better contract enforcement has a relatively stronger positive impact in industries that have higher contractual needs, when contractual needs are proxied with the Her…ndahl index of input use. In Columns 2 and 5, I add the cross-interactions between …nancial dependency and contractual enforcement, and …nancial development with contractual intensity. This is done in order to check whether …nancial markets development works through the appropriate channel, i.e. through external …nancial dependency. This is indeed the case. Interestingly, contract enforcement does not seem to work particularly through contractual needs, while instead the interaction with …nancial 36 With respect to the …rst measure of contractual needs, a similar procedure has been used by Nunn (2005). There is a positive correlation between the index of vertical integration and the variables contractual needs and Her…ndahl in input use for the 26 industries in our sample in the United States.

36

dependency is statistically signi…cant.37 Finally, in Columns 3 and 6, I test whether …nancial development and contract enforcement are picking up the e¤ects of broader institutions simply by interacting contractual intensity and external …nancial dependency with the GDP per capita. Results are again robust to this alternative speci…cation.38 Evidence is however mixed with respect to the role of contractual institutions interacted with contractual needs in the input markets. The model has an unambiguous prediction on the role of contractual institutions. Columns 1 to 3 support the prediction of the model, while columns 4 to 6 do not. Results thus hinge on the particular way contractual needs are measured. One possible interpretation for these …ndings goes as follows: …rms in countries with worse contract enforcement institutions may self-select into the production of simpler goods that require lower degrees of vertical integration. This unobserved substitution then reduces the e¤ect of poor contractual institutions on observed organizational form. This interpretation is consistent with the fact that the results for the contractual interaction improve when the interaction of contractual needs with GDP per capita is included in columns 3 and 6.39 Finally Table 4 investigates the relevance of further controls, in particular skills. There are several concerns here. A …rst concern is that higher …nancial needs could be correlated with higher skill intensity of the industry, and …nancial development with the availability of a skilled labor force. Besides that, our measure of vertical integration is quite closely related to value added, and could be in‡uenced by factor intensity or other determinants of productivity. Finally, these additional controls may be of some interest on their own, in providing some evidence on the relationship between human capital and vertical integration.40 37

This result makes sense since, as emphasized in the introduction, an important component of credit markets development relates to the development of contract enforcement institutions. 38 Contractual enforcement may also be measured with a measure of percentage costs needed to enforce a debt contract. When this is done, the contractual channel disappears, and the credit market variables improve their statistical signi…cance in all the speci…cation in Table 3. I have checked whether ethnic fragmentation and average level of trust in the society acted as substitutes for poorly functioning judicial systems by running the same set of regressions interacting the measures of contractual needs with social trust and ethnic fragmentation. As expected, we …nd a strong positive e¤ect of average level of social trust. Countries with higher level of social trust are relatively more vertically integrated in industries with higher contractual needs. The coe¢ cient on the interactions with ethnic fragmentation is negative, but not statistically signi…cant. Results are available upon request. 39 The theoretical section presents a transaction-costs-like theory of vertical integration. In a property rights framework however it is not clear that better contractual institutions would necessarily lead to lower integration. The prediction would depend on the relative improvement in investment incentives at the margin. 40 Vertical integration may require higher coordination of tasks, and this in turn may need workers with higher degrees of human capital.

37

I proxy skill intensity of the industry with the ratio of employees (non production workers) over total workers. Results are robust to the inclusion of these further controls. Moreover the interactions with the measure of skill intensity are signi…cant. Countries with higher levels of higher education are relatively more vertically integrated in industries that require a higher level of skills. I again perform the checks introducing the interactions of industry variables with GDP per capita and the cross interactions. Once more, results are robust to these alternative speci…cations. Interestingly, the e¤ect of skill intensity is not entirely captured by higher education.41 I also …nd that countries with higher …nancial development are relatively less integrated in industries that require higher skills. This points towards interesting interaction e¤ects of institutional variables, to be explored in further research.42

3.3

Further Robustness Checks and Measurement Issues

In this subsection I discuss some further robustness checks, and some issues relating to measurement error. The index of vertical integration su¤ers from some limitations arising from being computed at the industry level. In particular, I have not ruled out the possibility that our results are partially picking up the fact that, within industries, rich and poor countries are engaged in the production of di¤erent mixes of goods. Institutional variation could than be correlated with "what" is produced, and not only with "how " production is carried out. This concern may be particularly relevant to explain the lack of a strong relationship between the index of vertical integration and measures of contract enforcement in the data. Because of poor contract enforcement, within industries, …rms in less developed countries specialize in the production of goods that rely less on contractual enforcement. If this is the case, …rms in those countries may not be more vertically integrated simply because they engage in the production of goods for which non-integration does not reduce e¢ ciency.43 With more disaggregated data, one way of partially addressing this concern, would be 41

The interaction with GDP per capita is signi…cant, possibly due to the fact that higher education is strongly correlated with GDP pc., and in fact when skill intensity is measured with R&D expenditures, the interaction with GDP per capita eliminates the interaction with higher education. 42 Acemoglu et al. (2005) also …nds similar results. 43 Acemoglu et al. (2005b) however …nd that poorer countries have disproportionately more …rms in sectors that are more vertically integrated in the U.S. On the other hand, since their data include …rms in the service sector, their evidence does not rule out the possibility that within manufacturing sectors …rms substitute away from productions of good that require, for technological reasons, higher degrees of vertical integration. Another related problem of our data is that data are collected at establishment, rather than …rm, level.

38

to restrict attention to industries in which the product mix is relatively similar across countries. Since the manufacturing sector is classi…ed in only 26 industries, the choice of the sectors for which product mixes are not too di¤erent would be rather arbitrary. I pursue another route in Table 5. In Column 1 I add as additional controls the interactions of all the industry level variables used to estimate the coe¢ cients of interest, with GDP per capita. This is done in order to check whether results are driven by broader institutional characteristics correlated with …nancial development. Column 2 repeats the exercise, including all the previous industry variables with the square of the GDP per capita in case omitted institutional characteristics are not captured linearly by GDP. Results are robust. The magnitude of the coe¢ cients does not change signi…cantly, and statistical signi…cance is only marginally reduced. In Column 3, I add interactions between industry dummies and GDP per capita. This is done to capture the fact that industries are engaged in di¤erent production across countries, and more broadly that there may be broader omitted institutional factors that have di¤erential impact across industries, and that work through speci…c channels which are di¤erent from, but correlated to, external …nancial dependency. Results are again quite robust both in terms of statistical signi…cance, magnitude and interpretation of the coe¢ cients. Another concern with the index is its sensitiveness to the degree of intra-industry trade between vertically disintegrated …rms in the industry.44 Within industries, countries may di¤er in their product mixes. If the degree of intra-industry trade is correlated with other characteristics a¤ecting vertical integration through speci…c institutions, the coe¢ cients may be biased. In order to partially account for this issue, I further include interactions of industry dummies with GDP per capita and the industry level measure of intra-industry trade in the United States, which I compute from the input-output table. Results are again robust to the inclusion of these further interactions. Finally Columns 5a and 5b present the results for a pooled regression in which I allow the coe¢ cients that identify the credit channel to di¤er across two separate samples of countries. In 44

It can easily be shown that the extent to which intra industry trade generates measurement error depends on the dispersion of …rm size in the industry, but that if …rms use inputs in …xed proportions within the industry, intra-industry trade is not a source of measurement error. With a mild abuse of notation, assume that each …rm i in the industry purchases a proportion ij from …rm j 6= i in the industry, and uses inputs Ii from other industries. The Y [ index of vertical integration in industry and country c can be expressed as IN T c = 1 1 i ( j6=i ij j ) 1 i ( Ii ) N

Y

Yi

N

Yi

1 i Ii while the observed index is given by IN T c = 1 N1 i j6=iiYiij j . The measurement error thus depends on N i Yi the distribution of relative propensity to use inputs from the same industry. Assuming Ii = Yi and j6=i ij Yj = Yi ; it is easy to show that the two indexes give identical results.

39

particular, I split the sample into two groups: countries with GDP per capita above the median of the distribution of GDP per capita and countries below. The sample indicator is then interacted with all the credit markets channels. Results are again very robust, and coe¢ cient are larger for the group of richer countries. A second measurement issue is that the index is a weighted average (weighted by output) of the ratios of value added over output at the …rm level, i.e. it is not strictly speaking the measure of the average degree of vertical integration within an industry that was used to derive comparative statics from the model. In the Appendix, I show that the observed index of vertical integration can be written as [ IN T ic = IN Tic +

ic

where IN Tic is the index used to derive comparative statics in the model, while

ic

is an error term

related to the within-industry correlation between …rm size and vertical integration. Empirical studies …nd a positive correlation between vertical integration and measures of …rm size in the manufacturing sector, after controlling for industry …xed e¤ects. This source of measurement error biases our results if, the determinants of the within-industry correlation between …rm size and vertical integration in industry j and country c are correlated with our right hand side variables. The model in the previous section helps us to understand whether we should expect this source of measurement error to be (strongly) correlated with the independent variables related to credit markets or contractual enforcement. An explicit solution for the "weighted" index of vertical [ integration IN T jc can be derived under the assumption that G( ) is a Pareto distribution. When this is the case, we can show that Proposition 6 Assume G( ) = 1

1

: Then i)

ic

does not depend on

; and ii)

@ @

ic

<0

Proposition 6 has two important implications for the empirical analysis in the previous section. Under the assumption that …rm productivities are n from a Pareto distribution, the di¤erence between the weighted and the unweighted indexes of vertical integration depends on the relative e¢ ciency of non-integration versus vertical integration, v

and

e:

( ; ); and on the ratio of the thresholds

First of all, the measurement error does not depend on

; the degree of contractual

imperfection in the credit markets. In other words, the model itself suggests that the bias arising 40

from the correlation between the error term and the independent variables is likely to be very small.45 The second part of the proposition yields that the measurement error is instead systematically associated with the degree of contractual imperfections in input markets: higher contractual imperfections amplify measurement error. This suggests that the estimated coe¢ cient is likely to be biased upward, and the magnitude of the e¤ect in Table 3 should be interpreted with caution. The precise form of the bias due to measurement error does not explain why we do not …nd an e¤ect of contractual institutions when we measure contractual needs at the industry level with the index proxying input speci…city. Some of the issues raised by measurement error (at least those that have a more structural component, e.g. within industry composition of product mixes) could be accounted for by regressions including country-industry …xed e¤ects. I use data for several years (1990 - 1999). This enables me to run a panel speci…cation in which industry-country dummies are included. Some results from this speci…cation are reported in Table 6. In particular, I focus on the interaction of …nancial development, external …nancial dependency and the proxy for …rm’s size distribution. This is done since the measure of …nancial development, the ratio of bank credit over GDP, varies over time. The sign and economic interpretation of the interactions between the industry characteristics and …nancial development are highly robust to the inclusion of industry-country …xed e¤ects. The main concern with these speci…cations, are time trends in the variables. Column 1 does not include any speci…c trend. Columns 2 and 3 instead include industry and country trends separately. The coe¢ cients are smaller when industry trends are added. Finally, Column 4 includes the interactions of the industry variables with GDP per capita. While the coe¢ cients are barely una¤ected, statistical signi…cance is reduced.46 Overall, Table 6 shows that, while the statistical inference that can be n from the inclusion of industry-country …xed e¤ects depends on the treatment of the time component of the error term, the economic interpretation of the coe¢ cients is una¤ected.47 45

Moreover, it should be noted that, at least for the regressions investigating the role of institutions through speci…c channels, the identi…cation comes from the interaction between institutions and industry level variables in the United States. This should further reduce the correlation between the component of the error terms given by jc , and our independent variables. 46 However, it would be hard to imagine that changes from one year to the next in the country level variables contain high frequency variation predicting changes in vertical integration. For this reason I focus on what can be identi…ed through cross-sectional variation. 47 Finally, if value added is a nonlinear function of output, the index could simply pick up the e¤ects of institutions (and speci…c channels) on output. To partially address this concern I run all the regressions in the table 1 to 6

41

3.4

Further Evidence from Case Studies and Business History

Case studies from developing countries and American business history provide some micro level evidence on the subject. Here I focus on the evidence on the relationship between credit markets development and vertical integration. For ease of exposition and comparability I focus mainly on the textile industry.48 Banerjee and Munshi (2004) provide a very detailed study of the Tiruppur cotton industry. They analyze the e¤ects of better access to local capital markets by di¤erent groups of entrepreneurs on capital investment and vertical integration. Because of connections with the local elite, one social group has better access to local informal capital markets than newly migrated entrepreneurs. They conclude that di¤erences in credit market access translate into a clear di¤erence in the extent of vertical integration. The group of entrepreneurs with better access to local …nance are more vertically integrated and therefore have greater control over the production process.49 Porter and Livesay (1971) argue that in the early phase of industrialization, backward integration of rich merchants into the production stage represented an essential form of …nance. While their analysis includes a rich set of industries and business stories, the experience of the cotton textile industry in New England is especially informative. Porter and Livesay (1971) describe an environment which is very similar to that described by Banerjee and Munshi (2004).50 Imperfect credit markets can constitute a powerful barrier to entry, and may thus reduce comreplacing the index of vertical integration with the residual of a regression of vertical integration over interactions of industries and countries dummies with output. This set of interactions takes care of the possibility that our index of vertical integration simply picks up non-linearities in output, and I allow these non linearities to be industry speci…c. Results are not a¤ected by these robustness checks, and sometimes standard errors are improved. 48 The focus on the historical experience of the textile industry is not casual. As Haber (1991) pointed out, the cotton textile industry is a good setting to analyze the e¤ects of credit market imperfections on market concentration and vertical integration. In his comparative analysis of the textile sector in America, Brazil and Mexico, he notes that in the early textile cotton industry "the usual mechanisms by which …rms obtain market control were lacking" since "capital equipment was easily divisible" and "no signi…cant barriers to entry existed" in the industry. Moreover this is an industry with relatively simple production processes making it especially relevant for developing countries today. 49 Banerjee and Munshi (2004) …nd large di¤erences in productivity across the two groups of entrepreneurs: …rms in the credit constrained group, while less vertically integrated, they are more productive. While our model predicts that more productive …rms are larger and vertically integrated, it is possible to extend our theoretical framework to include heterogeneity in …rms access to credit. When this is done, the model perfectly matches Banerjee and Munshi (2004) …ndings. 50 Porter and Livesay (1971) interpret these …ndings as indicating that backward vertical integration may substitute for the lack of developed credit markets. The main problem with this conclusion is that the nineteenth century also witnessed the gradual development of capital markets and contract laws. At the same time no signi…cant trend towards a reduction in vertical integration has been observed, as, among others, La¤er (1969) and Tucker and Wilder (1977) have pointed out (quoted in Perry (1989)).

42

petition. This e¤ect can in turn reduce vertical integration, since …rms lose rents that can be used to integrate vertically.51 Beyond the cited work of Porter and Livesay (1971), further evidence on this point comes from a comparison of the textile industry in the nineteenth century England, United States and Germany. Temin (1988) notes that the large and vertically integrated …rms of the textile industry in New England were by no means more e¢ cient than the British textile industry. Interestingly, he notes that powerful interest groups in New England ensured very e¤ective trade protection to the textile industry. One can therefore argue that with respect to …rms in the relatively less concentrated and more competitive market in England, large American …rms enjoyed rents and market power. These rents may have been used to integrate vertically and acquire more control over the production process.52 Similar evidence can be found for the historical evolution of the automobile industry in the United States. Helper (1996) reports evidence that credit was a substantial problem in the organization and expansion at early stage of the industry, and therefore while the degree of vertical integration varied from …rm to …rm, virtually all automobile companies began as assemblers rather than manufacturers, implying a lower degree of vertical integration. Langlois and Robertson (1989) compare years immediately before and after the Depression, during which credit availability is likely to have changed signi…cantly. They report that in years before the big depression the industry was on an upward trend with respect to vertical integration, while during the depression the share of intermediate inputs purchased from external suppliers went up.

4

Conclusion

This paper uncovers novel patterns in cross-country di¤erences in the extent of vertical integration and on their institutional determinants, by focussing on the relative role of contract enforcement and …nancial market development. Contrary to conventional wisdom, I …nd some evidence of higher 51 There is evidence linking credit market imperfections to the concentration of industries, and more broadly on entry and …rm size (see references in Levine (2005)). One very interesting historical analysis is provided by Haber (1991). He provides a comparative analysis of the cotton textile industry in the nineteenth century United States, Brazil and Mexico. He …nds that Mexico and Brazil, with far less developed and more segmented capital markets, exhibited very high levels of concentration in the industry. At the turn of the century Brazil passed reforms which improved the working of the …nancial system. This led to a rapid growth of the textile industry in Brazil, and a reduction in the level of concentration. 52 Similarly Brown (1992) reports that the German textile industry, which in the late 19th century enjoyed a very high level of protection, was far more concentrated and integrated than its counterpart in England. Despite the high level of integration, the industry did not appear to be more productive.

43

vertical integration in developed countries. Furthermore, industries that are more dependent on external …nance tend to be relatively more vertically integrated in developed countries. I argue that these facts suggest that contractual imperfections in input and …nancial markets have radically di¤erent impact on vertical integration across industries, and that they are not consistent with existing theories of vertical integration. To explore these issues, I develop an industry equilibrium model of vertical integration with heterogeneous …rms. I introduce in this framework imperfect contracting in input and …nancial markets. The model predicts that better contract enforcement in speci…c input markets unambiguously leads to lower vertical integration. More interestingly, better contract enforcement in …nancial markets is associated with higher degrees of vertical integration in industries that are dominated by large …rms and lower degrees of vertical integration in industries that are dominated by small …rms. I test the predictions of the model using cross-industry-country data. The econometric evidence provides strong support for the predictions of the model with respect to the di¤erential impact of better …nancial markets on vertical integration across industries. I also …nd some evidence of better contract enforcement being associated with lower vertical integration in industries that are contract intensive. Much theoretical and empirical work remains to be done, both on the theoretical and on the empirical side. With respect to the theory, an important avenue for future research is to explore the general equilibrium implications of the di¤erent mechanisms underlined in the model of this paper, and their implications for cross-country patterns of industrial structure. In particular, the intra-industry e¤ects of institutions emphasized here (how institutions a¤ect "how " things are produced) may be o¤set by inter-industry e¤ects of institutions ("what" is produced). Along the same lines, it should be useful to consider the role of other institutional characteristics. For example, the theoretical prediction on the e¤ects of the average productivity of potential entrepreneurs on vertical integration, and the empirical …ndings on the interactions between proxies of human capital, skill intensity and …nancial development, suggest that the skills of the labor force may be a key determinant of organizational form decisions. These considerations immediately point towards the need for further work on the empirical front. More e¤ort should be devoted to the exploration of interactions between the institutional

44

characteristics considered in this paper and the role of other institutional variables such as, for example, trade openness and informal networks. Moreover, while the analysis of institutional determinants of di¤erential organizational forms across countries is interesting per se from the point of view of organizational economics, eventually it is of crucial importance to understand what can be learnt about the relationship between di¤erential organizational forms and productivity.

References [1] Acemoglu, D., P. Aghion and F. Zilibotti (2003) "Distance to Frontier and Vertical Integration", Journal of the European Economic Association Papers and Proceedings, vol 1, pp. 630-638. [2] Acemoglu, D., P. Aghion, R. Gri¢ th and F. Zilibotti (2004) Vertical Integration and Technology: Theory and Evidence, NBER working paper #10997 [3] Acemoglu, D., P. Antras and E. Helpman (2005a) Contracts and the Division of Labor, mimeo MIT and Harvard [4] Acemoglu, D., S. Jonshon and T. Mitton (2005b) Determinants of Vertical Integration: Finance, Contracts and Regulation, mimeo MIT [5] Adelman, M. (1955), “Concpet and Statistical Measurement of Vertical Integration,”in Business Concentration and Price Policy Princeton University Press, Princeton NJ. [6] Aghion, P. and J. Tirole (1994) ”On the management of innovation”, Quarterly Journal of Economics 109: 1185-1207. [7] Aghion P., M. Dewatripont and P. Rey (1994) "Renegotiation Design With Unveri…able Information", Econometrica 62(2): 257-282. [8] Andrabi, T., M. Ghatak and A. Khwaja (2004) "Subcontractors for Tractors: Theory and Evidence on Flexible Specialization, Supplier Selection, and Contracting", mimeo LSE [9] Antras, P. (2003) Firms, Contracts and Trade Structure, Quarterly Journal of Economics , 118, 1375-1418. 45

[10] Antras, P. and E. Helpman (2004) "Global Sourcing", Journal of Political Economy 112, 552580 [11] Baker, G., R. Gibbons, and K. Murphy (2004) "Contracting for Control." mimeo, Harvard Business School [12] Banerjee, A. (2004) "Notes Toward a Theory of Industrialization in the Developing World", mimeo MIT [13] Banerjee, A., Esther Du‡o, and Kaivan Munshi (2003) ”The (mis)allocation of capital”Journal of the European Economic Association 1(23), 484-494 [14] Banerjee, A. and K. Munshi (2004) ”How e¢ ciently is capital allocated? Evidence from the knitted garment industry in Tirupur”. Review of Economic Studies 71(1), 19-42 [15] Banerjee, A. and A. Newman (1993), "Occupational Choice and the Process of Development", Journal of Political Economy 101, 274-298. [16] Brown, J. (1992) "Market Organization, Protection, and Vertical Integration: German Cotton Textiles before 1914" The Journal of Economic History, Vol. 52, No. 2, pp. 339-351 [17] Burkart, M. and T. Ellingsen (2004) "In-Kind Finance: A Theory of Trade Credit", American Economic Review, 94 (3):569-590 [18] Carlton, D. (1979) “Vertical Integration in Competitive Markets Under Uncertainty”, Journal of Industrial Economics, 27: 189-209. [19] Djankov, S., R. La Porta, F. Lopez-de-Silanes and A. Shleifer (2002) "The Regulation of Entry", Quarterly Journal of Economics 117, 1-37 [20] Djankov, S., R. La Porta, F. Lopez-de-Silanes and A. Shleifer (2003) "Courts", Quarterly Journal of Economics 118, 453-517 [21] Faure-Grimaud, A. and R. Inderst (2005) “Conglomerate Entrenchment under Optimal Financial Contracting”, forthcoming American Economic Review [22] Fee, E., Hadlock, C. and S. Thomas (2005) "Corporate Equity Ownership and the Governance of Product Market Relationships", Journal of Finance, forthcoming 46

[23] Gerschenkron, A. (1962) Economic Backwardness in Historical Perspective, Harvard University Press, Cambridge MA. [24] Gibbons, R. (2004) ”Four Fornal(izable) Theories of the Firm ?”, mimeo MIT [25] Grossman, S. and O.D. Hart (1986) ”The costs and bene…ts of ownership: a theory of vertical and lateral integration”, Journal of Political Economy 98: 1119-1158. [26] Grossman, G. and E. Helpman (2002) "Integration vs. Outsourcing in Industry Equilibrium", Quarterly Journal of Economics 117, 85-120. [27] Grossman G. and E.Helpman (2004), "Managerial incentives and the international organization of production", Journal of International Economics 63: 237-262. [28] Grossman G. and E.Helpman (2005), "Outsourcing in a global economy", Review of Economic Studies 72: 135-159. [29] Haber, S. (1991), "Industrial Concentration and the Capital Markets: A Comparative Study of Brazil, Mexico, and the United States, 1830-1930”, Journal of Economic History, 51: 559-580. [30] Hart, O.D. (1995) Firms, Contracts and Financial Structure (Oxford University Press, London). [31] Hart, O. and J. Moore (1990) "Property rights and the nature of the …rm", Journal of Political Economy 98: 1119-1158. [32] Helper, S. and D. Hochfelder (1996) "Joint Product Development in the Early American Auto Industry", Business and Economic History, Winter 1996. [33] Hortaµcsu, A. and C. Syverson (2005) "Cementing Relationships: Vertical Integration, Foreclosure, Productivity and Prices", mimeo Chicago University. [34] Inderst, R. and H.Müller (2003) “Internal vs. External Financing: An Optimal Contracting Approach”, Journal of Finance, Volume 58: Issue 3, 1033 - 1062. [35] Khanna, T., and K. G. Palepu (1997), "Why Focused Strategies May Be Wrong for Emerging Markets", Harvard Business Review 75:4, 41-51. 47

[36] Khanna, T. and K. Palepu (2000), "Is Group A¢ liation Pro…table in Emerging Markets? An Analysis of Diversi…ed Indian Business Groups", Journal of Finance 55, 867-891. [37] Kumar, K. B., R. G. Rajan and L. Zingales (2001), “What Determines Firm Size”, University of Chicago, mimeo. [38] La Porta, R. , F. Lopez-de-Silanes, A. Shleifer and R. W. Vishny (1998), "Law and Finance", Journal of Political Economy, 106: 1113-1155. [39] Langlois, R. and P. Robertson (1989) "Explaining Vertical Integration: Lessons from the American Automobile Industry" The Journal of Economic History, Vol. 49, No. 2, pp. 361375 [40] Legros, P. and A. Newman (2004) ”Competing for ownership” ULB and UCL, mimeo. [41] Levine, R. (2005) “Finance and Growth: Theory and Evidence.” forthcoming in P. Aghion and S. Durlauf, eds. Handbook of Economic Growth. The Netherlands: Elsevier Science. [42] Levy, B. (1990) "Transactions costs, the size of …rms and industrial policy : Lessons from a comparative case study of the footwear industry in Korea and Taiwan" Jounral of Development Economics, vol. 34, issue 1-2, pages 151-178. [43] Livesay, H. and P. Porter (1969) "Vertical Integration in American Manufacturing, 1899-1948" The Journal of Economic History, Vol.29, No.3, September, pp.494-500. [44] Macchiavello, R. (2004) "Investors Protection and the Boundaries of the Firm", mimeo LSE. [45] Marin, D. and T. Verdier (2002) "Power Inside the Firm and the Market: a General Equilibrium Approach", mimeo DELTA. [46] Maskin, E. and J. Tirole (1999) “Unforeseen Contingencies and Incomplete Contracts”Review of Economic Studies, 66:1, pp. 83-114. [47] Melitz, M. (2003) The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity, Econometrica 71, 1695-1725 [48] Nunn, N. (2005) Rekationship Speci…city, Incomplete Contracts and the Pattern of Trade, mimeo UBC. 48

[49] Perry, M. (1989) "Vertical Integration: Determinants and E¤ects" in Handbook of Industrial Organization, North-Holland, 1989. [50] Porter, R. and J. Livesay (1971) "Merchants and Manufacturers: studies in the changing structure of nineteenth-century marketing" Baltimore, Md. Johns Hopkins Press. [51] Rajan, R. and L. Zingales (1998) Financial Dependence and Growth, American Economic Review 88, 559-586. [52] Rauch, J. (1999) "Networks Versus Markets in International Trade," Journal of International Economics 48 June: 7-35. [53] Schmitz, H. (1995) " Small shoemakers and fordist giants: Tale of a supercluster", World Development Pages 9-28. [54] Stein, J. (2004) "Agency, Information and Corporate Investment", forthcoming in eds. G. Constantinides, M. Harris and R. Stulz. Handbook of the Economics of Finance, Amsterdam: North Holland. [55] Stigler, G. (1951) “The Division of Labor is Limited by the Extent of the Market.”Journal of Political Economy 59: 195-93. [56] Temin, P. (1988) "Product Quality and Vertical Integration in the Early Cotton Textile Industry", The Journal of Economic History, Vol. 48, No. 4, pp. 891-907 [57] Williamson, O (1971) "The Vertical Integration of Production: Market Failures Considerations", American Economic Review, Vol. 61. [58] Williamson, O. (1975) "Markets and Hierarchies: Analysis and Antitrust Implications", Free Press, New York. [59] Williamson, O. (1985) "The Economic Institutions of Capitalism", Free Press, New York. [60] Whinston, M. (2003) "On the transaction costs determinants of vertical integration", Journal of Law, Economics and Organizations, 19: 1-23. [61] Woodru¤, C. (2002) "Non-contractible Investments and Vertical Integration in the Mexican Footwear Industry", International Journal of Industrial Organization 20, 1197-1224. 49

5

Appendix A

5.1

Proof of Lemma 1 and Discussion of Financial Constraints

Let us …rst consider the case of an entrepreneur borrowing K units of capital and signing a contract in which she commits to repay B out of her (variable) pro…ts fraction 1

( ): Of the K units of capital, a

has to be invested in the project, since the investors can perfectly monitor such

investments. The remaining amount

K can either be invested, or it can be diverted by the

entrepreneur. If the entrepreneur invests, she generates revenues

( ); and she repays B: If instead

she diverts cash, she obtains K: She abstains from diversion if and only if K

( )

B: When

this inequality is satis…ed, the entrepreneur repays the external investors with probability equal to one, since there is no uncertainty in production. Since external investors are risk neutral and on the long side of the market, B = K. The former inequality can be rewritten as (1 + )K

( )

In this environment, an entrepreneur does not have the incentive to borrow more than what is required to …nance the …xed costs to start production, and hence without loss of generality one can consider K = f + k for a vertically integrated …rm and K = f for a …rm entering the market as assembler. This proves the result for a vertically integrated …rm. I now turn to the case of a non-integrated …rm. The sequence of events is as follows. First the …nal assembler …nances the …xed costs f borrowing from the external investors, issuing an amount of debt equal to B = f . Then she is matched with an upstream supplier. Since suppliers are on the long side of the market, they compete in order to attract customers. Since they have deep pockets, they o¤er an ex-ante transfers T ( ) to an assembler with productivity . Ex-ante competition among suppliers, implies that the ex-ante transfers drive their pro…ts to zero. Ex-post, the match realizes revenues R( ); and the supplier retains a fraction

of these revenues. Denoting by C(I) the costs of producing the intermediate

input, ex-ante competition among suppliers implies that T ( ) + C(I) = R( ): External investors hold claims on the assembler’s ex-post pro…ts (1 on (1

)R( ) + T ( ) = R( )

C(I) =

o(

)R( ) and on the ex-ante transfer M ( ); i.e.

): This completes the proof for a non-integrated …rm.

50

I now brie‡y consider the case in which suppliers are also …nancially constrained. Consider the following timing of events. First the …nal assembler …nances the …xed costs f borrowing from the external investors, issuing an amount of debt equal to B = f . Once these …xed costs have been paid she is matched with an upstream supplier. The assembler has ex-ante bargaining power and asks for an ex-ante transfer M from the supplier: The supplier is liquidity constrained and thus needs to borrow M from external investors. The upstream supplier borrows an amount e She can either transfer M to the …nal assembler, in which case she of cash M; issuing debt B: enters a relationship with the assembler, and she will receive a share R B if

R

of the ex-post surplus

e Alternatively, she can divert the cash M: The upstream supplier abstains from diversion B: f

e B

M: In any subgame perfect equilibrium the latter inequality is satis…ed, and

e = M: The supplier can the supplier repays her debt with probability equal to one, implying B

thus borrow up to M

M =

1+

(R

f ): Since the assembler has ex-ante bargaining power, in

equilibrium the supplier borrows up to her limit M : The …rst implication is that in a subgame perfect equilibrium in which the assembler repays the debt B = f; the …nal payo¤ Vd ( ) of an assembler with productivity

and generating revenues R( ) is equal to Vd ( ) = (1

e = M we obtain Vd ( ) = and since M = B

residual claimant of a share constraint takes the form

1 1+

1 1+

(

1 1+

(

o(

)

)(R( )

f

e + M; B)

f ): The …nal assembler e¤ectively is the

< 1 of the total pro…ts generated by the relationship. The credit

o(

)

f ) > f . Contractual imperfections in …nancial markets are

ampli…ed by the fact that the …nal assembler is e¤ectively the residual claimant of only a fraction 1 1+

of the total pro…ts, due to …nancial constraints of upstream suppliers. Finally, I have framed the discussion using the premise that in order to integrate vertically, a

…rm must acquire a new machine. However, results are qualitatively similar if the …rm acquires a supplier. Under this scenario, the supplier could agree to be paid once revenues have been generated, thus relaxing the …nancial constraint of the …rm. The formulation in the text is equivalent to a case in which the supplier has no monitoring advantage over external investors. While recent contributions in the trade credit literature (see e.g. Bukart and Ellingsen (2004)) have emphasized di¤erent reasons for why suppliers may have a monitoring advantage over external investors, results would not be a¤ected by allowing for these di¤erences in our framework.

51

5.2

Derivation of Pro…t Functions and Related Results

Under vertical integration, the …rm chooses investments I to maximize pro…ts ) = A1

v(

I

C(I)

(16)

Since all elementary investments x(i) are symmetric, and pro…ts are a concave function of x(i); the R1 …rm optimally sets x(i) = x; for all i 2 [0; 1]: The intermediate input becomes I = exp 0 ln xdi = x; and hence pro…ts can be rewritten as

) = A1

v(

x

x

The …rst order condition with respect to x yields "

x( ) = Substituting into the pro…t function yields

v(

A

"

"A

)=

"

(1

) ; which is the expression in the

text. I now turn to the pro…ts of a non-integrated …rm. Denoting xc the contractible investment, and xn the non contractible investments, pro…ts can be written as

o(

) = A1

xc xn(1

)

xc

(1

)xn

(17)

The sequence of events is as follows. First, …rms contract on contractible tasks. Second, the upstream …rm take the non contractible investements decision as given, and, anticipating ex-post bargaining, maximizes with respect to xn her share of pro…ts. I solve for the subgame perfect equilibrium. The …rst order condition for the upstream …rm gives xn = (

)1

1 (1

1 )

A1

(1

52

)

1

(1

)

xc1

(1

)

Substituting this expression back into the pro…t function yields 1

o(

) = A1

(1

)

1

(1

)

xc1

(1

)

(

(1 ) (1 )

)1

(1

(1

)(

))

xc

(18)

The contract, anticipating the choice of xn picks up the optimal xc : The …rst order condition is gives "(1

xc =

1

(1

(1

))

A

)

"

(

)

(1

)"

(1

(1

))"(1

)(

(1

))

(19)

) ( ; )

(20)

and by further substitution in the pro…ts function, I obtain o(

When

)=A

"

"

(1

(1 (1

! 0 we obtain lim

o(

!0

while when

11 1

"

)

"

(1

)=A

"

)=A

! 1 we obtain lim

o(

!1

"

) )

+1 "

=A

)(

"

)

(1

"

(1

"

)

which are the pro…ts of a vertically integrated …rm. I …nally prove that pro…ts are monotonically increasing in : Taking the logarithm of the pro…t function, I obtain @ log sign @

Denoting

( ; )=

sign

1 (1 ) 1 (1 )

"

o(

+1

)

@ log = sign

"

1 (1 ) 1 (1 )

+1

@

; and taking the derivative with respect to ; gives

d (log ( ; )) = sign d

ln

1 1

(1 (1

) )

(1 1

) (1

)

0

where the inequality follows from the fact that

sign Note in fact that

@ 2 log ( ; ) = sign 1 @ @

@ 2 log ( ; ) @ @

< 0 implies that

1

d log ( ; ) d

53

1 (1

)

<0

reaches a minimum in

= 1; i.e. when

d(log ( ; )) d

= 0; and thus is positive everywhere else. I have proved that

this observation with the fact that lim

5.3

o(

!1

"

)=A

" (1

@

o(

)

0: Combining

@

) proves Lemma 2 in the text.

Proof of Proposition 3 f (k+f )

The condition

( ; ) ensures that

Since A is in equilibrium a function of

v

>

v

and

e,

e;

and hence that the equilibrium is interior.

the two thresholds

v

and

e

de…nes a system

of two equations in two unknown. Unicity of the equilibrium follows from the fact that the ratio v e

=

(1 ) k f (1 ) (1

"

)

"

> 1 is constant, and that, by totally di¤erentiating the expression for

e;

we obtain

e

d

e

= =

"

f A( e ; "

) " (1 f " ) (1 )

v) (

(

()

)

(21)

"

A( e ;

v)

"

[

@A( e ; @ v

v)

d

v

+

@A( e ; @ e

v)

d e]

which can be rewritten as d d since

@A( e ; @ v

When

v)

> 0 and

f (k+f )

>

@A( e ; @ e

v)

e

KA( e ;

=

v)

(1 + KA( e ;

v

> 0; if

( ; ) instead

v

<

dG( ) d e;

" @A( e ; v ) @ v @A( e ; " v) @ e

5.4

<0

(22)

> 0:

and only vertically integrated …rms enter the industry.

The unicity of the equilibrium follows from the fact that an increasing function of

v)

v

is decreasing in A; and that A is instead

v:

Proof of Proposition 4

Consider the (unweighted) average level of vertical integration in the industry given by the index IN T = 1

1

54

1 1

G( v ) G( e )

(23)

To prove the …rst part of the proposition, simply note that by taking the derivative of IN T w.r.t. M and denoting (1

)(M

@IN T @

sign

(1 =

1) = ; we obtain

@ @

)+

sign

v

[ + (

1)]

e

[ + (

[ + (

1)] (1

v

)+

@ @

e

1)]2

e

Hence @IN T sign @

1

1

1

1

0 ()

It is easy to check that lim

(1 (1

1 !0

1

1) 1)

1

1 @A vA @

+1

1

1 @A eA @

+1

1 @A +1 vA @ 1 @A +1 eA @

= lim

1 1

(1 (1

1 (1 !0 1 (1

) ) ) )

v e

v e

= 1: On the other

+1 ( 1 1) v A1 @A ) @ v = ve < lim !0 11 (1 = ve 11 !0 1 1 @A (1 ) e 1 ( 1) e A @ +1 To prove the second part of the propostition simply note that by taking the derivative w.r.t. 1

hand lim

we obtain @IN T sign = @ since

@ @

v

> 0 and

@ @

e

g( e ) @@ e [1 [1 G( e )] [1

g( v ) @@ v

sign

[1

G( e )]

G( v )] <0 G( e )]

< 0:

To prove the third part of the proposition, note that taking the derivative w.r.t. [ @ IN T @ First we note that d v d

=

@ @

v

+

@ v @A @A @

d d

e

implies

=

@ @

d v d

=

e

0 ()

+

@ e @A @A @

1

1

d e d d v d

g( e ) g( v )

implies

d e d d v d

e

G( e ) G( v ) 1

=

1 @A [email protected]

1 1+

v

d d

1 1

1

we obtain

(24)

1 @A [email protected]

1 1+

e

and similarly

; hence

e

=

v

Moreover, imposing that G( ) is distributed according to a generalized Pareto distribution with mean

and shape parameter , i.e. G( ) = 1 @IN T @

0 ()

e (1

(1 +

( (1

1 (1

) 55

1

)

)

(1

1) ) 1

1

; we obtain 1

(1

)

1

)

v

(25)

Since

5.5

v

>

e;

1

the inequality is satis…ed if and only if 1

(1

)

1

> 0; i.e. if 1

1

> :

Proof of Proposition 6 and Related Results

I …rst derive an expression for

ic

in the text. Consider …rm f in industry j in country c; and

de…ne the index of vertical integration at the …rm level as intf jc =

V Af jc Yf jc where

V Af jc and Yf jc

stem for value added and output respectively. Assuming that the industry is composed of N …rms

[ vertical integration is IN T jc =

N

N

1 N

f 2 f1; 2; :::; N g we would like to measure IN Tjc =

f =1

intf jc : Instead, the observed index of

N

f =1

V Af jc =

f =1 N

[ industry level of variable X; we have that IN T jc = Y [ IN T jc = IN Tjc +(intf jc IN Tjc ) ( Yf jc

1 N ):

jc

Yf jc :Denoting with X the aggregate at the

V Af jc Yf jc : f =1 Yf jc Y jc

In the text I set

jc

Some further simple algebra gives =

Y

f

(intf jc IN Tjc ) ( Yf jc jc

1 N ):

I now turn to the proof of proposition 6. 1

Assuming a Pareto distribution, i.e. G( ) = 1

; the model implies that the unweighted

index of vertical integration in the industry can be computed as

IN Tic = 1

2

1

v

41

3 5

e

(26)

The weighted industry level index of vertical integration observed in the data is instead given by "

[ IN T ic = 1

1

R1

Rv dG( ) R v R1 R ( )dG( ) + Rv dG( ) o e v v

#

(27)

Substituting in this equation the expression for the revenue functions of integrated and nonintegrated …rms, we obtain [ IN T ic = 1

It is well known that if

1

( ; )

R

v e

R1

"

dG( ) v R1 " dG( ) + v

"

dG( )

!

is distributed according to a Pareto with shape parameter ,

distributed according to a Pareto with shape parameter

56

1

"

is also

": In this case the index can be

rewritten as

0

1

B @1

[ IN T ic = 1

Finally note that

1

( ; ) + (1

1

ic

[ = IN T ic

e

( ; ))

0 B @

v

IN Tic =

+ "

v e

+ "

v e

1 C A 1

" v e 1

( ; ) + (1

( ; ))

+ "

v e

C 1A

(28)

which is the expression derived in the text. To prove the second part of the proposition, we have

sign

@ @

0

ic

= sign

[email protected]

1+ "

( ) +(1

)

1+ "

f +k f

d

1

1A

d d

which, after some algebra, gives

sign

5.6

@ @

ic

=

sign

"

1+

"

k f

(1

+

) 1

1

+

+

k f

"

<0

Derivation of Pro…t Functions with Labor

I derive the pro…ts functions of both organizational form for the more general production function

q( ) = ( For simplicity, I focus on the case for the case in which

L 1

)1

I ( )

(29)

= 0: The corresponding expressions in the text are then derived

! 1: The demand equation implies that for each …rm, q( ) = Ap

"

()

57

A1 q( )1

= p( )

The revenue function for each …rm

becomes

R( ) =

Let C(I) =

R1 0

A1 q( )1

q( ) = A1

q( )

x(i)di be the cost of producing intermediate input I:

Under vertical integration, the …rm chooses investments in L and I to maximize pro…ts v(

) = A1

(

L

(1

)

1

)

I ( )

L

C(I)

Since all elementary investments x(i) are symmetric, and pro…ts are a concave function of x(i); the R1 …rm optimally sets x(i) = x; for all i 2 [0; 1]: The intermediate input becomes I = exp 0 ln xdi = x; and hence pro…ts can be rewritten as v(

) = A1

(

L

)

1

(1

)

x ( )

L

x

The optimal investment x and labor demand are functions of the productivity : The system of …rst order conditions with respect to L and x respectively gives

F:O:C: =

8 < :

A1

) ( L( ) 1

(1

) ( x( ) )

A1

) ( L( ) 1

(1

) 1 ( x( ) )

1

=1 =1

implying the standard Cobb - Douglas result 1 L( ) = x( ) Substituting into the pro…t function yields v(

) = A1

and from the …rst order condition, we obtain v(

)=

( "A

"

A

58

x( )

"

)

x( )

= ( x( ) ); which …nally gives pro…ts "

(1

)

(30)

which is the expression in the text. Note that the …rm employs L( ) = (1

)

"A

"

units of

labor. The size of the …rm is thus increasing in the productivity parameter : I now turn attention to the (sum of the) pro…ts generated by a specialized assembler and her supplier. For simplicity, I assume that the decision regarding labor investments is contractible. The sum of pro…ts is given by o(

) = A1

(

L

(1

)

1

)

x ( )

L

x

With respect to the case of vertical integration there are two di¤erences. The investment decision over x is taken by the upstream …rm, and can not be speci…ed in an ex-ante contract. The upstream supplier, anticipating ex-post bargaining, knows that she will only capture a fraction of the surplus generated by the investments. Secondly, since it is possible to write contracts on L; investments in L will take into account the investment in x choosen by the upstream …rm. The …rst order condition for the upstream …rm gives x( ; L)

=(

)1

1

1

A1

(

1

L

)

1

(1 1

)

where I make explicit the fact that the investment of the supplier depends on the productivity parameter, and on the contractually speci…ed amount of labor. Substituting this expression into the pro…ts function yields o(

)=(

1

)1

A1

(

1

L 1

)

(1 1

)

(1

)

L

Taking the …rst order condition with respect to L yields 1 1

(

1

)

"

A

"

)(1

(1

)"

=(

L( ) ) 1

and …nally yielding pro…ts o(

)=A

"

(

1 1

)(1

59

)"

( )

"

"

(1

)

(31)

Taking the limit for

! 1 we obtain the expresion in the text, "

lim

!1

Note that

@ o( ) @ "

=

(1 (1

) )

( )

o(

)=A

" @ v( ) @ "

"

(1

@ v( ) @ " ;

)

since (1

)( )

"

is incresasing in ; and is

thus bounded above by 1: Moreover, as for the case of vertical integration, the optimal labor demand L( ) is increasing in the productivity parameter : The distribution G( ) induces a distribution in …rms size, when …rms size is proxied by employees.

6

Appendix B

6.1

Data Description

Data on industries in the manufacturing sector (at the 3-digit code ISIC second revision ) are available from UNIDO database. Data at the country level come from di¤erent sources. Data on GDP per capita and total GDP are from the World Penn Tables. Institutional variables are from the Doing Business dataset from the World Bank. Measures of …nancial development are from Levine et al. (2002). Measures of educational level are from the Barro and Lee (2001) dataset. The measure of external dependency in the United States is taken from Rajan and Zingales (1998). The variable small …rms (the share of employees working in establishments with less than 500 employees in the United States) is taken from the US Census of Manufacturing, and are reported in Beck et al. (2004). External Financial Dependency: is the average over the ’80s of the ratio EDi =

Cap Exp Cash Flows Cap Exp

for the median …rm in each industry. Data are from Compustat. I use the values reported in Rajan and Zingales (1998). Small Firms: is the share of employees working in establishments with less than 500 employees in the United States. Contractual Needs (1) is a measure of contractual needs computed as follows. From the

60

input-output table in the United States I compute for each (SIC 87 4-digit) industry i the share of inputs purchased from other industries j 6= i in the manufacturing sector, shij =

uij j6=i uij

. I use the

classi…cation in Rauch (1999) to compute a measure of speci…city at the industry level. I match the classi…cation in Rauch (1999) with the input-output table. For each (SIC 87 4-digit) industry i I compute the average degree of speci…city as the fraction si of products produced in that industry that are classi…ed by Rauch as speci…c. Contractual needs is than given by CN (1)i =

j shij

sj

I then aggregate at the ISIC code 3-digit level by taking the within industry median value. Contractual Needs (2): is constructed as Contractual Needs (1), but considers the Her…ndhal index for the use of inputs. The index is thus CN (2)i =

2 j shij

External Financial Dependency of Backward Industries: I construct an average measure of External Financial dependency for the backward industries of industry i: Denoting by EDj the fij the share of inputs external …nancial dependency in (3-digit ISIC code) industry j and by sh

purchased by (3-digit ISIC code) industry i from other (3-digit ISIC code) industries j; the measure is given by Ext_Dep_Backi =

6.2

f

j6=i shij

EDj

Cross-Country Di¤erences in Vertical Integration

In this subsection I discuss the results of estimating regressions of the form IN Tic =

+ Zc +

i

+ "ic

(32)

where IN Tic is the index of vertical integration in industry i in country c; Zc is a vector of country level variables and

i

is a set of industry dummies. Since I am primarily interested in the coe¢ cient

; I correct the standard errors adjusting for clustering by country. To provide a control for the average size of establishments in the industry, a measure of average output per establishment is

61

also included as additional control in the regressions.53 Results are reported in Table A1. Column 1 con…rms the …ndings presented in Figure 1. Once industry …xed e¤ects are controlled for, there is no evidence of higher vertical integration in less developed countries. In column 1 the GDP per capita enters positively and is statistically signi…cant.54 If …rms integrate vertically in order to exploit economies of scale, countries with larger markets may have more vertically integrated …rms. Column II addresses this issue by including the logarithm of population as proxy for domestic market size. The coe¢ cient is positive, but is not statistically signi…cant.55 Columns 3 and 4 explore the relationship between …nancial development and vertical integration. Financial development may a¤ect vertical integration through several channels. First of all, vertical integration may be considered as an additional investment that has to be …nanced. Better credit markets would then be associated with higher degrees of vertical integration. However, better …nancial markets may foster competition, and thus reduce the possibility of …rms to integrate vertically. Moreover, better credit markets can also foster the development of certain key backward industries, reducing the need for vertical integration in downstream industries.56 I use two measures of …nancial development. In Column 3 I proxy …nancial market development with the ratio of credit to the private sector over GDP. The coe¢ cient is negative, large, and highly signi…cant. The coe¢ cient on GDP per capita remains positive, and is statistically signi…cant.57 In Column 4, I proxy …nancial market development with an index of the degree of protection of creditors rights, from the Doing Business database at World Bank. Creditor’s Legal Rights measures the degree to which collateral and bankruptcy laws facilitate lending.58 While the sign of 53

The UNIDO dataset contains information on employees at the industry level. I can thus control for the average size measured as the average number of employees and for output per worker. Results are not a¤ected by the addition of these controls. However, because of pervasive missing values in the employees variable, I focus in the remaining part of the paper on regressions only controlling for average output per establishment. Results for the alternative speci…cation are available upon request. 54 The relationship between the level of development and the degree of vertical integration could be non-linear. For example, constraints originating from the underdevelopment of backward industries could push towards vertical integration only after a certain level of development. However, I do not …nd evidence of a non-linear relationship between the income of the country and the average propensity to integrate vertically. 55 Since in this speci…cation the e¤ect of country level variables is assumed to be identical across industries, estimates an average country-level propensity to vertically integrate. Since I do not include country …xed e¤ects, I interpret the results as correlations. 56 Finally, to the extent that …rms, at least partially, pursue a strategy of vertical integration to create internal capital markets substituting for poorly functioning (external) capital markets, better capital markets may be further associated with less vertical integration. 57 Similar results are obtained when …nancial development is proxied with the ratio of bank assets over total …nancial assets. 58 The methodology is developed in Djankov et. al (2005) and adapted from La Porta et. al. (1998).

62

the coe¢ cient goes in the same direction as the previous measure of …nancial development, it is not statistically signi…cant. Overall, I …nd evidence that countries that are more …nancially developed are less vertically integrated, although the statistical signi…cance of the results depends on the precise way …nancial development is measured.59 Column 5 explores the role of contractual enforcement. As emphasized in the theoretical section, vertical integration is a response to contractual imperfections. Countries with better contract enforcement have relatively less integrated …rms, since transactions across …rms are mediated by relatively more e¢ cient contracts. I proxy contract enforcement with the number of procedures mandated by law or court regulation demanding interactions between the parties or between them and the judge, from Djankov et al. (2003). I …nd some evidence that countries with worse enforcement of contracts tend to be more vertically integrated, however the coe¢ cient is not statistically signi…cant.60 Finally, in Column 6 I consider both the role of contract enforcement and …nancial development, and I also include controls for entry barriers and human capital. Arti…cial entry barriers (e.g. those induced by excessive regulation) may reduce competition, enabling surviving …rms with market power to integrate vertically. Using the Doing Business database, I measure entry barriers as the number of entry procedures, following the methodology in Djankov et al. (2002). I …nd no evidence that entry barriers a¤ect vertical integration. I measure human capital with average years of higher education in the adult population. If vertical integration requires higher coordination, and coordination requires higher skills, …rms will tend to be relatively more vertically integrated in countries with higher supply of skilled workers. I …nd that countries with higher levels of education are more vertically integrated, but the coe¢ cient is not statistically signi…cant. Column 6 also shows that the …nancial development variables remains highly statistically signi…cant. I have run similar regressions (results are omitted, but available on request) considering the role of trade openness, labor regulation, ethnic fragmentation and antitrust policies. The degree of openness to trade may a¤ect vertical integration giving …rms access to larger inputs markets and increasing competition on domestic markets. Both channels suggest that higher trade openness should be associated with less vertical integration. I …nd that openness to trade (measured as the 59 I have tried other measures such as bank concentration and creditor’s information indexes, or costs of forming collateral, obtaining very similar results. Acemoglu et al. (2005b) …nd very similar results on a di¤erent dataset. 60 The Doing Business database contains other measures of contract enforcement. I obtain very similar results with these alternative measures of contracts enforcement.

63

sum of imports and exports over GDP) is negatively associated with vertical integration. Strict labor regulation is often considered as one important constraint on …rm growth in less developed countries. In order to avoid strict labor regulation, especially with respect to …ring workers, …rms may decide to operate on a smaller scale. To the extent that …rm size is correlated with vertical integration, or that there are technological limits to small scale operation, labor regulation will reduce vertical integration by making commitments to internal supply relatively more costly (e.g. because of lack of ‡exibility in an uncertain environment). However, the index of labor regulation does not appear to be correlated to vertical integration. Countries with higher levels of trust may have relatively less integrated …rms, if other social networks provide informal ways of enforcing contracts. Where formal contracts cannot provide adequate incentives, …at or trust may substitute for the lack of enforcement. I …nd that ethnic fragmentation (which is negatively correlated with trust) is positively associated with vertical integration. On a much smaller sample, worse antitrust institutions are associated with more vertical integration. None of these coe¢ cients, however, is statistically signi…cant. Similarly, a regression simultaneously considering all the institutional variables does not change the main conclusions. To summarize, with the exception of …nancial development, the results in this section do not show strong patterns relating country level variables and institutions with the degree of vertical integration. Interestingly, using a di¤erent dataset, Acemoglu et al. (2005b) …nd that less developed countries are more vertically integrated. However they …nd that this result is entirely driven by industrial composition. Once industry composition is taken into accout (with the inclusion of industry …xed e¤ects, as it is done here), contracting costs, entry barriers and …nancial market development are not strong predictors of vertical integration patterns. The model in section 2 and the empirical evidence in section 3 suggest another explanation for the lack of correlation between (most of the) institutional variables at the country level and vertical integration. Institutional characteristics at the country level have di¤erential impacts across industries, which cancel out when the same average e¤ect across industries is imposed.61 61

My data do not allow for a careful decomposition of within industry composition e¤ects. As an example, consider the case of contract enforcement. The lack of a relationship between vertical integration and contractual enforcement may be due to the fact that …rms substitute away from production of goods which rely heavily on contracts towards goods that can be produced with more standardized inputs and thus are less dependent on contractual enforcement. For instance, …rms in the textile industry in India or Pakistan are mainly engaged in the production of T-shirts, while textile …rms in Italy may be mainly engaged in the design of new products. If the latter kind of activity is more exposed to hold-up problems, it is quite natural to think that production at the design and distribution stage

64

.1

Figure 9: Di¤erential Vertical Integration and Contractual Needs

transportation equipemen food products

beverages

wood products

Differential Vertical Integration -.2 -.1 0

nonferrous metal

textile

iron and steel paper and products printing and publishing non metal products

furniture machinery leather footwear electric machinery

other industries apparel

professional goods plastic pottery products

rubber products metal products

glass

-.3

other chemicals

.2

.3

.4 Contractual Needs

.5

.6

Differential Vertical Integration = Vertical Integration(non-OECD) - Vertical Integration(OECD)

carried out in Italy may be more vertically integrated than the production activities in India and Pakistan, despite Italy having better contract enforcement institutions.

65

Table 1: Descriptive Statistics of Main Variables Country Level Variables Variable

Obs

Mean

Industry Level Variables (in U.S.)

Std. Dev. Min

Max

Variable

Obs

Mean

Std. Dev. Min

Max

Vertical Integration Propensity

89

0.39

0.09

0.13

0.67

Vertical Integration

25

0.50

0.10

0.36

0.68

GDP Per Capita (in Logs)

86

8.61

1.04

6.20

10.18

External Financial Dependency

25

0.25

0.35

-0.45

1.14

Bank Credit over GDP Ratio

80

0.37

0.29

0.03

1.45

Contractual Needs

25

0.37

0.12

0.12

0.61

Legal Rights Index

84

5.10

2.25

0

10

Share of Small Firms

25

0.38

0.15

0.09

0.63

Number of Legal Procedure to Enforce a Contract

89

29.29

11.33

11

58

Externnal Financial Dependency of Upstream Industries

25

0.33

0.10

0.10

0.60

Notes: Country Level Variables: Vertical Integration Propensity is computed for each country as the unweighted average of the industry vertical integration indexes across the 25 industries in the sample (Data Source: UNIDO 2001 database). GDP Per capita is from the Penn Database. Bank Credit over GDP Ratio comes from 2000 World Bank data. Legal Rights index and Number of Legal Procedures to Enforce a Contract come from Doing Business Database at World Bank (2004, 2005 and Djankov et al. (2002, 2003)) Industry Level Variables: Vertical Integration is computed with data from UNIDO (2001) database. External Financial Dependency is from Rajan and Zingales (1998). Share of small Firms is the measured as the share of employees working in establishments with less than 500 employees, from the Census Bureau. Contractual Needs and External Financial Dependency of Upstream Industries are from author’s calculations. See Data Appendix for further details. Petroleum and Refineries is omitted. All industry variables are 3-Digit Code Isic, for the manufacturing sector in the United States.

Table 2: Disentangling Credit Markets Effects on Vertical Integration Dependent Variable: Vertical Integration External Financial Dependency × Financial Development

I

II

III

IV

V

0.034 [0.032]

0.181*** [0.069]

0.186*** [0.069]

0.186*** [0.077]

0.185*** [0.071]

-0.295*** [0.115]

-0.289** [0.114]

-0.291** [0.116]

-0.299*** [0.118]

-0.146* [0.081]

-0.202** [0.104]

-0.101 [0.087]

yes yes

yes yes yes

yes yes

External Financial Dependency × Financial Development × Small Firms

External Financial Dependency of Upstream Industries × Financial Development

Industry Dummies Country Dummies Industry Characteristics × GDP Per Capita Vertical Integration U.S. × GDP Per Capita

yes yes

yes yes

yes

Observations 1734 1734 1734 1734 1734 R-squared 0.52 0.53 0.53 0.54 0.54 *** Statistically significant at 1%, ** at 5%, * at 10%. Robust Standard Errors, corrected for clustering at the country level, reported in parenthesis. All variables are mean for the 1990-1999 years. Vertical Integration is the ratio of Value Added over Output in the Industry, computed from Data from UNIDO 2001 Database. Financial Development is measured as the ratio of Bank Credit over GDP from 2000 World Bank data. External Financial Dependency is for the U.S., from Rajan and Zingales (1998). Small Firms is measured as the share of employees working in establishments with less than 500 employees in the U.S., from the Census Bureau. External Financial Dependency of Upstream Industries is from author’s calculations. See Data Appendix for details. Industry characteristics include all the interactions of U.S. variables used to estimate the parameters of interest. Estimated, but unreported, the coefficient of the interaction of Financial Development with Small Firms. The U.S. are excluded from the sample. The regressions are estimated using the ranking of industry variables in the U.S. instead that the original measure. All the other variables are in Logs.

Table 3: Credit Markets and Contractual Enforcement Dependent Variable: Vertical Integration

Herfindal Inputs Use External Financial Dependency × Financial Development

External Financial Dependency × Financial Development × Small Firms

External Financial Dependency of Upstream Industries × Financial Development

Contractual Needs × Number of Legal Procedures

Specificity Inputs Use

I 0.152** [0.070]

II 0.137* [0.071]

III 0.153** [0.074]

IV 0.165** [0.071]

V 0.159** [0.075]

VI 0.174** [0.076]

-0.247** [0.115]

-0.254** [0.116]

-0.245** [0.117]

-0.260** [0.118]

-0.289** [0.124]

-0.278** [0.117]

-0.114 [0.080]

-0.119 [0.081]

-0.141* [0.083]

-0.147* [0.081]

-0.143* [0.081]

-0.155* [0.083]

-0.615** [0.280]

-0.574 [0.320]

-0.760*** [0.324]

-0.203 [0.183]

-0.117 [0.201]

0.284 [0.211]

Contractual Needs × Financial Development times ciao puppa e

-0.021 [0.134]

-0.027 [0.090]

-0.046* [0.027]

-0.058** [0.027]

External Financial Dependency × Number of Legal Procedures

Industry Dummies Country Dummies Vertical Integration U.S. × GDP Per Capita Industry Characteristics × GDP Per Capita

yes yes

yes yes yes

yes yes yes yes

yes yes

yes yes yes

yes yes yes yes

Observations 1734 1734 1734 1734 1734 1734 R-squared 0.52 0.53 0.53 0.52 0.53 0.53 *** Statistically significant at 1%, ** at 5%, * at 10%. Robust Standard Errors, corrected for clustering at the country level, reported in parenthesis. All variables are mean for the 1990-1999 years. Vertical Integration is the ratio of Value Added over Output in the Industry, computed from Data from UNIDO 2001 Database. Financial Development is measured as the ratio of Bank Credit over GDP from 2000 World Bank data. External Financial Dependency is for the U.S., from Rajan and Zingales (1998). Small Firms is measured as the share of employees working in establishments with less than 500 employees in the U.S., from the Census Bureau. External Financial Dependency of Upstream Industries and Contractual Needs variables are from Author’s Calculation. See Data Appendix for details. Industry characteristics include all the interactions of U.S. variables used to estimate the parameters of interest. Estimated, but unreported, the coefficient of the interaction of Financial Development with Small Firms. The U.S. are excluded from the sample. Columns Va and Vb report separately the coefficients estimated from a pooled regressions in which the sample is divided between OECD and non-OECD countries. The regressions are estimated using the ranking of industry variables in the U.S. instead that the original measure. All the other variables are in Logs.

Table 4: Financial Markets, Human Capital and Vertical Integration Dependent Variable: Vertical Integration External Financial Dependency × Financial Development

External Financial Dependency × Financial Development × Small Firms

I 0.170** [0.076]

II 0.170** [0.076]

III 0.198*** [0.077]

IV 0.141* [0.079]

-0.250** [0.127]

-0.207* [0.128]

-0.284** [0.128]

-0.269** [0.128]

0.362*** [0.117]

0.417*** [0.125]

-0.184** [0.003]

External Financial Dependency of Upstream Industries × Financial

Skill Intensity × Higher Education

0.254** [0.109]

Skill Intensity of Upstream Industries × Higher Education

0.262*** [0.109] -0.518 [0.343]

Skill Intensity × Financial Development

-0.269*** [0.0839]

External Financial Dependency × Higher Education

Industry Dummies Country Dummies Industry Characteristics × GDP Per Capita Vertical Integration U.S. × GDP Per Capita

0.001 [0.032] yes yes

yes yes

yes yes

yes yes yes yes

Observations 1369 1369 1369 1369 R-squared 0.55 0.56 0.56 0.56 *** Statistically significant at 1%, ** at 5%, * at 10%. Robust Standard Errors, corrected for clustering at the country level, reported in parenthesis. All variables are mean for the 1990-1999 years. Vertical Integration is the ratio of Value Added over Output in the Industry, computed from Data from UNIDO 2001 Database. Financial Development is measured as the ratio of Bank Credit over GDP from 2000 World Bank data. External Financial Dependency is for the U.S., from Rajan and Zingales (1998). Small Firms is measured as the share of employees working in establishments with less than 500 employees in the U.S., from the Census Bureau. Skill intensity is measured as the ratio of employees over total workers in the U.S., from the NBER manufacturing database. External Financial Dependency of Upstream Industries is from author’s Calculation. See Data Appendix for details. Industry interactions include all the interactions of U.S. variables used to estimate the parameters of interest. Estimated, but unreported, the coefficient of the interaction of Financial Development with Small Firms. The U.S. are excluded from the sample. Columns Va and Vb report separately the coefficients estimated from a pooled regressions in which the sample is divided between OECD and non-OECD countries. The regressions are estimated using the ranking of industry variables in the U.S. instead that the original measure. All the other variables are in Logs.

Table 5: Further Robustness Checks Dependent Variable: Vertical Integration External Financial Dependency × Financial Development

External Financial Dependency × Financial Development × Small Firms

External Financial Dependency of Upstream Industries × Financial Development

Industry Dummies Country Dummies Industry Interactions × GDP pc. Industry Interactions × (GDP pc)². Industry Dummies × GDP pc. Industry Dummies × Intra Indust × GDP pc. Breaking the Sample

I

II

III

IV

Va

Vb

0.157*

0.139*

0.149*

0.149*

0.182***

0.265***

[0.092]

[0.075]

[0.091]

[0.091]

[0.070]

[0.109]

-0.219

-0.207

-0.212

-0.212

-0.287***

-0.357**

[0.140]

[0.131]

[0.139]

[0.139]

[0.116]

[0.162]

-0.214**

-0.206*

-0.209**

-0.209**

-0.145*

-0.173*

[0.104]

[0.128]

[0.105]

[0.105]

[0.082]

[0.106]

yes yes yes

yes yes yes yes

yes yes

yes yes

yes yes

yes yes

yes

yes yes yes

yes

Observations 1715 1715 1715 1715 1715 R-squared 0.53 0.54 0.55 0.56 0.53 *** Statistically significant at 1%, ** at 5%, * at 10%. Robust Standard Errors, corrected for clustering at the country level, reported in parenthesis. All variables are mean for the 1990-1999 years. Vertical Integration is the ratio of Value Added over Output in the Industry, computed from Data from UNIDO 2001 Database. Financial Development is measured as the ratio of Bank Credit over GDP from 2000 World Bank data. External Financial Dependency is for the U.S., from Rajan and Zingales (1998). Small Firms is measured as the share of employees working in establishments with less than 500 employees in the U.S., from the Census Bureau. External Financial Dependency of Upstream Industries and Intra Industry are from author’s Calculation. See Data Appendix for details. Industry interactions include all the interactions of U.S. variables used to estimate the parameters of interest. Estimated, but unreported, the coefficient of the interaction of Financial Development with Small Firms. The U.S. are excluded from the sample. Columns Va and Vb report separately the coefficients estimated from a pooled regressions in which the sample is divided between OECD and non-OECD countries. The regressions are estimated using the ranking of industry variables in the U.S. instead that the original measure. All the other variables are in Logs.

Table 6: Panel Specification Dependent Variable: Vertical Integration I 0.090*** [0.029]

II 0.089*** [0.025]

III 0.038** [0.015]

IV 0.051 [0.033]

-0.161*** [0.060]

-0.157*** [0.051]

-0.097*** [0.034]

-0.117* [0.069]

Industry-Country Dummies Country Trends Industry Trends Industry Interactions × (GDP pc)

yes

yes yes

yes

yes

Clusters

i-c

Country

Industry

i-c

Observations R-squared

9961 0.55

9961 0.57

9961 0.63

9961 0.56

External Financial Dependency × Financial Development

External Financial Dependency × Financial Development × Small Firms

yes yes

*** Statistically significant at 1%, ** at 5%, * at 10%. Robust Standard Errors are reported in parenthesis. Vertical Integration is the ratio of Value Added over Output in the Industry, computed from Data from UNIDO 2001 Database. Financial Development is measured as the ratio of Bank Credit over GDP from 2000 World Bank data. External Financial Dependency is for the U.S., from Rajan and Zingales (1998). Small Firms is measured as the share of employees working in establishments with less than 500 employees in the U.S., from the Census Bureau. Industry interactions include all the interactions of U.S. variables used to estimate the parameters of interest. Estimated, but unreported, the coefficient of the interaction of Financial Development with Small Firms. The U.S. are excluded from the sample. Columns Va and Vb report separately the coefficients estimated from a pooled regressions in which the sample is divided between OECD and non-OECD countries. The regressions are estimated using the ranking of industry variables in the U.S. instead that the original measure. All the other variables are in Logs.

Table A1: Cross-Country Differences in Vertical Integration Dependent Variable: Vertical Integration GDP per capita

I 0.065* [0.037]

Population

II 0.082** [0.038] 0.003 [0.020]

Financial Development

III 0.132*** [0.038]

IV 0.068 [0.044]

V 0.07 [0.042]

VI 0.125** [0.052] 0.013 [0.027] -0.160*** [0.033]

0.022 [0.074]

-0.058 [0.073] -0.012 [0.049] 0.038 [0.039]

yes

yes

-0.164*** [0.030]

Investor's Legal Rights

-0.045 [0.052]

Contract Enforc. Procedures Entry Procedures Higher Years of Education

Industry Dummies

yes

yes

yes

yes

Observations 1734 1734 1734 1734 1734 1344 0.21 0.22 0.29 0.23 0.22 0.32 R-squared *** Statistically significant at 1%, ** at 5%, * at 10%. Robust Standard Errors, corrected for clustering at the country level, reported in parenthesis. All variables are mean for the 1990-1999 years. Vertical Integration is the ratio of Value Added over Output in the Industry, computed from Data from UNIDO 2001 Database. Financial Development is measured as the ratio of Bank Credit over GDP from 2000 World Bank data. Investors’ Legal Rights and Number of Entry Procedures are from Doing Business Database (World Bank). Years of Higher Education is from Barro and Lee database. All variables are in Logs.

Life Enjoy

" Life is not a problem to be solved but a reality to be experienced! "

Get in touch

Social

© Copyright 2013 - 2019 DOKUMENTIS.COM - All rights reserved.