Product Market Relationships and Cost of Bank Loans: Evidence from Strategic Alliances

Transcription

1 Product Market Relationships and Cost of Bank Loans: Evidence from Strategic Alliances Yiwei Fang Lally School of Management and Technology, Rensselaer Polytechnic Institute 110, 8 th Street, Pittsburgh Building Troy, New York Telephone: fangy2@rpi.edu Bill Francis Lally School of Management and Technology, Rensselaer Polytechnic Institute 110, 8 th Street, Pittsburgh Building Troy, New York Telephone: francb@rpi.edu Iftekhar Hasan Lally School of Management and Technology, Rensselaer Polytechnic Institute 110, 8th Street, Pittsburgh Building Troy, New York Telephone: hasan@rpi.edu Haizhi Wang Stuart School of Business Illinois Institute of Technology 565 W Adams St, Suite 460 Chicago, IL Phone: hwang23@stuart.iit.edu This version: December 30, 2008 Please do not quote with the authors permission! corresponding author: please send all correspondence to fangy2@rpi.edu 1

2 Abstract Product Market Relationships and Cost of Bank Loans: Evidence from Strategic Alliances This paper examines whether and to what extent product market relationships (i.e. strategic alliances) may affect nonfinancial firms cost of bank loans. We construct several measures to capture the frequency of a firm s strategic alliance activities and its position in the alliance network. We find that firms with active corporate alliance activities experience lower cost of debt from banks. We also document that allying with a prominent partner (i.e., S&P 500 firms) can provide the endorsement effect and benefit the borrower by reducing the cost of debt. Moreover, a firm positioning in the central of the alliance network enjoys the benefit of low cost of bank loans. Our findings are consistent with increased visibility and information flows reducing asymmetric information problem faced by a borrower and the likelihood for a repeatedly observed actor to behave opportunistically. Keyword: Cost of Bank Loans; Strategic Alliances; Product Market Relationships JEL Classification: G21 G30 D82 D85 2

3 Product Market Relationships and Cost of Bank Loans: Evidence from Strategic Alliances 1. Introduction For the past two decades, the importance of strategic alliances in American industry has been increasing sharply (Lerner & Rajan 2006). Commonly defined as voluntarily initiated organizational agreements between firms, corporate alliances bring together otherwise legally independent firms to share the costs and benefits of a mutually beneficial activity (Chan et al. 1997). Corporate alliances allow participating firms to gain access to complementary resources and strengthen their competitive positions (Gulati 1995b; Baum et al. 2000). More importantly, corporate alliances blur firms boundaries and provide another important option for firms to grow (Habib & Mella- Barral 2006; Robinson 2006; Lindsey 2008). While a large strategic management literature has explored the patterns, motivations and benefits for firms entering alliances (Gulati 1995b; Eisenhardt & Schoonhoven 1996; Das et al. 1998; Stuart et al. 1999; Lavie & Rosenkope 2006), knowledge is still scant with regard to the financial consequences of corporate alliance activities (Lerner & Rajan 2006). In this paper, we study whether and to what extent nonfinancial firms alliance activities may affect their cost of bank debts. More specifically, we examine the crosssectional difference of bank loan pricing, taking into consideration of the frequency of alliance activities conducted by nonfinancial firms. In addition, borrowing techniques from graph theory, we attempt to investigate how nonfinancial firms positions in the alliance network may affect the loan prices. There are several possible reasons why firms, Ceteris paribus, may obtain cheap bank financing if they are actively involved into strategic alliances. First, actively participation in alliances may help to reduce asymmetric information problem. Businesses experience a difficult time raising capital when it is difficult to make judgment on their future prospects by outsiders. This asymmetric information problem creates a dilemma for potential lenders (Strahan 1999). If a creditor can obtain sufficient information on a borrower s risk and potential profitability at the outset of a relationship, 3

4 the cost of extending credit to that borrower will be reduced. It has been documented that interfirm alliances greatly spur the knowledge flows between partner firms (Anand & Khanna 2000; Dessein 2005; Gomes-Casseres et al. 2006). Increased knowledge flows among firms may well inform potential lenders about the future of borrowing firms. Second, allying with a prominent partner could mitigate information problems regarding the quality of focal firms. For example, if asymmetric information has an adverse impact on the cost and availability of external capital, it may be less costly to form interfirm relationship with an informed partner. Under this argument, an informed and reputable partner can serve as a validation of the firm s investment opportunities to the capital market and other capital providers (e.g., commercial banks). Podolny (1993) argues that when information regarding the quality of a firm cannot be observed directly or is not sufficient to resolve the uncertainty, evaluators and resource holders routinely take into account the characteristics of its partners. Finance literature has accumulated quite a bit of evidence about the certification effects provided by prominent partners (Carter & Manaster 1990; Carter et al. 1998; Fang 2005; Nicholson et al. 2005). Third, there are network effects existing in corporate alliances (Robinson & Stuart 2000; Garmaise & Moskowitz 2003; Allen & Babus 2008). Network effects can be multi-fold. For example, interfirm alliances can function as mechanism for sharing technological knowledge (Gomes-Cassseres et al. 2006; Schoenmakers & Duysters 2006). Reliable information about actions of the agents in the network is not universally available to every participant (Robinson & Stuart 2000; Baker et al. 2002; Robinson 2006). Rather, crucial information is made accessible to other members through the ties in the network. Repeated alliances form a network with ties and nods among many participants. When a group of agents are connected by a set of overlapping relationships, transaction-based information is governed by the network structure as to who gets what information (Hochberg et al. 2007). For another instance, an active network participant will be observed repeatedly, hence be put under the spot light. In this situation, the focal firm is less likely to behave opportunistically since as since the non-cooperative activity is easily observed when surrounded by many network members. Thus, network can function as an effective governance mechanism. Network members are constrained to a 4

5 group of social mechanism such as power, influence, and reputation, which help support the operation of the open-ended network governance (Throrelli 1986; Lipparini and Lorenzoni 1993). Moreover, similar to evidence documented by Garmaise and Moskowitz (2003) in real estate markets, it is plausible that an information network exists among nonfinancial firms that can bridge potential pairs of debtors and creditors thus link product market and capital market together. Lastly, alliances are oftentimes used as a means to pursue related diversifications. Related diversification may well bring partners with complementary resources together and create a significant amount of wealth (Fan & Lang 2000; Fan & Goyal 2006). In addition, many alliances involving product market relationships other than joint R&D (e.g., marketing agreement, manufacturing agreement and licensing agreements). When firms operate in industries involving dynamic technological regimes, participating in alliances leading to related diversification can significantly reduce firms risk and consequently, their likelihood to fail and default. To study the role of corporate alliance in debt financing, we collect a sample of U.S. public firms that receive bank loans during 1991 to Relying on SDC Alliances database, we track the history of their alliance relationships with other firms in the SDC Alliance database for 3 years prior to the loan contract initiations. Based on that, we construct several measures to capture the frequency of a firm s strategic alliance activities. We also apply a network approach to evaluate each participating firm s relative position in the corporate alliance network. We first document the extent of our sample firms participation in alliance relationships, and provide comparison tests on bank loan prices between the group of firms entering into corporate alliances and the group of firms that do not. Next, conditional on firms having alliance activities, we examine the frequency of strategic alliances and its effect on cost of bank debt. Borrowing tools from graph theory, we then examine the relation between sample firms network positions and their cost of debt. We investigate whether a sample firm s location central in the network will reduce its cost of bank debt. Consistent with our prediction, we find that firms actively involved in alliance activities experience lower cost of bank debt, compared to the group of firms without any 5

6 alliances experience. However, infrequent participation in interfirm agreements do not warrant such benefit of lowered cost of external financings. Firm with active involvement in corporate alliance activities enjoy cheaper bank loans. It is plausible certain industry characteristics and firm size are naturally linked to higher propensity for firm to form alliances. Therefore, we use residuals from regressing a firm s alliances activities on industry characteristics and firm size as alternative measures for a firm s network position. Ideally, the residual represents the portion of alliance activity which cannot be explained simply by industry characteristics and firm size. Nonetheless, using these alternative measures does not change our main findings materially. We further look at whether allying with a prominent partner can mitigate the information problem and reduce the cost of bank loans because of the endorsement effects. We do find that if the alliance partner is an S&P 500 company, the loan pricing is lower. In addition, this effect is more prominent if the borrower itself is not an S&P 500 company at the same time. Our centrality measures of a firm s relative position in the alliance network offer striking results. We argue these results are unlikely driven by reverse causality (i.e., the argument that because of lower cost of external finance, firms will be able to gain more partnerships and thus be the central points of the interfirm network) because of the way we construct the centrality measures for our sample firms (Hochberg et al. 2007). For a sample firm, measures of network centrality are calculated from its alliance activities for the 3 preceding years. Thus, we relate a nonfinancial firm s past network position to its future cost of bank debt. Our findings indicate that firms central in the network receive lower cost of loans. Using similar logic, we regress our measures of network centrality on industry characteristics and firm size. We take residuals as alternative measures of network centrality and find robust results. We think our study contributes to existing literature in several ways. First, our findings manifest a robust link between product market and firm financing cost. Previous literature has documented the strategic purposes of corporate alliances (Gulati et al. 2000). Our evidence provides additional insights on the financial consequence for firms entering into interorgainzational agreements. Second, consistent with existing literature (Stuart et al. 1999), allying with prestigious partners will benefit borrowers in terms of cheaper 6

7 financing. Third, we demonstrate that there are informal social networks existing in corporate alliances activities (Garmaise & Moskowitz 2003). Being in the central of the corporate network will increase a firm s visibility and reduce the information asymmetry problem, which lead to lower cost of bank loans. The remainder of this paper is organized as follows. In section 2, we provide a brief review of existing literature to place our study in an appropriate context. Section 3 details our data collection and sample construction procedure as well as our measurements of alliances activities and sample distributions. We report our empirical findings in section 4. Section 5 summarizes and concludes. 2. Literature Review According to the National Bureau of Economic Research (NBER) conference held in 2002, corporate alliances can be defined as a cooperative agreement between two or more firms, involving substantial investment, and lying between one extreme of full ownership by on firm of the other others and the other extreme of a short-term, arm s length contract between the firms. (Lerner & Rajan 2006) If one thinks of market transactions on a spectrum from arm s length to relational, strategic alliances are typically viewed as the latter case. They involve substantial relation-specific investments and long standing cooperative mechanisms, blurring firms boundaries through a network of relationships that can be an important source of value (Garvey 1995; Baker et al. 2002; Gay & Dousset 2005); Jensen and Meckling 1991). A large strategic management has explored causes and consequences of corporate alliances and generated significant insights on this ongoing phenomenon. For example, literature has documented that strategic alliances can be motivated by gaining marketing power, accessing to complementary resources, exploiting firm-specific competencies, or reducing environmental uncertainties (Kogut 1988); Burger, Hill and Kim, 1993). By participating in a strategic network, firms can gain advantage from learning, scale and scope economies; it also allows firms to achieve strategic objectives, such as risk sharing and outsourcing value-chain stages (Gulati et al. 2000). Organizational scholars have 7

8 also emphasized the signaling value of young firms having prominent affiliations to key external resource holders (Podolny 1993; Podolny et al. 1996; Stuart et al. 1999; Gulati & Higgins 2003). Much of the research argues that the value that accrues to firms not only depend on its own endowment, but also derived from its partner as well as the social network where it is embedded (Baum et al. 2000; Das & Teng 2000). Finance literature has increasingly recognized the importance of strategic alliances and offered some insights on the nature of corporate alliances. Among a small number of studies, Chan et al. (1997) are the first to document the wealth creation effects of announcements of strategic alliances. Allen and Phillips (2000) find block ownership purchasing can add value to the target company when combined with product market relationships. Other research has shown that strategic alliances involving equity stakes can be used an effective way to deter entrance (Chen & Ross 2000; Mathews 2006). For biotech companies, strategic alliances are an important source of finance to fund their R&Ds, especially when public market financing is diminished (Lerner et al. 2003). Furthermore, Nicholson et al. (2005) develops a signaling model of strategic alliances and demonstrates that in the imperfect information environment alliance deals could signal firm quality and future growth opportunities. Consequently, with established alliance relationships, biotech companies receive substantially higher valuation from venture capitalist and public equity markets (Nicholson et al. 2005). Facing uncertain and dynamic conditions, potential investors are likely to rely on the prominence of the affiliates of those companies. A recent paper by Lindsey (Lindsey 2008) find that two entrepreneurial firms are more likely to form alliances if they share a common venture capitalist. Despite a few pieces of above mentioned findings, we are still lacking of understanding with regard to the financial consequence corporate alliance activities. In this study, we are trying to offer some new insights by relating a firm s alliance activity to its cost of bank loan and attempt to establish a link between product market and capital market. 8

9 3. Data and Methodology 3.1 Sample construction We rely on three main databases to construct our sample, namely: Dealscan (Loan Pricing Corporation Database), Compustat, and SDC M&A/Joint Venture Database. Dealscan is our primary source of bank loan information (see., (Strahan 1999). It contains U.S. and foreign commercial loans made to both public and private corporations. The data on Dealscan are recorded as Deal or Facility. Each deal is a loan contract signed between a borrower and a lender at a given time with most deals having just one facility (tranch). Therefore, our unit of observation is one each facility (tranch). Bank loans are priced according to default risk. Though a typical loan contract involves the usage of collateral requirements and covenants, and considers loan size and maturity, there are still considerable risks associated with loans that vary across firms (Strahan 1999). The drawn all-in-spread generally reflect the riskness of the loan and is prices as a mark-up over a market interest rate (i.e., LABOR). For each facility or tranche, we, following Strahan (1999), retrieve drawn all-in-spread information and take natural logarithm to normalize the distribution as our measure of loan price. We focus on U.S. public borrowers that are recorded in Dealscan to have at least one time loan origination during the period from 1991 to Since Dealscan does not provide detailed information about the borrowers, we then match Dealscan with Compustat to obtain enough firm-specific accounting information. We exclude bank institutions or bank holding companies (with two digit SIC codes of 60-depository institution, 61-nondepository institution, 62-investment banks and 67-bank holding companies). Other types of financial firms are kept. The matching process generates 3160 unique U.S. public firms with loan facilities. As a robust check, we eliminate all financial industry with SIC All regression results (sign and significance of key variables) remain the same. 3.2 Strategic alliance activities 9

10 We employ Thomson Financial M&A/Joint Venture Database to trace sample firms alliance activities. This dataset covers world-wide alliances and joint ventures formed from 1987 onward. The coverage is not only restricted to public firms and is documented to be very accurate and comprehensive (Anand & Khanna 2000; Lindsey 2008). Two types of data are recorded in the alliance database. Deal specific information includes the announcement date of alliances and a broad description about alliance purpose (e.g., marketing agreement, joint R&D and licensing agreement). Participant specific information contains company name, SIC code, public status and location for each participating company, etc. For each borrowing firm in our sample, we identify its alliance activities based on the following criteria. (1) The alliance is announced in a 3-year window preceding borrower s loan facility origination. Since our borrower sample starts from 1991 to 2004 and for each of them we track back for three years, our alliance activities data span from 1988 to (2) Partnerships with both private and public firms are included as the background to construct alliance network and alliance activity measures. Foreign firms are excluded. (3) Both equity (e.g. joint venture) and non-equity alliances (e.g. buyer or supplier relationships) are included in our sample. After we track down the interfirm agreements entered by our sample firms 3 years preceding their loan originations, we construct separate measures to capture their alliance activities Measuring alliance activities We generate several measures based on the actual alliance agreements made by our sample firms. First, we define a strategic alliance (SA) dummy which takes the value one if a sample firm has ever announces at least one alliance agreement during the 3 years preceding the loan facility origination, and zero otherwise. Second, we count the number of alliance agreements made by our sample firms to create a variable as the number of SA to measure the frequency of firm alliance activities. We also use the squared term of the number of SA to capture the potential nonlinear effect. We take natural logarithm to both measures to normalize the distribution. It is 10

11 plausible that certain industry characteristics and firm characteristics may lead to a firm to choose more alliances. In addition to the raw measures of alliance frequency, we use residuals from regressing our raw measures on certain industry characteristics and firm characteristics as our alternative measures. With regard to industry characteristics, we argue that the market completion and technologic instability are two key driving forces for firms to enter into collaborative activities. We borrow the idea from organization theory (Dess & Beard 1984) to objectively measure the industry environment. In order to capture the market competition, we use the average market share change over a ten-year period as a proxy. Demmert and Klein (2003) showed that "substantial changes in market share indicate high levels of competition." As a consequence, it is more difficult for entrepreneurial companies to penetrate a highly competitive market. We obtain market share information from the 1990, 1993, 1997, 1999, 2001, and 2002 editions of Ward s Business Directory, which ranked firms by sales within a 4-digit SIC codes. We manually calculate the market share change and firms presented in certain industries over the period in order to finally calculate the average market share change. As to the measurement of technology instability, Sharfman and Dean (1991) measured technological instability as the average number of patents in an industry. Ensley (2003) argues this measurement cannot capture the unpredictable changes, and therefore extends this methodology for measuring technological instability by using the standard error of research and development intensity (Tosi et al. 1973; Snyder & Glueck 1982): RNDt b0 b1y t t,...(1) WhereY time, RND researchdevelopment, residual t The industry research and development (R&D) was regressed with year dummy for 1972 to 2002 for each 4-digit SIC code industry. We obtain the standard errors of the slope and then divide the standard errors by mean of industrial R&D. The use of standard errors as measures of instability is common to both the Dess and Beard (1984) and Sharfman and Dean (Sharfman & James W. Dean 1991) environmental measurement constructs. The standard errors capture the unpredictable change, and therefore, the higher the standard errors, the more difficult to predict the technological change. 11

12 It is natural to think that larger firms and older firms may have more alliance activities. Therefore, we use firm size proxied by log of assets in the regress. 1 Our alternative measures are residuals obtaining from regressing our raw measures on firm size, industry market completion and technological instability. Third, we consider the partner prominence in a particular alliance agreement. A sample firm allying with a prestigious firm may well enjoy the benefit of endorsement or certification effect (Carter et al. 1998; Stuart et al. 1999). We consider whether the borrower and its partner are S&P 500 firms with the intuition that if a borrower itself is a prominent firm, the endorsement effect from its prominent partner will be weaker. Consequently, we define a dummy called borrower prominence which takes the value one if the borrower itself has been ranked as S&P500 during 3 years preceding loan origination and zero otherwise. Similarly, we define partner prominence as a dummy which takes the value one if a borrowing firm has allied with an S&P 500 firm during 3 years preceding the loan origination Network measures Strategic alliances are a hybrid organizational form which Jensen and Meckling (1991) refer to as network organizations. Relative network position is of strategic significance and influential in determining the cumulative economic outcomes (Gulati et al. 2000). Some firms, for instance, occupy influential network locations and enjoy better-quality relationships. To capture firms relative network positions, we apply a network approach browed from graph theory. Though, methodology based on graph theory is well established in social network analysis, its application in finance literature is still occasional (Allen & Gale 2000; Robinson & Stuart 2000; Hochberg et al. 2007; Allen & Babus 2008). In this study, we define an alliance network as a set of inter-firm relationships including all types of alliance agreements. The network is not static since each year there 1 We also use log of age in the regression. Since firm age and assets are highly correlated. We obtain similar results. For the sake of brevity, we only report results based on firm assets. 12

13 are new alliance formation and old alliance dissolvement. Normally, the duration for alliance agreement lasts for 3 years. We therefore construct our network measures for each year by looking back over 3-year window. 2 The advantage of applying network measures lies in its ability to identify the position of a firm in the network by considering multiple relationships simultaneously in the entire network (Gulati 1995a). Moreover, an examination of alliance network requires measures of how well networked a firm is. Network literature uses centrality, which, by intuition, gives a sense of how central and how well-connected a focal actor is in his/her environment. It indicates the social power of an actor based on how well it "connects" the network. Similar to (Hochberg et al. 2007), we calculate the following three centrality measures which are widely used in the network analysis. 3 A. Degree The simplest and most straightforward way to measure centrality is by the degrees of the various actors in the network. The degree refers to the number of other actors to which an actor is adjacent. Formally, it is calculated as the number of direct links between a focal actor and other actors in the network. An actor is central if it has a high degree. In intuition, it means well-connected or in the thick of things (Scott, 2000). In our context, the higher degree a firm has, the more opportunities for resource exchange with others, and so the greater visibility or name recognition it could gain. Because degree centrality is a function of network size, which in our data set varies over time due to new formation of partnerships, we normalize the degree measure by dividing by the maximum possible degree in an n-actor network. B. Betweenness A network position can be also characterized as whether it has high or low betweenness. Betweenness captures the extent to which a particular actor lies between various other actors and makes connections within the network. An actor with high betweenness is assumed to play an important intermediary role and therefore to be very 2 As robustness checks, we also use 1-year base and 5-year base. The results are similar to 3-year window. 3 See, Wasseman and Faust (1994)) for a comprehensive introduction of social network analysis. 13

14 central to the network (Freeman, 1979). In our context, firms of high betweenness centralities can play the part of a broker or gatekeeper by bringing potential pairs of debtors and creditors. By having a potential for control over others, they will be able to exploit more information, negotiate better agreements, and in general be more powerful and successful than those that do not occupy such a bridging role. Formally, an actor i betweenness is calculated as the proportion of all paths linking the other two actors (e.g. j and k) that pass through actor i. Again, we normalize by dividing the maximum betweeness in an n-actor network. C. Eigenvector A good network position not only accounts for the number of connections but also considers the attributes and quality of linkages. To take relationship quality into account, we use Bonacich eigenvector measure. It is calculated as the sum of an actor s connections to other actors weighted by their respective centralities. According to this definition, an actor who is connected to central actor has its own centrality boosted, and this, in turn, boosts the centrality of the other actors to which it is connected (Bonacich, 1972, 1987). In our context, eigenvector gives a sense of the quality of the linkages. Higher eigenvector corresponds to connecting with centrally-networked firms. Given different network sizes, our eigenvector measurement is also normalized by the highest possible eigenvector in an n-actor network. 3.3 Control variables In addition to our main measures of sample firms alliance activities, we enter into two sets of control variables in the regression analysis to control for firm characteristics and loan features documented to be important determinants for loan price (Strahan 1999); Bradley and Roberts, 2004) A. Firm characteristics We control for firm leverage calculated as total debt divided by total assets. Firms with higher leverage are, all else equal, facing a greater likelihood of future 14

15 insolvency and are subject to more severer moral hazard problems (Strahan 1999). Tobin s Q is widely used to capture the market s perception of the current and potential profitability for a particular company, and hence is used a proxy for firm growth potential (Morck & Yeung 1991). Himmelberg and Morgan (1995) argue that tangible assets reduce firm opaqueness and thereby increase a firm s access to external capital. To account for asset opaqueness, we include asset tangibility, a ratio of tangible assets, property, plant & equipment net of depreciation plus inventories, to total assets. We also control firm profitability by using profit margin calculated as the ratio of earnings before interest, taxes, depreciation and amortization (EBITDA) to total sales. B. Loan characteristics Banks usually use non-pricing loan terms as complements in dealing with borrower risk. Strahan (1999) has documented that firms facing more restrictive nonprice terms pay higher interest rate. In our study, we look at the following loan terms: loan size, maturity, secured loans, syndicated loans, and borrowers public ratings. Loan size is measured as the natural log of loan amount, and maturity uses the natural log of loan maturity in months. Secured loans, syndicated loans, and borrower ratings are controlled using a series of dummy variables. Secured dummy takes value of one if a loan deal is secured by collaterals, zero otherwise. Similarly, syndication dummy is equal to one when the loan is syndicated. Unrated dummy indicates whether the borrower has a public rating. Clearly rated firms are likely face less severe asymmetric information problems than unrated firms due to the efforts of the ratings agencies. It is also important to consider whether the borrower has previous banking relations with the bank. Many empirical studies have documented borrowers benefits from previous lending relationship with the banks. For example, Petersen and Rajan (1994) and Berger and Udell (1995) find that the stronger the relationship, the greater credit availability and the lower the collateral requirements. We use bank relationship dummy to account for previous lending relationships that a borrower has borrowed from the same lead bank before. [See Appendix A for Variable Definitions] 15

16 3.4 Sample description and univariate tests Our sample includes all U.S. public non-financial firms that have obtained bank loans during the time span from 1991 to Table 1 reports summary statistics of firm characteristics and loan contract terms. Variable definitions are reported in Appendix A. [Insert Table 1 about here] We divide our sample into alliance and non-alliance firms. Table 2 shows the distribution of two groups of firms across years, and the univariate comparisons of loan spread between them. In general, the total borrowing firms increased dramatically from 1991 to The number of alliance firms maintains a rough range from 1/10 to 1/3 of the total firms of our borrower sample. The mean of loan spread is lower for alliance firms than non-alliance firms. The difference is highly significant in each year. It indicates a strong effect of alliance on bank loan price. We show in the regression section, after controlling for other determinants, the effect remains but with a lower level of significance. [Insert Table 2 about here] Our sample contains 8 major industries classified by one digit SIC code. Table 3 presents the distribution of alliance firms and non-alliance firms across industries. Manufacturing sector has the most number of alliance firms, while financial sector shows the smallest number. This is because we dropped banking firms during the sample selection procedure. The left financial firms are mostly consisted of real estates and leasing firms. Univariate analysis compares the mean bank loan spread between two groups. The difference is highly significant in each industry. As a robust check, we also eliminate all financial industry with SIC All univariate and multivariate regression remain robust. [Insert Table 3 about here] 16

17 4. Empirical Results In this section, we report our regression analysis results. Essentially, we have an unbalanced panel with multiple firms receiving bank loans at in different time points. Modern econometric theory points out that OLS and White standard errors will be biased when the residuals are not independent (Petersen 2009). Simply using fixed effect or random effect model produces unbiased standard errors but only when the firm effect is permanent. However, we are analyzing the dynamic effect of a firm s alliance activities, and a firm s network positions may change over the time. Consequently, unobserved firm effect may change over the time as well or even decay at some point of time. In dealing with the time-series dependence of residuals induced by the firm effect, the standard errors clustered by firm are unbiased whether the firm effect is permanent or temporary (Petersen 2009). Moreover, as suggested by Petersen (2009), adding year dummies can control for the economic-wide shocks and timely trends, thus takes care of the other dimension of residual dependence induced by time effect. Therefore, for all regression analysis reported in this section, we use standard errors clustered by firm and enter year dummies into our empirical models. In addition, we add indicator variables for industry at 2-digit SIC code level. 4.1 The effect of strategic alliance involvement on the cost of bank loan Table 4 presents regression analysis relating loan price to SA dummy capturing whether a sample firm has any alliance experience 3 year preceding a particular loan origination. For all model specifications, the dependent variable is loan price measured as drawn all-in-spread in logs (Strahan 1999). Our main explanatory variable is a dummy which equals one if the borrower has at least one alliance arrangement in the three years before loan facility starts, and zero otherwise. Column 1-3 report our findings using raw measure of alliance activities by sample firms. We find that the coefficient is significantly negative indicating that participating in 17

18 alliance activities does reduce cost of bank financing. It is less likely that our results are driven by reverse causality in the sense that obtaining cheap debt lead firms to enter into alliances. However, it is plausible that some unobservable factors that influence both the likelihood for firms to choose alliance agreements and obtaining lower cost of bank loans. In other words, SA dummy can be endogenous even we control for the firm effect by clustering standard errors by firm. It has been documented that alliances are largely driven by the industry dynamics in order for firms to gain better competitive positions (Williamson 1985; Kogut 1988) Burger, Hill and Kim, 1993). We thereby adopt instrument variable approach by using two industry dynamism measures, namely market competition and technological instability, which are detailed in the data section. They are highly correlated with the propensity of a firm to enter into alliance. In addition, we add firm size to reflect the idea that a large firm may have more alliances. Predicted likelihood of choosing alliance is obtained from the first stage logit regression, and is entered with all variables into the second stage regression as reported in column 4. However, using instrument variable approach generates consistent result. We still find that the predicted probability of having some alliance activities is negatively and significantly correlated with loan price. With regard to other control variables, we generally find consistent results with existing literature (Strahan 1999; Graham et. al. 2008). For example, firm leverage is positively associated with cost of bank loans because firms with higher leverage ratios generally have reduced debt capacity and higher likelihood of default. Consequently, firms with higher leverage pay more to obtain bank loans. Tobin s q, as a proxy for firm growth potential, is negatively affecting loan spread in the sense that having better growth opportunities may generate more future cash flow and therefore reduce the default risk. Yet, another argument, as pointed out by Strahan (1999), is that growth firms may have more uncertainty since their value depends more on profit growth than on current cash flow. However, following Graham et. al. (2008) s explanation, given that we control for other characteristics like tangibility of book assets, market-to-book may affect the loan spread negatively if market-to-book represents the additional value over book assets that debt holders can access in the event of default. Higher assets tangibility and higher profit performance tend to be associated with lower cost of bank debt. For the set 18

19 of control variables related to loan features, we find that loan size is negatively related to loan price, which is consistent with Strahan (1999). We do not enter firm size in the regression model to avoid multicollinearity problem because firm size is highly correlated with loan size. In addition, longer loan maturity reflects higher uncertain, hence leads to higher cost of debt. Indicators of higher uncertainties, secured dummy and unrated dummy, are positively correlated with loan price. From the point of view of risk sharing and second opinion effects (Lerner 1994), we find that syndicated loans generally have lower loan spread. Previous relationship with lead lender has a negative coefficient but not always significant across different model specifications. [Insert Table 4 about here] 4.2 The effects of frequency of alliance activities on the cost of bank loans In this section, we further explorer the relation between the frequency of a firm s alliance activity and its cost of loans. We directly count the number of alliance agreements for a particular firm 3 years preceding the loan origination in a given year, and take natural logarithm of the count as our measure of alliance frequency. In order to detect any non-linear effect, we also add squared term of alliance frequency. Table 5 reports our regression results relating the cost of bank loan for our sample firms to their frequency of alliance activities. In column 1, we only enter the number of alliance agreements as our main explanatory variable along with other control. However, we only find an insignificantly negative coefficient for the frequency of alliance activities. In column 2, we enter both the number of strategic alliances and its squared term. We document an inverse U shape between cost of bank loans and alliance frequency. For our sample firms that only occasionally collaborate with other firms through alliance agreement, the effect of alliance frequency is marginally positive (significant at 10%). In contrast, for firms actively involved in alliance activities, increasing the number of alliances reduce cost of bank debt significantly. We think our finds make sense to some extent. Those firms only infrequently participate in alliances are likely to be firms seeking for opportunistic behaviors, and hence are associated with higher uncertainty. 19

20 They do not get enough exposure so as to let banks obtain sufficient information to determine their credit worthiness. In contrast, for firms actively involved in interfirm alliances, they are likely to be those with higher reputation or popularity, and frequently participate in alliance activity can gain better visibility or name recognition. It is possible that more analysts will follow these firms and market will screen them more carefully. With reduced information asymmetry, they are more likely to achieve lower cost of external financing. Additionally, we want separate out the effect of more alliances induced by larger firm size and drastic industry dynamism. We take residuals from regressing the number of corporate alliances on firm size, industry level market share change and technological instability. Again, we enter both the first order effect and squared term of alliance frequency into the regression with other controls. From column 3, we find quite similar result that only firms actively involved in alliance activities enjoy the benefit of lower cost of capital. With regard to our other control variables, the results are consistent with previous findings and existing literature. [Insert Table 5 about here] 4.3 The prominence of alliance partner and cost of bank loan Some literature has documented that when a firm is judged by a third party, outsiders tend to examine with whom the firm is affiliated with, especially when the firm is associated with asymmetric information problem. For example, prestigious underwriters can be a certification to IPO firms (Carter & Manaster 1990; Carter et al. 1998; Higgins & Gulati 2003). Bio-tech firms associated with big pharmaceutical companies get higher valuation from investments made by venture capitalists or other investors because subsequent investors use that alliance as a signal of firm assets and quality (Stuart et al. 1999; Nicholson et al. 2005). Therefore, in this section, we attempt to explore this line of research by linking cost of bank loans to the prominence of alliance partners. 20

21 Table 6 reports regression analysis relating bank loan spread to the prominence of alliance partners. In order to disentangle the confounding effect of a borrower s own prominence, we add to dummy variables into the regression with other controls. One dummy captures the prominence of alliance partners, and the other dummy indicates the prominence of loan borrowers. We define that a borrower or a partner to be prominent if it is an S&P 500 firm. Column 1 shows that the status of both the partner and the borrower matters. Being an S&P 500 firm or having an S&P 500 partner do reduce cost of loans significantly. As a next step, we partition our sample according the prominence of borrowing firms. Column 2 reports regression results based on the subsample that the borrowing firms are not S&P 500 firms, and column 3 uses the rest of the sample where borrowers are S&P 500 firms. We find only when the borrowing firms are not S&P 500 firms, the prominence of partner firms has a significant and negative coefficient. This suggests that S&P 500 firms do not necessarily rely on alliances partner to certify their credit worthiness because they themselves are enough reputational capital at stake. However, small and less reputational firms do achieve lower cost of bank loan from allying with S&P 500 firms. According to other control variables, our findings are quite constituent and robust. [Insert Table 6 about Here] 4.4 Firm network position and cost of bank loans Social network analysis based on graph theory has been well established and applied in other business disciplines. However, its application in finance literature is still rare. Alliance networks-broadly understood as a collection of nodes and links between modes-can be a useful representation of product market relationships among firms. By modeling economic interactions, network analysis can better explain certain economic phenomena (Allen & Babus 2008). The networks of relationships in which firms are embedded can profoundly influence their conduct and performance. Moreover, positioning pattern of a firm in the network has a unique value (Hochberg et al. 2007) 21

22 and the potential to confer competitive advantage (Gulati, 2000). Consequently, an entity s network positive has important implication to its performance and other decisions. In this section, we intend to shed further light on the relation between a firm s alliance activities and its cost of debt from the social network perspective. Network positions are measured by centrality. Intuitively, centrality captures how well-connected a firm is in the surrounding environment. We borrow three centrality measures from social network analysis literature (Wasserman & Faust 1994), namely degree centrality, betweenness centrality, and eigenvector centrality. The method of network construction and calculation of centrality measures are detailed in the data section. Table 7 reports regression results relating a firm s cost of bank debt to its network measures. [Insert Table 7 about here] In column 1-3, we enter raw measures of degree, betweenness and eigenvector separately with other controls. As we discuss before, degree centrality measures the unique ties a firm has in the network. Betweeness captures the effect that a particular firm serves as the bridge of other network actors. Eigenvector gauge the centrality of a particular firm by taking into consideration the importance of other participants. We predict that a well-positioned firm will benefit from increased visibility and ties to other prominent partners. Moreover, centrally located firms will be observed repeatedly by other participants thus have less incentive to behave opportunistically (Larson 1992; Robinson & Stuart 2000). Hence, a firm with higher centrality measures is expected to have lower cost of external financing because of reduced information asymmetry problems and mitigated moral hazard problems. In line with our predication, we find all of our measures of network centrality are negatively and significantly correlated with loan spread, which support our conjecture that a centrally located firm in the network enjoys the benefit of lower cost of debt. Following the same logic we adopt in previous sections, we use alternative measures obtained from regressing our network measures on firm size and industry characteristics, and take the residuals to proxy the portion of centrality measures cannot 22

23 explained by firm size and industry characteristics. In column 4-6, we enter the residual measures of degree centrality, betweenness centrality and eigenvector centrality into the model specifications, respectively, along with other controls. We obtain qualitatively similar results except that the residual measure for eigenvector is negative but not significant. Recall that eigenvector centrality considers the importance of other network participants. We argue the lacking of significance of the residual measure is due to the prominence of the firm itself. We have documented that if a firm itself is a prominent firm, the importance of its partners is diminished. Hence, the result reported in this section is quite consistent. Again, for other control variable, we generally find consistent results. 5. Summary and Conclusion With increased competition in product market, the collaboration among nonfinancial firms has experienced sharply growth during the past a few decades. Strategic alliances not only allow participating firms to gain access to competitors resources and remain competitive, but also create a network to allow information diffuse rapidly. A well-connected firm in this network can ally with a prominent partner and increase its visibility, hence reduce its information asymmetry. Furthermore, it is less possible for such a firm to behave opportunistically since it is observed repeatedly by other participants. Corporate finance literature has increasingly recognized the importance of corporate alliances but is still lacking of evidence with regard the financial consequences of such interfirm collaboration activities. Aiming at providing more insights to this important phenomenon, this paper takes an initial step towards understanding how a firm s corporate alliance activities may affect its cost of bank loans. We find that firms actively involved in alliance activities experience low cost of bank debt, all else equal. Further, the networking effect is substantial for acquisition of bank loans. Loan spread is negatively associated with all three measures capturing the centrality of a particular firm in the network. We take necessary steps to address the endogeneity issues. We argue that our findings are unlikely to be driven by reverse 23

24 causality or simply because larger (older) firms and firms in industry possessing certain characteristics tend to have more alliances. In this study, we establish a robust link between product market relationship and firms cost of external financing. By providing new evidence on the financial consequence of corporate alliance activities, we shed light on this important economic phenomenon. 24