What Determines Firm Size?

Krishna B. Kumar

Marshall School of Business,

University of Southern California

Raghuram G. Rajan

International Monetary Fund & NBER

Luigi Zingales∗

Graduate School of Business,

University of Chicago, NBER, & CEPR

Abstract

In this paper we study how industry-specific and institutional factors affect the sizeof corporations. We find that countries that have better institutional development, asmeasured by the efficiency of their judicial system, have larger firms. Once we correctfor institutional development, there is little evidence that richer countries or countrieswith better developed capital markets have larger firms. Interaction effects are perhapsof greatest use in discriminating between theories. We find the efficiency of the judicialsystem has the strongest relationship with firm size in industries with low physical cap-ital intensity, a finding consistent with a broad class of theories emphasizing “CriticalResources” as central to determining the boundaries of firms.

∗We thank Eugene Fama, Doug Gollin, Ananth Madhavan, John Matsusaka, Andrei Shleifer, and work-

shop participants at Chicago, Copenhagen, Norwegian School of Management-BI, and USC for comments.

Kumar acknowledges support from the US Department of Education and USC’s CIBEAR. Rajan and Zingales

acknowledge financial support from the Sloan Foundation (officer grant) and the Stigler Center.

An interesting aspect of economic growth is that much of it takes place through thegrowth in the size of existing organizations.1 What are the potential determinants of thesize of economic organizations? What are the constraints to size and, hence, potentialconstraints to growth?

Even if one were totally uninterested in economic growth, one would still encounter firmsize in many other sub-fields of economics. For instance, firm size has often been usedas an exogenous variable in explaining financing decisions (see Barclay and Smith (1995 aand b)), managerial compensation (Jensen and Murphy (1990)), and even required rate ofreturns (Banz (1981), Fama and French (1992)). While this approach has produced greatinsights, there are reasons to open the black box enveloping firm size: as we have come tobetter understand financial decisions taking the size of firms as fixed, the intellectual payoffsto developing a better understanding of the determinants of the boundaries of firms haveincreased. After all, they may not be orthogonal to financial decisions (see Zingales (2000)).Moreover, the boundaries of corporations are themselves changing rapidly (see, for example,Pryor (2000) or Rajan and Zingales, 2000) increasing the urgency for such studies.

While our focus is on firm size, we face a problem that studies on organizational formhave faced, that is of taking the theory of the boundaries of firms to the data.2 One approachresearchers have taken to tackling this problem is to focus on a particular industry and studyit intensely (see, for example, Berger et al. (2001), Brinckley et al. (2001), Mullainathanand Scharfstein (2000) or Eldenberg, et al. (2001)). While this focus allows cleaner testsof the theories, it makes it harder to identify generalizable patterns. What is a persistentorganizational form in an industry in a country may be a result of simply historical accident.

The alternative is to take a broader brush approach to identifying common robust pat-terns across industries and countries. While such an aggregate analysis can, at best, tell uswhat theories may be worth investigating further, when combined with the more detailedmicro-analysis, it can give us greater confidence in our understanding. Obviously, bothapproaches are necessary.

It is the latter approach that we take in this paper. More precisely, we examine the size offirms as a measure of where organizational boundaries are set. We do this across a variety ofcountries and industries. We use the lens of broad classes of models we label “technological,”“organizational,” or “institutional,” to sharpen our focus and develop some understandingon what might account for the immense observed variation in firm size. Specifically, we askthe following questions: Is it sufficient to think of the firm as a black box, as technological

1For instance, in the sample of 43 countries they study, Rajan and Zingales (1998a) find that two-thirds

of the growth in industries over the 1980s comes from the growth in the size of existing establishments, and

only the remaining one-third from the creation of new ones.

2See, for example, the work on joint ventures and alliances by Desai and Hines(1999), Lerner and Merges

(1998), and Robinson (2001).

2

theories typically do, or do we need to be concerned with features such as asset specificity andthe allocation of control that are at the center of organizational theories? Which regressorswill aid in discriminating between these theories? What broad patterns hold in the data,and where should we concentrate our future search for determinants? Since some of theproposed determinants of firm size are common across theories, our econometric exercisecannot distinguish among specific theories unless they have opposite predictions for thesame variable. Our approach is best suited to assess the broad success of classes of theoriesthan those of competing models within a single class.

We attempt to establish the nature of relationships between likely determinants suggestedby theory and firm size using data on the size distribution of firms across industries in15 European countries. This data set is particularly well suited for our purposes becausethese are all fairly well-developed countries, so the minimum conditions for the existenceof firms such as a basic respect for property rights, the widespread rule of law, and theeducational levels to manage complex hierarchies exist. This enables us to focus on moresubtle institutional factors which exhibit considerable variation even in this sample of well-developed countries.3 We also have a large number of industries, and the variation acrossindustries in their use of different factors can provide an understanding of the effects ofproduction technology on firm size. Most importantly, we are able to study the interactionsbetween institutional and technological effects suggested by theories that provide specificson the mechanism by which size is affected. By examining these interactions we get theclearest insights into the possible determinants of firm size.

We first examine patterns in firm size across industries and countries. At the industrylevel, we find physical capital intensive industries, high wage industries, and industries thatdo a lot of R&D have larger firms.4 We also find that, on average, firms facing largermarkets are larger. At the country level, the most salient findings are that countries withefficient judicial systems have larger firms, though, contrary to conventional wisdom, thereis little evidence that richer or higher human capital countries have larger firms. Similarly,while countries that are more financially developed have larger firms, this seems to reflectthe greater institutional development of those countries (such as greater judicial efficiency)rather than the effects of finance per se.5 The cross-industry results are broadly consistent

3The wide variation seen in the regressors of our sample is discussed in greater detail in Section 2.3.

4Some of these cross-industry correlations are not surprising given the past literature on intra-industry

patterns (see Audretsch and Mahmood (1995), Caves and Pugel (1980), Klepper (1996), Sutton (1991), for

example). There are also a number of cross-industry studies on the determinants of industry concentration

(see Schmalensee (1989) for a survey). While we do not believe there is a one-to-one correspondence between

concentration and firm size, some of the cross-industry correlations (for example, that between R&D and

size) may be viewed as reinforcing the findings of that literature.

5This finding is not surprising. The effects of financial development on average firm size are, as we will

3

with all three classes of theories, but at the country level organizational and institutionaltheories fare better, with the efficiency of the judicial system emerging as a correlate worthyof further study.

These broad correlations, however, can take us only so far in identifying theories withthe most promising structural determinants of firm size. For this reason, we study the effectsof interactions between an industry’s characteristics and a country’s environment on size,after correcting for industry and country effects. The analysis of these interactions, whichare best able to discriminate between theories, is perhaps the most novel aspect of thispaper. Organizational theories, which model the micro mechanisms in greater detail, aremore amenable to such tests. A central finding is that the relative size of firms in capitalintensive industries diminishes as the judicial system becomes more efficient. This is in largemeasure because the average size of firms in industries that are not physical asset intensiveis larger in countries with better judicial systems. One way to interpret this finding is thatfirms in consulting and advertising, which are based on intangible assets such as reputations,client relationships, or intellectual property, tend to be much bigger in countries with a betterlegal system because a sophisticated legal system is necessary to protect these assets. Bycontrast, only a crude legal system is necessary to enforce the property rights associated withphysical capital, so basic physical capital intensive industries like steel or chemicals will haverelatively larger firms in countries with an underdeveloped legal system. In further support ofthis argument, we find that R&D intensive industries have larger firms in countries that havebetter patent protection, suggesting that the judicial system may be particularly importantin protecting intangible resources, enabling larger firms to emerge in industries that dependon such resources.

We address issues of endogeneity by using instruments for variables such as the marketsize, predetermined values for institutional variables such as judicial efficiency, and averagesover several countries for industry variables such as investment per worker. We also useinteractions of technological and institutional variables, which are likely to be more exogenousthan level variables with respect to firm size.

Our evidence suggests that theories emphasizing the fundamental importance of own-ership of physical assets in determining the boundaries of the firm, and those that suggestmechanisms other than ownership can expand firm boundaries when the judicial system im-proves – organizational theories we refer to as “Critical Resource” theories – are promisingframeworks with which to explore the determinants of firm size. It also appears that fur-ther attention can be more fruitfully focused not on technological variables or institutionalvariables by themselves, but on interactions between these variables.

A caveat is in order. There is a large literature on how the forces affecting competitionbetween firms (see Sutton (1991) for an excellent review) determine the concentration of

argue, ambiguous.

4

the market and, indirectly, firm size. While there may be some overlap between our cross-industry findings and the vast literature on the cross-industry determinants of concentration,not all the patterns we establish for firm size can simply be inferred from the literature onconcentration. More important, following the spirit of Sutton (1991) our most persuasiveresults are on interactions, which are hard to explain using theories of market structure.Nevertheless, we acknowledge that the focus on firm size necessitated by the macro leveldata we have, which ignores the forces of competition, is likely to offer as partial an answeras one resulting from an analysis of competition, which ignores other internal or institutionalconstraints on firm size.

The rest of the paper proceeds as follows. In section 1, we provide a candidate list ofdeterminants of firm size suggested by the theories, and in section 2 we present the dataand discuss the broad patterns. We report partial correlations of size with industry specificcharacteristics (section 3), country specific variables (section 4), and interactions betweenthe two (section 5) correcting for industry and country characteristics. We discuss the resultsin section 6 and conclude with suggestions for future research.

1 Existing Theories

There exists an immense literature that, directly or indirectly, deals with firm size. Tosimplify our analysis we classify theories broadly as technological, organizational, and insti-tutional, based on whether they focus on the production function, the process of control,or influences of the economic environment. Clearly, the classification will be somewhat ar-bitrary because some theories combine elements of different approaches. Moreover, spaceconstraints do not permit us the luxury of being exhaustive, and the theories discussed be-low should only be viewed as representative with those more amenable to proxies we haveaccess to receiving special attention.

1.1 Technological Theories

Technological theories go back to Adam Smith (1776), who suggests that the extent ofspecialization is limited by the size of the market. If a worker needs to acquire task-specifichuman capital, there is a “set-up” cost incurred every time the worker is assigned to anew task. Workers perform specialized tasks and a firm needs to hire more workers when itsproduction process becomes more specialized. Therefore, not only the extent of specializationbut also the size of firms is limited by the size of the market being served. However, Smith’sprediction has not remained unchallenged. Becker and Murphy (1992) point to the existenceof multiple firms serving most markets to argue that coordination costs play a major role inlimiting the size of firms before the size of the market becomes binding. Hence, the effect ofmarket size is ultimately an empirical question.

5

An important formulation of the neoclassical theory of firm size is Lucas (1978). Heidentifies a firm with a manager (and the capital and labor under the manager’s control)and assumes that the “talent for managing” is unevenly distributed among agents. If theelasticity of substitution between capital and labor is less than one, an increase in per capitacapital raises wages relative to managerial rents, inducing marginal managers to becomeemployees, and increasing the ratio of employees to managers. Therefore, greater capitalintensity, proxied by investment per worker or R&D intensity, should be associated withlarger firms.

Human capital is suggested as a positive correlate of firm size in the theories in Rosen(1982) and Kremer (1993). Rosen (1982) considers a hierarchical organizational structure,where improved labor productivity at any given level filters through lower levels. The tradeoffbetween managerial scale economies and the loss of control associated with larger organiza-tions results in determinate firm sizes. In equilibrium, persons with the highest skills areplaced in the highest positions of the largest and deepest firms. The multiplicative pro-ductivity interactions mentioned above make the equilibrium distribution of firm size moreskewed than the underlying distribution of talent.

Kremer (1993) models human capital as the probability that a worker will successfullycomplete a task. Each task is performed by one worker, so the output of the firm (a sequenceof tasks) depends on the product of the skill levels of all workers. Firms using technologiesthat need several tasks will employ highly skilled workers because mistakes are more costlyto such firms. Kremer then assumes that the number of tasks and number of workers arelikely to be positively correlated, and finds his model consistent with the stylized fact thatricher (higher human capital) countries specialize in complicated products and have largerfirms. An ancillary implication of the model is that firm size should be positively correlatedwith the wage per worker, since higher wage implies a higher quality worker.6

While we refine some of these hypotheses in our empirical discussion, a broad summaryof the empirical implications of this class of theories can be stated as follows:

• There should be a positive connection between market size and firm size according toSmith (1776), but not according to Becker and Murphy (1992).

• Firms in richer countries should be larger. Investment per worker and R&D intensityshould be positively associated with firm size.

6The lack of a hierarchical structure in Kremer’s model makes the relationship between average human

capital of the workforce and average firm size more explicit, unlike the models of Lucas and Rosen where the

human capital of managers, who typically constitute a small fraction of the workforce, is what matters.

A recent contribution in this genre is Garicano (2000). In his model the size of a managerial hierarchy is

determined by a trade-off between communication and knowledge acquisition costs. The theory is promising

but the data requirements for shedding light on whether it matters are beyond the scope of an analysis like

ours (see, however, Rajan and Wulf (2002) for an initial attempt).

6

• Firm size should increase with human capital and wage per worker.

1.2 Organizational Theories

Organizational theories fall into three broad categories. One set of theories – often referredto as “Contracting Cost” theories (see, for example, Alchian and Demsetz (1972) and Jensenand Meckling (1976)) – suggest little difference in the contracts at work ‘inside a firm’ and inthe market. Instead, the firm is simply a particular configuration of contracts characterizedby a central common party to all the contracts, who also has the residual claim to the cashflows. While these theories are extremely useful in explaining contractual arrangementsinside the firm, they do not provide sharp predictions on the boundaries of the firm.

A second set of theories, mostly associated with Williamson (1975, 1985)) and Klein,Crawford and Alchian 1978, can be loosely labeled “Transaction Cost” theories. Thesetheories offer more guidance as to whether a transaction should take place between twoarm’s length entities or within a firm, where a firm can be loosely defined as an entitywith a common governance structure. The advantage of doing a transaction within a firmis that the firm brings incentives to bear that cannot be reproduced in an arm’s lengthmarket. Unfortunately, factors that typically determine the extent of integration in thesetheories, such as asset specificity and informational asymmetry, are hard to proxy for evenwith detailed data, let alone in a relatively macro-level study such as ours.

The third set of theories, which can collectively be described as “Critical Resource”theories of the firm (see Grossman and Hart (1986) for the seminal work in this area andHart (1995) for an excellent survey of the early literature), however, offer greater grist forempirical mills. The starting point of these theories is that transactions can either be fullygoverned by contracts enforced by courts or by other mechanisms that confer power in non-contractual ways to one or the other party to the transaction. The non-contractual sourceof power is usually a critical resource that is valuable to the production process. A varietyof (non-contractual) mechanisms attach the critical resource to one of the parties in a waythat maximizes the creation of surplus.

For example, the Property Rights approach (see Grossman and Hart (1986), Hart andMoore (1990)) emphasizes physical assets as the primary critical resource, and ownershipas the mechanism that attaches this resource to the right agent. According to this view,ownership differs from ordinary contracts because it confers on the owner the residual rightsof control over the asset (the right to decide in situations not covered by contract). Thepower associated with the common ownership of physical assets is what makes the firmdifferent from ordinary market contracting.

More recent developments of this theory (see Rajan and Zingales (1998b) and (2001),Holmstrom (1999), and Holmstrom and Roberts (1998) for example) emphasize that the crit-ical resource can be different from an alienable physical asset, and mechanisms other than

7

ownership, which work by fostering complementarities between agents, and between agentsand the critical resource, can be sources of power within the firm. For instance, Rajan andZingales (2001) analyze a model where an entrepreneur’s property rights to the critical re-source are allowed to vary. This model has the implication that as long as the governmentrespects property rights broadly, physical assets are hard to make away with and thus phys-ical capital intensive firms will typically be larger. But as legal institutions improve, otherforms of protection become available to the entrepreneur – patent rights protect intellectualproperty, while non-compete clauses restrain employees from departing. Thus firms that relyon intangible forms of critical resources, such as brand names, intellectual property, or inno-vative processes, should become larger as the legal environment improves. This implicationforms the cornerstone of our analysis of interactions involving judicial efficiency and capitalintensity.7

In summary, this class of theories gives rise to the following empirical implications:

• Higher levels of capital – a critical resource – should be associated with greater firmsize.

• By offering better protection for critical resources, a better legal environment shouldlead to larger firms.

• Firms that rely on intangible critical resources should become disproportionately largeras the legal environment improves.

1.3 Institutional theories

There are many channels through which institutions can affect firm size beyond what ispredicted by the technological and organizational theories. We group these channels intotwo main categories: regulatory and financial.

On the regulatory side, many costly regulations apply only to larger firms (for examplethe obligation to provide health insurance in the United States or Union Laws in Italy). Thistilts the playing field towards small firms. Other regulations, such as strong product liabilitylaws, favor the creation of separate legal entities that can avail themselves of the protectionafforded by limited liability. This should lead to smaller firms. For instance, each taxi cabin New York is a separate firm. This effect has been found to be important in explainingthe time variation of size of firms in the United States (Ringleb and Wiggins (1990)). In the

7Of course, courts are not the only enforcement mechanism. In repeated relationships contracts can be

self-enforcing. Unfortunately, we do not have a good proxy for the repeated nature of a relationship. It is

reasonable to assume, though, that the frequency of interaction varies by industry and as such should be

captured by our industry dummies.

8

same way, a strict enforcement of anti-trust laws can have an effect in reducing the averagesize of firms.

If the availability of external funds is important for firms to grow, firm size should bepositively correlated with financial development and, more generally, with factors promot-ing the development of financial markets. Rajan and Zingales (1998a), however, find thatfinancial development affects growth in both the average size of existing establishments andin the number of new establishments in industries dependent on external finance (thoughdisproportionately the former). With the development of financial markets, more firms willbe born, reducing the average size of firms; however, existing firms will be able to growfaster, increasing the average size of firms. Whether the average size of firms in industriesthat rely on external finance is larger is, therefore, ultimately an empirical question, whichwe will try to resolve.

La Porta, Lopez-De-Silanes, Shleifer, and Vishny (1997a) find that a country with aCommon Law judicial system, and having strict enforcement of the law, has a more developedfinancial system. This would suggest that there is also an indirect channel through whichsound laws and judicial efficiency affect firm size - through their effect on financial marketdevelopment.

The empirical implications of the institutional theories can be summarized as follows:

• Stricter liability and anti-trust laws give rise to smaller firms.

• Financial development exerts an ambiguous effect on firm size.

• A higher quality of the legal environment should result in larger firms.

We have attempted in the preceding paragraphs to describe some important technical,legal, and institutional influences on firm size. Several of these effects are country specific.Given the large number of possible effects and the limited number of countries we have datafor, we will not attempt to capture them all, but will typically control for them by insertingcountry fixed effects.

1.4 Existing Empirical Literature

1.4.1 Market Structure

Much of the evidence we have comes indirectly from the vast literature on market struc-ture. Most of the early cross-industry work appealed to the structure-conduct-performanceparadigm of Bain (1956). However, this literature came under criticism, especially after theextensive work on game theory in the 1980s, when it became clear that the theoretical pre-dictions were much more nuanced than allowed for in the empirical work. As a result, much

9

work since has been within-industry because it is easier to map the underlying characteristicsof specific industries to particular theoretical models.8

However, more recently Sutton (1991) has argued that “too rigid adherence to such aview runs the risk of abandoning a central part of the traditional agenda of the subject,which concerns the investigation of regularities of behavior that hold across the general runof industries”. Instead, he argues for investigating robust predictions that are likely to holdacross a class of models. In particular, he argues when up front sunk costs are exogenous,and the degree of price competition is fixed, a robust prediction is that the size of the marketand the concentration of the market should be negatively related. Intuitively, the larger themarket, the greater the ability to amortize fixed costs, and the greater the amount of entry.

The results on concentration (e.g., see the early work by Pashigian (1968)and Stigler(1968)) do not, however, map one to one onto results on firm size. Bresnahan and Reiss(1991) argue that price competition does not change much with entry. If so, the number offirms should expand linearly with the market, and there should be no predicted relationshipbetween market size and firm size. By contrast, other theories of firm size predict a positiverelationship between the size of the market and the size of firms, and the evidence, we willsee, is consistent with this prediction.

Because our focus is on technological, managerial, and institutional determinants of firmsize and not on competition within a market for a specific product, we can afford to examineindustries defined fairly broadly so long as the underlying technological characteristics offirms within each broad industry group are strongly correlated. To this extent, we alsoreduce the importance of a particular product’s market structure for our findings. Andby emphasizing interaction effects that are unlikely to come from theories of competition,we hope to establish the independent validity of our focus. Of course, ultimately, theoryshould address both the determinants of firm size arising from external competition, andthose arising from factors like the capabilities of the firms themselves (see Penrose (1959)).Pending such a theory, a focus on either the competitive forces determining market structureor on the technological, managerial or institutional forces determining firm size, can only bepartial, however careful the empirical work.

1.4.2 Institutional differences and firm size

There have been few cross-country studies focusing on the effects of institutional differenceson firm size. Davis and Henrekson (1997) find that the institutional structure in Swedenforces a tilt (relative to the US) towards industries that are dominated by large firms. Theirinterest, however, is in the relative distribution of employment across sectors, not in thedeterminants of firm size per se. La Porta, Lopez-De-Silanes, Shleifer, and Vishny (1997b)

8Though see Caves and Pugel (1980), Audretsch and Mahmood (1995), and Mata (1996)

10

find a positive correlation between an indicator of the level of trust prevailing in a countryand the share of GDP represented by large organizations (defined as the 20 largest publiclytraded corporations by sales). Their focus, however, is on the relative importance of largeorganizations, while our focus is on the determinants of the absolute size of organizations.

2 The Data

In 1988, Directorate-General XXIII of the European Commission and Eurostat launcheda project to improve the collection and compilation of statistics on small and medium en-terprises. This project led to the publication of Enterprises in Europe by the EuropeanCommission in 1994. This data set contains statistics on enterprises, employment, and pro-duction for all economic sectors (except agriculture) in the European Union (EU) and theEuropean Free Trade Agreement (EFTA) countries.

One of the explicit purposes of this effort was to assemble “comparable and reliablestatistics which make it possible to identify the relative importance of different categories ofenterprises.” (Enterprises in Europe, Third Report, v.I, p. xxii). In spite of the effort tohomogenize the statistical criteria Eurostat warns that several methodological differences inthe classification and coverage of units remain. Thus, the cross-country analysis should beinterpreted with more caution, while the cross-sector analysis or interactions, which controlfor country fixed-effects, are less sensitive to this problem. With all its limitations this is thefirst source we know of that provides comparable data on firm size across countries.

Enterprises in Europe provides us with the size distribution of firms (number of employ-ees) in each NACE two-digit industry (we use “industry” interchangeably with “sector”).9

We exclude from the analysis Iceland, Luxembourg, and Liechtenstein because of their ex-tremely small size. We also drop Ireland, because data are not consistently reported. Thisleaves us with 15 countries. For all these countries, data are available for either 1991 or 1992.Further details on the data are presented in the data appendix.

We have data on how many firms and employees belong to each firm size bin (e.g., thereare 30,065 employees and 109 firms in the bin containing 200-499 employees in “Electricity”in Germany).10

2.1 The Theoretical Unit of Interest and the Empirical Unit

Technological theories focus on the allocation of productive inputs across various activitiesand the effect this has on the size of the production process, rather than on the ways in which

9NACE is the general industrial classification of economic activities within the EU. Two-digit NACE

industry roughly corresponds to two-digit SIC sectors.

10We use the terms “bin” and “size class” interchangeably.

11

hierarchical control is exerted. The length of the production process is thus the theoreticalsize of the firm. These theories would not, for example, make much of a distinction betweenToyota and its supplier network, and General Motors (a much more vertically integratedfirm) and its supplier network, since both feed into broadly similar production processes.

Organizational theories, on the other hand, focus on how hierarchical control is exerted.The Property Rights view asserts that control exerted through an arm’s length contract(General Motors over its suppliers) is not the same thing as control exerted through ownership(General Motors over its divisions). According to this view, the economic definition of a firmcorresponds to the legal view – a firm is a set of commonly owned assets. However, morerecent developments (see Rajan and Zingales (1998b, 2001), for example) go further andsuggest that if the economic distinction between transactions that are firm-like and market-like turns on whether hierarchical, non-contractual control is exerted, common ownership isneither necessary nor sufficient to define the economic limits of a firm. Toyota may exertmuch more control over its suppliers who are tied to it by a long history of specific investmentthan General Motors, which puts supply contracts up for widespread competitive bidding.Toyota and its suppliers, although distinct legal entities, could thus be thought of as a firmin the economic sense, while General Motors and its suppliers are distinct economic and legalentities. Thus, in general, the firm described by the theories does not correspond to the legalentity.

However, data are available only for the legal entity. The unit of analysis is the enterprise,defined as “the smallest combination of legal units that is an organizational unit producinggoods or services, which benefits from a certain degree of autonomy in decision-making,especially for the allocation of its current resources. An enterprise carries out one or moreactivity at one or more location” (Enterprises in Europe, Third Report, v.II, p. 5).

The reference to “organizational unit producing goods or service”, however, suggeststhat the definition takes into consideration also the length of a production process. Theemphasis on the “smallest combination of legal units” is also important. Conversations withEurostat managers indicates that subsidiaries of conglomerates are treated as firms in theirown right, as are subsidiaries of multinationals.11 Some subjectivity, however, is introducedbecause Eurostat looks also for autonomy in decision making in drawing the boundary ofthe enterprise. To this extent, the enterprise corresponds to the economic entity discussedabove – the economic realm over which centralized control is exercised.

In sum, the definition used for the enterprise is a combination of the “legal”, and whatwe call the “economic”, firm. To make some headway, we have to assume that it is also agood proxy for the length of the production process. All we need is the plausible assumptionthat factors permitting a longer chain of production should also increase the average size of

11Foreign subsidiaries of a multinational with headquarters in a particular country do not, therefore, add

to the enterprise’s size in the country of the parent.

12

the autonomous legal entity called the firm.

2.2 What Do We Mean by Average Size?

Should we measure the size of a firm in terms of its output, its value added, or the numberof its employees? Value added is clearly preferable to output, because the complexity of theorganization is related to the value of its contribution rather than to the value of the outputsold. Enterprises in Europe reports that value added per employee is fairly stable acrossdifferent size-classes, which implies a measure of firm size based on the number of employeesis likely to be very similar to one based on value added. Yet, coordination costs, which arepresent both in the technological and the organizational theories of the firms, are in termsof number of employees, not their productivity. This argues for a measure based on numberof employees, which is what we use in the rest of the paper. Such a measure has a longintellectual tradition (e.g., Pashighian (1968)).12

What is the most appropriate measure of average firm size for our data? The simpleaverage, obtained by dividing the total employment in the country-sector combination bythe total number of firms in that combination, is inappropriate for two reasons. First, itignores the richness of the data on the distribution of firm size. Second, it could give us anumber that has little bearing on the size of the firm that is “typical” of the sector or has thegreatest share in the sector’s production. For instance, in the automobile sector in Spain, 78%of the employees work for 29 firms which employ 38,302 employees. There are, however, 1,302self-employed people, who account for an equal number of firms. Taken together with theintermediate categories, the simplest measure would suggest that the average firm has only57 employees. The theories we surveyed in Section 1 have little to say about the degeneratefirm of size one; therefore, using the simple average to test the empirical implications of thesetheories would be misguided.

We instead calculate the size of the typical firm in the following way. We locate the binin which the median employee of a country-sector combination works. We divide the totalemployment in that bin by the number of firms in that bin to get the average firm size. Forthe remainder of this paper when we refer to firm size without qualification, we mean the“typical” firm’s size calculated in this fashion (or the log of this size in the regressions).13 In

12The number of employees is available for many more country-sector combinations than is value added,

which further motivates our choice.

13In addition to focusing on firms that carry out the bulk of the economic activity in a sector, this measure

is also more likely to capture firms at or near their optimal scale. The theories we have described suggest the

determinants of the optimal scale of firms. Likewise, the measure mitigates the problem of not knowing firm

entry and exit details in our cross-sectional data. When there is entry and exit, firms grow substantially when

young, and there is inter-industry variation in these rates of growth (see Caves (1998) and Pakes and Ericson

(1998)). Despite this problem, since we have countries at relatively similar stages of development, industries

13

the above example of automobile manufacturing in Spain, this measure gives an average firmsize of 3,820 employees.14 The correlation in our sample between the average firm size andthe “typical” firm size is 0.39; more relevant for the regressions is the correlation betweenthe log of these measures, which is 0.86.15

2.3 Summary Statistics

In Table 1, we present statistics on firm size by country. The firm size is calculated eitheras a simple average or as the “typical” firm’s size described above for each country-sectorcombination; the mean across sectors for a given country is presented. The correlationcoefficient between the two measures of size is 0.25 and the Spearman’s rank correlation is0.4.

Greece, Portugal, and Austria have the smallest typical firm size. The UK and Italyhave firms with the highest typical size, in fact, substantially higher than the remaining

should be at similar stages of the product life cycle across countries (see Klepper (1996) for a formalization

of the effects of the PLC). Hence, the cross-country and interaction effects should indeed reflect the effects of

institutional differences on average size. More important, Sutton (1997, p52) argues that most entry and exit

has relatively little effect on the largest firms in the industry (which are likely to have achieved the optimal

scale).

14Our measure is similar in spirit to the coworker mean suggested by Davis and Henrekson (1997) to

emphasize the number of employees at the average worker’s place of employment. It is the size-weighted

mean of employer size; by squaring the number of employees in each firm, this measure clearly emphasizes

the larger firms. If we were to adapt this measure to our data, which are available only at the level of firm

size bins, we would compute a weighted average size as follows. The average firm size in each size bin is first

calculated by dividing the number of employees by the number of firms. The average size for the entire sector

is then calculated as the weighted sum of these bin averages, using as weights the proportion of the total

sectoral employment in that bin. This produces a “employee-weighted” average of firm size.

Employee Weighted Average Number of Employees =

n∑bin=1

(NEmp

bin

NEmpSector

)(NEmp

bin

NFirmsbin

)where NEmp

bin is the total number of employees in a bin, NEmpSector is the total number of employees in the

sector, and NFirmsbin is the total number of firms in a bin. In contrast to the firm-weighted simple average,

the employee-weighted average emphasizes the larger firms; note the squaring of bin employment in the

numerator.

Our results are virtually unchanged, qualitatively, if we were to use this employee-weighted measure than

the size of the “typical” firm we have adopted.

15When we do use the simple average, which we have argued is a noisy measure of firm size, the significance

of the cross-industry coefficients are preserved but most coefficients in the other regressions lose significance.

However, the results are not overturned in the sense that a positive and significant coefficient with the

“typical” firm size does not become negative and significant with the simple average.

14

countries. Since Italy is reputed to have many small firms, it might come as a surprise thatit has the second highest average size. This is mainly due to the composition of industries inItaly, the utility sectors in particular. When we control for industry fixed effects (reported inthe second to last column), Italy’s firms are smaller; only Greece and Portugal have smallerfirms.16 Spearman’s rank correlation between the typical size and the relative size controlledfor industry effects is 0.36, suggesting differences in size are not simply driven by a differentindustry composition.17

We present in Table 2 the summary statistics for the explanatory variables used in thesubsequent analysis as well as their cross-correlations. Two variables that we will focusattention on are an index of the efficiency of the judicial system, henceforth termed “judicialefficiency”, and an index of patent rights termed “patent protection”.

Judicial efficiency is an assessment of the “efficiency and integrity of the legal environmentas it affects business, particularly foreign firms”. It is produced by the country risk ratingagency Business International Corp. We use the measure to capture the degree to whichthe rights of contractual parties are protected by the courts. Because it emphasizes foreignfirms, we believe it also reflects the ease of access to the judicial system by all parties, notjust insiders. We have the average between 1980 and 1983 on a scale from zero to 10 withlower scores reflecting lower efficiency levels.

Patent protection is from Ginarte and Park (1997). They compute an index for 110countries using a coding scheme applied to national patent laws. The five aspects of patentlaws examined were (1) the extent of coverage, (2) membership in international patent agree-ments, (3) provisions for loss of protection, (4) enforcement mechanisms, and (5) durationof protection. Each of these categories was scored on a scale of 0 to 1, and the sum was thevalue of the index of patent protection, with higher values indicating greater protection. Weuse value of the index computed in 1985. The definitions of all these variables are containedin the Data Appendix.18

16When we look at the total number of firms in the country, Italy has by far the most firms. One must be

careful, though, because the number of sectors reported differ by country. This might explain the relatively

low number of firms in Austria and Denmark. The anomalous observation is Greece, which reports a sizeable

number of sectors, but appears to have very few firms. The reason is that only enterprises with more than 10

employees are reported in Greece. While this biases the average size upwards, our typical firm size minimizes

the effect. Nevertheless, we verify that our results are robust to the exclusion of Greece.

17These are the residuals across countries from the regression:

log (Employee weighted average sizec,s) = cons + βs · ds + εc,s

where, ds is a vector of sector dummies.

18The use of institutional variables from an earlier time period than the one for which firm size data is

available might work in our favor. Using these “pre-determined” variables might partially address issues

15

Two facts are worth pointing out. First, despite the homogeneity of the sample (all thecountries are European and would be classified as developed), there is enough variation inmost of the explanatory variables. For example, per capita income varies between $6,783(Greece) and $16,245 (Switzerland) and the measure for human capital varies between 3.8years (Portugal) and 10.4 (Norway). Likewise, judicial efficiency, an important variable inour analysis, ranges from 5.5 to 10.

The second fact is that some of these measures are highly correlated. For example,judicial efficiency has a correlation of 0.9 with human capital. However, using interactionsto test the specific mechanics of the theories will shed some light on their individual effectson firm size.

3 Cross-Industry Correlations

As mentioned earlier, the empirical results involving interactions between industry-level andcountry-level variables are the most important and novel aspect of our study. However, webriefly present our analysis of the correlation between firm size and each of these sets ofvariables first, to see what they reveal about each type of theory and to identify the broadcorrelations that hold in data.

3.1 Proxies

As a proxy for the size of the actual market we use the log of total employment in theindustry in that country.19 There are two problems with this measure. First, theories (e.g.,Smith (1776)) obviously refer to the potential market. Second, and following from the first,there may be spurious correlation between average firm size and our measure of the marketsize. For example, in the case of monopolies, there will be a one-to-one correspondencebetween average firm size and our proxy for the size of the market. To correct for this, weinstrument our measure of market size with the logarithm of GDP, the country population,and the ratio of exports to GDP (1960-85 average). These country level variables should beuncorrelated with industry level constraints, but should be correlated with the size of thepotential market, hence they should be good instruments.20

of endogeneity such as larger firms lobbying for more efficient judicial systems and a high level of patent

protection.

19Whenever we require the logarithm of a variable, we always add one before taking logs.

20Note that we are using log total GDP and population individually as instruments instead of per capita

GDP. Indeed, we use log per capita GDP as a regressor while studying cross-country correlations, which will

require us to drop either total GDP or population as an instrument there.

16

While we do not have a direct measure of capital stock in an industry to enable us tocalculate physical capital intensity, we have the gross investment in an industry. Dividingthis by the number of workers in that industry, we have investment per worker. In order toobtain a more exogenous measure, we take the mean of this variable across all the countriesfor which we have this data.21

To measure an industry’s dependence on intangible capital, we use the R&D done in thatindustry as a share of the country’s total R&D. Unlike physical capital, we do not computeper worker intensity of R&D because R&D is a nonrival input that impacts the whole firm.22

The R&D data is from OECD’s ANBERD database, with appropriate conversion done fromISIC to NACE. Again, the mean across all countries is used.

Finally, Rajan and Zingales (1998a) compute an industry’s dependence on external fundsas the fraction of capital expenditure in that industry in the United States funded fromexternal sources. We use the Rajan and Zingales measure of external dependence and weightit by the investment per worker in an industry to get the amount per worker that has to beraised from external sources.

The maintained assumption in using these proxies is that there are technological char-acteristics of certain industries that carry over countries.23 In other words, if Drugs andPharmaceuticals is more research intensive in the United States than Leather goods, it willcontinue to be so in Italy. While this assumption allows us to use independent variablesthat are likely to be more exogenous, and also overcome the paucity of data, a failure of thisassumption means, of course, that some of our independent variables are noise and shouldhave little explanatory power.

3.2 Results

We report in Table 3, the estimates from a regression of log typical firm size on industrycharacteristics, when the size of the market (measured as total sectoral employment) isinstrumented. The partial correlation of market size with firm size is positive, though notstatistically significant.24

21This includes the United States. Dropping the United States in the calculation does not change the

results qualitatively. Neither does using the median across countries. All industry variables are winsorized

at the 5% and 95% levels to reduce the effects of outliers.

22The intensity with respect to sales, rather than to employees, is an alternate measure, but given the

paucity of sales data we chose the sectoral R&D share.

23This is also the assumption made in Rajan and Zingales (1998a). Since we are primarily interested in

the sign of coefficients, what we really require is that the relative relationship between industries carry over

rather than the precise levels.

24Henceforth, “significant” will denote significance at the 10% level or better.

17

We also include investment per worker, R&D intensity, wages per worker, and amountfinanced externally per worker. The first three explanatory variables are positively and sig-nificantly correlated with size, while the amount financed externally is negatively correlated,but is not statistically significant. While we expected investment, wages, and R&D expen-diture to be positively correlated, as discussed in Section 1.3, we had no strong prior on thesign of the correlation between the amount financed externally and size.

The magnitudes of the effects are also considerable. According to the estimates in Table3, an increase in investment per worker, R&D intensity, and wage per worker from the 25thpercentile to the 75th percentile of the variable are associated with an increase in log firmsize of 18%, 14%, and 43% of its inter-quartile range respectively. The explanatory power ofthe regression is also considerable (R2 = 0.45).

These significant correlations are robust to the inclusion of country fixed effects (whichnecessitates dropping country level instruments), and including the explanatory variablesone at a time. These robustness results have been omitted from the paper in the interestof brevity, and to focus attention on the section on interactions. These and other omittedresults can be obtained from the authors on request.

Which theories are these correlations consistent with? Recall our discussion of the em-pirical implications of existing theories in Section 1. If industries are located in specific areasand there is a high cost to labor mobility, or if agents have industry specific human capital,then the Lucas (1978) model could be applied industry by industry in a country.25 Thepositive partial correlation between investment per worker and size is consistent with Lucas.It is also consistent with Critical Resource theories of the firm where a firm of larger size iseasier to control when the critical resource is physical capital.

Lucas does not distinguish between investment in physical capital and investment inR&D. Thus we should expect a positive correlation between size and R&D expenditure.Similarly, from the perspective of Critical Resource theories, if the critical resource is intel-lectual property and it is protected to some extent by patent laws, we should expect such acorrelation.

The positive correlation between wages per worker and size is consistent with the thrustof Lucas’ (1978) notion that the incentive to become an entrepreneur is relatively small whenwages are high. It is also consistent with Rosen’s view that large firms have better managerswho are paid more, or Kremer’s (1993) view that if wages are a proxy for the quality of aworker, higher wages should be associated with larger firms.

In this regression, we are mainly constrained by the availability of data on investment per worker and wage

per worker; the number of observations is thus smaller than the one in the regressions in Table 4.

25The immobility of labor is particularly relevant in the European context. For instance, Decressin and

Fatas (1995) show that economic shocks in Europe are mostly absorbed by changes in the participation rate

while, in the US, it is immediately reflected in migration.

18

Finally, the insignificant correlation between the amount financed externally and sizesuggests the adverse effects on average size of financial constraints on the growth of existingfirms may offset the positive size effect caused by reduced entry of new firms in financiallydependent industries.

Before we conclude this section, we should point out that we are not the first to findsome of these patterns. For example, Caves and Pugel (1980) find that size is correlatedwith capital intensity within industry, while the relationship between firm size and totalfirm R&D investment within an industry is well known (see Cohen and Klepper (1996)for references). However, our finding is across industries and countries and, therefore, weestablish the greater generality of these patterns.

In summary, the cross-industry correlations are consistent with multiple theories and donot provide enough room to discriminate between technological and organizational theories.Nevertheless, the correlations are not irrelevant in that they provide a minimum set ofpatterns that theories should fit.

4 Cross-country correlations

Let us now examine the partial correlations of country level variables with firm size.

4.1 Results

The first variable we include in the regression is the log of per capita income. This can beinterpreted as a proxy for per capita capital (as in Lucas (1978)) or, more generally, as ameasure of a country’s wealth. As a measure of human capital, which is suggested by thetechnological theories, we use the average years of schooling in the population over age 25.This comes from the Barro-Lee database and is measured in the year 1985.

In Table 4, column I, we present partial correlations of size with the cross-country ex-planatory variables. We include industry fixed effects and the size of the market, which isinstrumented by the log of GDP and the ratio of exports to GDP.26 The coefficient on themarket size is now positive and highly significant; an increase in the market size from the25th percentile to the 75th percentile is associated with an increase in log firm size by 26%of its inter-quartile range. Other than market size, the only variable with a positive andhighly significant coefficient is the efficiency of the judicial system. An increase in judicialefficiency from the 25th percentile to the 75th percentile is associated with an increase inlog firm size by 45% of its inter-quartile range. The coefficient on log per capita income is

26Since we use log per capita GDP as an explanatory variable, of the three instruments used earlier, log

GDP, log population, and ratio of export over GDP, we can use only two otherwise they would be perfectly

collinear with log per capita GDP.

19

significant but negative. However, when included alone (column II), the coefficient on percapita income is positive, while when human capital is included (column III), the coefficienton human capital is positive and significant, while the coefficient on per capita income turnsnegative.

We also inserted a number of other variables in our specification, one at a time, in placeof judicial efficiency in the specification in column I. We include Ginarte and Park (1997)’sindex of the protection given to patent rights in different countries, the quality of accountingstandards which Rajan and Zingales (1998a) argue is a good proxy for the extent of financialsector development in a country, as also proxies for regulatory constraints and a measure ofproduct liability. It turns out that none of these is significant alone or when judicial efficiencyis included. The coefficient estimate for judicial efficiency, however, always remains positiveand highly statistically significant, and that for log per capita income is negative.

The “stylized fact” that richer countries have larger firms seems true only when weexamine the obvious difference between the size of firms in really poor countries where thereis little industry to speak of, and those in the rich developed countries, and when we do notcorrect for differences in institutions.27 Within the set of industrialized countries, however,there seems to be little evidence of a significant positive partial correlation between percapita GDP and size, or human capital and size.28

With reference to the theories discussed in Section 1, the large positive correlation offirm size with judicial efficiency is consistent with multiple classes of theories. An efficientlegal system eases management’s ability to use critical resources other than physical assetsas sources of power. This leads to the establishment of firms of larger size (see Rajan andZingales (2001)). It also protects outside investors better and allows larger firms to befinanced (see La Porta et al. (1997a,1998)). Finally, it reduces coordination costs and allowslarger organizations (Becker and Murphy (1992)).

This cross-country analysis should be interpreted with caution because it is most sen-sitive to differences in the definition of enterprise across countries. Moreover, some of theexplanatory variables are strongly correlated. For example, judicial efficiency has a correla-tion of 0.90 with our measure of human capital, 0.65 with accounting standards, and 0.52with patent protection. This is a traditional problem with cross-country regressions – allmeasures of institutional and human capital development are typically highly correlated.

27See Gollin (1998) for an example of a study that focuses on economic development and firm formation.

28One could argue that our dependent variable is a proxy for labor intensity, and high per capita income

countries could be substituting cheap capital for labor. This would yield a negative correlation between

size and per capita income. To check that this is not driving our result, we would need data on sales.

Unfortunately, data on sales present in Enterprises in Europe are sporadic. Nevertheless, we experimented

with the limited sample available. Even with sales weighted average sales as dependent variable, the coefficient

estimate for per capita income is never significantly positive.

20

This is one of the reasons we stress our results on interactions in the next section, where wecontrol for both country and industry effects.

Nevertheless, two results seem to emerge from the cross-country analysis. First, thecorrelation between per-capita income and firm size is not as clear as previously thought andmay, in fact, be a proxy for institutional development. Second, judicial efficiency seems tohave the most clear cut correlation with firm size and deserves more detailed study.

5 Interactions

Cross-country regressions could be biased by differences in the definition of the enterprise aswell as effectively have fewer degrees of freedom. Hence, it is hard to estimate anything withaccuracy, and know whether something is really a proxy for what it purports to be. Oneway to reduce both these problems is to test predictions that rely on an interaction betweencountry and industry characteristics, after controlling for both country and industry effects.By doing so, we not only use more of the information in the data, but also test a more detailedimplication of the theory, which helps distinguish theories that have the same prediction fordirect or level effects. Organizational theories, which model the micro mechanisms in greaterdetail, are more amenable to such tests.

5.1 Predictions and Proxies

As we have seen, greater judicial efficiency is correlated with bigger firms. In all the countriesin our sample, basic property rights over physical assets are protected, and guarantee ownersa certain degree of power. However, the increased protection afforded to intellectual property,management techniques, firm-client relationships, etc., by a more efficient judicial systemshould allow more resources (even inalienable ones) to come into their own as sources ofpower. Therefore, according to Critical Resource theory, we should expect judicial efficiencyto particularly enhance management’s control rights for firms with relatively few physicalassets resulting in larger firms in such industries. This will imply the interaction betweenjudicial efficiency and investment per worker (a proxy for physical asset intensity) should benegatively correlated with size.29

Indeed, when we divide our sample into “knowledge” intensive and non-intensive indus-tries and rerun the cross-country regression in Table 4, column I, both the coefficient on andsignificance of judicial efficiency are higher for the knowledge-intensive sample. This provides

29Note that this is not a direct implication of theories like Lucas (1978), which emphasize the rents created

for workers by physical capital but not its control properties, or Becker and Murphy (1992), which, refers to

the coordination benefits of a better judicial system without emphasizing specific channels through which it

works.

21

some confirmation of the above reasoning as well as a greater incentive to understand themechanism by which judicial efficiency increases firm size.30

5.2 Results

One of the advantages of looking at interaction effects is that we can include both countryand industry indicators to absorb all the direct effects. Thus we do not need to worry aboutwhich country or industry variables to include. The problem, however, is that we cannotinstrument the size of the market because the instruments are collinear with the countryindicators. This is why we will try specifications both with and without country indicators.

In Table 5, we report estimates for the coefficient of the interaction variable (includedalong with market size, country indicators, and industry indicators). Capital intensive in-dustries have relatively smaller firms in countries with better judicial systems. In column I,an increase in judicial efficiency from its 25th percentile to its 75th percentile is associatedwith a decrease in the difference in log average size between firms in industries at the 75thpercentile of capital intensity and firms in industries at the 25th percentile of capital intensityof approximately 19% of the inter quartile range of log size. Put differently, in countries witha better legal system, the evidence indicates that the difference in size between automobilemanufacturers and consulting firms is smaller.

The negative correlation between the judicial efficiency-capital intensity interaction andfirm size is therefore consistent with Critical Resource theories of the firm. The importanceof the interaction effect is in giving us greater assurance that the main effects (such as theeffect of capital intensity or judicial efficiency on firm size) are correlated, at least in part, forthe particular theoretical reasons we attribute to them. Perhaps a greater reason to focus oninteraction effects is their value in distinguishing otherwise hard-to-disentangle level effects.31

Recall that both physical investment per worker and R&D intensity were positively corre-lated with firm size. Critical Resource theory, however, predicts different interaction effects.

30The knowledge-intensive industries are: Chemicals, Manufacturing of office machinery, Electrical engi-

neering, Instrument engineering, Communication, Banking & finance and auxiliary businesses, Insurance,

Education, R&D, and Health.

31Could judicial efficiency be a proxy for low levels of tax evasion, given that a better judicial system is

generally associated with better tax enforcement? This would be consistent with the finding that countries

with a better judicial system (less easy to evade taxes) have bigger firms, since the rationale for staying

small to evade taxes is absent in such economies. However, if tax evasion were the real determinant of size,

capital intensive firms, which are more easily detectable, have a particularly strong incentive to stay small

and evade taxes when enforcement is ineffective. These are the firms most likely to find the rationale for

staying small disappear with improved judicial efficiency, and the interaction effect of capital intensity and

judicial efficiency should be positive in size regressions. But this is contrary to the results presented in Table

5; thus tax evasion is unlikely to be the real reason behind the regression results on judicial efficiency.

22

The theory would suggest that R&D intensive firms, which rely more on intangible assets,should benefit much more from protections offered by the law. So R&D intensive firms shouldbe larger in countries with better patent protection. In column II, we include the interactionbetween R&D and patent protection. The interaction effect is positive, suggesting that firmsin R&D intensive industries are larger in countries with better patent protection, but it issignificant at only the 20th percent level.

The problem with including country indicators is that we cannot instrument for the sizeof the market, and we know there may be significant biases in including it directly. So incolumn III, we drop country indicators but include instruments, and also the level of judicialefficiency and the level of patent protection. The interaction between judicial efficiencyand capital intensity is again negative and statistically significant, while the direct effect ofjudicial efficiency is positive and significant. The interaction between patent protection andR&D intensity is again positive, but now statistically significant. Interestingly, the directcorrelation between firm size and patent protection is now negative. This correlation wasobscured in earlier specifications.

A high level of patent protection can be the result of the presence of large R&D-intensivefirms, which lobbied for it, rather than the cause. To mitigate the effects of this possiblereverse causation, in column IV we instrument patent protection (both as a level and in theinteraction with R&D intensity) with the rule of law index, which captures how stronglylaws in general are enforced in a country. The results are essentially unchanged. The effectof patent protection is quantitatively bigger and still statistically significant, both as a leveland interacted with R&D intensity.

While the interaction effects seem consistent with Critical Resource theory the directeffect of patent protection seems puzzling. Why, as the estimates indicate, is greater patentprotection associated with a lower size of firm outside R&D intensive industries while it isassociated with a higher size of firm in R&D intensive industries? One explanation consistentwith Critical Resource theory is the following: In countries where R&D is poorly protected,it may be that firms relying on R&D alone cannot command critical resources sufficientto achieve scale. This is why firms from industries based on other, more appropriable,resources, have the scale and thus a comparative advantage, in undertaking R&D. Whenpatents become better protected, firms in R&D intensive industries can themselves reachthe necessary scale. Thus R&D activity migrates away from firms outside R&D intensiveindustries to firms in those industries, reducing the size of the former, and increasing thesize of the latter.

While this is admittedly an ex post-facto rationalization, there is some evidence consistentwith it. If, in fact, there is migration of activity across firms in different industries, we shouldsee the share of employment in R&D intensive industries increase in countries with betterpatent protection. This is indeed the case. When we run a regression of an industry’s shareof total employment in a country against country indicators, industry indicators, and the

23

interaction between the industry’s R&D intensity and patent protection in that country, wefind a positive and significant coefficient on the interaction. Thus R&D intensive sectors area greater fraction of total employment in countries with better protection of R&D.

5.3 Why Does a Better Judicial System Reduce the Impact of Capital

Intensity on Size?

Thus far, the results do not tell whether the negative interaction effect between improvedjudicial efficiency and capital intensity comes from firms with relatively few physical assetsbecoming larger or from capital intensive firms becoming smaller. Being able to distinguishbetween these effects would provide more insight into why we observe this correlation. Wehave argued that improved judicial efficiency will enable management to gain control fromlegal devices other than ownership rights over physical assets. This suggests that the effectshould largely come from the increase in size of firms in industries that are not physical capitalintensive (this will also provide additional support for the causal link we want to draw forthe R&D intensity-patent protection interaction). There is, however, another possibility. Ifthe residual rights arising from ownership of physical assets are what distinguish firms frommarkets, as judicial efficiency improves, contractibility improves, and the rights associatedwith physical assets become less important. Also, physical assets become easier to finance fordeparting employees, therefore becoming less unique and well protected, and again residualcontrol rights associated with them diminish. This implies that capital intensive firms shouldbecome smaller with improvements in judicial efficiency, which could also explain the result.

To test this, we replace the interaction variable with judicial efficiency multiplied byindicators if an industry is in the highest or lowest tertile of physical capital intensity inTable 5B. As column I shows, the coefficient on the interaction between judicial efficiencyand the highest tertile indicator is negative but not significant. Most of the action comesfrom the lowest tertile indicator which is positive and significant. This suggests that theeffects of improvements in judicial efficiency come primarily from the growth in the size offirms that are not physical capital intensive. In column II, we re-estimate the same regressioninstrumenting for patent protection with the rule of law index. The results are essentiallyunchanged.

Therefore, the interaction effect seems consistent with the direct influence of capitalintensity and judicial efficiency on firm size as postulated by the Critical Resource theory,suggesting that these theories might be worthy of further investigation.

24

6 Discussion

6.1 Relevance of Technological Theories

The cross-industry results are consistent with the technological theories. The cross-countryevidence, however, is mixed. For example, log per capita income has, if anything, a negativecorrelation with firm size after correcting for institutional variables that should be irrelevantin technological theories. Similarly, the level of human capital has an insignificant correlationwith firm size after we correct for the efficiency of the judicial system.

This result is to be expected, as technological theories often abstract from institutionalfeatures that vary across countries and affect firm formation. They instead focus on theorganization of a typical sector within a given economy, and are thus most likely to beconsistent with time-series evidence, as in Lucas (1978), or cross-industry evidence, as in theearly part of our study.

One way to get at more detailed implications of technological theories may be to examinethe influences on the dispersion of firm size within industry. For example, Kremer’s (1993)results on assortative matching based on human capital, and higher human capital firmsundertaking more complex processes (larger firms) can be combined to get the implicationthat greater inequality in human capital would be correlated with greater dispersion in firmsize. Rosen’s (1982) model has a similar implication - with more skilled managers runninglarge firms and less skilled managers running small firms. As a measure of inequality inhuman capital we compute the coefficient of variation in educational attainments.32 InTable 6, we regress the employee-weighted coefficient of variation of firm size against thismeasure of inequality.33 We find, as predicted, the coefficient on inequality to be positiveand highly statistically significant.

One could, however, argue that institutional development reduces the importance offactors like talent and incumbency for the exercise of managerial control, and levels theplaying field. For example, greater judicial efficiency could help even small entrants securetheir property, thus ensuring they reach optimal scale for production. Therefore, we would

32Barro and Lee (1993) have data on the percentage of population over 25 in 4 categories of educational at-

tainment – none, primary, secondary, higher – for each country. They also have years of education attainment

of each type – PYR25, SYR25, and HYR25. We assign the following years of education to the four-category

frequency distribution mentioned above: 0, PYR25, PYR25+SYR25, and PYR25+SYR25+HYR25. The

coefficient of variation is then computed the usual way and used as a measure of human capital inequality.

All human capital data is for the year 1985.

33Since we need to account for the whole distribution, the typical firm size of the earlier regressions cannot

be used. Therefore, we resort to the employee-weighted coefficient of variation calculated using the employee-

weighted average firm size outlined in footnote 14 and an analogous measure of employee-weighted standard

deviation.

25

expect measures of institutional development to be negatively correlated with dispersion.This is, in fact, the case. When we include judicial efficiency in column II, it is stronglynegatively correlated with the weighted coefficient of variation of firm size, and inequality inhuman capital, while still positive, is no longer statistically significant. Finally, the inclusionof per capita income in column III does not change our conclusions.

As done for organizational theories, we could conduct more convincing tests of techno-logical theories if we could exploit the way differences in institutional environments acrosscountries affect the use of particular technologies or the organizational structures. One couldtease out of the technological theories tests for such interactions, even if it is hard to do so.For instance, in Kremer (1993) when the value added in a particular technology is high,human capital will be more important. For such technologies, firm size will be larger whenthe human capital available in the country is better. This suggests a positive correlationbetween firm size and the interaction of human capital in the country with value added perworker. This correlation is actually negative, but insignificant, when we include this interac-tion in the model in Table 5A, column III. Of course, our measures are noisy, so more workis warranted before definite conclusions are drawn.

Even if one can devise more careful tests that discriminate between technological andorganizational explanations of the size of firms, it is likely that they will be biased in favorof the latter. The firm’s definition in our dataset is closer to the legal one, which is the oneorganizational theories focus on. Clearly, they should have more explanatory power. Bycontrast, technological theories focus more on the technological limits to production. Theunit of observation to test such theories should be the length of a production process fromraw material input to final output. This may extend across several firms. Unfortunately, wedo not have data on the length of processes, hence the bias.

6.2 Vertical vs. horizontal integration

Firms could become larger because they become more vertically integrated or because theyexpand horizontally. A few theories are very specific about which of these forces leads togreater size. For example, Adam Smith’s view that size is related to the extent of the marketclearly refers to a firm expanding by doing more of the same when the market grows, i.e.,horizontal expansion. By contrast, the Property Rights view of Grossman and Hart was firstset in the context of vertical integration. In general, however, with this and most theoriesof firm size, similar predictions can be obtained regardless of whether expansion comes fromincreased vertical, or horizontal, integration. For this reason, we have purposefully avoidedthe question thus far of whether differences in size arise from differences in the level ofvertical, or horizontal, integration.

Nevertheless, it is of some interest to examine how size and the extent of vertical inte-gration are related. A traditional measure of vertical integration is the value added to sales

26

ratio (Adelman, 1955). More vertically integrated firms have higher value added per unit ofsales and, thus, will have a higher ratio. Unfortunately, Enterprises in Europe report thedata on value added and sales only for a subset of sector-country observations (249 overall).Furthermore, these do not coincide perfectly with the sector-country observations for whichwe have data on size distribution (only 146 sector-country observations have both data).Thus, our analysis is, by necessity, exploratory.

Table 7A reports the average value-added-to-sales ratio across countries (column I). Withthe exception of Denmark and Greece, all countries have a similar level of vertical integration,where the value added generated by an enterprise corresponds to roughly 35% of sales.34

The pattern is unchanged if we correct for the different industry composition in the variouscountries, as shown in column II. A possible explanation of this remarkable homogeneityacross the European countries is the long-standing presence of a valued added tax systemamong members of the European Union. Unlike sales taxes, a valued added tax is neutralwith respect to the integration decision. Thus, our sample of European countries appearsparticularly suitable to explore other determinants of size, because it is unaffected by biasesstemming from taxes on sales.

Table 7B reports the average value-added-to-sales ratio across industries (column I). Herewe do observe a lot of cross sectional variation. In fact, in an analysis of variance, sectordummies account for 58% of the overall variability, while country dummies only 10%. Asseems quite plausible, the least vertically integrated sectors are Agents, Wholesale Distri-bution, Coke Oven, Retail Distribution, and Oil Refining. At the other extreme, the mostvertically integrated sectors are Railways, Services, and Research & Development.

How much of the differences in size across industries and countries are driven by differ-ences in the degree of vertical integration? To examine this, we determine the correlationbetween our measure of the “typical” firm’s size and the degree of vertical integration. Sur-prisingly, the correlation is approximately zero (0.006). This continues to be true if wecontrol for country fixed effects and industry fixed effects. This is consistent with Jovanovic(1993), who documents empirical evidence on the joint trend toward greater size and diver-sification (i.e., horizontal integration) of firms in the United States and other countries. Iffirms achieve greater size largely through horizontal integration we may indeed find that sizeand the degree of vertical integration are uncorrelated.

We then substitute the measure of firm size with the value-added-to-sales ratio and re-estimate all the above specifications in Tables 3 and 4 (estimates not reported). While thecoefficient signs are generally the same as in the tables, none of the coefficients is statistically

34The different levels of integration in Denmark and Greece are probably attributable to differences in the

definitions of enterprise. As reported by Eurostat (Enterprises in Europe 3 vol. I, p. xxvi) Greece uses a

narrower definition of enterprise, which is more similar to the notion of establishment, while Denmark uses a

broader definition, based on the legal entity.

27

significant. The only exception is the log of per capita income, which enters positively andsignificantly when the dependent variable is vertical integration, and negatively and insignifi-cantly when the dependent variable is size. While the limited sample at our disposition (146observations) and the likely noisiness of the measure prevent any strong conclusions, theresults seem to indicate that the differences in size we capture are not driven by differencesin the extent of vertical integration.

7 Conclusion

We believe we have uncovered a number of regularities that can provide some answers to thequestions that started this paper as well as offer guidance for future empirical and theoreticalwork on the determinants of firm size. The broad patterns are that firms in capital-intensiveindustries are larger, as are firms in countries with efficient judicial systems. The cross-industry results are consistent with both technological and organizational theories, whilethe cross-country results are understandably less supportive of the technological theories.The most novel aspect of this study is the use of interactions between institutional andtechnological variables to test detailed mechanics of the theories; organizational theoriesare particularly amenable to such tests. A central finding is that firms in capital intensiveindustries are relatively smaller when located in countries with efficient judicial systems,which lends support to organizational theories. Moreover, R&D intensive industries havelarger firms when patents are better protected, though firms outside R&D intensive industriesare smaller. We have addressed the issue of causality to some extent by investigating thedetailed implications of theories.

Our work suggests that more detailed investigation of theories such as the Critical Re-source theory is warranted. Such an analysis, we hope, will eventually enable us to shedmore light on the following question: Which mergers make sense from an organizational per-spective? Which projects should be left to a separate entity, as opposed to being undertakenas a joint venture, a cooperative alliance, or an in-house venture? Much remains to be done.

28

Table 1

Firm Size by Country

The source of the data is Enterprises in Europe which provides us with the distribution of firm size

in each NACE two-digit industry in a number of European countries in 1991-92. The average number of

employees is calculated for each country-sector combination; the average of this measures across sectors is

presented for each country as the average size. We then locate the bin in which the median employee of a

country-sector combination works. We divide the total employment in that bin by the number of firms in that

bin to get the “typical” firm size; the average of this measures across sectors is presented for each country as

the “typical” size. Relative size correcting for industry effects is the average residual for that sector obtained

after regressing the logarithm of typical firm size on industry indicators. It can be interpreted as the relative

deviation of the size of firms in that country after purging industry effects.

A: Different Measures of Size

Relativesize Number

Number of correcting of sectorsAverage firms in the “Typical” for industry withfirm size country firm size effect data

AUSTRIA 77.5 8,124 286.7 0.46 8BELGIUM 322.8 177,973 1,366.8 0.46 53DENMARK 21.9 53,400 302.7 -0.05 8FINLAND 166.3 199,942 1,234.9 0.18 47FRANCE 39.8 1,555,064 1,222.0 0.32 29GERMANY 69.2 2,159,489 949.2 0.62 33GREECE 47.0 8,342 226.1 -0.82 22ITALY 100.7 3,242,062 2,346.4 -0.51 54NETHER 32.0 154,364 379.0 -0.09 17NORWAY 96.4 83,018 305.4 -0.23 33PORTUGAL 21.1 280,185 233.7 -0.74 17SPAIN 70.5 2,373,379 1,136.5 -0.45 49SWEDEN 28.4 132,662 508.6 0.39 16SWITZ 63.0 268,653 1,378.0 -0.13 50UK 37.8 1,165,907 2,674.0 0.95 25

29

Table 2

Summary Statistics

A detailed description of all the variables as well as their sources is contained in the data appendix.

A: Summary statisticsCountry variables Mean Median Std. Deviation Minimum Maximum NPer capita income 12,642 13,281 2,698 6,783 16,245 15Human capital 7.949 8.572 1.905 3.827 10.382 15Inequality in human capital 0.295 0.265 0.135 0.127 0.625 15Judicial efficiency 8.767 9.500 1.616 5.500 10.000 15Patent protection 3.489 3.710 0.621 1.980 4.240 15Accounting standards 64.600 64.000 11.488 36.000 83.000 15Sectoral variablesInvestment per worker 0.009 0.006 0.007 0.002 0.025 36R&D intensity 0.031 0.016 0.044 0.0002 0.164 33Sector wage 0.028 0.028 0.010 0.013 0.046 36External dependence 0.004 0.002 0.004 -0.001 0.015 34

Variables that changeby country and sectorLog of size of firms 4.804 4.575 2.238 0.629 11.118 463Size of the market 10.535 10.786 2.186 1.099 14.940 463Judicial efficiency x 0.067 0.049 0.052 0.010 0.254 348investment per workerPatent protection x 0.114 0.048 0.164 0.0004 0.695 316R&D intensity

B: Correlation across country variablesCountry variables log typical Log per capita Human Inequality in Judicial Patent

firm’s size income Capital human capital Efficiency ProtectionLog per capita income 0.464Human capital 0.520 0.699Inequality in human capital -0.637 -0.572 -0.744Judicial efficiency 0.661 0.760 0.901 -0.711Patent protection 0.432 0.790 0.422 -0.438 0.520Accounting standards 0.551 0.606 0.696 -0.609 0.651 0.397

C: Correlation across sectoral variablesSectoral variables log typical Capital R&D Sector

firm’s size intensity intensity wageInvestment per worker 0.349R&D intensity 0.088 0.124Sector wage 0.593 0.651 0.371External dependence 0.023 0.532 0.431 0.465

30

D: Correlation across variables that changeboth by country and sector

Variables that change log typical Market Jud. eff. x Pat. prot.by country and sector firm’s size size cap. int. R&D int.Size of the market 0.021Judicial efficiency x 0.326 -0.361investment per workerPatent protection x 0.216 -0.023 0.162R&D intensity

31

Table 3

Cross-Industry Determinants of Firm Size

The dependent variable is the logarithm of the typical firm’s size in each NACE two-digit industry in

each country. The size of the market is measured as the logarithm of total employment in a NACE two-digit

industry in a country. Investment per worker is from OECD’s ISIS Database. The average across European

countries and the US is used. Wage per worker is from the same database, and the average across European

countries and the US is used. For an industry’s R&D intensity, we use the R&D done in that industry as

a share of the country’s total R&D. The R&D data is from OECD’s ANBERD database, with appropriate

conversion done from ISIC to NACE. Again, the average across European countries and the US is used.

Amount financed externally is the product of capital expenditures that U.S. companies in the same NACE

two-digit sector finance externally (see Rajan and Zingales, 1998a) and investment per worker calculated

above. Estimates have been obtained by instrumental variable estimation, where the instruments for the size

of the market are the logarithm of population, GDP, and the ratio of export to GDP. Heteroskedasticity-

robust standard errors are reported in parenthesis. The standard errors are also adjusted for the possible

dependence of observations in the same industry across different countries.

Size of the market 0.13( 0.08 )

Investment per worker 64.32( 28.87 )

R&D intensity 15.33( 3.95 )

Sector wage 122.86( 36.94 )

External dependence -121.48( 86.31 )

R-squared 0.45N 266

32

Table 4

Cross-Country Determinants of Firm Size

The dependent variable is the logarithm of the typical firm’s size in each NACE two-digit industry in

each country. The size of the market is measured as the logarithm of total employment in a NACE two-

digit industry. Per capita income is the log of per capita income in 1990. The data are from the Penn World

Table, Mark 5.6. Human capital is measured as the average number of school years (Barro-Lee, 1993). Judicial

efficiency is an assessment of the “efficiency and integrity of the legal environment as it affects business”. It is

an average of the index between 1980 and 1983. The table reports the instrumental variable estimates, where

the instruments for the size of the market are the logarithm of population and the ratio of export to GDP.

Heterosckedasticity-robust standard errors are reported in parenthesis. The standard errors are also adjusted

for the possible dependence of the errors of observations in the same countries across different industries.

Industry fixed effects are included in all columns.

I II III

Size of the market 0.31 0.05 0.27( 0.05 ) ( 0.12 ) ( 0.12 )

Per capita income -1.04 1.18 -0.40( 0.43 ) ( 0.40 ) ( 0.70 )

Human capital -0.06 0.27( 0.12 ) ( 0.11 )

Judicial efficiency 0.47( 0.12 )

R-squared 0.78 0.72 0.76N 463 463 463

33

Table 5

Interaction Effects

The dependent variable is the logarithm of the typical firm’s size in each NACE two-digit industry in

each country. The size of the market is measured as the logarithm of total employment in that NACE

two-digit industry in a country. Per capita income is the log of per capita income in 1990 as reported by

the Penn World Table, Mark 5.6. Judicial efficiency is an assessment of the “efficiency and integrity of the

legal environment as it affects business.” Investment per worker is from OECD’s ISIS Database. The average

across European countries and the US is used. Patent protection is an index that measures the strength of

patent laws, with higher values indicating greater protection. The source is Ginarte and Park (1997). We

use value of the index computed in 1985. For an industry’s R&D intensity, we use the R&D done in that

industry as a share of the country’s total R&D. The R&D data is from OECD’s ANBERD database, with

appropriate conversion done from ISIC to NACE. Again, the average across European countries and the US

is used. Columns I and II of Panel A are estimated by OLS and contain industry fixed effects and country

fixed effects. Column III of Panel A is estimated using the logarithm of population and the ratio of export to

GDP as instrumental variables for the size of the market, and does not include country fixed effects. Column

IV additionally instruments patent protection with the rule of law index. In panel B, low investment per

worker is a dummy equal to one for sectors in the lowest tertile of investment per worker. High investment

per worker is a dummy equal to one for sectors in the highest tertile of investment per worker. Industry fixed

effects and instrumental variables, as in column III of Panel A, are used. Heteroskedasticity-robust standard

errors are reported in parenthesis.

Panel A

I II III IVSize of the market 0.68 0.60 0.32 0.43

( 0.08 ) ( 0.10 ) ( 0.09 ) ( 0.15 )Judicial efficiency X -22.71 -28.72 -26.72 -29.88investment per worker ( 7.04 ) ( 9.51 ) ( 10.25 ) ( 10.63 )Patent protection X 4.15 4.57 10.5R&D intensity ( 2.87 ) ( 2.60 ) ( 4.25 )Judicial efficiency 0.55 0.62

( 0.11 ) ( 0.12 )Patent protection -0.54 -1.04

( 0.24 ) ( 0.44 )R-squared 0.78 0.80 0.76 0.76N 348 266 266 266

34

Panel B

I IISize of the market 0.33 0.42

( 0.09 ) ( 0.14 )Patent protection X 4.42 9.98R&D intensity ( 2.65 ) ( 3.99 )Judicial efficiency 0.32 0.35

( 0.09 ) ( 0.09 )Patent protection -0.50 -0.93

( 0.23 ) ( 0.40 )Judicial efficiency X -0.05 -0.05high investment per worker ( 0.11 ) ( 0.11 )Judicial efficiency X 0.16 0.21low investment per worker ( 0.06 ) ( 0.08 )R-squared 0.76 0.76N 266 266

35

Table 6

The Determinants of Dispersion in Firm Size

The dependent variable is the employee-weighted coefficient of variation of the number of employees

per firm in each NACE two-digit industry in each country. Inequality in human capital is measured as the

product of the percentage of the population with the highest educational achievement and percentage of

the population with the lowest educational achievement. The numbers are from Barro-Lee, 1993. Judicial

efficiency is an assessment of the “efficiency and integrity of the legal environment as it affects business.” Per

capita income is the log of per capita income in 1990 as reported by the Penn World Table, Mark 5.6. All

estimates are obtained by OLS and contain industry fixed effects. Heteroskedasticity-robust standard errors

are reported in parentheses. The standard errors are also adjusted for the possible dependence of the errors

of observations in the same countries across different industries.

I II IIIInequality in human capital 1.22 0.25 0.09

( 0.51 ) ( 0.56 ) ( 0.45 )Judicial efficiency -0.12 -0.19

( 0.03 ) ( 0.05 )GDP per capita 0.60

( 0.24 )R-squared 0.68 0.71 0.72N 463 463 463

36

Table 7

Vertical Integration

The source of the data is Enterprises in Europe which, for a subset of countries, provides us with the

value added in each NACE two-digit industry. The measure of vertical integration used is the ratio of the

value added to sales. The first column in Panel A is the average of this ratio across industries in the same

country. The second column is the average residual for that country obtained after regressing the logarithm

of the value added to sales ratio on industry indicators. It can be interpreted as the percentage deviation of

the ratio in that country after purging industry effects. The first column in Panel B is the average of the

value added to sales ratio across countries. The second column is the average residual for that sector obtained

after regressing the logarithm of the value added to sales ratio on country indicators. It can be interpreted

as the percentage deviation of the ratio in that sector after purging country effects.

A: At the country level

Value added/Turnover

Average correctingValue added/ for sector

Turnover effects N

AUSTRIABELGIUMDENMARK 0.495 0.343 36FINLANDFRANCE 0.397 -0.024 54GERMANY 0.337 -0.032 30GREECE 0.263 -0.240 16ITALY 0.340 -0.023 23NETHER 0.376 -0.028 36NORWAYPORTUGAL 0.353 -0.063 37SPAINSWEDEN 0.322 -0.139 17SWITZUK

37

B: At the industry level

Average correctingValue added/ for country

Turnover effects N

Extraction solid fuels 0.507 0.328 2Coke ovens 0.164 -0.836 2Extraction petroleum 0.510 0.372 1Oil refining 0.208 -0.886 4Nuclear fuels 0.453 0.305 2Electricity 0.509 0.327 4Water supply 0.567 0.515 4Extraction-metal 0.378 0.084 4Production-metal 0.295 -0.289 3Extraction-other 0.425 0.150 4Manufacture-non metal 0.387 0.145 7Chemical industry 0.300 -0.139 7Man-made fibers 0.271 -0.205 2Manufacture- metal 0.358 0.024 7Mechanical engineering 0.378 0.119 7Office machinery 0.385 0.098 7Electrical engineering 0.347 0.030 7Motor vehicles 0.238 -0.344 8Other means transportation 0.317 -0.058 7Instrument engineering 0.435 0.245 7Food and tobacco 0.243 -0.343 8Textile 0.316 -0.067 7Leather goods 0.316 -0.053 7Footwear and clothing 0.335 -0.001 7Timber and furniture 0.335 0.012 8Paper products and publishing 0.378 0.129 8Rubber and plastic 0.319 0.025 7Other manufacturing 0.352 0.106 5Building and civil engin. 0.354 0.014 6Wholesale distrib. 0.145 -0.902 6Scrap and waste 0.223 -0.573 3Agents 0.134 -1.045 3Retail distribution 0.188 -0.641 6Hotels and catering 0.436 0.184 5Repair of consumer goods 0.334 -0.178 5Railways 0.954 0.999 1Other land transport 0.495 0.342 6Inland water transport 0.522 0.263 3Sea transport 0.298 -0.241 5Air transport 0.433 0.041 2Supporting services 0.737 0.704 4Travel agents 0.318 -0.220 7Communication 0.619 0.600 4Banking and finance 0.534 0.402 3Insurance 0.258 -0.269 3Auxiliary services 0.456 0.169 4Renting, leasing 0.450 0.108 3Letting of real estate 0.588 0.516 1Public administration 0.708 0.692 5Sanitary services 0.578 0.498 1Education 0.467 0.285 1Research and development 0.643 0.458 2Medical and others 0.430 0.092 3Recreational services 0.622 0.518 4Personal services 0

38

A Appendix

A.1 Formulas for Statistical Measures

Let ei denote the employment in bin i, and nidenote the number of firms in bin i for any givencountry-sector combination. Omit country-sector subscripts for convenience. Let E denote the totalemployment for the country-sector and N the total number of firms. The simple average is given by:

av =∑

i

(ni

N

) (ei

ni

)=

E

N.

The employee-weighted average is given by:

ewav =∑

i

(ei

E

) (ei

ni

).

The simple variance is given by:

var =∑

i

(ni

N

) [(ei

ni

)− av

]2

sd =√

var.

The analagous quantity for the employee-weighted variance is:

ewvar =∑

i

(ei

E

) [(ei

ni

)− ewav

]2

ewsd =√

ewvar.

The simple coefficient of variation is given by:

cv =sd

av.

The analagous one for the employee-weighted measure is:

ewcv =ewsd

ewav.

The “employee-weighted mode” is defined by:

ewmode = arg max((ei

E

) (ei

ni

)).

The average size of the modal bin is reported.

The employee-weighted median will be the first i for which 1ewav

∑i

(ei

E

) (ei

ni

)becomes ≥ 0.5.

Pearson’s measure for employee-weighted skewness is (ewmean−ewmode)ewsd .

A.2 Data Appendix

In this appendix, we discuss the data sources and the construction of our regression variables. In the

interest of brevity, we present only the main points on the construction of a unified database from

disparate sources. More comprehensive notes are available from the authors.

Employee-weighted average firm size:

39

• Data source(s): Enterprises in Europe (1994), which provides us with the distribution of

size of firms in each NACE two-digit industry (sector) of selected European countries. Data is

available for either 1991 or 1992 for all countries.

• Construction of variable: The average firm size (number of employees) in each size class

is first calculated by dividing the number of employees by the number of firms. The average

size for the entire sector is then calculated as the weighted sum of these bin averages, using as

weights the employees in the size class as a fraction of total sectoral employment.

• Scope of Variable: Country-sector

• Notes:

– The size classification varies considerably across countries, as does coverage. All sectors

for which the bin employment does not add up to total employment are eliminated.

– When data for multiple, related sectors are combined in the source, we attribute the

combined figures to the smallest sector number. For instance combined data for Chemical

industry (NACE 25) and Man-made fibres industry (NACE 26) are presented as the

former.

– Total employment numbers are used. If data for total employment is either not available

or sparsely available, we use salaried employment.

Size of the market:

• Data source(s): Enterprises in Europe (1994).

• Construction of variable: The logarithm of the

total employment in a NACE two-digit industry is used.

• Scope of Variable: Country-sector

Investment per worker:

• Data source(s): OECD: Industrial Structure Statistics (ISIS) (1997), which provides us with

data on production, value added and investment in each ISIC sector of selected OECD countries.

Data is available from 1970 to 1995, though coverage is not uniform. Employment data used

to calculate per worker quantities is also from the same database.

• Construction of variable: The total sectoral investment (gross capital formation) from ISIS

is divided by total employment from Enterprises in Europe to get the investment per worker,

and adjusted by the exchange rate to get the figure in US dollars. The average value across all

countries is computed for each sector, to get a sector-level variable.

40

• Scope of Variable: European & U.S. Sector

• Notes:

– The use of investment per worker for capital per worker is based on the assumption that

the economies studied were at their steady states. Direct data on capital per worker, at

the sectoral level, is difficult to obtain for the countries being studied.

– The investment data is for the year closest to the one for employment data.

– Data for ISIC – Revision 2 classification is used wherever available; else, the data is for

ISIC – Revision 3. These ISIC codes were translated to NACE codes; the translation used

is available upon request. The ISIC data is available at the 3-digit level, and sometimes

even at the 4-digit level; this facilitates a fairly complete translation of codes.

– The market exchange rate (average over the year) provided by ISIS is used to convert

investment in local currency to US dollars.

Value added to sales:

• Data source(s): Enterprises in Europe (1994).

• Construction of variable: The total value added in a country-sector combination (thousands

of ECUs) is divided by reported turnover for that combination (also in thousands of ECUs).

• Scope of Variable: European Sector

Wages per worker:

• Data source(s): OECD: Industrial Structure Statistics (ISIS) (1997).

• Construction of variable: The total wages from ISIS is divided by total employment from

Enterprises in Europe to get the wages per worker, and adjusted by the exchange rate to get

the figure in US dollars. The average value across all countries is computed for each sector, to

get a sector-level variable.

• Scope of Variable: European & U.S. Sector

• Notes:

– The items on year of data, classification conversion, and the use of market exchange rates

mentioned under “capital intensity” apply here as well.

R & D intensity:

• Data source(s): OECD’s ANBERD Database.

41

• Construction of variable: The share of R&D done in that industry as a share of the country’s

total R&D is computed for the analogue of the NACE 2-digit sectors.

• Scope of Variable: European & U.S. Sector

• Notes:

– Unlike physical capital, we do not compute per worker intensity of R&D because R&D is

a nonrival input that impacts the whole firm.

External dependence:

• Data source(s): COMPUSTAT – Business Segment File, OECD: Industrial Structure Statis-

tics (ISIS) (1997).

• Construction of variable: The fraction of capital expenditures that is financed externally for

the analogue of the NACE 2-digit sectors among publicly traded U.S. companies is multiplied

by the investment per worker in that sector for the European country. In other words, this is

an estimate of external financing per worker in the given country-sector combination.

• Scope of Variable: Country-sector

• Notes:

– See Rajan and Zingales (1998a) for further details.

Utility & Transport sector:

• Data source(s): Enterprises in Europe (1994).

• Construction of variable: This is a dummy variable that is set to 1 if the NACE sector is

one of Production and Distribution of Electricity (16), Railways (71), Air Transport (75), or

Communication (79).

• Scope of Variable: Sector dummy

Per capita income:

• Data source(s): Penn World Table (Mark 5.6), which provides us with country level data on

real GDP, goverment expenditure, and human capital.

• Construction of variable: The logarithm of per capita GDP (“RGDPCH” in PWT) is used

for the relevant year.

• Scope of Variable: Country

42

Human capital:

• Data source(s): Barro and Lee’s dataset for a panel of 138 countries (1994).

• Construction of variable: The average schooling years in the total population over age 25,

for the year 1985, is used (“human85” in Barro and Lee).

• Scope of Variable: Country

• Notes:

– This is an attainment variable that changes very slowly over time. Therefore, the use of

1985 data is not seen as a serious limitation.

Human capital inequality:

• Data source(s): Barro and Lee’s dataset for a panel of 138 countries (1994).

• Construction of variable:

Barro and Lee (1993) have data on the % of population over 25 in 4 categories of educational

attainment – none, primary, secondary, higher – for each country. They also have years of

education attainment of each type – PYR25, SYR25, and HYR25. We assign the following

years of education to the four-category frequency distribution mentioned above: 0, PYR25,

PYR25+SYR25, and PYR25+SYR25+HYR25. The coefficient of variation is then computed

the usual way and used as a measure of human capital inequality. All human capital data is

for the year 1985.

• Scope of Variable: Country

Judicial efficiency:

• Data source(s): Business International Corporation.

• Construction of variable: This variable is an assessment of the “efficiency and integrity

of the legal environment as it affects business, particularly foreign firms.” Data is found in a

scale form, with scores ranging from zero to ten, where the lower scores mean lower efficiency

levels.

• Scope of Variable: Country

• Notes:

– It is an average of the index between 1980 and 1983.

Patent protection:

43

• Data source(s): Ginarte and Park (1997).

• Construction of variable: Patent protection is an index that measures the strength of patent

laws, with higher values indicating greater protection.

• Scope of Variable: Country

• Notes:

– Ginarte and Park (1997) compute the index for 110 countries using a coding scheme

applied to national patent laws. The five aspects of patent laws examined were (1) the

extent of coverage, (2) membership in international patent agreements, (3) provisions for

loss of protection, (4) enforcement mechanisms, and (5) duration of protection. Each of

these categories was scored on a scale of 0 to 1, and the sum was the value of the index of

patent protection, with higher values indicating greater protection. We use value of the

index computed in 1985.

Financial development:

• Data source(s): Center for International Financial Analysis and Research.

• Construction of variable: Financial development is proxied by a measure of accounting

standards. This variable measures the transparency of annual reports.

• Scope of Variable: Country

• Notes:

– See Rajan and Zingales (1998a) for further details.

44

References

Adelman, Morris A. 1955, “Concpet and Statistical Measurement of Vertical Integra-tion,” in Business Concentration and Price Policy Princeton University Press, Prince-ton NJ.

Alchian, A. and H. Demsetz, 1972, “Production, Information Costs and Economic Or-ganization,” American Economic Review 777-705.

Audretsch, David B. and Talat Mahmood 1995, “New Firm Survival and the Techno-logical Regime,” Review of Economics and Statistics 77:1, 97-103.

Bain, Joe S., 1956, Barriers to New Competition Harvard University Press, Cambridge,MA.

Banz, Rolf W., 1981, “The Relationship Between Return and the Market Value of Com-mon Stocks,” Journal of Financial Economics 9, 3-18.

Barclay M, Smith C (1995). “The Priority Structure Of Corporate-Liabilities,” Journalof Finance 50 (3): 948-949

Barclay M, Smith C (1995). “The Maturity Structure Of Corporate-Debt,” Journal ofFinance 50 (2): 609-631.

Barro R. and J. Lee, 1993, “International Comparisons of Educational Attainment,” NBERWorking Paper.

Becker, Gary S., and Murphy, Kevin M. (1992). “The Division of Labor, Coordina-tion Costs, And Knowledge,” The Quarterly Journal of Economics, November 1992,1137-1160.

Bresnahan, Timothy, F. and Peter C. Reiss (1991). “Entry in Monopoly Markets,” .Review of Economic Studies.

Brickley , James, James S.Linck And Clifford W.Smith, Jr. (2001), “Boundaries OfThe Firm: Evidence From The Banking Industry,” Working Paper, University ofRochester.

Caves, Richard E. (1998). “Industrial Organization and New Findings on the Turnoverand Mobility of Firms,” Journal of Economic Literature, December 1998, 1947-1982.

Caves, Richard E. and Thomas A. Pugel (1980). Intra-Industry Differences in Con-duct and Performance Monograph Series, New York University.

Coase, R., 1937, “The Nature of the Firm,” Economica 4:386-405.

45

Cohen, Wesley M. , and Klepper, Steven (1996). “Firm Size and the Nature of In-novation within Industries: The Case of Process and Product R&D,” The Review ofEconomics and Statistics.

Davis, Steven J., and Henrekson, Magnus. (1997). “Explaining National Differencesin the Size and Industry Distribution of Employment,” Working Paper, August 1997,The University of Chicago Graduate School of Business.

Decressin, Jorg and Antonio Fatas (1995). “Regional labor market dynamics in Eu-rope,” European Economic Review, December 1995, 1627-1655.

Desai MA, Hines JR (1999). “‘Basket cases’: Tax incentives and international jointventure participation by American multinational firms,” Journal of Public Economics71 (3): 379-402

Eldenburg ,L., Benjamin E.Hermalin , Michael S.Weisbach And Marta Wosinska(2001). “Hospital Governance, Performance Objectives, And Organizational Form ”,Working Paper, University of Illinois at Urbana Champaign.

Fama, E. and M. Jensen (1983). “Separation of Ownership and Control,” Journal ofLaw and Economics , June.

Garicano, Louis (2000). “Hierarchies and the Organization of Knowledge in Production,”Journal of Political Economy, 108: 874-905.

Gertler, Mark and Simon Gilchrist (1994). “Monetary Policy, Business Cycles, andthe Behavior of Small Manufacturing Firms,” Quarterly Journal of Economics, 109:309-341.

Gertner R, Scharfstein D, Stein J (1994). “Internal Versus External Capital-Markets,”Quarterly Journal of Economics 109 (4): 1211-1230.

Ginarte, Juan Carlos and Walter G. Park (1997). “Determinants of Patent Rights:A Cross-National Study,” mimeo, American University, Washington D.C.

Gollin, Douglas. (1998). “Nobody’s Business but My Own: Self Employment and SmallEnterprise in Economic Development,” Working Paper, June 1998, Williams College.

Grossman, S. and Hart, O. 1986, “The Costs and the Benefits of Ownership: A Theoryof Vertical and Lateral Integration,” Journal of Political Economy 691-719.

Hart, Oliver Firms, Contracts, and Financial Structure, (Oxford: Clarendon Press, 1995).

Hart, O. and J. Moore 1990, “Property Rights and the Nature of the Firm,” Journal ofPolitical Economy 1119-1158.

46

Holmstrom, Bengt, 1999 “The Firm as a Subeconomy,” Journal of Law, Economics andOrganization,15, no 1, 74-102.

Holmstrom, Bengt and J.Roberts 1998 “ The Boundaries of the Firm Revisited,” Jour-nal of Economic Perspectives , 12: 73-94.

Jensen M (1993). “The Modern Industrial-Revolution, Exit, And The Failure Of InternalControl-Systems,” Journal of Finance 48 (3): 831-880

Jensen M. C. and W. Meckling. 1976, “Theory of the Firm: Managerial Behavior, AgencyCosts and Capital Structure,” Journal of Financial Economics 3: 305-360.

Jensen,M. and K.J. Murphy 1990, “Performance Pay and Top-Management Incentives,”Journal of Political Economy 98, 225-264.

Jovanovic, Boyan 1993, “The Diversification of Production,” Brookings Papers: Microe-conomics, 197-247.

Klein, Benjamin, Robert Crawford , and Armen Alchian. (1978). “Vertical Integra-tion, Appropriable Rents and the Competitive Contracting Process,” Journal of Lawand Economics, XXI (1978), 297-326.

Klepper, Steven (1996). “Entry, Exit, Growth, and Innovation over the Product LifeCycle,” American Economic Review, 86:3, 562-583.

Kremer, Michael. (1993). “The O-Ring Theory of Economic Development,” The Quar-terly Journal of Economics, August 1993, 551-575.

La Porta, R., Florencio, Lopez de Silanes, Andrei Shleifer, and Robert Vishny ,1997a, “The Legal Determinants of External Finance,” Journal of Finance, 52, 1131-1150.

La Porta, R., Florencio, Lopez de Silanes, Andrei Shleifer, and Robert Vishny ,1997b, “Trust in large organizations,” American Economic Review, 87:2, 333-338.

La Porta, R., Lopez-de-Silanes, F., Shleifer, A. and R. Vishny , 1998, “Law and Fi-nance,” Journal of Political Economy, 106, 1113-1155.

Lerner J, Merges RP (1998). “ The Control Of Technology Alliances: An EmpiricalAnalysis Of The Biotechnology Industry,” J IND ECON 46 (2): 125-156.

Lucas, Robert E., Jr. (1978). “On the Size Distribution of Business Firms,” Bell Journalof Economics, vol 9, 508-523.

Mata, Jose. (1996). “Markets, Entrepreneurs, and the Size of New Firms,” EconomicLetters, 52:1, 89-94.

47

Mullainathan, S. and D. Scharfstein (2001). “Do Firm Boundaries Matter?” Ameri-can Economic Review 91 (2): 195-199

Pagano, Marco, and F. Panetta and L. Zingales, 1998, “Why Do Companies Go Pub-lic? An Empirical Analysis,” Journal of Finance, 53: 27-64

Pashigian, B. Peter, , 1968, “Market Concentration in the United States and Great Britain,”Journal of Law and Economics,, 11: 299-319.

Pakes, Ariel, and Richard Ericson, , 1998, “Empirical Implications of Alternative Mod-els of Firm Dynamics,” Journal of Economic Theory, 79: 1-45.

Penrose, E. T., 1959, Theory of the Growth of the Firm J. Wiley & Sons, New York.

Rajan, R. and J. Wulf, 2002, “The Flattening Firm: Evidence from Panel Data on theChanging Nature of Firm Hierarchies,” Working Paper, University of Chicago.

Rajan, R. and L. Zingales, 1998a, “Financial Dependence and Growth,” The AmericanEconomic Review 88: 559-586.

Rajan, R. and L. Zingales, 1998b, “Power in a Theory of the Firm,” Quarterly Journalof Economics CXIII: 387-432.

Rajan, R. and L. Zingales, 2000, “The Governance of the New Enterprise,” in CorporateGovernance X. Vives ed., p. 201-227, Cambridge University Press.

Rajan, R. and L. Zingales, 2001, “The Firm as a Dedicated Hierarchy: A Theory of theOrigins and Growth of Firms,” Quarterly Journal of Economics, CXVI: 805-852.

Ringleb and Wiggins (1990) “Liability and Large-Scale, Long-Term Hazards,” Journalof Political Economy 98.

Robinson, D (2001). “Strategic Alliances and the Boundaries of the Firm,” Working Pa-per, Columbia University.

Rosen, Sherwin. (1982). “Authority, Control, and the Distribution of Earnings,” BellJournal of Economics, Autumn 1982, 311-323.

Smith, Adam (1776). The Wealth of Nations : An Inquiry into the Nature and Causes

Smith, C (2001). “Organizational Architecture And Corporate Finance,” Journal of Fi-nancial Services Research

Stigler, Geroge, 1968, The Organization of Industry, Irwin, Homewood: IL.

Sutton, John, 1991, Sunk Costs and Market Structure: Price Competition, Advertising,and the Evolution of Concentartion, MIT Press, Cambridge: MA.

48

Sutton, John (1997). “Gibrat’s Legacy,” Journal of Economic Literature, 35:1, 40-59.

Williamson, O., 1975, Market and Hierarchies: Analysis and Antitrust Implications, TheFree Press, New York.

Williamson, O., 1985, The Economic Institutions of Capitalism, The Free Press, NewYork.

Zingales, L. 2000, “In Search of New Foundations,” Journal of Finance, 55: 1623- 1653.

49