You are on page 1of 45

ARTICLE IN PRESS

Journal of Financial Economics 83 (2007) 321365 www.elsevier.com/locate/jfec

Corporate capital structure and the characteristics of suppliers and customers$


Jayant R. Kalea, Husayn Shahrurb,
a

J. Mack Robinson College of Business, Georgia State University, Atlanta, GA 30303, USA b Department of Finance, Bentley College, Waltham, MA 02452, USA

Received 2 June 2005; received in revised form 29 November 2005; accepted 19 December 2005 Available online 20 November 2006

Abstract We investigate the link between a rms leverage and the characteristics of its suppliers and customers. Specically, we examine whether rms use decreased leverage as a commitment mechanism to induce suppliers/customers to undertake relationship-specic investments. We nd that the rms leverage is negatively related to the R&D intensities of its suppliers and customers. We also nd lower debt levels for rms operating in industries in which strategic alliances and joint ventures with rms in supplier and customer industries are more prevalent. Consistent with a bargaining role for debt, we nd a positive relation between rm debt level and the degree of concentration in supplier/customer industries. r 2006 Published by Elsevier B.V.
JEL classications: G32; G33; L13; L14; L22; L24 Keywords: Capital structure; Relationship-specic investments; Implicit contracts; Market power; Buyer power

We have beneted from discussions with Vikas Agarwal, Anwar Boumosleh, Sudip Datta, Gerry Gay, Marty Grace, Atul Gupta, Srini Krishnamurthy, Kartik Raman, Stefan Ruenzi, Chip Ryan, Ajay Subramanian, Anand Venkateswaran, Lingling Wang, Chip Wiggins, an anonymous referee, and seminar participants at the 2004 Financial Management Association meetings, the 2005 European Financial Management Association meetings, Bentley College, the Indian Institute of ManagementBangalore, Institut dAdministration des Entreprises de ParisUniversite Paris (Sorbonne), Laval University, the Lebanese American University, and University of Pittsburgh. We appreciate the research assistance of Lin Hai and Jian Wen. We thank Joseph Fan and Larry Lang for providing the conversion table for the IO-SIC codes. Corresponding author. E-mail address: hshahrur@bentley.edu (H. Shahrur). 0304-405X/$ - see front matter r 2006 Published by Elsevier B.V. doi:10.1016/j.jneco.2005.12.007

ARTICLE IN PRESS
322 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

1. Introduction Since Modigliani and Millers (1958) capital structure irrelevance result, researchers have searched for capital structure explanations primarily within the context of rm boundaries that are determined by explicit contracts among stakeholders including shareholders, debtholders, managers, and the government. The research in this stream of literature provides important insights into the effects of taxes, bankruptcy costs, information asymmetries, agency issues, and other frictions on corporate leverage decisions. Building on this work, another body of research (see, e.g., Titman, 1984; Maksimovic and Titman, 1991) analyzes a rms capital structure decision in a setting in which the rms boundaries include implicit as well as explicit contracts. We contribute to this latter stream by investigating how the inclusion of suppliers and customers as stakeholders affects a rms leverage choice. We focus on two aspects of the relation between a rms debt level and its dealings in the input and output markets. First, we hypothesize that a rm can use a lower level of debt in its capital structure to induce its suppliers and customers to undertake relationship-specic (RS) investments. Our hypothesis is based on the work by Titman (1984) and Maksimovic and Titman (1991). Titman (1984) suggests that a rm with a unique product may require its customers to undertake investments that lose value if the rm goes into liquidation. In this setting, lower leverage commits the rm to a liquidation policy that takes into account the effects on its customers. Further, customers may not be willing to deal with a highly levered rm, which is less likely to worry about its reputation (Maksimovic and Titman, 1991). We apply this intuition to RS investments by suppliers and customers and hypothesize that rms that expect their suppliers/customers to undertake R-S investments will carry lower levels of debt. Our second hypothesis considers the relation between a rms choice of debt level and its bargaining position relative to its suppliers/customers. The intuition for our hypothesis follows from the extant literature on the role of debt in managementlabor union bargaining (see, e.g., Bronars and Deere, 1991), which suggests that raising the debt level ` -vis a labor union by reducing the increases the managements bargaining power vis-a amount of rm surplus available for sharing with labor. Specically, we hypothesize that a rm may choose a higher debt level when it faces suppliers/customers who have relatively higher bargaining power. The empirical implication of this hypothesis is a positive relation between a rms debt level and measures of supplier/customer negotiation power. In order to test our hypotheses, we construct two separate data sets. The rst data set identies suppliers and customers at the industry level and the second consists of key customer and supplier rms. The industry-level data offer three distinct benets relative to the rm-level data. First, the sample of rms in the industry-level data set is much larger. Second, endogeneity issues that are endemic to corporate nance research are likely to be signicantly less severe in tests that relate a rms capital structure to variables measured for supplier and customer industries than to variables measured for supplier and customer rms. Third, industry-level data allow us to relate rm nancing decisions to important variables such as the levels of buyer and supplier power that need to be measured at the industry level. The main advantage of the rm-level data set, on the other hand, is that it identies supplier and customer rms more precisely and thus the inferences based on the ndings from this data are cleaner. Further, it allows us to examine the effect of a rms leverage on the RS investments of its key suppliers and customers.

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 323

In order to test our rst hypothesis, we measure RS investment by two sets of variables. Allen and Phillips (2000) suggest that R&D-intensive industries are more likely to create relationship-specic assets. Accordingly, we base our rst set of variables for RS investments on supplier and customer R&D investments. Next, as in Fee, Hadlock, and Thomas (2005), we conjecture that rms are more likely to establish strategic alliances (SAs) and joint ventures (JVs) with suppliers/customers when there is a greater need for RS investments. Thus, the intensity of SAs and JVs that are established between the sample rms and suppliers/customers forms our second set of RS investments variables. We document signicant support for the negative relation between rm debt levels and the above measures of RS investments by suppliers/customers. We nd that a rms leverage is decreasing in the R&D intensities of its supplier and customer industries. We also nd that rms operating in industries characterized by high intensities of SAs and JVs with rms in supplier and customer industries tend to have lower debt levels. We show that the economic signicance of these relations is comparable to that of traditional determinants of capital structure. Next, we nd that the negative relations between leverage and the supplier and customer R&D variables are weaker for vertically integrated rms, which provides additional support for the RS investments hypothesis. Further, we show that the negative relation between customer R&D investments and rm leverage is more pronounced for rms that belong to a concentrated industry and for rms with high market share. This nding implies that the effect of customer RS investments on rm leverage is stronger when the rm produces an input for which there are few alternate suppliers. The analysis using rm-level data further supports our hypothesis that rms lower debt levels to induce RS investments. We nd that rm leverage is negatively related to the R&D intensity of its key suppliers and customers and that leverage is lower for rms that have established SAs or JVs with their key suppliers and customers. We then estimate simultaneous equation models that treat the rms leverage and the R&D of key suppliers and customers as endogenous variables. In addition to addressing endogeneity concerns, this specication also tests whether the rms leverage and the R&D investment decisions of its suppliers and customers are simultaneously determined. We nd that the negative relation between rm leverage and the R&D intensities of its suppliers and customers is robust to the simultaneous equation specication. In the R&D regressions, we nd negative relations between the supplier and customer R&D and expected leverage. Our ndings therefore add to the literature that studies the effect of leverage on the rms dealings in the product markets (see, e.g., Opler and Titman, 1994; Phillips, 1995). Our second set of hypotheses relates to the use of debt in improving a rms bargaining ` -vis its suppliers/customers, where we use industry concentration to measure position vis-a the negotiation power of suppliers/customers. Consistent with a bargaining role for debt, we nd that rms that face concentrated supplier and customer industries tend to have higher levels of debt. Further, the positive relation between leverage and the concentration of supplier/customer industries is weaker for rms with high market shares in their own industry. The latter nding implies that when a rm has a higher market share in its own ` -vis its suppliers/customers is greater and, hence, it has industry, its bargaining power vis-a a lower incentive to use debt as a bargaining mechanism. The rest of the paper proceeds as follows. In Section 2, we review the literature and develop the empirical hypotheses. In Section 3, we provide details of the sample and the

ARTICLE IN PRESS
324 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

methodology we use to identify suppliers and corporate customers. We present the results in Section 4. Section 5 concludes the paper. 2. Related literature and testable hypotheses Our work is part of a growing body of research that links capital structure and product market strategy; see Maksimovic (1995) for a review of this literature. Theoretical studies suggest that leverage can affect the intensity of product market competition (see, e.g., Brander and Lewis, 1986; Maksimovic, 1988; Chevalier and Scharfstein, 1996; Dasgupta and Titman, 1998). Empirical studies such as Opler and Titman (1994), Chevalier (1995a, b), Phillips (1995), Kovenock and Phillips (1997), Zingales (1998), Khanna and Tice (2000, 2005), and Campello (2003, 2005) nd evidence consistent with leverage affecting product market strategy. Our study contributes to this literature by examining the relation between the rms capital structure and the characteristics of its key suppliers and customers. Understanding this link is important since the ability of a rm to compete in the product market depends on its relations with its suppliers and customers. 2.1. Relationship-specic investments and capital structure Our work closely relates to research that studies the relation between nancial structure and implicit contracting. Titman (1984) demonstrates that the rm can commit to a liquidation policy that takes into consideration the effect of rm liquidation on customers by choosing a lower debt level. Developing this idea further, Maksimovic and Titman (1991) show that customers may be unwilling to conduct business with a highly levered rm because higher debt reduces the rms willingness to invest in its reputation and produce high-quality products. We build on the insights from the implicit contracting studies to formulate our hypothesis regarding the relation between a rms debt level and RS investments by its suppliers and customers. Suppose that a rms operations depend on its ability to induce its suppliers or customers to undertake RS investments, and further, that these investments lose value if the rm goes into liquidation. Then the rms capital structure will affect the incentives of suppliers and customers to make RS investments as long as the rms liquidation decision is causally linked to its bankruptcy status (Titman, 1984). In addition, because high debt can reduce the rms incentives to invest in its reputation (Maksimovic and Titman, 1991), and hence reduces the willingness of suppliers and customers to undertake RS investments, a rm that expects its customers/suppliers to undertake RS investments should choose lower debt levels. On the other hand, if some rms are expected to remain highly levered due to exogenous factors, one would expect the suppliers and customers of such rms to undertake less RS investments. Further, the effect of RS investments by customers on the rms leverage can be more pronounced if the rm produces an input for which there are few alternative suppliers. For instance, if a customer rm has many potential suppliers, it may decide to diversify its RS investments and deal with many suppliers that produce the same input, thereby reducing its exposure to a loss given the liquidation of any particular supplier. In addition, the loss in the value of RS investments undertaken with any given supplier may be lower if there are similar rms in the supplier industry that can salvage some of these investments in the case of liquidation. Using the concentration of the rms industry and the rms market

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 325

share as proxies for the importance of the rm to its customer, the above arguments suggest that the effect of RS investments on leverage will be weaker for rms in a disperse supplier industry and for rms with lower market shares. The above discussions relate to the use of debt as a mechanism to induce RS investments. We recognize that debt is not the only tool available to a rm for this purpose, that is, that a rm can induce the desired behavior from its trading partners through other mechanisms. For example, Williamson (1985) argues that rms contemplate vertical integration when incentive problems with suppliers/customers are severe. This observation leads to an interesting test of our hypothesis. If vertical integration is a solution to the incentive problem, then debt is less likely to be used as the incentive mechanism for vertically integrated rms. We therefore expect that the negative relations between the debt level and the proxies for RS investments will be weaker when rms integrate vertically. The intensity of R&D expenditure of suppliers and customers is our rst proxy for the extent of RS investments. The use of R&D intensity as a proxy for asset specicity is prevalent in the empirical literature on transactions cost economics; see Boerner and Macher (2001) for a recent review of this literature. Levy (1985) argues that researchintensive industries tend to involve specialized inputs that require transaction-specic investments by suppliers. In addition, vertical chains that are R&D intensive are likely to have complex interstage interdependencies (Armour and Teece, 1980). Allen and Phillips (2000) suggest that R&D-intensive industries are more likely to create relationship-specic assets. In addition to the R&D-based measures, we construct variables from data on SAs and JVs as alternative measures of RS investments between a rm and its suppliers/ customers. Fee, Hadlock, and Thomas (2005) nd that a rm is more likely to establish SAs with trading partners that are expected to undertake RS investments. As a result, RS investments are likely to be higher if the rm and its suppliers/customers have established such alliances.1 To summarize, the central hypothesis described in this section is: Hypothesis 1:. Firms that deal with R&D-intensive suppliers/customers and rms with high intensities of SAs and JVs with suppliers/customers choose lower leverage. We also test several related subhypotheses, namely: (i) the negative relations between the rms leverage and the proxies for RS investments by customers (customer R&D, and the intensity of SAs and JVs with customers) are weaker for rms in disperse industries and for those with low market shares; (ii) the negative relations between debt levels and the proxies for RS investments by supplier/customers are weaker for vertically integrated rms; and (iii) suppliers and customers of a highly levered rm undertake lower levels of R&D investments.

1 An alternative proxy is one that only includes SAs. An argument for excluding JVs is that, once a JV is formed, partners in the JV may be less concerned with nancial distress in the parent companies. On the other hand, if establishing a JV indicates that the partners are likely to cooperate in the future, then nancial distress at the parent level becomes more important. To address this issue, we repeat our analysis by excluding JVs from our measures. The results of this analysis are qualitatively similar to those reported here.

ARTICLE IN PRESS
326 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

2.2. Debt and bargaining Bronars and Deere (1991) investigate the use of debt as a negotiating tool by rms in reducing the rm surplus that workers can extract through union formation. Since a rm is committed to pay a portion of its future surplus to debtholders, debt obligations reduce the part of the surplus that a union can extract without driving the rm into bankruptcy. Bronars and Deere show that an optimal capital structure obtains when the marginal wealth gain from the reduction in the union wage is equal to the marginal increase in bankruptcy costs. Dasgupta and Sengupta (1993), Perotti and Spier (1993), and Subramanian (1996) show that a rm can use debt to shield a portion of its future surplus from workers and/or other input suppliers.2 Bronars and Deere (1991) present empirical evidence that rms facing greater threats of unionization have higher leverage ratios (see also Hanka, 1998). We propose that debt can play a similar bargaining role in a rms dealings with its suppliers/customers. Specically, we conjecture that a rm will choose a higher debt level when it faces suppliers/customers who enjoy greater bargaining power. In our empirical tests we proxy supplier/customer bargaining power by the degree of concentration in the supplier/customer industry, where greater concentration indicates greater bargaining power. The use of industry concentration as a proxy for bargaining power has signicant support in the literature. In the case of suppliers, researchers argue that more concentrated suppliers have greater seller power over their customers (see Stigler, 1964; Scherer and Ross, 1990 for a review of this literature). Thus, a rm that faces a concentrated supplier industry is at a bargaining disadvantage, whereas the presence of numerous alternative suppliers empowers the rm because it can make a credible threat to withhold future business from its existing suppliers (see, e.g., Holmstrom and Roberts, 1998). Similar arguments apply to the degree of concentration in the rms customer industry. In particular, Maskin and Riley (1984), Stole and Zwiebel (1996), and Snyder (1996) propose models that suggest that more concentrated customers enjoy greater buyer power over ` -vis its their suppliers. The buyer power theory predicts that a rms bargaining power vis-a customers should be decreasing in the concentration of the downstream industry. To summarize, our second central hypothesis described in this section is: Hypothesis 2:. A rms leverage is positively related to the concentration levels in its supplier and customer industries. Since bargaining power is increasing in industry concentration, a rm with a dominant ` -vis its trading partners. position in its industry should enjoy a bargaining advantage vis-a Therefore, a rm with a high market share should have less incentive to use debt as a bargaining tool. The ensuing subhypothesis, therefore, is that the positive relation between a rms debt level and supplier/customer industry concentration is weaker for rms with higher market shares.

2 The business press recognizes the ability of a rms suppliers and customers to expropriate some of the rms prots. For example, Fitzgerald (1996) suggests that suppliers of US automakers resist opening their books to their customers because of their fear that the customers would signicantly chip away at their prot margins.

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 327

3. Data, sample construction, and variable descriptions To construct our sample, we identify all rms covered by Compustat during the period 19842003. As is common in cross-sectional studies of capital structure determinants (see, e.g., Berger, Ofek, and Yermack, 1997), we exclude nancial rms (SIC codes between 6000 and 6999) and utilities (SIC codes between 4900 and 4999). We also exclude all rms that are headquartered in a foreign country (Compustats state variable equals 99) since the supplier and customer characteristics we capture reect the US market. From this master sample, we construct different subsamples to examine supplier and customer effects at both the industry and rm levels. In Section 3.1, we discuss the supplier and customer variables and the restrictions we employ to create the subsamples used in our industry-level analysis. In Section 3.2, we describe the variables and the subsamples used in our analysis of key suppliers and customers. Section 3.3 discusses our leverage measures and control variables. 3.1. Samples and variable construction at supplier- and customer-industry levels As we mention earlier, we rst relate a rms debt level to the characteristics of the rms supplier and customer industries. The advantages of the industry-level data, relative to the rm-level data, are: signicantly larger sample size, less severe endogeneity problems, and greater suitability for constructing the concentration measures. 3.1.1. Supplier- and customer-industry R&D intensities Since most rms use a large number of inputs, we measure supplier R&D intensity as the weighted average of the R&D intensities of all supplier industries, where the weight represents the importance of the input bought from each supplier industry in the production of the rms output. Specically, the supplier R&D intensity for a rm in the ith industry is given by
Supplier Industries R&D
n X
j 1; j ai

Supplier Industry R&Dj Industry Input Coefficientji ;

(1)

where n is the number of supplier industries, Supplier Industry R&Dj is the jth supplier industrys R&D expenditures divided by its total assets, and Industry Input Coefcientji is the dollar amount of the jth supplier industrys output used as an input to produce one dollar of the output of the ith industry. Note that the supplier R&D intensity measure will be high if the rm outsources a signicant part of its inputs from R&D-intensive supplier industries. Analogously, the R&D intensity of customer industries, Customer Industries R&D, is given by
Customer Industries R&D
n X
j 1 j ai

Customer Industry R&Dj Industry Percentage Sold ji ;

(2)

where n is the number of customer industries, Customer Industry R&Dj is the R&D intensity of the jth customer industry measured by the ratio of the industrys total R&D expenditures to its total assets, and Industry Percentage Soldji is the percentage of the ith industrys output that is sold to the jth customer industry. Intuitively, Industry Percentage

ARTICLE IN PRESS
328 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Soldji measures the importance of the jth customer industry as a buyer of the output of the rms industry. Thus, Customer Industries R&D is a weighted average of the R&D intensity of all customer industries, where the weight is the percentage of the output of the rms industry that is sold to each customer industry. 3.1.2. Supplier/customer industry concentration To capture the concentration of supplier (customer) industries, we follow the literature and use a weighted average of the concentrations of all supplier (customer) industries (see, e.g., Ravenscraft, 1983; Scherer and Ross, 1990). For each rm in the ith industry, the supplier concentration measure is dened as: Supplier Concentration
n X
j 1 iaj

Herfindahl Indexj Industry Input Coefficientji ,

(3)

where n is the number of supplier industries, Herndahl Indexj is the sales-based Herndahl index of the jth supplier industry, and Industry Input Coefcientji is as dened above. Similarly, for each rm in the ith industry, the concentration of customers is: Customer Concentration
n X
j 1 j ai

Herfindahl Indexj Industry Percentage Sold ji ;

(4)

where n is the number of customer industries, Herndahl Indexj is the Herndahl index of the jth customer industry, and Industry Percentage Soldji is as dened above. 3.1.3. Growth in supplier/customer industries It is possible that the supplier and customer R&D variables can capture some aspects of the rms own growth that is related to growth in its upstream and downstream industries. In addition to rm-specic control variables that we use to control for the rms growth opportunities (see Section 3.4), we construct variables that control for the effect of growth in supplier and customer industries on the rms leverage. Following the literature (see, e.g., Shea, 1993; Bartelsman, Caballero and Lyons, 1994), we dene the two variables Supplier Change in Sales and Customer Change in Sales as follows:3 Supplier Change in Sales
n X
j 1 j ai

Change in Salesj Industry Input Coefficientji ;

(5)

Customer Change in Sales

n X
j 1 j ai

Change in Salesj Industry Percentage Sold ji ;

(6)

Our measures are more in line with the variables used in Bartelsman, Caballero, and Lyons (1994). Shea (1993) uses a measure that is similar to Customer Change in Sales as an instrument for demand shock. However, Shea imposes restrictions on the downstream industries that are included in his measure to ensure both the absence of correlation between the variable and industry supply (exogeneity), and the presence of correlation between the variable and industry output (relevance). While these restrictions are crucial to Sheas objective of estimating the elasticity of supply, they are less relevant to our study.

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 329

where n is the number of supplier (customer) industries, Change in Salesj is the one-year change in sales for the median rm in the jth supplier (customer) industry, and Industry Input Coefcientji (Industry Percentage Soldji) is as dened above.4 3.1.4. Construction of supplier/customer R&D, concentration, and change in sales variables We rely on two data sources, the Use table of the benchmark input-output (IO) accounts for the US economy and the Compustat database, to construct the supplier and customer industry variables described above (see Fan and Lang, 2000; Shahrur, 2005 for recent papers that use this data set). For any pair of supplier and customer industries, the Use table reports estimates of the dollar value of the supplier industrys output that is used as an input in the production of the customer industrys output. The Use table enables us to identify the rms customer and supplier industries and the importance of each supplier/ customer industry to the rm. We use the 1987, 1992, and 1997 Use tables for the periods 19841989, 19901994, and 19952003, respectively. We compute the R&D intensities, industry concentrations, and change in sales of supplier/customer industries from Compustat. Appendix A describes the variable construction process in greater detail. 3.1.5. Labor as an input In addition to the inputs bought from suppliers, labor is also a major input to the rms production process. The supplier R&D variable described above does not take into consideration differences across rms with respect to their reliance on labor. To control for this factor, we construct the variable Compensation of Employees as the dollar amount spent on employee compensation in the rms industry divided by the industrys total output. This variable is computed using data from the input-output accounts. Further, since some rms may require that their employees undertake rm-specic investments, we use the product of Compensation of Employees and the rms own R&D intensity as a proxy for the importance of rm-specic investments by employees. 3.1.6. Industry-level strategic alliances and joint venture variables We conjecture that industries with higher levels of RS investments between the industry rms and suppliers/customers are also likely to have a higher intensity of SAs and JVs with their suppliers/customers (see, e.g., Fee, Hadlock, and Thomas, 2005). For each rm in our samples, we identify all the SAs and JVs announced over the sample period (19842003) in which the sample rm is a participant from the Securities Data Company (SDC) strategic alliance and joint venture database. Using the input-output tables and the rms four-digit SIC code, we then identify whether at least one of the other participants in the SA/JV arrangement operates in a supplier or a customer industry. For the purposes of this selection, we include only those supplier industries that sell more than 1% of their total output to the rms industry. We also include a customer industry if the total dollar amount spent on the input bought from the rms industry represents more than 1% of the industrys total output. For each four-digit SIC code industry, we dene Supplier (Customer) Alliance and JV Intensity as the number of SAs and JVs formed by rms in
Using a three-year change in sales (t3 to t) in the construction of Supplier and Customer Change in Sales yields similar results. As an alternative measure for growth in supplier (customer) markets, we use a weighted average of the market-to-book ratios of the median rms in supplier (customer) industries. The results from this specication are similar to those reported below.
4

ARTICLE IN PRESS
330 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

that industry with rms in a supplier (customer) industry divided by the number of rms in the industry. 3.1.7. Vertical integration variables Firms can also resort to vertical integration to address contracting problems that arise with their suppliers and customers. As a result, the effect of supplier- and customer-related variables on leverage should be less pronounced for vertically integrated rms. From Compustats segment tapes, we determine whether a rm in our sample has a segment in a supplier/customer industry SIC code. In order to capture the extent of vertical integration for rms in our sample, we construct a Backward (Forward) Integration Dummy, which equals one if the rm has at least one segment in a supplier (customer) industry, and zero otherwise. In the construction of these variables, we include only those supplier industries that sell at least 1% of their total output to the rms industry, and those customer industries whose total dollar amount spent on the input bought from the rms industry represents at least 1% of their total outputs. 3.1.8. Samples used in the analyses Not every rm has identiable suppliers and customers whose characteristics can affect its leverage. In order to test our hypotheses on the largest number of rms possible, we examine upstream and downstream effects separately on different data subsamples. We also discuss in Section 4.2.4 an alternative approach whereby we use the same sample to examine supplier and customer effects. To construct the two subsamples, we impose the following restrictions. First, since the inputs bought by rms in retail and wholesale industries (SIC codes between 5000 and 5999) are sold to other customers without any signicant processing, that is, they are not used as intermediate inputs, we analyze supplier effects on a subsample that excludes rms in these industries. This subsample consists of a panel of 76,290 rm-year observations for 10,310 rms for which we can construct the industry-level supplier variables.5 We also impose restrictions to obtain a sample for examining customer effects. Some rms sell to nal users and have no identiable customer rms. Because customer characteristics are not observable for these rms, in our analysis of customer effects a rm is included in the subsample if no more than 25% of its output is sold to nal users.6 Based on the Use table, we dene sales to nal users as uses classied under the IO system as:
5 We verify whether the inclusion of rms that mainly use raw materials, such as steel producers, adds noise to our analysis. Noise may result, for instance, from including rms that buy raw materials from R&D-intensive suppliers, whose R&D investments are not likely to be RS investments (in this case, the supplier R&D variable will be a noisy measure of RS investments by suppliers). We nd that most rms in the agriculture and mining industries do not undertake any R&D investments, and thus their R&D intensity is not a noisy measure of their limited RS investments. Nonetheless, we repeat the analysis of supplier effects excluding all rms that buy at least 50% of their input from rms in the agriculture and mining industries; the results are qualitatively similar to those reported here. 6 The Use table shows the ow of commodities as if they move directly to nal users, that is, even if they reach nal users by means of wholesalers or retailers (which are treated as service industries whose primary service is the distribution of goods). If trade were shown as buying and reselling commodities, then many industries would have wholesalers and retailers as their suppliers and customers. This method of treating trade is crucial to our study since we are able to capture the actual consuming industries or nal users even if a commodity is sold through retailers and wholesalers. However, this leads to a downward bias in the customer concentration for industries selling consumer products, since they sell some of their goods through concentrated wholesale and retail

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 331

Personal Consumption Expenditures, Gross Private Fixed Investments, and Government Consumption Expenditures and Gross Investment.7 Appendix A contains the assumptions made in the construction of the customer variables regarding the characteristics of nal users for rms included in the sample. Appendix B presents a list of selected industries along with the percentage of the industry output sold to nal users, which ranges from 100% to 0%. For example, the industry Motor Homes (SIC code 3716) sells all its output to nal users (mainly individuals), and thus rms in this industry are not included in the customer subsample. On the other hand, a rm in the industry Primary Production of Aluminum (SIC code 3334) is included in the subsample since all its output is used as intermediate input in the production of other products. The restrictions imposed on the customer subsample result in an unbalanced panel of 26,300 rm-year observations for 4,065 rms for which we can construct the industry-level customer variables. The median industry included in this subsample sells approximately 6.5% of its output to nal users. 3.2. Customer and supplier variables using rm-level data The supplier and customer variables described above are at the industry level. We also investigate the relation between corporate debt levels and supplier/customer characteristics that are measured at the rm level. We compute these rm-level characteristics from the subsamples of rms for which we can identify the key suppliers and customers. In accordance with the Statement of Financial Accounting Standards (SFAS) no. 14 and 131, public rms have to disclose the identity of any customer that contributes at least 10% to the rms revenues, although some rms choose to report customers that contribute less than 10%. While these data are available on Compustats industry segment les, the database reports only the name of the customer (not CUSIP or other identiers) and, further adding to the difculty, sometimes it reports only the abbreviated versions of the names. As a result, we use a combination of automated and manual procedures to identify the customer rms for our sample rms.8 We impose the same restrictions used to construct the industry-level subsamples. We exclude nancial rms and utilities, rms in the retail and wholesale industries, and rms headquartered in a foreign country.9 Further, we exclude rms whose main supplier (main customer) is a retailer or a wholesaler from the sample used to examine supplier (customer) effects. Appendix C shows a list of selected supplier-customer industry pairs along with an example from each pair.
(footnote continued) industries. Since such industries are not included in the sample used to examine customer effects, the effect of this bias on the relation between customer concentration and leverage is greatly mitigated. 7 Personal Consumption Expenditures represent purchases by individual consumers. Gross Private Fixed Investments consist of purchases of residential and nonresidential structures and of equipment and software by private businesses. Government Consumption Expenditures and Gross Investment consist of consumption expenditures and investments by federal, state, and local governments. 8 The procedure we follow to identify the customer rms is similar to that used in Fee and Thomas (2004) (see pp. 436437 of their study for more details). 9 Recall that in the industry-level analysis, a rm is included in the sample if at most 25% of the outputs of the rms industry are sales to nal users. We do not impose this restriction in this part of the analysis since we have detailed data about actual customers. Thus, a rm that sells its output to other rm is include in the sample even if the rms industry sells most of its output to nal users. Imposing the 25% restriction does signicantly affect our results.

ARTICLE IN PRESS
332 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Since some rms report many customers for any given year, we construct a weighted average of the R&D intensities of the rms key customers as follows:
Key Customers R&D
n X j 1

Key Customer R&Dj Key Customer Percentage Sold j ;

(7)

where n is the number of customer rms, Key Customer R&Dj is equal to the R&D expense of the jth customer divided by its total assets, and Key Customer Percentage Soldj is the percentage of the rms sales to the jth customer. For the period 19842002, the resulting sample is a panel of 9,452 supplier-year observations for which we are able to compute Key customers R&D and the other control variables. A signicant number of the customers in our sample are also the key customers to more than one supplier rms. Thus, for each customer rm in the sample, we compute the R&D intensity of its suppliers as follows:
Key Suppliers R&D
n X j 1

Key Supplier R&Dj Key Customer Input Coefficientj ;

(8)

where n is the number of suppliers, Key Supplier R&Dj is equal to the ratio of the R&D expense of the jth supplier to its total assets, and Key Customer Input Coefcientj is the ratio of the customers purchases from the jth supplier to the customer total sales.10 For the period 19842002, this procedure results in a panel of 4,122 customer-year observations for which Key Suppliers R&D and the other control variables are available. Finally, using the data from the SDC database, we construct a dummy variable for SAs and JVs between a rm and its suppliers/customers. This dummy variable, Customer (Supplier) SA and JV Dummy, equals one if the rm has at least one SA or JV established with one of its key customers (key suppliers), and zero otherwise. 3.3. Leverage measures and control variables We use two measures of nancial leverage, Market Leverage and Book Leverage, where Market Leverage is equal to the sum of the book values of long-term debt and debt in current liabilities (Compustat items 9 and 34) divided by the sum of the book value of debt and the market value of common equity (item 25item 199), and Book Leverage is the sum of the book values of long-term debt and debt in current liabilities divided by the book value of assets (item 6). Following the literature (see, e.g., Kale, Noe, and Ramirez, 1991; Berger, Ofek, and Yermack, 1997; Mackay and Phillips, 2006), we use the following control variables: 1. Firm Size is the log of total assets (item 6). 2. Return on Assets is operating income (item 13) divided by total assets.
10 Note that unlike the respective variables that are constructed using industry-level supplier and customer data, the Key Customer Percentage Sold and Key Customer Input Coefcientj variables do not sum to one for each rm since not all customer and supplier rms are reported. Thus, by construction, the key supplier and customer R&D variables are biased downward. To ensure that our results are not affected by this bias, we repeat our analysis of supplier (customer) effects after using the R&D intensity of the main supplier (customer). The results of this specication are qualitatively similar to those reported below. Section 4.3 show estimates of simultaneous equation models that use this specication.

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 333

3. Asset Collateral Value is net property, plant, and equipment (item 8) divided by total assets. 4. Volatility is the standard deviation of operating income divided by total assets. We require at least three consecutive observations to construct this variable. 5. Nondebt Tax Shields is equal to investment tax credits (item 51) divided by total assets. 6. R&D Intensity is equal to total research and development expenditures (item 46) divided by total assets (item 12). 7. SE Intensity is equal to selling, general, and administrative expenses (SE) (item 189) divided by total assets. 8. Tobins q is equal to the book value of assets plus the market value of common equity minus the book value of common equity (item 60) divided by the book value of assets. 9. Industry Concentration is the sales-based Herndahl Index of the rms primary industry.11 10. Industry and year dummy variables. The industry dummy variables are based on the rms two-digit historical SIC code.12 In the construction of our variables, we assume that the rm spends zero dollars on R&D (SE) if the R&D (SE) expense for a rm is missing.13 We nd that for most of the control variables, some rms have extremely high values. For example, while the median value of the selling expense variable is 0.23, its maximum value is 2,343. We winsorize all the dependent and independent variables (including the supplier and customer variables) at the 1st and 99th percentiles in order to reduce the effect of outliers on our results.14 We present descriptive statistics for the dependent and independent variables in Table 1. The mean and median values for Market Leverage are 0.187 and 0.132, respectively. The mean and median values for Book Leverage are 0.268 and 0.210, respectively. The descriptive statistics of the independent variables presented in the table show wide ranges of values for all these variables, suggesting that the rms in our sample differ considerably with regard to the characteristics of their suppliers and customers. 4. Empirical ndings 4.1. Univariate analysis In the univariate analysis, for each variable of interest we divide our sample of rms into two groupsthose with variable values greater than the median and those below. We then compare the debt levels of above-median rms with the debt levels of rms that are below the median and present the ndings in Table 2. The average values of Market Leverage and
11 Mackay and Phillips (2006) nd that rms in concentrated industries tend to rely more on debt nancing. Lyandres (2006) also nds that leverage is negatively related to the number of rivals in the industry. 12 We discuss the effect of including rm xed effects on our results in Section 4.5. 13 We follow Loughran and Ritter (1997) and verify the validity of this assumption by checking the R&D of biotech rms and retailers. We nd that almost all biotech rms had a nonzero value for the R&D variable while most retailers had missing values for this variable. 14 Our results are not especially sensitive to winsorization. We nd qualitatively similar results without winsorizing the variables, although some of the results pertaining to the control variables that have extreme outlier values (such as R&D Intensity and SE Intensity) become statistically insignicant.

ARTICLE IN PRESS
334 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 Table 1 Descriptive statistics This table reports descriptive statistics for the dependent and independent variables. The sample period is from 1984 to 2003. Market Leverage is the book value of long-term debt and debt in current liabilities divided by the sum of the the book value of debt and the market value of common equity. Book Leverage is the book value of long-term debt and debt in current liabilities divided by the book value of assets. Supplier Industries R&D (Customer Industries R&D) is the weighted average of the R&D intensities of all supplier (customer) industries. Supplier (Customer) SA and JV Intensity is the number of SAs and JVs established between industry rms and rms in supplier (customer) industries, divided by the number of rms in the industry. Compensation of Employees is total compensation paid to employees divided by total output, constructed at the industry level. Supplier Concentration (Customer Concentration) is the weighted average of the Herndahl indices of all supplier (customer) industries. Supplier (Customer) Change in Sales is a weighted average of the change in sales for supplier (customer) industries. Key Suppliers R&D (Key Customers R&D) is the weighted average of the R&D intensities of key suppliers (key customers). Suppliers(Customer) SA and JV Dummy is a dummy variable that equals one if the rm has at least one alliance or a SA established with one of its suppliers (customers), and zero otherwise. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. All variables are winsorized at the 1st and 99th percentiles.
Variable Market Leverage (overall sample) Book Leverage (overall sample) Supplier Industry-Level Variables (76,290 obs.) Supplier Industries R&D Supplier SA and JV Intensity Compensation of Employees Supplier Concentration Supplier Change in Sales Customer Industry-Level Variables (26,300 obs) Customer Industries R&D Customer SA and JV Intensity Customer Concentration Customer Change in Sales Supplier Firm-Level Variables(4,122 obs.) Key Suppliers R&D Supplier SA and JV Dummy Customer Firm-Level Variables(9,452 obs) Key Customers R&D Customer SA and JV Dummy Control Variables(overall sample) Industry Concentration Firm Size (log of total assets) Asset Collateral Value Volatility Nondebt Tax Shield Return on Assets Tobins q R&D Intensity SE Intensity Mean 0.187 0.268 0.009 0.486 0.303 0.090 0.098 0.017 0.493 0.167 0.081 0.002 0.283 0.015 0.162 0.223 4.280 0.294 0.201 0.001 0.015 2.283 0.058 0.332 Median 0.132 0.210 0.006 0.218 0.299 0.085 0.096 0.009 0.176 0.167 0.082 0.000 0.000 0.007 0.000 0.179 4.146 0.235 0.091 0.000 0.104 1.459 0.005 0.250 Maximum 0.750 1.882 0.040 5.166 0.818 0.250 0.308 0.081 6.469 0.321 0.217 0.059 1.000 0.104 1.000 0.865 10.115 0.917 2.963 0.017 0.439 18.561 0.732 2.374 Minimum 0.000 0.000 0.000 0.000 0.022 0.029 0.050 0.000 0.000 0.068 0.030 0.000 0.000 0.000 0.000 0.043 1.139 0.001 0.010 0.000 2.284 0.528 0.000 0.000 Std dev. 0.189 0.290 0.007 0.795 0.116 0.039 0.066 0.017 0.910 0.058 0.047 0.008 0.450 0.021 0.368 0.163 2.188 0.226 0.381 0.002 0.344 2.558 0.115 0.354

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 335

Book Leverage for rms with above median values for Supplier and Customer Industries R&D, Key Suppliers and Customers R&D, Supplier and Customer SA and JV Intensities, and Supplier and Customer SA and JV Dummies are lower than those for rms with belowmedian values. All these ndings offer preliminary evidence consistent with the RS investments hypotheses (Hypothesis 1). For example, the average Market (Book) Leverage for rms with above-median values for Supplier Industries R&D is 0.143 (0.226) and for those with values below the median, the average Market Leverage is 0.231 (0.311). Further, both Market and Book Leverage appear to be lower for rms in the sample with abovemedian Compensation of Employees than for rms with below-median values for this variable. Consistent with Hypothesis 2, the mean Market Leverage for rms with abovemedian Supplier (Customer) Concentration is greater than the mean leverage for rms whose suppliers (customers) are less concentrated. Overall, the results of the univariate analysis support our hypotheses that relate the rms leverage to the characteristics of suppliers and customers. 4.2. Multivariate analysis with industry-level supplier and customer variables 4.2.1. Analysis of supplier effects Table 3 shows the results of three ordinary least squares (OLS) regressions of Market Leverage for the subsample with industry-level supplier variables. In regression Models 1 and 2, we test Hypothesis 1 by examining the relation between debt level and supplier R-S investment by Supplier Industry R&D and Supplier SA and JV Intensity, respectively. In Model 3, we include both the RS variables as well as interaction variables. Consistent with Hypothesis 1, the coefcients on Supplier Industries R&D and Supplier SA and JV Intensity reported in Models 1 and 2 are both negative and statistically signicant. For example, the coefcient on Supplier SA and JV Intensity is equal to 0.011, signicant at the 1% level. In the all-inclusive specication in Model 3, the coefcients on both the RS investment variables continue to be negative. Further, the coefcient on Backward Integration DummySupplier Industries R&D is positive and signicant, which suggests that the negative relation between supplier R&D and leverage is weaker for rms with some level of backward integration.15 The coefcient on Compensation of Employees in each reported model is negative and signicant. Further, in Model 3, the interaction variable Compensation of EmployeesR&D Intensity is also negative and signicant. It appears that R&D-intensive rms with high employee compensation are likely to require signicant rm-specic investments by their employees, and thus they have lower debt levels. The coefcients of Supplier Concentration are positive and signicant in Models 1 and 3, while the coefcients of Supplier ConcentrationMarket Share is negative and signicant in Model 3. These ndings are consistent with Hypothesis 2 and indicate that leverage is greater when rms face suppliers who are at a bargaining advantage, and this relation is weaker if the rm is a dominant player in its own industry. The coefcient on Suppliers Change in Sales is negative and signicant at the 1% level. As mentioned earlier, we include this variable to capture aspects of rm growth that are related to its input markets but are not captured by the rm-specic variables that measure growth opportunities such as Tobins q. Among the other control variables, the coefcients
In unreported results, we include Backward Integration DummySupplier SA and JV Intensity and nd a positive but statistically insignicant coefcient on this variable.
15

336

Table 2 Univariate analysis This table reports univariate analysis for the difference in means of the leverage variables for different subsamples formed based on the median values of the supplier and customer variables. The sample period is from 1984 to 2003. Market Leverage is sum of the book values of long-term debt and debt in current liabilities divided by the sum of the book value of debt and the market value of common equity. Book Leverage is the book value of long-term debt and debt in current liabilities divided by the book value of assets. Supplier Industries R&D (Customer Industries R&D) is the weighted average of the R&D intensities of all supplier (customer) industries. Key Suppliers R&D (Key Customers R&D) is the weighted average of the R&D intensities of key suppliers (key customers). Supplier (Customer) SA and JV Intensity is the number of SAs and JVs established between industry rms and rms in supplier (customer) industries, divided by the number of rms in the industry. Suppliers(Customer) SA and JV Dummy is a dummy variable that equals one if the rm has at least one alliance or a SA established with one of its suppliers (customers), and zero otherwise. Compensation of Employees is total compensation paid to employees divided by total output. Supplier Concentration (Customer Concentration) is the weighted average of the Herndahl indices of all supplier (customer) industries. The symbol *** indicates statistical signicance at the 1% level. Mean Market Leverage Below median 0.231 0.219 0.207 0.236 0.217 0.244 0.189 0.172 0.198 0.178 0.209 0.246 0.265 0.211 0.168 0.253 0.265 0.288 0.262 0.222 0.284 0.289 0.282 0.298 0.290 0.306 0.268 0.238 0.283 0.271 0.286 0.226 0.311 Difference (t-value) Above median Below median Mean Book Leverage Difference (t-value) 76,290 26,300 4,122 9,452 76,290 26,300 4,122 9,452 76,290 76,290 26,300 N

Variables

Above median

Supplier Industries R&D

0.143

Customer Industries R&D

0.211

Key Suppliers R&D

0.141

Key Customers R&D

0.196

ARTICLE IN PRESS

Supplier SA and JV Intensity

0.157

Customer SA and JV Intensity

0.183

Supplier SA and JV Dummy

0.135

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Customer SA and JV Dummy

0.093

Compensation of Employees

0.176

Supplier Concentration

0.196

Customer Concentration

0.221

0.089*** (66.42) 0.009*** (3.77) 0.065*** (14.10) 0.04*** (7.09) 0.06*** (43.97) 0.061*** (25.99) 0.054*** (10.46) 0.079*** (16.28) 0.021*** (15.38) 0.019*** (13.80) 0.011*** (4.99)

0.085*** (40.92) 0.005 (1.36) 0.06*** (9.66) 0.036*** (7.09) 0.045*** (21.21) 0.042*** (12.25) 0.057*** (8.23) 0.070*** (9.82) 0.03*** (14.37) 0.006*** (2.77) 0.002 (0.67)

Table 3 Ordinary least squares regressions of market leverage on supplier industry-level variables The sample period is from 1984 to 2003. The dependent variable is Market Leverage, which is the sum of the book values of long-term debt and debt in current liabilities divided by the sum of the book value of debt and the market value of common equity. Supplier Industries R&D is the weighted average of the R&D intensities of all supplier industries. Backward Integration Dummy is a dummy variable that equals one if the rm has a segment in a supplier industry. Supplier SA and JV Intensity is the number of SAs and JVs established between industry rms and rms in supplier industries, divided by the number of rms in the industry. Compensation of Employees is total compensation paid to employees divided by total output. Supplier Concentration is the weighted average of the Herndahl indices of all supplier industries. Supplier Change in Sales is a weighted average of the change in sales for supplier industries. Market Share is the ratio of the rms sales to total sales of its primary industry. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. All variables are winsorized at the 1st and 99th percentiles. Lagged values of rm-specic control variables are used. The reported t-values reect Whites heteroskedasticity correction. The symbols ** and *** indicate statistical signicance at the 5%, and 1% levels, respectively. Model 1 Coeff. 0.183*** 1.472*** 25.66 13.99 t-value Coeff. 0.184*** Model 2 t-value 25.59 Coeff. 0.181*** 1.208*** 0.786*** Model 3 t-value 24.86 10.48 3.90

ARTICLE IN PRESS

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

0.024*** 0.074*** 0.102***

2.70 3.00 6.40

0.011*** 0.025*** 0.035 0.105***

12.09 2.79 1.42 6.58

Intercept Supplier Industries R&D Supplier Industries R&DBackward Integration Dummy Supplier SA and JV Intensity Compensation of Employees Comp. Of EmployeesR&D Intensity Supplier Concentration Supplier ConcentrationMarket Share Suppliers Change in Sales

0.009*** 0.031*** 0.161*** 0.085*** 0.354*** 0.098***

9.03 3.41 6.85 3.14 3.66 6.14

337

338

Table 3 (continued ) Model 1 Coeff. Coeff. 0.002 0.026** 0.009** 0.004*** 0.193*** 0.008*** 4.435*** 0.110*** 0.015*** 0.232*** 0.032*** Yes 76,290 25.57% 2.24 12.71 47.31 4.06 18.83 37.96 56.08 34.63 15.14 0.011*** 0.005*** 0.196*** 0.008*** 4.451*** 0.111*** 0.015*** 0.244*** 0.031*** Yes 76,290 25.54% 2.69 14.25 48.61 4.00 18.86 38.48 56.20 36.95 14.76 0.020*** 0.007*** 0.191*** 0.006*** 4.364*** 0.109*** 0.015*** 0.188*** 0.029*** Yes 76,290 26.01% t-value Coeff. t-value t-value 0.66 2.07 4.62 15.73 46.80 2.71 18.54 37.71 55.61 21.33 13.69 Model 2 Model 3

Backward Integration Dummy Market Share Other Control Variables Industry Concentration Firm Size Asset Collateral Value Volatility Nondebt Tax Shield Return on Assets Tobins q R&D Intensity SE Intensity Year and Industry Dummies Number of observations Adjusted R-squared

ARTICLE IN PRESS

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 339

on Size, Asset Collateral Value, and Industry Concentration are signicantly positive. The positive coefcient on Industry Concentration is in keeping with the evidence in Mackay and Phillips (2006) and the predictions in Brander and Lewis (1986) and Maksimovic (1988). The signicantly negative coefcients on the remaining control variables are consistent with the literature on the determinants of capital structure. Finally, the negative and signicant coefcients on the rms R&D and SE Intensity measures are similar to those documented in Titman and Wessels (1988) and are consistent with the predictions in Titman (1984) and Maksimovic and Titman (1991). Armour and Teece (1980) suggest that vertical chains that are R&D intensive are likely to have complex interstage interdependencies (see also Williamson, 1975). Therefore, it is more likely that the R&D intensity of suppliers will capture RS investments by suppliers when the rm itself is R&D intensive. To investigate this possibility, we estimate our regressions separately on the subsample of rms with positive R&D expense and the subsample for which it is zero. Table 4 presents the ndings. Consistent with our expectation, we nd that the negative relation between Supplier Industries R&D and a rms debt level is signicant only for rms with positive R&D intensity. Further, the coefcient on the supplier concentration variable is positive and signicant only for the subsample of rms with positive R&D. This result suggests that the effect of bargaining on leverage is only pronounced when there are relationship-specic investments between the rm and its suppliers. 4.2.2. Analysis of customer effects Table 5 presents the results of the regression analysis for Market Leverage for the subsample with industry-level customer variables. Consistent with the RS investments hypothesis (Hypothesis 1), in Models 1 and 2, the coefcients on the variables Customer Industries R&D and Customer SA and JV Intensity, respectively, are both negative and signicant. Further, the coefcients on Customer Industries R&DMarket Share and Customer SA and JV IntensityMarket Share are negative and signicant. These ndings are consistent with the hypothesis that the effect of RS investments by customers on leverage is stronger for rms that are important to their customers. In unreported regressions, we nd insignicant coefcients on Customer Industries R&DIndustry Concentration and Customer SA and JV IntensityIndustry Concentration. While the coefcient on Customer Industries R&DForward Integration Dummy is insignicant in Model 1, the coefcient on Customer SA and JV IntensityForward Integration Dummy is positive and signicant in Model 2. Thus, the effect of RS investments on leverage appears to be less pronounced for vertically integrated rms. In Models 1 and 3, the coefcients on Customer Concentration are positive and signicant, consistent with the bargaining role of debt (Hypothesis 2). However, we nd that the coefcient on Customer ConcentrationMarket Share, while negative, is not statistically signicant at conventional levels. Next, we nd a negative relation between leverage and Customer Change in Sales, the variable that measures growth in downstream industries. Finally, except for the volatility measure, the results pertaining to the control variables are in keeping with our results in the supplier regressions and with the extant literature on capital structure.16
Unlike the specication for investigating supplier effects, we do not include R&D and SA/JV variables together in Model 3 in our analysis of customer effects because these two variables are highly positively correlated
16

ARTICLE IN PRESS
340 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 Table 4 Ordinary least squares regressions of market leverage on supplier industry-level variables: R&D-based subsamples The sample period is from 1984 to 2003. The dependent variable is Market Leverage, which is the sum of the book values of long-term debt and debt in current liabilities divided by the sum of the book value of debt and the market value of common equity. Column 1 (2) shows OLS regression results for rms with positive (zero) R&D Intensity, which is the rms R&D expenditures divided by its total assets. Supplier Industries R&D is the weighted average of the R&D intensities of all supplier industries. Compensation of Employees is total compensation paid to employees divided by total output. Supplier Concentration is the weighted average of the Herndahl indices of all supplier industries. Supplier Change in Sales is a weighted average of the change in sales for supplier industries. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. All variables are winsorized at the 1st and 99th percentiles. Lagged values of rm-specic control variables are used. The reported t-values reect Whites heteroskedasticity correction. The symbols *, **, and *** indicate statistical signicance at the 10%, 5%, and 1% levels, respectively. Firms with Positive R&D intensity Coeff. Intercept Supplier Industries R&D Compensation of Employees Supplier Concentration Suppliers Change in Sales Other control variables Industry Concentration Firm Size Asset Collateral Value Volatility Nondebt Tax Shield Return on Assets Tobins q R&D Intensity SE Intensity Year and Industry Dummies Number of observations Adjusted R-squared 0.179*** 1.395*** 0.047*** 0.083** 0.099*** 0.007 0.003*** 0.192*** 0.001 3.768*** 0.086*** 0.011*** 0.199*** 0.012*** Yes 40,685 22.12% t-value 18.30 12.11 3.42 2.56 4.60 1.46 6.94 32.08 0.58 16.21 27.21 43.41 27.22 5.12 Firms with Zero R&D intensity

Coeff. 0.146*** 0.08 0.021* 0.001 0.117*** 0.016** 0.009*** 0.185*** 0.016*** 4.660*** 0.140*** 0.023*** 0.050*** Yes 35,605 20.42%

t-value 13.62 0.26 1.83 0.04 5.17 2.48 15.52 33.77 4.21 8.76 25.05 33.55 12.67

In Table 6, we show the results of regression analysis that examines the relation between leverage and the R&D of customers for different subsamples of rms. In the rst column of the table, we report results for rms in the Agriculture & Mining (SIC codes between 0100 and 1499) industries. For these rms we nd an insignicant relation between Customer Industries R&D and market leverage. This result is expected given that many of these rms sell raw materials and are unlikely to engage in RS investments with their
(footnote continued) in this sub-sample (correlation equals 0.5). Including both variables does not signicantly affect our results, although it reduces the statistical signicance of both variables.

Table 5 Ordinary least squares regressions of market leverage on customer industry-level variables The sample period is from 1984 to 2003. The dependent variable is Market Leverage, which is the sum of the book values of long-term debt and debt in current liabilities divided by the sum of the book value of debt and the market value of common equity. Customer Industries R&D is the weighted average of the R&D intensities of all customer industries. Customer SA and JV Intensity is the number of SAs and JVs established between industry rms and rms in customer industries, divided by the number of rms in the industry. Forward Integration Dummy is a dummy variable that equals one if the rm has a segment in a customer industry. Customer Concentration is the weighted average of the Herndahl indices of all customer industries. Customer Change in Sales is a weighted average of the change in sales for customer industries. Market Share is the ratio of the rms sales to total sales of its primary industry. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. All variables are winsorized at the 1st and 99th percentiles. Lagged values of rm-specic control variables are used. The reported t-values reect Whites heteroskedasticity correction. The symbols ** and *** indicate statistical signicance at the 5% and 1% levels, respectively. Model 1 Coeff. 0.119*** 0.832*** 1.442*** 0.043 13.13 5.97 3.39 0.30 t-value Coeff. 0.123*** Model 2 t-value 13.62 Coeff. 0.118*** 0.875*** Model 3 t-value 12.79 6.38

ARTICLE IN PRESS

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Intercept Customer Industries R&D Customer Industries R&DMarket Share Customer Industries R&DForward Integration Dummy Customer SA and JV Intensity Cust. SA and JV IntensityMarket Share Customer SA and JV IntensityForward Integration Dummy Customer Concentration Customer ConcentrationMarket Share Customer Change in Sales Forward Integration Dummy Market Share 0.101*** 0.171*** 0.008** 0.047***

3.92 8.45 2.26 4.10

0.015*** 0.070*** 0.012*** 0.033 0.168*** 0.004 0.048***

9.18 7.43 5.69 1.24 8.33 1.51 4.62

0.114*** 0.198 0.171*** 0.026

4.01 1.59 8.48 1.03 341

342

Table 5 (continued ) Model 1 Coeff. t-value Coeff. Coeff. t-value t-value Model 2 Model 3

Other control variables Industry Concentration Firm Size Asset Collateral Value Volatility Nondebt Tax Shield Return on Assets Tobins q R&D Intensity SE Intensity Year and Industry Dummies Number of observations Adjusted R-squared 0.044*** 0.006*** 0.187*** 0.006 3.663*** 0.134*** 0.019*** 0.264*** 0.044*** Yes 26,300 21.21% 5.47 8.39 28.72 1.31 8.29 21.89 29.68 15.76 9.25 0.050*** 0.007*** 0.183*** 0.006 3.432*** 0.133*** 0.018*** 0.251*** 0.043*** Yes 26,300 21.40% 6.22 9.69 28.29 1.18 7.80 21.69 29.37 14.92 9.08 0.044*** 0.006*** 0.187*** 0.006 3.661*** 0.135*** 0.019*** 0.265*** 0.044*** Yes 26,300 21.17%

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 5.43 8.98 28.70 1.33 8.30 21.98 29.69 15.85 9.18

ARTICLE IN PRESS

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 343

customers. On the other hand, we nd (column two) that for rms producing durable manufacturing products, the relation between leverage and customer R&D is negative.17 In unreported results, we nd that this relation is statistically insignicant for rms producing nondurable manufacturing products. This result is consistent with the conjecture that there are more RS investments between rms in durable good industries and their customers. The last three columns of the table report results on the subsample of rms that have positive R&D expenses. In all three specications, we nd that the relation between leverage and customer R&D is negative and signicant. In untabulated results, we nd that for zero-R&D rms this relation is not signicant. Thus, as in the case of suppliers, we nd support for the conjecture that the R&D of customers is more likely to capture RS investments when the rm itself is R&D intensive. Results in columns four and ve also offer support for the hypothesis that the effect of RS investments by customers on a rms debt level is more pronounced if the rm produces output for which there are few alternative suppliers. In column 4, the coefcient on the interaction variable Customer Industries R&DIndustry Concentration is negative and signicant. Further, in column ve, we nd that the coefcient on Customer Industries R&DMarket Share is also negative and signicant. 4.2.3. Economic signicance In order to determine the economic signicance of the effects of the supplier and customer variables, we use the estimated coefcients reported in Tables 36 to compute the change in leverage if a rm moves from the 5th percentile value to the 95th percentile value of a specic independent variable, holding all the other independent variables at their sample means. In Table 7, we report the levels of economic signicance for selected specications and variables, owing to space constraints. For the supplier variables, Supplier Industries R&D, Supplier SA and JV Intensity, and Supplier Concentration, we nd that moves from the 5th to 95th percentile change leverage by 0.026, 0.016, and 0.008 (14.05%, 8.23%, 4.17%), respectively. For changes in the customer variables, Customer Industries R&D, Customer SA and JV Intensity, and Customer Concentration, leverage changes by 0.054, 0.056, and 0.023 (23.59%, 25.31%, and 11.44%), respectively. For the positive-R&D supplier (customer) subsample, the change in leverage for similar moves in Supplier and Customer Industries R&D are 0.042 and 0.083 (27.87% and 41.31%), respectively. The levels of economic signicance of the supplier and customer effects are even greater when we consider the effect of the interacted variables. For example, among the reported results, the largest change in leverage of 0.138 (or 67.73%) results from a move from the 5th to 95th percentile of Customer Industries R&D for a rm in the 95th percentile of market share in the positive R&D subsample. Are these changes in a rms leverage from changes in the supplier and customer variables economically signicant? It turns out that an increase in rm size from the 5th to the 95th percentile value is associated with a change in leverage of 0.051 (or 30.83%). Overall, it appears from the results reported in Table 7 that the levels of economic
17 We follow the US Census Bureau and classify manufacturing industries into durable and nondurable good industries. Durable (Nondurable) good industries are those with the following two-digit SIC codes: 24,25,32,33,34,35,36,37,38, and 39 (20,21,22,23,26,27, 28,29,30, and 31).

344

Table 6 Ordinary least squares regressions of market leverage on customer industry-level variables: various subsamples The sample period is from 1984 to 2003. The dependent variable is Market Leverage. Column 1 show OLS regression results for rms with historical primary SIC code between 0100 and 1499, while Column 2 show OLS regression results for rms producing durable goods. The last three columns show regression results for rms with positive R&D Intensity, which is the rms R&D expenditures divided by its total assets. Customer Industries R&D is the weighted average of the R&D intensities of all customer industries. Customer Concentration is the weighted average of the Herndahl indices of all customer industries. Customer Change in Sales is a weighted average of the change in sales for customer industries. Market Share is the ratio of the rms sales to total sales of its primary industry. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. All variables are winsorized at the 1st and 99th percentiles. Lagged values of rm-specic control variables are used. The t-values reported in parentheses reect Whites heteroskedasticity correction. The symbols ** and *** indicate statistical signicance at the 5%, and 1% levels, respectively. Agriculture and mining Durable manufacturing Positive R&D rms Positive R&D rms Positive R&D rms

Intercept

ARTICLE IN PRESS

Customer Industries R&D

Customer Industries R&DIndustry Concentration

0.141*** (3.56) 0.982 (0.95)

0.288*** (9.00) 0.841*** (5.29)

0.179*** (13.46) 1.122*** (6.94)

Customer Industries R&DMarket Share

0.168*** (12.15) 0.807*** (4.24) 1.723*** (3.21)

0.169*** (12.42) 1.012*** (6.11)

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Customer Concentration

Customer Change in Sales

0.184*** (5.19) 0.041 (1.24)

Industry Concentration

0.219*** (2.66) 0.315*** (5.43) 0.007 (0.28)

0.137*** (3.62) 0.118*** (3.70) 0.015 (1.36) 0.005

0.160*** (4.11) 0.114*** (3.58) 0.048*** (3.03)

Market Share

2.362*** (5.36) 0.164*** (4.23) 0.117*** (3.67) 0.025** (2.08) 0.014 (0.87)

Other Control Variables Firm Size

Asset Collateral Value

Volatility

Nondebt Tax Shield

Return on Assets

Tobins q

R&D Intensity

SE Intensity

ARTICLE IN PRESS

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Year and Industry Dummies Number of observations Adjusted R-squared

0.020*** (4.27) 0.199*** (14.25) 0.018 (1.43) 8.079*** (4.33) 0.087*** (5.41) 0.015*** (8.44) 0.117 (1.53) 0.073*** (3.50) Yes 4,593 18.71%

(0.42) 0.00 (1.57) 0.09*** (7.22) 0.03*** (3.27) 2.15*** (3.40) 0.19*** (15.83) 0.02*** (15.35) 0.35*** (9.80) 0.04*** (2.78) Yes 9,046 20.06% 0.001 (0.93) 0.193*** (18.26) 0.005 (0.87) 3.555*** (7.50) 0.120*** (15.38) 0.015*** (21.28) 0.249*** (12.77) 0.033*** (5.34) Yes 10,883 23.27% 0.001 (0.91) 0.195*** (18.36) 0.005 (0.94) 3.616*** (7.63) 0.120*** (15.39) 0.015*** (21.29) 0.248*** (12.72) 0.033*** (5.29) Yes 10,883 22.54% 0.000 (0.30) 0.194*** (18.27) 0.004 (0.79) 3.634*** (7.67) 0.120*** (15.47) 0.015*** (21.19) 0.250*** (12.82) 0.033*** (5.20) Yes 10,883 23.27%

345

346

Table 7 Economic signicance of regression results The table reports the predicted values of Market Leverage at the 5th and 95th percentiles of each of the independent variables, holding all other independent variables at their sample means. For models with interaction variables, the overall effect for each variable by itself (the Interaction Effects column is set to None) is computed by using the mean of the interacted variable(s). For variables with interaction effects, the predicted values of leverage are also computed at the 5th and 95th percentiles of the interacted variable. Change is the difference between the predicted value at the 95th percentile and that at the 5th percentile. The last two columns show the number of the table that contains the model used to compute the predicted values and the number of the model in the respective table. The sample period is from 1984 to 2003. Supplier (Customer) Industries R&D is the weighted average of the R&D intensities of all supplier (customer) industries. Supplier (Customer) SA and JV Intensity is the number of SAs and JVs established between industry rms and rms in supplier (customer) industries, divided by the number of rms in the industry. Compensation of Employees is total compensation paid to employees divided by total output. Supplier (Customer) Concentration is the weighted average of the Herndahl indices of all supplier (customer) industries. Supplier (Customer) Change in Sales is a weighted average of the change in sales for supplier (customer) industries. Market Share is the ratio of the rms sales to total sales of its primary industry. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. Predicted Leverage at the 5th and 95th Percentiles of Independent Variable 5th perc. 95th perc. Change % Change Table Model

ARTICLE IN PRESS

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Independent variable

Interaction effects

Supplier variables Supplier Industries R&D Supplier Industries R&D Supplier SA and JV Int. Supplier Concentration Supplier Concentration Supplier Concentration Comp. of Employees Comp. of Employees Comp. of Employees Sup. Change in Sales 0.197 0.191 0.192 0.184 0.187 0.174 0.194 0.209 0.140 0.199 0.169 0.181 0.171 0.192 0.197 0.171 0.181 0.199 0.116 0.177

None Vertically Integrated Firms None None 5th perc. of Market Share 95th perc. of Market Share None 5th perc. of R&D Intensity 95th perc. of R&D Intensity None

0.028 0.010 0.016 0.008 0.010 0.003 0.013 0.010 0.024 0.022

14.05 5.07 8.23 4.17 5.39 1.73 6.70 4.77 17.20 11.27

3 3 3 3 3 3 3 3 3 3

3 3 3 3 3 3 3 3 3 3

Customer variables Customer Industries R&D Customer Industries R&D Customer Industries R&D Customer SA and JV Int. Customer SA and JV Int. Customer SA and JV Int. Customer SA and JV Int. Customer Concentration 0.229 0.232 0.215 0.223 0.226 0.225 0.210 0.205 0.151 0.201 0.194 0.216 0.201 0.204 0.191 0.184 0.164 0.137 0.188 0.191 0.250 0.210 0.198 0.197 0.173 0.194 0.214 0.279 0.185 0.168 0.158 0.124 0.147 0.169 0.018 0.010 0.051 0.142 0.004 0.022 0.091 0.086 0.051 0.028 9.50 5.54 30.83 103.09 2.05 11.59 36.63 40.77 25.81 14.13 0.109 0.118 0.131 0.093 0.130 0.066 0.042 0.083 0.063 0.123 0.071 0.138 27.87 41.31 32.68 56.87 35.22 67.73 4 6 6 6 6 6 3 3 3 3 3 3 3 3 3 3 1 4 4 4 5 5 3 3 3 3 3 3 3 3 3 3 0.175 0.184 0.138 0.167 0.201 0.184 0.087 0.228 0.054 0.049 0.078 0.056 0.025 0.041 0.123 0.023 23.59 20.95 36.18 25.31 11.04 18.36 58.49 11.44 5 5 5 5 5 5 5 5 1 1 1 2 2 2 2 1

None 5th perc. of Market Share 95th perc. of Market Share None Vertically Integrated Firms 5th perc. of Market Share 95th perc. of Market Share None

Positive-R&D samples Supplier Industries R&D Customer Industries R&D Customer Industries R&D Customer Industries R&D Customer Industries R&D Customer Industries R&D

None None 5th perc. of Industry Conc. 95th perc. of Industry Conc. 5th perc. of Market Share 95th perc. of Market Share

ARTICLE IN PRESS

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Control variables Market Share Industry Concentration Firm Size Asset Collateral Value Volatility Nondebt Tax Shield Return on Assets Tobins q R&D Intensity SE Intensity

None None None None None None None None None None

347

ARTICLE IN PRESS
348 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

signicance of many of the supplier and customer variables are generally comparable to the traditional determinants of capital structure. 4.2.4. Robustness checks of industry-level analysis We conduct several tests with alternative measures and regression specications to determine the robustness of our above ndings on the relation between corporate debt levels and supplier/customer characteristics. (1) We repeat the supplier and customer regression analysis with Book Leverage as the dependent variable. The ndings are untabulated because of space constraints. Recall from Table 5 that the coefcient on the variable Customer ConcentrationMarket Share is negative but not signicant at conventional levels. In the Book Leverage regressions, the coefcient on this variable is 0.450, which is statistically signicant at the 1% level (t-statistic equals 2.73). The inferences from the book leverage regressions regarding all the other supplier and customer variables are identical to those when Market Leverage is the dependent variable. (2) We utilize the same regression model specications as above but estimate betweenrm regressions, which addresses concerns that time-series observations may not be independent (see, e.g., Berger, Ofek, and Yermack, 1997). In these regressions, the rm leverage ratios and all the independent variables are averaged across the years in the sample and the regression is estimated using these averages. Model 1 (2) of Table 8 presents the results from estimating a between-rm regression model on the subsample with supplier (customer) industry-level variables. All of the ndings from the earlier regressions continue to hold. For instance, in Model 1 (2), the coefcient on Supplier (Customer) R&D Intensity is negative and signicant, while the coefcient on Supplier (Customer) Concentration is signicantly positively related to debt levels. Overall, the ndings from the between-rm regressions are qualitatively similar to those in the OLS regressions. We also nd qualitatively similar results using the FamaMacBeth (1973) estimation procedure, which addresses the serial correlation problem. (3) Since it is possible that the results for the supplier (customer) variables are signicant in our regressions because these variables are correlated with the characteristics of customers (suppliers), in column of Table 8 we present results of an OLS regression of an extended model that includes both the supplier as well as the customer variables. We use the supplier sample in these regressions. Recall that in the construction of the customer sample, we restrict our sample to rms in industries that sell at most 25% of their output to nal users, and we make some assumptions regarding the characteristics of these nal users (as detailed in Appendix A). Here, we use the same assumptions to construct the customer variables for rms that are included in the supplier sample but not in the customer sample (rms in industries that sell more than 25% of their output to nal users). The ndings reported in the third column of Table 8 from the regression model that includes all supplier and customer variables are similar to our earlier results. We also conduct similar regression analysis with the sample used to examine customer effects (26,300 rm-year observations) and obtain qualitatively similar results. (4) For the sake of completeness, we use the lagged values of supplier and customer industry variables to check whether our estimates in Tables 38 suffer from an endogeneity bias and nd qualitatively similar results. This result is not surprising, as

Table 8 Ordinary least squares and between-rm regressions of leverage on supplier and customer industry-level variables The sample period is from 1984 to 2003. The dependent variable is Market Leverage, which is the sum of the book values of long-term debt and debt in current liabilities divided by the sum of the book value of debt and the market value of common equity. Between-rm regressions represent OLS regressions of the mean of the rms market leverage ratios on the means of independent variables. Supplier (Customer) Industries R&D is the weighted average of the R&D intensities of all supplier (customer) industries. Compensation of Employees is total compensation paid to employees divided by total output. Supplier (Customer) Concentration is the weighted average of the Herndahl indices of all supplier (customer) industries. Supplier (Customer) Change in Sales is a weighted average of the change in sales for supplier (customer) industries. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. All variables are winsorized at the 1st and 99th percentiles. Lagged values of rm-specic control variables are used. The reported t-values in the OLS regressions reect Whites heteroskedasticity correction. The symbols *, **, and *** indicate statistical signicance at the 10%, 5%, and 1% levels, respectively. Between-rm regressions Coeff. 0.114*** 1.94*** 0.013 0.18** 0.092* 0.13** 6.55 0.67 2.32 1.87 2.19 0.885** 0.220*** 0.164** 2.24 3.11 2.26 4.01 t-value Between-rm regressions Coeff. 0.087** t-value 2.13 OLS Regressions Coeff. 0.177*** 1.01*** 0.023** 0.17*** 0.048* 0.08*** 0.329*** 0.126*** 0.083*** t-value 23.5 7.94 2.54 7.24 1.86 4.63 6.15 5.72 4.59

ARTICLE IN PRESS

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Intercept

Supplier variables Supplier Industries R&D Compensation of Employees Compensation of EmployeesR&D Intensity Supplier Concentration Supplier Change in Sales

Customer variables Customer Industries R&D Customer Concentration Customer Change in Sales

349

350

Table 8 (continued ) Between-rm regressions Coeff. t-value Coeff. Coeff. t-value t-value Between-rm regressions OLS Regressions

Other control variables Industry Concentration Firm Size Asset Collateral Value Volatility Nondebt Tax Shield Return on Assets Tobins q R&D Intensity SE Intensity Year and Industry Dummies Number of observations Adjusted R-squared 0.004 0.01*** 0.222*** 0.003 5.819*** 0.10*** 0.016*** 0.17*** 0.023*** Yes 10,310 24.39% 0.39 6.81 24.60 0.68 7.63 14.95 22.10 5.63 4.52 0.030* 0.007*** 0.205*** 0.010 5.841*** 0.118*** 0.019*** 0.265*** 0.033*** Yes 4,065 19.77% 1.90 4.83 13.98 1.32 4.67 9.42 13.55 7.66 3.32 0.00 0.004*** 0.19*** 0.007*** 4.35*** 0.108*** 0.01*** 0.189*** 0.03*** Yes 76,160 25.51%

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 0.99 12.17 46.91 3.47 18.53 37.47 55.92 21.55 15.15

ARTICLE IN PRESS

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 351

endogeneity issues are less relevant in the regressions that use the industry-level supplier and customer variables because it is unlikely that concentration and R&D and alliance intensities at the industry level are determined by the debt level of a particular rm. Indeed, one can think of the industry-level R&D and alliance variables as instruments for the variables that measure these characteristics at the rm level. 4.3. Multivariate analysis with rm-level key supplier and customer variables The earlier tests use supplier/customer variables at the industry level. In this section, we use supplier/customer variables measured at the rm level. As we describe earlier, we identify (whenever possible) key suppliers and customers for each sample rm and construct RS investment variables for these key trading partners. This data set allows us to examine not only the effect of RS investments on leverage, but also whether the rms nancing choice is related to the decisions of its suppliers and customers to undertake such investments. 4.3.1. Firm leverage and R&D of key suppliers Table 9 presents results of OLS regressions for the sample rms for which we have the data needed to construct the variables for key suppliers. Our focus in these regressions is on the RS investments variables, but we also include industry-level supplier variables as controls. An extended model that uses controls for customer characteristics measured at the industry level yields similar results. In the rst column of Table 9, the coefcients on Key Suppliers R&D and Supplier SA and JV Dummy are negative and signicant, which is consistent with our earlier results and with Hypothesis 1. The results of the regressions presented in the second and third columns suggest that the relation between the RS investments variables and leverage is signicantly negative only for the subsample of rms with positive R&D intensity. Among the control variables, the coefcient on Supplier Change in Sales is negative but not signicant at conventional levels.18 4.3.2. Are rm debt levels and supplier R&D jointly determined? To address endogeneity concerns and to test whether the rms leverage and R&D of its key suppliers are jointly determined, we report in Table 10 the two-stage least squares (2SLS) estimates of a simultaneous equation model in which we treat the R&D of a supplier rm and the leverage of its main customer as endogenous variables. Since some suppliers report more than one customer for any given year, we consider the leverage of the main customer, the one that buys the largest percentage of the suppliers output. If a customer is the main customer for more than one supplier in any given year, we only include the most important supplier to the customer. We use this approach to avoid including a large number of observations for few rms, which may lead to inferences that are driven by these rms and not by the average rm in the sample.19 In addition, we include a supplier-customer observation only if the customer contributes at least 10% to
This variable is measured at the supplier-industry level. When it is constructed using key supplier data, the coefcient is also insignicant. 19 For example, General Motors was the main customer of 35 different suppliers in 2002. In this case, we only include General Motors and its most important reported supplier for that year, which is Lear Corp, the thirdlargest autoparts maker in North America. We also repeat our analysis not excluding any observation. We nd that the results that pertain to the variables of interest are similar to those reported in Table 10.
18

ARTICLE IN PRESS
352 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 Table 9 Ordinary least squares regressions of market leverage on key supplier variables The sample period is from 1984 to 2002.The dependent variable is Market Leverage, which is the sum of the book values of long-term debt and debt in current liabilities divided by the sum of the book value of debt and the market value of common equity. Column 2 (3) show OLS regression results for rms with positive (zero) R&D Intensity, which is the rms R&D expenditures divided by its total assets. Key Suppliers R&D is the weighted average of the R&D intensities of key suppliers. Suppliers SA and JV Dummy is a dummy variable that equals one if the rm has at least one alliance or a SA established with one of its suppliers, and zero otherwise. Compensation of Employees is total compensation paid to employees divided by total output. Supplier Concentration is the weighted average of the Herndahl indices of all supplier industries. Supplier Change in Sales is a weighted average of the change in sales for supplier industries. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. All variables are winsorized at the 1st and 99th percentiles. Lagged values of rm-specic control variables are used. The reported t-values reect Whites heteroskedasticity correction. The symbols *, **, and *** indicate statistical signicance at the 10%, 5%, and 1% levels, respectively. All customers Coeff. Intercept Key Suppliers R&D Suppliers SA and JV Dummy Control variables Compensation of Employees Supplier Concentration Supplier Change in Sales Industry Concentration Firm Size Asset Collateral Value Volatility Nondebt Tax Shield Return on Assets Tobins q R&D Intensity SE Intensity Year and Industry Dummies Number of observations Adjusted R-squared 0.237*** 1.632*** 0.014*** 0.135*** 0.027 0.070 0.000 0.002 0.088*** 0.014 4.653*** 0.282*** 0.023*** 0.336*** 0.101*** Yes 4,122 41.01% t-value 6.71 5.68 3.15 4.09 0.37 1.59 0.34 1.21 5.38 0.67 5.37 12.41 14.03 8.31 8.15 Positive R&D customers Coeff. 0.242*** 1.687*** 0.038*** 0.187*** 0.220** 0.068 0.030* 0.006*** 0.07*** 0.013 2.751*** 0.30*** 0.015*** 0.337*** 0.09*** Yes 2,932 40.31% t-value 5.17 6.03 5.62 5.32 2.47 1.26 1.95 3.59 3.29 0.46 3.24 12.74 11.27 8.26 7.62 Zero R&D customers Coeff. 0.217*** 1.070 0.046 0.013 0.395*** 0.068 0.066* 0.002 0.140*** 0.017 11.615*** 0.262 0.029*** 0.089** Yes 1,190 36.48% t-value 3.48 1.35 1.63 0.18 2.95 0.89 1.72 0.67 5.01 0.36 3.64 5.04 6.63 2.08

the suppliers revenues. The simultaneous equation model we estimate is as follows: Customer Market Leverage b0 b1 Supplier R&D gX , Supplier R&D d0 d1 Customer Market Leverage ZY m. (9) (10)

The two vectors X and Y include our control variables. Note that estimating both b1 and d1 is the principal objective of conducting the 2SLS procedure. In Eq. (7) above, we add the other determinants of Customer Market Leverage.

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 353 Table 10 Simultaneous equations: supplier R&D and leverage of main customer This table reports the second stage estimates of a 2SLS estimation of a simultaneous equation model that treats the supplier R&D and the leverage of its main customer as endogenous variables. The sample period is from 1984 to 2002. Customer Market Leverage is the market leverage of the customer that buys the highest percentage of the suppliers output, where market leverage is the sum of the book values of long-term debt and debt in current liabilities divided by the sum of book value of debt and market value of common equity. Supplier R&D is suppliers R&D expenditures divided by total assets. Compensation of Employees is total compensation paid to employees divided by total output. Supplier Concentration is the weighted average of the Herndahl indices of all supplier industries. Supplier Change in Sales is a weighted average of the change in sales for supplier industries. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. Supplier Size is the log of the suppliers total assets. Supplier Book-to-Market is the suppliers total assets divided by the sum of the book value of debt and the market value of common equity. Supplier Surplus Cash is the suppliers cash from assets-in-place divided by total assets. Supplier Sales Growth is the suppliers one-year growth in total sales. Supplier Stock Return is the suppliers one-year stock return. Supplier Leverage is the suppliers book leverage. All variables are winsorized at the 1st and 99th percentiles Lagged values of rm-specic control variables are used. The symbols *, **, and *** indicate statistical signicance at the 10%, 5%, and 1% levels, respectively. Customer market leverage Coefcient Dependent variable Intercept Supplier R&D Customer Market Leverage Control variables Compensation of Employees Supplier Concentration Supplier Change in Sales Industry Concentration Firm Size Asset Collateral Value Volatility Nondebt Tax Shield Return on Assets Tobins q R&D Intensity SE Intensity Supplier Size Supplier Book-to-Market Supplier Surplus Cash Supplier Sales Growth Supplier Stock Return Supplier Leverage Year Industry Dummy Variables Number of observations Adjusted R-squared t-value Supplier R&D Coefcient t-value

0.277*** 0.164**

5.93 2.13

0.174*** 0.202***

5.11 4.95

0.129*** 0.049 0.076 0.007 0.002 0.039* 0.031 4.903*** 0.248*** 0.017*** 0.333*** 0.114***

3.47 0.49 1.23 0.35 0.90 1.93 1.22 3.47 10.89 9.05 5.66 8.23 0.017*** 0.084*** 0.026** 0.003 0.018*** 0.010*** Yes 2,185 27.27% 12.07 11.19 2.07 1.19 6.34 2.98

Yes 2,185 38.95%

ARTICLE IN PRESS
354 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

In the Supplier R&D equation, we follow Coles, Daniel, and Naveen (2006) and use the following variables as determinants of the supplier R&D investments: 1. Supplier Size is the log of the suppliers total assets (item 6). 2. Supplier Book-to-Market is the suppliers total assets (item 6) divided by the sum of the book value of debt and the market value of common equity (item 6 item 60 + item 199item 25). 3. Supplier Surplus Cash is the suppliers cash from assets-in-place (item 308-item 125+item 46) divided by total assets (item 6). 4. Supplier Sales Growth is the suppliers one-year growth in total sales (item 12). 5. Supplier Stock Return is the suppliers one-year stock return (collected from CRSP). 6. Supplier Leverage is the equal to the sum of the book values of long-term debt and debt in current liabilities (items 9 and 34) divided by the book value of assets. 7. Industry and year indicator variables. The nal sample for this analysis consists of 2,185 rm-year observations. The results reported in the rst column of Table 10 for customer market leverage are consistent with the results of the earlier regressions. In the Customer Market Leverage regression the coefcient on Supplier R&D is negative and statistically signicant, and, interestingly, in the Supplier R&D regression, leverage is negatively related to Supplier R&D. Thus, we nd evidence that the leverage decision by the rm and the R&D level decision by the supplier are simultaneously determined. The results regarding the other determinants of supplier R&D are largely consistent with those reported in Coles, Daniel, and Naveen (2006). 4.3.3. Firms leverage and R&D of actual customers Table 11 presents the results of estimating our base models for the subsamples of rms for which the data required to construct the variables for the key customers are available. Consistent with Hypothesis 1, the coefcients on Key Customers R&D and Customer SA and JV Dummy are negative and statistically signicant. As columns 25 of Table 11 show, the negative relation between leverage and the R-S investments variables is pronounced only for rms with positive R&D and for rms producing durable manufacturing products. We next estimate a simultaneous equation model in which the R&D investments of a customer rm and the leverage of its main supplier are treated as endogenous variables. Since many of the customer rms in our sample are reported as customers by many supplier rms, we consider only the main supplier, or the supplier rm from which the customer buys most of its input (as a percentage of the customers total sales). If a supplier is the main supplier of more than one customer for a given year, we only include the customer who is the most important customer to the supplier. Further, for a suppliercustomer pair to be included, we require that the customer purchases from the supplier represent at least 10% of the customers sales.20 The resulting sample for this analysis consists of 385 rm-year observations.
We repeat the analysis on the larger sample that includes all customers, that is without these two restrictions, and obtain similar results.
20

Table 11 Ordinary least squares regressions of market leverage on key customer variables The sample period is from 1984 to 2002. The dependent variable is Market Leverage, which is the sum of the book values of long-term debt and debt in current liabilities divided by the sum of the book value of debt and the market value of common equity. Column 2 (3) show OLS regression results for rms with positive (zero) R&D Intensity, which is the rms R&D expenditures divided by its total assets. Column 4 show OLS regression results for rms with historical primary SIC code between 0100 and 1499. Column 5 show OLS regression results for rms producing durable goods, based on the Census Bureau classication. Key Customers R&D is the weighted average of the R&D intensities of key customers. Customers SA and JV Dummy is a dummy variable that equals one if the rm has at least one alliance or a SA established with one of its customers, and zero otherwise. Customer Concentration is the weighted average of the Herndahl indices of all supplier industries. Customer Change in Sales is a weighted average of the change in sales for supplier industries. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. All variables are winsorized at the 1st and 99th percentiles. Lagged values of rm-specic control variables are used. The t-values reported in parentheses reect Whites heteroskedasticity correction. The symbols *, **, and *** indicate statistical signicance at the 10%, 5%, and 1% levels, respectively. Positive R&D suppliers Zero R&D suppliers Agriculture and mining Durable manufacturing

All suppliers

ARTICLE IN PRESS

Intercept

Key Customers R&D

Customer SA and JV Dummy

0.096*** (5.63) 0.209*** (2.86) 0.029*** (8.37)

0.06*** (3.41) 0.209*** (2.78) 0.02*** (5.79)

0.127*** (4.32) 0.108 (0.47) 0.019 (1.52)

0.093 (1.52) 0.985 (0.70) 0.059* (1.83)

0.206*** (3.56) 0.226** (2.00) 0.019*** (3.87)

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Control variables Customer Concentration

Customer Change in Sales

Industry Concentration

Firm Size

0.145*** (3.02) 0.115** (2.14) 0.032*** (2.73) 0.007*** (6.76)

0.211*** (3.55) 0.16** (2.34) 0.018 (1.38) 0.003*** (3.54)

0.133 (1.51) 0.082 (1.04) 0.081*** (3.55) 0.017*** (8.65)

0.233 (1.35) 0.009 (0.07) 0.090 (1.37) 0.022*** (5.63)

0.147*** (2.63) 0.167** (2.02) 0.032** (2.13) 0.002 (1.32)

355

356

Table 11 (continued ) Positive R&D suppliers Zero R&D suppliers Agriculture and mining Durable manufacturing

All suppliers

Asset Collateral Value

Volatility

Nondebt Tax Shield

Return on Assets

Tobins q

R&D Intensity

SE Intensity

ARTICLE IN PRESS

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Year and Industry Dummies Number of observations Adjusted R-squared

0.246*** (21.07) 0.000 (0.06) 5.044*** (9.32) 0.105*** (13.36) 0.014*** (22.03) 0.197*** (12.22) 0.001 (0.21) Yes 9,452 31.28%

0.254*** (19.11) 0.00 (0.44) 4.380*** (9.15) 0.09*** (11.32) 0.011*** (19.47) 0.14*** (8.94) 0.003 (0.40) Yes 6,175 30.38%

0.203*** (9.90) 0.013 (0.85) 6.258*** (3.61) 0.133*** (6.68) 0.024*** (9.68) 0.013 (0.79) Yes 3,277 24.86%

0.140*** (3.04) 0.038 (0.89) 5.639 (1.48) 0.069 (1.60) 0.023* (1.85) 0.133 (0.50) 0.035 (0.51) Yes 9,07 16.33%

0.260*** (14.67) 0.014* (1.74) 4.422*** (6.71) 0.139*** (11.82) 0.015*** (14.69) 0.325*** (9.69) 0.007 (0.52) Yes 4,645 30.20%

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 357 Table 12 Simultaneous equations (2SLS): customer R&D and leverage of main supplier This table reports the second stage estimates of a 2SLS estimation of a simultaneous equation model that treats the customer R&D and the leverage of its main supplier as endogenous variables. The sample period is from 1984 to 2002. Supplier Market Leverage is the market leverage of the main supplier, where market leverage is the book value of long-term debt and debt in current liabilities divided by the sum of book value of debt and market value of common equity. Customer R&D is customers R&D expenditures divided by total assets. Customer Concentration is the weighted average of the Herndahl indices of all customer industries. Customer Change in Sales is a weighted average of the change in sales for customer industries. Industry concentration is the sales-based Herndahl index of the rms industry. Firm Size is the log of total assets. Asset Collateral Value is net property, plant, and equipment divided by total assets. Volatility is the standard deviation of operation income divided by total assets. Nondebt Tax Shields is investment tax credits divided by total assets. Return on Assets is operation income divided by total assets. Tobins q is book value of assets minus book value of common equity plus market value of common equity divided by book value of assets. R&D Intensity is R&D expenditures divided by total assets. SE Intensity is selling, general, and administrative expenses divided by total assets. Customer Size is the log of the customers total assets. Customer Book-to-Market is the customers total assets divided by the sum of the book value of debt and market value of common equity. Customer Surplus Cash is the customers cash from assets-in-place divided by total assets. Customer Sales Growth is the customer one-year growth in total sales. Customer Stock Return is the customer one-year stock return. Customer Leverage is the supplier book leverage. All variables are winsorized at the 1st and 99th percentiles. Lagged values of rm-specic control variables are used. The symbols ** and *** indicate statistical signicance at the 5% and 1% levels, respectively. Supplier Market Leverage Coefcient Dependent Variable Intercept Customer R&D Supplier Market Leverage Control variables Customer Concentration Customer Change in Sales Industry Concentration Firm Size Asset Collateral Value Volatility Nondebt Tax Shield Return on Assets Tobins q R&D Intensity SE Intensity Customer Size Customer Book-to-Market Customer Surplus Cash Customer Sales Growth Customer Stock Return Customer Leverage Year and Industry Dummy Variables Number of observations Adjusted R-squared t-value Customer R&D Coefcient t-value

0.328*** 0.585**

4.87 2.16

0.258*** 0.258***

5.39 2.86

0.234 0.207 0.005 0.016 0.164*** 0.113*** 9.412** 0.113** 0.013*** 0.172 0.111**

1.51 1.12 0.08 0.64 3.66 2.81 2.18 2.48 2.96 1.00 2.18 0.015*** 0.108*** 0.035 0.005 0.014 0.022 Yes 385 28.65% 4.35 5.10 1.02 0.67 1.50 0.73

Yes 385 27.05%

ARTICLE IN PRESS
358 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

The results from estimating the simultaneous equation model are in Table 12. As in the case of suppliers (Table 10), the ndings offer support for the conjecture that the leverage decision of the rm and the R&D investment decision by the customer are made simultaneously. In the Supplier Market Leverage equation, the coefcient on the customer R&D variable is negative and signicant, which is consistent with our previous ndings. In the Customer R&D regression, the coefcient on the supplier leverage variable is negative and signicant. Finally, the coefcients for the other variables are largely in line with the results reported in this paper. 4.4. Robustness checks We perform the following tests to ensure that our ndings are robust. (1) The subsample of rms for which we have both supplier and customer information at the rm level consists only of 818 rm-year observations. We conduct the same analysis on this subsample using all rm-level supplier and customer variables. The results of this analysis are qualitatively similar to those reported here although the statistical signicance levels are lower. (2) Recall that we do not include rm xed effects in our regressions. In unreported regressions, we nd that including rm xed effects results in ndings that are qualitatively similar to those pertaining to the supplier and customer R&D variables. However, we nd that the coefcients on the supplier and customer concentration variables, although positive, are not signicant at conventional levels. We believe that these coefcients are insignicant because the changes in the supplier and customer concentration variables from one year to the next are very small relative to the crosssectional variations. For example, the standard deviation for Supplier Concentration for 2003 is 0.05 compared to 0.01, the standard deviation of the change in this variable between 2002 and 2003. Zhou (2001) shows that when time-series variation in the data is not signicant relative to the cross-sectional variation, including rm xed effects may result in insignicant results even if the actual relation is signicant. (3) In the construction of all the R&D-based variables, we normalize R&D expense by the rms total assets. We repeat our analysis after normalizing R&D by the rms total sales and nd similar results. 5. Conclusion In this article, we investigate how a rms capital structure decision is determined in a setting that includes the rms suppliers and customers as stakeholders. Our study contributes to the literature on how the rm may use its choice of debt level to affect relationship-specic investments by the rms stakeholders. Extant theory suggests that a rm may rely less on debt nancing in order to induce its suppliers and customers to undertake relationship-specic investments. Consistent with this hypothesis, we show that a rms leverage is negatively related to the intensity of R&D investments by its suppliers and customers. We also nd that leverage is negatively related to the rms intensity of SAs and JVs with its suppliers and customers. Further, we nd that the negative relation between leverage and customer RS investments is more pronounced for rms operating in a concentrated industry and for those with high market share.

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 359

We obtain these empirical results when supplier/customer R&D intensity is measured at either the industry or the rm level. On the rm-level sample of key suppliers/customers, we estimate simultaneous equation models that treat the rms leverage and the R&D of key suppliers and customers as endogenous variables. We nd evidence to suggest that the rms leverage and the R&D investments of its suppliers and customers are simultaneously determined. In both the leverage as well as the R&D regressions, we nd a negative relation between the rms leverage and the R&D investments of its suppliers and customers. We also propose that if a rm has fewer suppliers or customers, then it is at a bargaining disadvantage relative to these two stakeholders. If this is true, then theory suggests (e.g., Bronars and Deere, 1991) that the rm will offset this bargaining disadvantage by taking on greater debt levels. Our empirical analysis provides considerable evidence to support this hypothesis. In particular, we nd that rms leverage ratios are signicantly higher when supplier and/or customer industries are more concentrated.

Appendix A. Construction of the supplier and customer industry-level variables The construction of the supplier and customer industry-level variables requires both the identication of the supplier and customer industries for each rm in our sample and the measurement of the R&D intensity, concentration, and change in sales for each of these industries. We rely on the Use table of the benchmark inputoutput (IO) accounts to identify the supplier and customer industries as follows. For each rm in our sample, we rst identify its primary SIC code using Compustats Historical SIC Code data item. Since Compustat reports the historical primary SIC code from 1987 onward, we use the 1987 historical SIC code for years prior to 1987. Given that the Use table is constructed based on the IO six-digit coding system, we use the IO-SIC codes conversion table employed by Fan and Lang (2000) to identify the rms IO code. We then use the rms IO code to identify its supplier and customer industries from the Use table. We use the 1987, 1992, and 1997 Use tables for the periods 19841989, 19901994, and 19952003, respectively. Note that, unlike previous tables, the 1997 benchmark accounts are constructed based on the North American Industry Classication System (NAICS) and not the SIC coding system. Thus, to be consistent across the sample period, we use the 1997 annual updates of the 1992 benchmark tables for the period 19952003. The next step in the construction process is to compute the R&D intensity, concentration, and change in sales for each identied supplier and customer industry. Toward this end, we identify all Compustat rms that operate in each of these industries. We compute the R&D intensity of each industry as the sum of the R&D expenses of all industry rms divided by total industry assets. Industry concentration is measured using the sales-based Herndahl index. Change in sales in computed as the one-year change is sales for the median rm in the industry. If there are no Compustat rms in a particular six-digit IO industry, we use rms at the ve-digit, four-digit, two-digit, and one-digit level, respectively. We also repeat our analysis by only considering six-digit IO industries with available Compustat data and nd qualitatively similar results. The nal step is to compute the weights we assign to each supplier and customer industry, and to make assumptions regarding the characteristics of some particular supplier and customer industries. The details of this last step are as follows.

ARTICLE IN PRESS
360 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

A.1. Customer industry-level variables The customer industry-level variables are dened as follows:
Customer Industries R&D
n X
j 1 j ai

Customer Industry R&Dj Industry Percentage Sold ji ,

Customer Concentration

n X
j 1 j ai

Herfindahl Indexj Industry Percentage Sold ji ;

Customer Change is Sales

n X
j 1 j ai

Change in Salesj Industry Percentage Sold ji :

In the above three equations, Industry Percentage Soldji is equal to industry is output that is sold to industry j divided by the total output of industry i. Total output of industry i is computed by summing the output sold across all customer industries, excluding industry i itself, and including the following nal uses: Personal Consumption Expenditures, Gross Private Fixed Investments, and Government Consumption Expenditures and Gross Investment. Note that following this method, Industry Percentage Soldji will be equal to the output sold to industry j divided by total output of industry i that is consumed domestically, this is, excluding exports. The results reported in the paper are robust to the inclusion of exports in the above computation. However, if one assumes that exports are sold to industries that have characteristics that are similar to domestic customer industries, then excluding exports as we do in this paper from the analysis should be a better choice. Including the percentage of the industrys output that is exported in our regressions does not affect our results. Below we show below the assumptions we make regarding the R&D intensity, concentration, and change in sales of nal users. We nd qualitatively similar results using the assumptions shown in parentheses below. Note that due to our restriction regarding the inclusion of industries that have at most 25% of their output sold to nal users, the assumptions below have little effect on our analysis. In fact, the median industry included in the industry-level customer analysis has about 6.5% of its output sold to nal users. The assumptions we make are as follows: Personal Consumption Expenditures:

R&D 0. Concentration 0. This assumption follows Schumacher (1991). (Robustness: concentration median concentration of retail industries). Change in Sales median change in sales of all four-digit SIC code industries. Gross Private Fixed Investments:

R&D median R&D intensity of all Compustat rms. (Robustness: R&D mean R&D intensity of all Compustat rms).

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

361

Concentration median concentration of all four-digit SIC code industries). (Robustness: Concentration mean concentration of all four-digit SIC code industries). Change in Sales median change in sales of all four-digit SIC code industries. Government Consumption Expenditures and Gross Investment:

R&D median R&D intensity of all Compustat rms. (Robustness: R&D mean R&D intensity of all Compustat rms). Concentration 1. This assumption follows Schumacher (1991). (Robustness: Concentration mean concentration of all four-digit SIC code industries). Change in Sales median change in sales of all four-digit SIC code industries.

A.2. Supplier industry-level variables The supplier industry-level variables are dened as follows: Supplier Industries R&D
n X
j 1 j ai

Supplier Industry R&Dj Industry Input Coefficientji ,

Supplier Concentration

n X
j 1 j ai

Herfindahl Indexj Industry Input Coefficientji ,

Supplier Change in Sales

n X
j 1 j ai

Change in Salesj Industry Input Coefficientji :

In the above equations, Industry Input Coefcientji is the dollar amount of the jth supplier industrys output used as input to produce one dollar of the output of the ith industry. Industry is total output is computed by summing the purchases from all supplier industries and adding Total Value Added, which includes compensation of employees, indirect business tax and nontax liability, and other value added. In the analysis reported in the paper we control for the compensation of employees. Thus, we construct the variable Compensation of Employees as the dollar amount spent on employee compensation in the rms industry divided by the industrys total output. For robustness, we repeat our analysis after dening Industry Input Coefcientji as the dollar amount bought from the jth supplier industry divided by total purchases from all supplier industries. Note that this method of dening Industry Input Coefcientji excludes Total Value Added from the denominator. The results of the analysis under this specication are qualitatively similar to those reported in the paper. Appendix B. Percentage of output sold to nal users For a number of selected industries, Table B.1 shows the percentage of an industrys output that is used as intermediate input in the production of other goods and services, and the percentage sold to nal users. Final uses are of three types: Personal Consumption

ARTICLE IN PRESS
362 Table B.1 Industry name SIC Code % used as intermediate input 0 1 2 52 63 83 93 96 % sold to nal users 100 99 98 48 37 17 7 4 Main customer industry/ Another signicant nal users customer industry/nal users Personal Consumption Expenditures Federal Government Personal Consumption Expenditures Personal Consumption Expenditures Aircraft Gross Private Investments Motor Vehicles and Passenger Car Bodies Motor Vehicle Parts and Accessories N/A N/A N/A Air Transportation Federal Government J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

Motor Homes

3716

Small Arms Ammunition 3482 Motor Vehicles and 3711 Passenger Car Bodies Petroleum Rening 2911 Aircraft Engines and Engine Parts Motors and Generators 3724 3621

Motor Vehicle Parts and 3714 Accessories Sheet Metal Work 3444

Screw Machine Products 3451 Nonferrous Foundries (castings) Primary Smelting and Rening of Copper Primary Production of Aluminum Plastics Materials 3360

98 99

2 1

3331

100

3334 2820

100 100

0 0

Refrigeration and Heating Equipment Automotive Repair Shops And Services Industrial and Commercial Buildings Construction Motor Vehicles and Motor Vehicle Parts and Passenger Car Bodies Accessories Motor Vehicle Parts and Industrial and Accessories Commercial Machinery and Eqpt. Rolling, Drawing, and Industrial and Extruding of Copper Commercial Machinery and Equipment Aluminum Rolling and Aluminum Castings Drawing Miscellaneous Plastics Paints and Allied Products Products

Expenditures, Government Consumption Expenditures and Gross Investment, and Gross Private Fixed Investments. We obtain the data from the 1992 Use table of the benchmark inputoutput accounts. The details of the treatment of exports are included in Appendix A.

Appendix C. Examples of supplier-customer industry pairs Table C.1 reports selected supplier-customer industry pairs drawn from the 2002 Compustat industry segmented les. For each supplier, we select the main customer, which is the rm that buys the largest percentage of the suppliers output. The rst two columns show the primary industries of the supplier and main customer rms with the respective SIC codes. The number of pairs shown in the third column is the number of supplier-main customer pairs across the two industries. The number of unique customers shown in the fourth column is different from the number of pairs since some suppliers have the same main customer. The last column shows an example of a supplier-customer pair. The table

Table C.1 Customer industry SIC code 3711 16 3 Dana Corp Supplier Number of pairs Example Main customer Ford Motor Co

Supplier industry

Name

SIC code

Name

Number of unique customers

3714

Motor Vehicles and Passenger Car Bodies 2834 4813 11 8 Tellabs Inc 12 11 Medarex Inc

2836

Pharmaceutical Preparations

Lilly (Eli) & Co Verizon

Motor Vehicle Parts and Accessories Biological Products Telephone and Telegraph Apparatus 3674 3663 6 3 8 5

3661

Telephone Communications

3559

Intel Corp Motorola Inc

3674

ARTICLE IN PRESS

3841

Communications Semiconductors and Related Devices Radio and Television Broadcasting and Communications Eqpt. Pharmaceutical Preparations 2834 5 3

Novellus Systems Triquint Semiconductor Inc Micro Therapeutics Inc 5 Healthetech Inc

Abbott Laboratories

3845

Pharmaceutical Preparations

2834

Natures Sunshine Prods

J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

3825

Semiconductors and Related Devices

3674

Ltx Corp

Texas Instruments Inc

Special Industry Machinery Semiconductors and Related Devices Surgical and Medical Instruments and Apparatus Electromedical and Electrotherapeutic Apparatus Instruments for Measuring and Testing of Electricity and Electrical Semiconductors and Related Devices 3571 4 2

3674

Electronic Computers

Tripath Technology Inc

Apple Computer

363

ARTICLE IN PRESS
364 J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365

shows the top nine customer-supplier industry pairs, ranked based on the number of supplier-main customer pairs.

References
Allen, J.W., Phillips, G.M., 2000. Corporate equity ownership, strategic alliances, and product market relationships. Journal of Finance 55, 27912815. Armour, H.O., Teece, D.J., 1980. Vertical integration and technological innovation. Review of Economics and Statistics 62, 470474. Bartelsman, E.J., Caballero, R.J., Lyons, R.K., 1994. Customer- and supplier-driven externalities. American Economic Review 84, 10751084. Berger, P.G., Ofek, E., Yermack, D.L., 1997. Managerial entrenchment and capital structure decisions. Journal of Finance 52, 14111438. Boerner, C.S., Macher, J.T., 2001. Transaction cost economics: an assessment of empirical research in the social sciences. Working paper. UC Berkeley and Georgetown University. Brander, J.A., Lewis, T.R., 1986. Oligopoly and nancial structure: the limited liability effect. American Economic Review 76, 956970. Bronars, S.G., Deere, D.R., 1991. The threat of unionization, the use of debt, and the preservation of shareholder wealth. Quarterly Journal of Economics 106, 231254. Campello, M., 2003. Capital structure and product markets interactions: evidence from business cycle. Journal of Financial Economics 68, 353378. Campello, M., 2005. Debt nancing: does it boost or hurt rm performance in product markets? Forthcoming, Journal of Financial Economics. Chevalier, J.A., 1995a. Capital Structure and product-market competition: empirical evidence from the supermarket industry. American Economic Review 85, 415435. Chevalier, J.A., 1995b. Do LBO supermarkets charge more? An empirical analysis of the effects of LBOs on supermarket pricing. Journal of Finance 50, 10951112. Chevalier, J.A., Scharfstein, D.S., 1996. Capital-market imperfections and countercyclical markups: theory and evidence. American Economic Review 86, 703725. Coles, J.L., Daniel, N.D., Naveen, L., 2006. Managerial incentives and risk taking. Journal of Financial Economics 79, 431468. Dasgupta, S., Sengupta, K., 1993. Sunk investment, bargaining and choice of capital structure. International Economic Review 34, 203220. Dasgupta, S., Titman, S., 1998. Pricing strategy and nancial policy. Review of Financial Studies 11, 705737. Fama, E.F., MacBeth, J.D., 1973. Risk, return, and equilibrium: empirical tests. Journal of Political Economy 81, 607636. Fan, J.P.H., Lang, L.H.P., 2000. The measurement of relatedness: an application to corporate diversication. Journal of Business 73, 629660. Fee, E.C., Hadlock, C.J., Thomas, S., 2005. Corporate equity ownership and the governance of product market relationship. Journal of Finance 61, 12171250. Fee, E.C., Thomas, S., 2004. Sources of gains in horizontal takeovers: Evidence from customer, supplier, and rival rms. Journal of Financial Economics 74, 423460. Fitzgerald, K.R., 1996. Proprietary information-should supplier share it? Purchasing, October 3, 5556. Hanka, G., 1998. Debt and the term of employment. Journal of Financial Economics 48, 245282. Holmstrom, B., Roberts, J., 1998. The boundaries of the rm revisited. Journal of Economic Perspectives 12, 7394. Kale, J.R., Noe, T., Ramirez, G., 1991. The effect of business risk on corporate capital structure: theory and evidence. Journal of Finance 46, 16931715. Khanna, N., Tice, S., 2000. Strategic responses of incumbents to new entry: the effect of ownership structure, capital structure, and focus. Review of Financial Studies 13, 749779. Khanna, N., Tice, S., 2005. Pricing, exit and location decisions of rms: evidence on the role of debt and operating efciency. Journal of Financial Economics 75, 397427. Kovenock, D., Phillips, G.M., 1997. Capital structure and product market behavior. Review of Financial Studies 10, 767803.

ARTICLE IN PRESS
J.R. Kale, H. Shahrur / Journal of Financial Economics 83 (2007) 321365 365 Levy, D., 1985. The transactions cost approach to vertical integration: an empirical examination. Review of Economics and Statistics 67, 438445. Loughran, T., Ritter, J., 1997. The operating performance of rms conducting seasoned equity offerings. Journal of Finance 52, 18231850. Lyandres, E., 2006. Capital structure and interaction among rms in output markets: theory and evidence. Journal of Business, Forthcoming. Mackay, P., Phillips, G., 2006. How does industry affect nancial structure? Review of Financial Studies, Forthcoming. Maksimovic, V., 1988. Capital structure in repeated oligopolies. RAND Journal of Economics 19, 389406. Maksimovic, V., 1995. Financial structure and product market competition Chapter 27. In: Jarrow, R.A., Maksimovic, V., Ziemba, W.T. (Eds.), Hanbooks in Operations Research and Management Science, vol. 9. North-Holland, Amsterdam, pp. 887920. Maksimovic, V., Titman, S., 1991. Financial policy and reputation for product quality. Review of Financial Studies 4, 175200. Maskin, E., Riley, J., 1984. Monopoly with incomplete information. RAND Journal of Economics 15, 171196. Modigliani, F., Miller, M.H., 1958. The cost of capital, corporation nance and the theory of investment. American Economic Review 48, 261297. Opler, T.C., Titman, S., 1994. Financial distress and corporate performance. Journal of Finance 49, 10151040. Perotti, E.C., Spier, K.E., 1993. Capital structure as a bargaining tool: the role of leverage in contract renegotiation. American Economic Review 83, 11311141. Phillips, G., 1995. Increased debt and product-market competition. Journal of Financial Economics 37, 189238. Ravenscraft, D., 1983. Structure-prot relationship at the line of business and industry level. Review of Economics and Statistics 65, 2231. Scherer, F.M., Ross, D., 1990. Industrial Market Structure and Economic Performance. Houghton-Mifin, Boston. Schumacher, U., 1991. Buyer structure and seller performance in US manufacturing industries. Review of Economics and Statistics 73, 277284. Shahrur, H., 2005. Industry structure and horizontal takeovers: analysis of wealth effects on rivals, suppliers, and corporate customers. Journal of Financial Economics 76, 6198. Shea, J., 1993. Do supply curves slope up? Quarterly Journal of Economics 108, 132. Snyder, C.M., 1996. A dynamic theory of countervailing power. Rand Journal of Economics 27, 747769. Stigler, G.J., 1964. A theory of oligopoly. Journal of Political Economy 72, 4461. Stole, L.A., Zwiebel, J., 1996. Organizational design and technology choice under intrarm bargaining. American Economic Review 86, 195222. Subramanian, V., 1996. Underinvestment, debt nancing, and long-term supplier relations. Journal of Law, Economics and Organization 12, 459477. Titman, S., 1984. The effect of capital structure on a rms liquidation decision. Journal of Financial Economics 13, 119. Titman, S., Wessels, R., 1988. The determinants of capital structure choice. Journal of Finance 43, 119. Williamson, O., 1975. Markets and Hierarchies: Analysis and Antitrust Implications. Free Press, New York. Williamson, O., 1985. The Economic Institutions of Capitalism. Free Press, New York. Zhou, X., 2001. Understanding the determinants of managerial ownership and the link between ownership and the link between ownership and performance: comment. Journal of Financial Economics 62, 559571. Zingales, L., 1998. Survival of the ttest or the fattest? Exit and nancing in the trucking industry. Journal of Finance 53, 905938.